Audio Critic the 21 r schematic

Issue No. 21

Display until arrival of

Issue No. 22.

Retail price: U.S. $7.50, Can. $8.50

Unconventional deployment of conventional drivers,
with superior results. (See the loudspeaker reviews.)

In this issue:

We present the definitive article for audiophiles on
the subject of digital jitter, written by one of the
world's top experts in response to the appalling
misinformation spread by the high-end audio press.

Your Editor reviews an unusually interesting and
varied assortment of loudspeaker systems.

We take a first look at the Sony MiniDisc system.
David Rich dissects analog and digital electronics

by Harman Kardon, Krell, Parasound, Sentec, et al.
Plus many other test reports, all our

regular columns, letters to the Editor,
and CD reviews by David Ranada.

pdf 1

Contents

10

Clock Jitter, D/A Converters, and Sample-Rate Conversion

By Robert W. Adams, Analog Devices, Inc., Wilmington, MA

25

Loudspeaker Systems Using Forward-Firing Cones and
Domes: How Good Can They Get?

By Peter Aczel, Editor and Publisher

25 B&W Matrix 803 Series 2
26 NHT Model 3.3
28 Sequerra Model NFM-PRO
29 Velodyne DF-661 (quick preview)
30 Magneplanar MG-1.5/QR (dipole)
31 Bag End ELF Systems S10E-C and S18E-C (subwoofers)
33 MSB Acoustic Screens (acoustical accessory)

34

Analog Electronics: More Power Amplifiers,
Preamp/Control Units, and Mild Surprises

By David A. Rich, Ph.D., Contributing Technical Editor

34 Stereo Power Amplifier: Harman Kardon PA2400
36 Full-Function Preamplifier: Harman Kardon AP2500

38 Line-Level Remote Preamplifier: Krell KRC-2
41 Stereo Power Amplifier: Parasound HCA-2200II
44 Line-Level Preamplifier: Rotel RHA-10 (quick preview by the Editor)
44 Stereo Power Amplifier: Rotel RHB-10 (quick preview by the Editor)

45

Digital Electronics: More CD Players, D/A Processors,
Transports, and a First Look at the Sony MiniDisc System

By Peter Aczel, Editor and Publisher
& David A. Rich, Ph.D., Contributing Technical Editor

45 Outboard D/A Converter: EAD DSP-90000 Pro (Reviewed by Peter Aczel)
46 Compact Disc Player: Harman Kardon HD7725 (Reviewed by David Rich)
48 Outboard D/A Converter: Krell Studio (Reviewed by David Rich)
50 DAC/Line Amplifier: Monarchy Audio Model 33 (Reviewed by Peter Aczel)
53 CD/Videodisc Transport: Monarchy Audio DT-40A (Reviewed by Peter Aczel)
53 Outboard D/A Converter: Sentec DiAna (Reviewed by David Rich)
55 Compact Disc Player: Sony CDP-X707ES (Reviewed by Peter Aczel)
56 2nd-Generation MiniDisc Recorder: Sony MDS-501 (Reviewed by Peter Aczel)

57

High-Definition Thinking in Small, Furry Mammals

or The Weasel's Guide to Maximum Satisfaction

By Tom Nousaine

60

Hip Boots Wading through the Mire of Misinformation in the Audio Press

Six commentaries by the Editor

62

Recorded Music

A

Miscellany of CDs and Musical Videodiscs

By David Ranada (with a few additional capsule reviews by the Editor)

3 Box 978: Letters to the Editor

ISSUE NO. 21 • SPRING 1994 1

pdf 2

Issue No. 21 Spring 1994
Editor and Publisher Peter Aczel

Contributing Technical Editor David Rich
Contributing Editor at Large David Ranada
Technical Consultant Steven Norsworthy
Columnist Tom Nousaine
Cartoonist and Illustrator Tom Aczel
Business Manager Bodil Aczel

The Audio Critic® (ISSN 0146-4701) is published quarterly for $24

per year by Critic Publications, Inc., 1380 Masi Road, Quakertown, PA

18951-5221. Second-class postage paid at Quakertown, PA. Postmaster:
Send address changes to The Audio Critic, P.O. Box 978, Quakertown,
PA 18951-0978.

The Audio Critic is an advisory service and technical review for
consumers of sophisticated audio equipment. Any conclusion, rating,
recommendation, criticism, or caveat published by The Audio Critic
represents the personal findings and judgments of the Editor and the
Staff, based only on the equipment available to their scrutiny and on
their knowledge of the subject, and is therefore not offered to the reader
as an infallible truth nor as an irreversible opinion applying to all extant
and forthcoming samples of a particular product. Address all editorial
correspondence to The Editor, The Audio Critic, P.O. Box 978,
Quakertown, PA 18951-0978.

Contents of this issue copyright © 1994 by Critic Publications, Inc. All rights
reserved under international and Pan-American copyright conventions. Repro­duction in whole or in part is prohibited without the prior written permission of
the Publisher. Paraphrasing of product reviews for advertising or commercial
purposes is also prohibited without prior written permission. The Audio Critic
will use all available means to prevent or prosecute any such unauthorized use
of its material or its name.

Subscription Information and Rates

First of all, you don't absolutely need one of our printed subscrip­tion blanks. If you wish, simply write your name and address as legibly
as possible on any piece of paper. Preferably print or type. Enclose with
payment. That's all. Or, if you prefer, use VISA or MasterCard, either
by mail or by telephone.

Secondly, we have only two subscription rates. If you live in the
U.S., Canada, or Mexico, you pay $24 for four consecutive issues
(scheduled to be mailed at approximately quarterly intervals). If you
live in any other country, you pay $38 for a four-issue subscription by
airmail. All payments from abroad, including Canada, must be in U.S.
funds, collectable in the U.S. without a service charge.

You may start your subscription with any issue, although we feel
that new subscribers should have a few back issues to gain a better un-

derstanding of what The Audio Critic is all about. We still have Issues
No. 11, 13, 14, and 16 through 20 in stock. Issues earlier than No. 11
are now out of print, as are No. 12 and No. 15. Please specify which
issues you want (at $24 per four).

One more thing. We don't sell single issues by mail. You'll find

those at somewhat higher cost at selected newsdealers, bookstores, and
audio stores.

Address all subscriptions to The Audio Critic, P.O. Box 978,

Quakertown, PA 18951-0978. VISA/MasterCard: [phone].

From the Editor/Publisher:

Issue No. 20 was optimistically dated Late
Summer 1993 but was mailed in the second
week of the fall. Unforeseen delays resulted
in one omitted quarter; hence the more
realistic Spring 1994 dating of this issue.
The staff expansion I so fondly previewed
is still in the incipient stage. With all the
reviews in this issue, it was suggested to
me that I split it down the middle to make it
into two issues (maybe I should have), or
label it a double issue in fulfillment of two
quarters of a subscription (I would never
do that). Yes, there will be a Summer 1994
issue, possibly even sooner than you think.

2 THE AUDIO CRITIC

pdf 3

Box 978

Letters to the Editor

In the last issue your Ed. griped in this space about uninteresting letters seeking advice on purely

private purchasing plans (alliteration unintended). Now I want to gripe about reasonably interesting

but unpublishable letters that ramble on for seven or eight illegibly handwritten pages, propounding

the correspondent's opinions on eleven different audio subjects. What is the purpose of such a letter?

What am I supposed to do with it? Get a life, guys—or get a word processor. Letters printed here
may or may not be excerpted at the discretion of the Editor. Ellipsis (...) indicates omission. Address
all editorial correspondence to the Editor, The Audio Critic, P.O. Box 978, Quakertown, PA 18951.

The Audio Critic:

When making the suggestion that
Berlioz was worthy of joining "the 3 B's"
[Issue No. 17, p. 57], you likely never
knew that Liszt's disciple Peter Cornelius
coined the formula with Berlioz as the
third! Bülow signaled his revolt from
Wagner (who had earlier stolen his wife
Cosima, née Liszt) by purloining the
phrase to glorify Brahms.

Actually "the 3 B's" may be held
nonsense by Magyars, or other non­Teutons (such as I). So is your warning
that Berlioz's worst is probably worse
than Bach's, Beethoven's, or Brahms's
worst. Since Berlioz is the lone great
composer who was never granted full
"canonization" (which can account for
your electing him so shyly to "3 B" status,
and the rare event of Inbal's semicycle
recordings: he'll never be subjected to the
fashion), he still suffers from such pre­posterous slights, even from renowned
musicians (e.g., Celibidache and Nigel
Kennedy), which only an exceedingly bold
critic would make against the others.

Can you recall Tovey's remarks—of
which Haggin was so fond—that "neither
Shakespeare nor Schubert will ever be
understood by any critic or artist who re-

gards their weaknesses and inequalities
as proof that they are artists of less than
the highest rank," and that "the highest
qualities attained in important [my em­phasis] parts of a great work are as inde­structible by weaknesses elsewhere as if
the weaknesses were the accidents of
physical ruin." Tovey wrote that to exalt
Schubert against the shallow regard then

rampant but now all but extinct....

I don't overlook the fact that the mu­sic of Berlioz pleases you greatly. (A vis­itor to Wagner reported that he first said
that Romeo and Juliet's Love Scene, just
as wisely called the Adagio of Berlioz's
3rd Symphony, was "the most beautiful
music ever composed.") A superior critic
(and Nobel-winning novelist) of France,
Romain Rolland (also friend and corre­spondent of Freud!) enrolled Bach, Han­del, Mozart, Beethoven, Schubert, and
Wagner as the finest of all composers and
claimed that after them he knew no other
who was superior, even equal, to Hector
Berlioz. Berlioz seemed incapable of the
vulgarity (or kitsch or bathos) which
afflicts German composers, and Brahms
especially (ever seen Cary Grant conduct
the Academic Festival Overture in Peo-

ple Will Talk?), and so became approved

and emulated in all Western music...

...While comparing one and an­other's worst and best, what is gained
from tallying pages, measuring the stacks
of bad and good? Yet you felt compelled
to indulge this pointless fancy at Berlioz's

expense....

Sincerely,
Owen M. Feldman
Elkins Park, PA

Ha! Fooled you all—didn't I?—by
beginning this column with a music­oriented letter rather than audio talk. My
main reason for doing so was actually to
contradict my preamble above with, shall
we say, the exception that proves the
rule. This letter came written in a small,
compressed hand on what appears to be
a legal-size yellow pad, then very badly
Xeroxed and the copy sent instead of the
original. I chewed my way through 3¼
legal-size pages of smeared, streaked,
gray mess and decided to publish about a

fourth of it because I found it interesting

and entertaining. I'm not taking back a
single word of my preambulary com­ments, but I guess I am a sucker for
knowledgeable music talk. It's certainly a
nice change from tweako audio talk. I'll

ISSUE NO. 21 • SPRING 1994 3

pdf 4

even apologize for my churlishness anent
Berlioz's small lapses; on second thought
they're probably no worse than Bach's or
Beethoven's. Besides, a man who is even
slightly underwhelmed by Brahms can't
be all bad and should be humored.

-Ed.

The Audio Critic:

..."Accountability in audio journal­ism" and your no-nonsense, rational ap­proach to product reviews are a refresh­ing change and a source of continuing
entertainment for me. David Rich is a su­perb find.

Regarding the MTM [mid/tweet/mid]
driver geometry, you are correct in that I
was not the first to use it (Issue No. 20,
page 42). To my knowledge the earliest
commercially successful use was by Koss
in a small 2-way system with 4" mid/bass
drivers. The choice of this geometry by
Koss, however, appeared to be largely
cosmetic. Meridian also made such a sys­tem about the time my paper appeared
["A Geometric Approach to Eliminating
Lobing Error in Multiway Loudspeakers,"
74th Convention of the AES, New York,
8-12 October 1983, Preprint 2000], but
again no mention of its superior polar re­sponse was made by them.

With Linkwitz's 1976 paper the
problem of polar-axis frequency­dependent wander or lobing error became
widely appreciated. Linkwitz's solution
to the problem was to use inphase cross­over networks. I was the first to demon­strate in the open literature that the MTM
geometry automatically eliminates lobing
error and to show the relationship be­tween polar response and crossover order
for this geometry. I also designed several
commercially successful loudspeaker sys­tems and system kits using this geometry.

Two of these systems were featured in
Speaker Builder magazine. The "Auditor
Point Source Aria Five," a Focal/JML
product which sold exlusively in Europe,
won the best loudspeaker of the year
award for 1991 from Hifi Vidéo (Paris,
March 1991). The MTM geometry is
now widely used and several manufactur­ers have attributed the concept to me in
their promotional literature. I believe it is
for these reasons that the MTM geometry
has become associated with my name.

Yours truly,

Joseph D'Appolito, Ph.D.

Andover, MA

Thank you for the compliments. Isn't

it remarkable that technologists with the
highest credentials, such as you, always
like us and that the scattered little en­claves of hostility out there are invariably

peopled by the technically untutored?

As for the MTM geometry, I myself
was the grunt of a design team (Bruce
Zayde, now with Hewlett-Packard, was
the whiz) that developed such a speaker
in 1984-85. It was called the Fourier 44
(because of the two 4½" mid/bass driv­ers) and shown at the 1985 Summer CES.
A few studio types and broadcasters are
still using it as a small monitor. We
didn't attach the D'Appolito appellation
to it, but conceptually the crossover was
along the Linkwitz/D'Appolito guidelines.
The speaker is currently extinct.

Audio designers have been known to
claim credit for work done by others, but
you are the first in my experience to dis­claim, or heavily circumscribe, credit for
something the world has already fully
credited to you. That's what I call a class
act!

—Ed.

The Audio Critic:

Sorry, Charlie! This $24 is going to
Stereophilel I am sitting in the smallest
room of my house with your so-called
magazine in front of me. It will soon be

behind me. I'm sorry I ever wasted a cent
on your rag.

[—Unsigned]

The above anonymous and untrace-

able message was scribbled on a blank
copy of our pink form soliciting renewal
of an expired subscription and returned
to us in our business reply envelope at
our expense. I am publishing it as a clue
to the sociocultural/intellectual profile of
those who opt for Stereophile in prefer­ence to our publication. Such class! Such
wit! But such an awkward seat for letter
writing! (Needless to say, I washed my
hands after handling the form.)

—Ed.

The Audio Critic:

In Issue No. 20, Drew Daniels' letter
cited a transmission line's electrical
length as analogous to the phase shift in a
length of speaker cable. This is not true.
A speaker cable is in effect a lowpass
filter—your own curves in Issue No. 16
clearly show this. Depending on the LCR
of the cable, the driving impedance (am­plifier), and the load impedance (speak­er), the cutoff frequency could take place

within or, hopefully, above the audio
band.

The phase shift within the audio
band is determined by all of the above,
but generally will be capacitive (negative
degrees) at low frequencies, pass through
resistive (zero degrees), then become in-

ductive (positive degrees) at higher fre­quencies. In any case, even modest
lengths of any of the cables sold today
will exibit many degrees of phase shift at
the speaker terminals as a function of fre­quency. The only way to avoid power
loss, high-frequency rolloff, reduced
damping factor (drastic in some cases),
and phase shift (although I don't know
that this is important) is to use no speaker
cable. Your own suggestion to use mono
amps at the speakers' backs with short

jumpers is a most valid one.

Another subject: May I respectfully

decline to accept your request to contrib-

ute for the advancement of Bob Harley's

technical education? I find his "jittery"
stepping through technical issues very

amusing and entertaining. With more ed­ucation he could become dangerous—
another Martin Colloms!

Don't give in to those who would
like you to compromise with the witch
doctors. Someone has to bring all the
hype to the forefront and it appears you
are the only one.

Sincerely,
Jefferson P. Lamb

Incline Village, NV

It seems to me that Drew Daniels
and you are talking about two different
things. He talks about phase shift as re­lated to propagation speed in an untermi­nated wire, which is an abstraction; you
talk about a real-world hookup with an
amplifier output impedance plus wire
characteristics plus a complex load impe­dance presented by the speaker. Yes, of
course, with your givens your conclu­sions apply, but then the issue becomes
the sound of the amplifier/wire/speaker,
not just the sound of the wire due to its
length, which is the "moronic" subject
that incenses Drew.

Anent the SHEESH (Send Harley to

E.E. School in a Hurry) Fund, see the

"Hip Boots" column in this issue. Now,
what you don't seem to realize is that
Martin Colloms is actually two persons
existing in parallel universes. There is
the techie Martin Colloms, made of mat­ter. There is the tweako cultist Martin
Colloms, made of antimatter. The two

4 THE AUDIO CRITIC

pdf 5

cannot possibly get together because that
would cause a cataclysmic explosion an­nihilating Hi-Fi News & Record Review
and maybe even Stereophile. That's why

he is dangerous.

—Ed.

The Audio Critic:

"Dear" Peter,

Your continued hysterical, personal
attacks on me are entertaining as usual.
As is Ken Pohlmann's letter, which ex­plains his affinity for you: he too is a pet­ulant, name-calling infant. Why don't
you do this: survey the top LP and CD
masterers in the field—the folks who
have access to the master tapes and the
CD transfers. Ask them, as I have,
whether the CDs—especially those made
from an analogue tape—sound like the
original. And ask them whether they pre­fer the sound of a CD or a properly man­ufactured and played-back LP.

Ask Doug Sax, Bernie Grundman,
Greg Calbi, Bob Ludwig, Ted Jensen,
Stephen Marcussen, Steve Hoffman,
Howie Weinberg, Bill Inglot, etc. Ask
Grammy-award-winning engineer Roy
Halee (Paul Simon, etc.) about digital re­cording, or any of the dozens of veteran
recording engineers I've surveyed on the
subject. Some like digital, many don't.
I'd be happy to provide you with a list. [I
thought you just did.—Ed..]

Here's a good one: call veteran jazz
producer Michael Cuscuna of Mosaic
Records and ask him what happened
when, without alerting his customers, he
started releasing LPs generated from
"perfect" digitally remastered analog
source material. He'll tell you this: virtu­ally all of them heard the deleterious ef­fects of digitization and called to com-

plain.

Or better yet, reprint the following
so all five of your new readers [currently

a minimum of five new readers a day, Mi-

chael, but usually more—Ed.] can read

it:

Help Stop the Digital Epidemic!

It has become a mindlessly parrot­ed truism in the world of commer­cial audio that digital recording is
the state of the art and the wave of
the future. At the same time, there
isn't a single audiophile-oriented
equipment reviewer, record pro­ducer or music critic who finds the
treble range of current digital re­cordings musically natural and en­joyable. The present technology of
50,000 samples per second with

16-bit encoding/decoding is sim-

ISSUE NO.21 • SPRING 1994

ply inadequate and mustn't be al­lowed to become the world stan­dard. If you agree with us, start
writing letters to the record compa­nies and commercial magazines
before it's too late.

Let's see, who wrote that? Could it
be...SATAN? NO. It's from The Audio
Critic, Spring through Fall 1980, and

judging by the tone I'd say you wrote it,

you opportunistic slug.

Cheers,
Michael Fremer
Senior Editor

The Absolute Sound

Sea Cliff, NY

If listing Senator Robert Dole as one
of the anti-Clinton Republicans consti­tutes a hysterical, personal attack on him,
then listing you as one of the antidigital
audio journalist—which was all I did (Is-

sue No. 20, p. 13)—constitutes a hysteri-

cal, personal attack on you. It seems to

me, however, that once again (see Issue
No. 14, pp. 9 and 51) you are the pot

calling the kettle black. Isn't calling Ken

Pohlmann "a petulant, name-calling in-

fant" an act of name-calling—both hys-

terical and personal? Weren't your re-

marks about him in the Winter 1993 issue
of The Absolute Sound, threatening to
vomit on him at a CES dinner, infantile,
personal, and name-calling? Let me do
some fact-calling: Ken Pohlmann has a
graduate degree in E.E.; you do not. Ken
Pohlmann is a professor at a major uni­versity; you are not. Ken Pohlmann is the
author of a basic textbook on digital
audio; you are not. Ken Pohlmann is a

Vice President of the Audio Engineering

Society; you are not. Shall I go on? Shall
I insult you with a spelled-out conclu­sion? I think that will be unnecessary.

As for your list of names—a couple
of them well-known, others less so—you
say "some like digital, many don't," so
what's your point? Which of them don't?
I can trump each digitophobic name you
bring up with three world-class names to
the contrary; it's a silly game. The point
is not who likes digital; the point is the
inherent accuracy, or lack thereof, of the
linear PCM technology and of A/D and
D/A conversion. That's an important sub-

ject that needs to be addressed repeatedly

and is the main reason I am answering
your trivial crank letter at such length.

No one claims, I least of all, that
everything that has ever appeared on CD
sounds great. There are plenty of oppor­tunities to mess up between the first A/D

and last D/A stage in the recording and
playback chain; it used to happen often
but now it's much rarer. The basic ques­tion, and the only intelligent one, is this:
between a state-of-the-art DDD compact
disc and a state-of-the-art all-analog vi­nyl disc, each free from all the possible
technical goofs, which reproduces with
greater accuracy and fewer spuriae the
signal from the microphones? If anybody
thinks the answer is uncertain (as it is
not), let me add one other reasonable
constraint: 50 playbacks. Have I made
my point? If you eliminate the vinyl, the
picture changes; today's best analog and
digital master tapes can sound quite com­parable overall, but the signal-processing
possibilities are much more limited with
analog, and as a storage format digital is
considerably more stable.

That brings me to my 1980 com­ments on "the digital epidemic." There
were no CDs in those days; the only
readily available examples of the new
digital audio technology were vinyl LPs
cut from digital master tapes. They
sounded pretty bad. I mistakenly attribut­ed the bad sound to the digital process. I
was dead wrong, as later CD releases of
those same digital master tapes clearly
showed. It was the transfer to the LP me­dium that was bad because the engineers
who did the transfers initially refused to
deal with the differences in high-frequency
energy and dynamic range between ana­log and digital master tapes. I had no
idea how the digital tapes sounded in the
control room at the recording sessions. I

jumped to the wrong conclusion, and my

advisors at the time were no smarter.

Have you ever jumped to the wrong con­clusion, Michael?

As for "opportunistic slug"—boy,
that's a muddleheaded, inept insult. What
opportunities does a snail without a shell
exploit unfairly? What undeserved advan­tages did I gain by repudiating and cor­recting my early perceptions of digital
audio? You, sir, are a very lightweight

polemist.

—Ed.

* * *

Epilogue: At the Winter CES in Las

Vegas early in January, Michael Fremer

deliberately staged a loud, public,

fishwifely confrontation with me for the

benefit of visitors to the Velodyne exhibit.
It was uncalled-for and embarrassing; I
later had to clear the air with David
Hall, Velodyne's boss, who is a soft­spoken gentleman unaccustomed to the

5

pdf 6

streets-of-New-York type of vulgarity. I
tried to calm Michael by telling him that I
did not consider him to be an evil person
but just someone with absurd ideas, but
he kept loudly accusing me of "ad homi­nem attacks" (his editor, Harry Pearson,
also loves that phrase) and looking
around for group approval. I then made
the mistake of tossing off a small pedan­tic joke. I asked, "Who was the homo in

hominem?" Then I added, "You know,
hominem is the accusative of homo, re-

quired by the preposition ad." Michael
did not get the Latinist jest. His confused
reaction indicated that all he had heard
was the word homo, and I'm not even
sure he knew it means man. That will
teach me to make gratuitous scholarly
noises where the cultural tone is New
York candy store. Indeed, that will teach

me to have any kind of commerce with

the Fremeroid element in the audio world.

The Audio Critic:

It's been roughly two years or so

since my departure from "tweako/voodoo

salon" land, and I owe thanks to you and

your staff for the insight and knowledge
you share through The Audio Critic. It's
amazing what science and a little com­mon sense can do for the soul. As the
song states, "I was once blind but now I
see!" My only regret is that some of the
information is a little technical for those
"laymen" who are not a part of the engi­neering kingdom. I would be grateful if
you could dilute some of the techno-lingo
from time to time.

Throughout the years, I've noticed
that many audio magazines rarely even
mention the name McIntosh. The compa­ny has been around since 1949 and has a
solid reputation for reliability and quality,

just like Krell, Bryston, etc., yet the

tweaks hardly touch it. Why? The design
(external appearance) may be a bit out of

date for some, but the internal compo­nents hardly seem archaic. According to
the principles presented in your publica­tion, if the McIntosh amps operate within
their given parameters, then they should

sound no different than a Krell or Boul-

der. Thus, these amps should be highly
regarded and recommended, unless
there's something I've missed. How
about The Audio Critic, in a future issue,
taking the opportunity to test a McIntosh
amp. I would love to see how it compares
to some of the other big boys on the

block....

Please keep up the excellent work.

6

Your magazine lights the path for many

in a confused and disturbed audio world.

Sincerely,

Mark S. Williamson

Washington, D.C.

/ blush because your words of praise

make me appear to be something close to
a spiritual leader, a responsibility I re-

fuse to shoulder. (Lenny Bruce once said

that anyone who calls himself a religious
leader and owns more than one suit is a
hustler. I own two suits.)

McIntosh is indeed an interesting
company. You are quite right; they make,
and always made, beautifully engineered
equipment. During their long years under
the leadership of the late Gordon Gow,
they kept their dialogue with the high-end

audio press to a minimum, probably be-

cause they felt that with their thoroughly
established reputation and highly suppor­tive dealer network they had nothing to

gain from a good review but something to

lose as a result of an irresponsible tweako

hatchet job. Hey, they were probably

right.

About two years ago the picture
changed somewhat. The firm is now
owned by the Japanese; the new manage­ment is not from the high-end audio
world and is gung ho on marketing, PR,
the whole big-business canon. They have
a new car-audio line, among other
things. So far, from where I'm sitting, I
can discern no compromise whatsoever
with traditional Mcintosh engineering or

product integrity, and the party line is

that there will be none. Amen. They are
definitely cozier with the audio press,
however; you will undoubtedly see more
reviews, and I think that will include re­views in this publication. Based on what I

already know, I expect their power ampli-

fiers, especially, to do very well in engi-

neering shootouts with some of the sacred
cows of the High End.

—Ed.

The Audio Critic:

I just read my first issue of The Audio
Critic (No. 20). I can only say that I have

been looking for this type of coverage of

the audio industry for several years and
have found it only in this one publication.
I first became involved in audio as a teen­ager in 1966, working in what was then a
"high-end" audio store in San Francisco.
I have remained interested and involved
ever since.

I have been an avid reader of all the

popular audio publications over the years
except for Stereophile and The Absolute
Sound (by the time I discovered these last
two, they had become too involved in be­liefs in the superiority of older tube and
analog equipment for my taste). I have
had a subscription to Audio since the late
'60s. In recent years I have become high­ly disillusioned with the direction of most
publications and audio salons. So-called
"high-end" audio is becoming more and
more an exercise in frustration rather than
the source of pleasure it should be. We
are not told what sounds good but rather
what is wrong with the sound of just
about everything out there except maybe
one of those $150,000 all-out high-end
systems. There is far too much emphasis

on the cost of components and how cost
is related to "sound quality," even though
most high-enders will deny it.

At one time I was going to be an au-

dio engineer. I chose instead to become a

psychologist but have never left my
scientific orientation. As a result, I have

become what might be termed a psychol­ogy critic. I have remained true to only
empirically based studies of behavior and
have taken many courses in research de­sign and statistics. There are some inter­esting parallels between psychology and
high-end audio, the most obvious being
the lack of empirical support for the as­sertions that are so commonly made. Psy­chology always has been and continues to
be filled with interesting but often worth­less theories that become the basis for in­teresting but often worthless therapies.

As in high-end audio, the public can spend

hundreds of thousands of dollars on tech-

nologies (therapies) that are of dubious

value.

I had become so frustrated with the
audio scene over the last year or so that I
was hardly even reading anything any­more. In trying to purchase some audio
products during this same time, I was
convinced to buy some products by an
audio salesman, which turned out to be a
big mistake. Luckily the store that had

sold them to me was happy to give me
my money back when I returned them un-

happy. However, this was for accessories

like cables, not big-money items like
speakers, amplifiers, or preamps. I even

got into a big argument with a salesman

about the purchase of a Toslink cable for

copying CDs onto DAT. I was told that

several other products were far superior
and "what I really wanted was..." What I

really wanted was a Toslink cable! Dur-

THE AUDIO CRITIC

pdf 7

ing the argument, I heard things like

"...but according to Robert Harley...." I

am not an engineer but I was skeptical of
Mr. Harley's qualifications based on
many statements he had made in Stereo-
phile. The salesman exhibited his own ig­norance when it became clear that he did
not know what a gound loop was. I be­came so angry at the salesman's persis­tence and ignorance that I left the store
and returned later to buy my Toslink
cable from another salesman (this was the
only store in town that carried the cable).

Some of your readers have written
to suggest that you cease your "tweak
bashing" and "ignoramus hunting." I
have written to both The Absolute Sound
and Stereophile over the years to criticize
their views on various topics, but mostly
to point out their lack of understanding of
scientific methods (mostly Stereophile).
The Absolute Sound published two of my
letters but "disqualified" my point of
view in print by pointing out how inferior
my equipment was and claiming that I
would not be able to hear the things they
were talking about with such equipment.
Stereophile has never published or ac­knowledged any of my letters. Like The

Audio Critic, I have tried to write letters

to Stereophile that were "unanswerable,"
after my experience with The Absolute

Sound.

The Audio Critic appears to be the

only publication I am aware of that takes

a truly serious approach toward the eval-

uation of so-called high-end components.
Given the prices some of these compo­nents currently have, such a serious

approach is really needed. Just as impor­tantly, given the pervasive misconcep­tions about high-end audio components

that exist today, the "tweak bashing" and
"ignoramus hunting" in The Audio Critic
probably does not go far enough. I'm not
suggesting that more attacks are needed,
but that more clarification is needed and
somehow outside of this one publication.
Audio should not be the "sport" that An­thony Cordesman so often refers to it as
(sport in that context seems to denote the
constant trading of usually expensive
equipment in the never-ending quest for
perfection, and the fun of debating the
theoretical issues) but rather a means to
enjoy music outside the concert hall with
as much realism as possible at an afforda­ble price and without having to have a de­gree in electrical engineering. As an

aside, I can no longer wade through what

I call the "high-end babble" that so per-

ISSUE NO.21 • SPRING 1994

vades Mr. Cordesman's reviews in Audio.

Sincerely,
Chuck Butler
Kalamazoo, MI

Your comparison of psychology and
high-end audio is right on the money.
You haven't told us, however, how a lay­man in search of effective therapy can
distinguish, and navigate, between the so-

phisticated empiricists and doctrinaire

theoreticians in your profession—it's a
tough question, isn't it? In audio, the an­swer to the analogous question is implicit
in your complimentary remarks: every

man, woman, and child who owns more

than three CDs should have a subscrip­tion to The Audio Critic, right?

—Ed..

The Audio Critic:

Tom Nousaine, for whatever reason,
seems incapable of accurate reportage
about me or my views. Having given to

the readers of The Audio Critic a distort­ed picture of our not unpleasant encoun­ter at the Stereophile High-End Show last
year, he then compounds his errors by
perverting some perfectly clear state­ments from Professors Greiner and Lip­shitz on the audibility of absolute polari­ty. I beg to set the record straight on these
latter facts.

"[Lipshitz's] results were significant
only when trials using test tones were in­cluded in the analysis," Nousaine says.
On the contrary! Listen up:

[Here follows a series of referenced
quotations from the writings of Lipshitz,
Greiner, Richard Heyser, et al., with
pithy anti-Nousaine comments by the let­ter writer. The trouble is that the quota­tions are heavy-handedly selected, ex­cerpted, and edited with massive ellipses,
omitting the qualifying words, phrases
and sentences, falsifying the context as
well as the chronology, and not contra­dicting Nousaine's highly specific state­ments at all. It would take an additional
page, or more, to include this obviously
manipulative "documentation" here, and
I refuse to do so.—Ed..]

...Such results [as quoted above]
are difficult for those in the Nousaine
camp to swallow, for commonly they
judge others by what their own ears can
hear, or cannot. Always a mistake. Yet
one may still enjoy beholding their ver­bal contortions as they chew the truth
about polarity served up by every legiti­mate researcher on record. Dead to rights,

we have them here: in denial, every one,
about a fabulous free fix. Will Tom throw
in the towel at last? How about all the au­dio critics he has helped lead astray? Stay
tuned! Find out whether "the muffling
distortion" (my term) will later be pro­claimed over this same station.

Finally, regarding reader Donald
Scott's recent letter dissing the under­signed's "cranky" advocacy of polarity as
"the cow chip effect": the errors in his
experiment to disprove polarity are so
rife and obvious, I must invite the poor
soul to write for a free copy of my ex­planatory book The Wood Effect, and
that's no bull.

Clark Johnsen
The Listening Studio
Boston, MA

/ have made an exception here to my

"no further soapbox opportunities for

Clark Johnsen" policy as stated in Issue

No. 18 because your letter—in which
you're up to your usual tricks of creative
documentation—provides me with an op-

portune lead-in to unedited quotations

from Lipshitz and Greiner for the benefit

of our readers, showing them what these

researchers really meant.

The following is an uninterrupted
and unedited quotation from S. P. Lipshitz,
M. Pocock, and J. Vanderkooy, "On the
Audibility of Midrange Phase Distortion
in Audio Systems," J. Audio Eng. Soc. 30
(September 1982): 580-95 (your own fa­vorite reference).

...On normal musical material
heard via loudspeakers in an aver­age listening room, we have not
thus far detected the effect of mid­range phase distortions of up to
two cascaded all-pass networks of
Q 2 2. We do not have evidence
to conclusively demonstrate wheth­er phase distortions of this amount
can be heard in normal reverberant
loudspeaker listening to normal
musical or acoustic transients
(which are largely oscillatory in
nature), but it is clear that the ef­fect, if audible, is extremely subtle.

More work needs to be done in
this area, so that transducer design­ers can make an intelligent deci­sion on the significance (not the
existence) of phase effects.

Finally, we wish to caution
most strongly against quoting our
results out of context. All the ef­fects described can reasonably be
classified as subtle. We are not, in
our present state of knowledge, ad­vocating that phase linear trans­ducers are a requirement for high­quality sound reproduction. More

7

pdf 8

research is necessary. We do not
wish the research outlined above
to suffer the same misunderstand­ings, distortions, and misapplica­tions as have occurred in recent
years with transient intermodula­tion distortion. We feel that listen­ing rooms will become more
anechoic as more sophisticated re­production systems become avail­able. Thus the increased audibility
of the phase effects which we have
found with headphones may in the
future apply also to loudspeaker
listening.

So much for what Lipshitz, unfiltered

through Johnsen, wants us to understand.
As for Greiner, here is an uninterrupted
and unedited quotation from R. A. Grein­er and D. E. Melton, "A Quest for the Au­dibility of Polarity," Audio 77 (December

1993): 40-47.

While polarity inversion is not
easily heard with normal, complex
musical program material, as our
large-scale listening tests showed,
it is audible in many select and
simplified musical settings. Thus,
it would seem sensible to keep
track of polarity and to play the
signal back with the correct polari­ty to insure the most accurate pos­sible reproduction of the original
acoustic waveform.

Authors' Addendum: The work
presented here was done in 1991.
(It is now September 1993.) Since
then, there has been some, but not
much, progress made in establish-

ing polarity standards in the re­cording industry. This work is con­tinuing at the present time. There
has been some discussion in hi-fi
publications and much anecdotal
reporting, in various publications,
on the audibility of acoustical po­larity inversion. There has been
nothing noteworthy in the profes­sional literature, however, that cla­rifies the issue or "proves" that au­dibility of polarity inversion is a
major factor in listening enjoy­ment. While it is not clear why this
is the case, several factors might
be: The difficulty of doing the ex­periments in a controlled way, as
evidenced by this work; the fact
that the effect of polarity inversion
is small in most program material,
or the fact that the effect seems to
be small compared to the many
other variables in the recording/
reproduction processes (micro­phone use, room acoustics, elec­tronic processing, and the like).
Nevertheless, it seems reasonable
that at some point another step to­ward achieving greater audio
fidelity will be maintaining polari­ty of the signals throughout the
record/reproduction chain.

To sum up, what Lipshitz and Grein-

8

er are really telling us is: yes, it's better
to pay attention to polarity than to ignore
it, but no, it isn't a big deal from a listen­ing point of view. In other words, Clark,

the horse you are trying to ride to the
higher reaches of the audio world, while
a real horse and not a donkey, is a rather
slow and somewhat lame mount, unlikely
to get you there. Why don't you trade it in ?

Lastly, the perpetrator of the phony

"triple-blind" listening test (see Issue
No. 17, pp. 44 and 47) is in no position to
reprimand Donald Scott or anyone else
about experimental errors. Get your own
act together, Clark, and until you do,

please don't write us again. (Maybe you

should move to Warsaw, where you could
really experience Absolute Polarity.)

—Ed.

The Audio Critic:

.. .I have been pleased by the profes­sionalism of your reviews, and I am re­newing [my subscription] primarily be­cause I am interested in seeing how your
publication evolves with the contribu­tions of your new editors. I have more
than enough things to read in my profes­sional life, and I am less interested in rig­id publication schedules and first-class
mailings than I am in high-quality, ana­lytic criticism. If you were to conduct a
reader survey, I would give the follow­ing answers:

I became aware of your publication

from an ad in Audio. I subscribed because

I had received a gift subscription to Ste-

reophile and I felt that I needed an anti-

dote to the logorrhea and what you refer
to as "tweakism." That subscription has
now lapsed, and I am continuing with
you, partly to understand better which
parts, circuits, and features are essential
to quality construction and which are not.
Most of my equiptment was purchased
between '77 and '82 and is still perform­ing reliably. I am not "looking for some­thing to buy."

I have followed with some interest
the pleas for "tolerance of opposing
views" and finding "a halfway ground" of
agreement. I am a Radiation Oncologist. I
would not recommend any course of ther­apy without first evaluating prospective
double-blind studies. This is how I gener­ate my professional opinions, and I will
not give any weight to the opinions of
those who do not go through a similar
process. Subjectivism in medicine can be
deadly. It has no place in any of the phys­ical sciences. Some understand this; some

do not. I do not tolerate unprofessional
opinions in my field and would never ask

you to do so in yours.

Vincent Capostagno, M.D.
Merced, CA

Readers of The Audio Critic are

rather sharply divided into the categories
of those who want to learn something
(such as you) and those who want to buy
something. We try to satisfy both mind­sets; what we refuse to do (although
there is a demand for it) is to rattle off a
long laundry list of buy-this, don't-buy­that recommendations without document­ing the reasons.

I would like to expand on your

"some understand this, some do not" ob­servation regarding scientific objectivity.
I have come to the tentative conclusion
that some have a natural gift for unde­standing the basic essence of the scien­tific method, requiring perhaps only a
good junior-high-school course in Gener­al Science to assimilate the idea, and that
others are color-blind or tone-deaf, so to
speak, to the scientific mode of thinking,
although otherwise intelligent and re-

peatedly exposed to the best scientific

influences. That would explain the im­mense stubbomess and impenetrability of
some far-from-stupid members of the
audio community on the subject of "I can
hear it" vs. "I can prove I can hear it,"
when the distinction is obvious to enlight­ened twelve-year-olds.

—Ed.

* * *

/ also want to respond here collec-

tively to a whole bunch of correspon-

dence—long, well-written, laser-printed

letters as well as illegibly scribbled,

barely coherent ones, both long and
short—in which the common theme is
audible differences between various piec­es of equipment and the common failure
is a disregard for the need to eliminate
observer bias and the placebo effect.

There is no point in publishing any of

these letters; we would just be going
around in circles, repeating over and
over again that without blind listening
tests at precisely matched levels all such
discussions are meaningless. How many
times do we have to reiterate that self­evident precept before it sinks in, without
any of these "yes, but" arguments? Or is

there anyone out there who actually be­lieves that unmatched levels and peeking

at the nameplates will get you closer to

the truth? That I can't deal with. •

THE AUDIO CRITIC

pdf 9

ISSUE NO. 21 • SPRING 1994 9

In Your Ear

pdf 10

A Moderately Technical Tutorial for the Serious Audiophile

Clock Jitter,

D/A Converters, and

Sample-Rate Conversion

By Robert W. Adams

Analog Devices, Inc., Wilmington, MA

Forget everything you have read in the "alternative" audio journals
on the subject of digital jitter and start from scratch here with the
correct scientific foundation.

Foreword by the Tech. Ed.:

"The Jitter Game"

I have used the same title for this
foreword to an important article on
jitter as Stereophile's Robert Harley

uses for his articles on jitter, but with
a different meaning. Harley is play­ing a game of pretend engineering
when he attempts to analyze the jitter
of a CD player and correlate the re­sultant measurement to the sound
quality he perceives. Why does Har­ley spend so much time on jitter? Be­cause he thinks that it strongly corre­lates with the sound quality of the
equipment. Since open-loop (i.e.,
nonblind) listening tests are subject
to externally originating listener
bias, it is easy to see how he can de­lude himself to arrive at such a con­clusion. Stereophile is unfortunately
quite influential, and jitter has thus

become in the early '90s what TIM
(transient intermodulation distortion)
was in the late '70s and early '80s.

But Harley has a huge problem
because clock jitter cannot be mea­sured directly at the output of a black
box. The effects of jitter can be as­sessed indirectly from black-box
measurements, but in a correctly de­signed CD playback system these
effects are commingled with, and

usually swamped by, noise and dis­tortion products. Indeed, in exotic de­signs, the loony-tune analog stages
are so riddled with noise and distor-

tion that even large amounts of jitter
would have little effect on the mea­surements. To overcome this prob-

lem, Harley plays his little game. He
takes off the cover, gets inside the
unit, and attempts to measure the jit­ter on the internal clock line. Now,
two problems exist when he does
that: (I) since jitter is an internal pa­rameter, its effect on the external

performance of the system is depen-

dent on other aspects of the system's
design, so it is not possible to com-

pare the measured results directly

between two models under test; and
(2) measuring clock jitter is a non­trivial task, subject to many errors
even when conducted by one skilled
in the art.

Note that Harley could continue
to play his game and make other
measurements while he has the cover
open, such as I/V settling time, pow­er-supply rejection ratio, the amount

of closed-loop feedback, power­supply output impedance, etc. All

these parameters could affect the

sound quality, and under Harley's

rationale—that being unable to ob-

serve the effect of such parameters at

the output of the system does not

mean they do not affect the sound
quality—one could ask why he
doesn't make these additional mea­surements. My contention is that
Harley would indeed make these
measurements, and then delude him-

self into thinking they were remark­ably revealing of sound quality, if
some manufacturer delivered to him
a test system for a given parameter
and showed him step by step how to
use it.

Jitter and its effect on the perfor­mance an electrical system is a
difficult subject, truly understood by
only a few experts involved in the de­sign of systems sensitive to jitter. As
a result, much misinformation on jit-

ter has been circulated in the press,

originating from manufacturers' press

releases reproduced without any
competent review. In an attempt to
clear the air on the subject, we com­missioned an article by a genuine ex-

pert in the field of digital audio, Rob-

ert W. Adams, of Analog Devices,
Inc. This article is based on a paper
Adams presented at the 95th Conven­tion of the Audio Engineering Society
in New York last October. (The pre-

print number was 3712.) Bob Adams

is perhaps the youngest Fellow of the
AES (the highest honor awarded in
the field of audio engineering), and
his many pioneering achievements in
digital audio at AD, and before that
at dbx, are too numerous to be sum­marized here. His investigations in
the field of jitter reduction have re­sulted in a new method to attenuate

jitter, a practical asynchronous sam­ple-rate converter chip, which is ex­plained in his article. Before this

10 THE AUDIO CRITIC

pdf 11

chip, asynchronous sample-rate con-

verters could be had only at very

high prices and in many cases did
not perform very well. Since the new

Analog Devices ASRC chip is all-

digital, it offers the potential for the
easier and cheaper implementation
of a jitter attenuator than a multiple

phase-locked-loop S/PDIF decoder.

As I said, this is not an easy sub­ject, and the article below is not sim­ple. The problem again is that we

are trying to explain how an internal
system parameter affects the total
system performance. If Harley had
not made jitter his hobbyhorse, we
might not have found it necessary to

run such a complex article. But given
the current trendiness of jitter in au­dio journalism, I think it is important
that the serious audiophile try to go
through the article in order to separ­ate the facts from the fictions the
high-end charlatans are trying to le­gitimize. If nothing else, this article
will acquaint you with the complex

interrelationships involved in the dig­ital design process. Note that while
Bob Adams has simplified as much as
was possible without leaving out the
essentials, anybody who attempts to
measure the performance of an
S/PDIF decoder, let alone design
one, had better have a much more
complete knowledge of the subject.

Indeed, it is not possible to read the
professional literature in this field
without a strong background in sig-

nal processing, modulation theory,
and random process.

Unfortunately, it is not clear

whether Robert Harley understood

Adams's AES paper from which the
present article is derived. In the Jan-

uary 1994 issue of Stereophile he
wrote that "the paper stated that a
converter's jitter sensitivity is a func­tion of the clock frequency and over­sampling rate. This conclusion con-

firms the validity of our technique of

expressing clock jitter as a propor­tion of the clock frequency." Read

Adams's simplified but thorough ex­planation below and judge for your­self whether that's what he is really

saying.

This article does not discuss the
S/PDIF encoding and decoding pro­cess itself; that discussion is planned

for a future issue.

—David Rich

ISSUE NO. 21 • SPRING 1994

0 Introduction

Although clock jitter has re­ceived a great deal of recent attention
in the popular audio press, its effect
on signal fidelity is poorly under­stood by most journalists, and many
inaccurate statements have appeared
in print. The purpose of this article is
to introduce the fundamentals of
clock jitter and to demonstrate how
it actually affects final signal quality
for various types of D/A converters.

We also will cover an exciting
new development in sample-rate con­version and show how it will influence
the next generation of digital audio
equipment.

It has become popular practice to
measure clock jitter in commercial

outboard D/A converters using an
FM demodulator attached to the

clock pin. The output of the demodu-

lator is fed to a spectrum analyzer,

so discrete components present in the

jitter waveform may be analyzed.

Unfortunately, the amount of degra-

dation a particular jitter spectrum will

cause in the output signal depends on
the type of D/A converter used. To
interpret the results of such a mea-

surement, one has to take into ac-

count at least the following signif­icant variables:

(a) Converter type—conventional

resistive ladder, sigma-delta, or
MASH.

(b) Clock frequency applied to

the converter.

(c) Output filter type—switched-

capacitor, active RC, or a combina­tion.

(d) Any digital dividers between

the measured clock pin and the inter­nal D/A clock rate.

(e) Interpolation ratio.
(f) The frequency and amplitude

of the input signal.

(g) Jitter introduced internally to
the D/A converter (not measurable
except by its effect on the signal it­self).

These variables have more than
a minor effect on the jitter sensitivity.
With the worst combination, phase

jitter may have to be lower than 20

ps rms to obtain signals of 16-bit
quality, as opposed to more than 1

ns for the best case. Clearly, the rela­tionship between clock jitter and the
analog output is complex enough that
one should understand the fundamen-

tals before making any judgments
based on jitter about the quality of a
particular piece of equipment.

Another complicating factor will
soon be introduced commercially: a
new chip from Analog Devices
called an "asynchronous sample-rate
converter," rapidly making its way
into outboard D/A processors. This
chip acts as a universal digital buffer
between an input at one sample rate
and an output at any other sample
rate. As a byproduct of the algorithm
employed in the chip, jitter on either
the input or output sample clocks is
largely eliminated. While most engi­neers understand how a conventional
analog PLL may be used to remove
clock jitter, it is not obvious how an
all-digital sample-rate converter can
accomplish the same task. Later in
this article we will discuss how use
of this chip affects jitter in D/A con­verters.

1 Review of Clock Jitter

Clock jitter may be defined as

the time displacement of a clock sig-

nal relative to an ideal clock signal

with no jitter. Note that all the infor­mation about jitter is contained in the
edges of the clock signal, and it is
common to specify jitter in the time
domain as either the p-p or rms devi­ation of any edge from its ideal posi­tion over many thousands of clock
cycles. Most digital systems will
change state only on one edge of the
clock signal (the rising or falling
edge), in which case the jitter is mea­sured on the clock edge to which the
system responds.

In practical systems, the com-

mon types of clock jitter are:

(a) Random variations in the ar­rival of clock edges relative to their
ideal positions. For advanced read­ers, this type of jitter is referred to as
white phase jitter, as it may be pro­duced by feeding a random-noise
(i.e., white-noise) signal into a phase­controlled oscillator.

(b) Random variations of the
width of a clock pulse. This type of

jitter differs from (a) in that each

edge is referenced to the previous
edge rather than to a hypothetical
ideal clock signal. Again, for E.E.
types, this jitter is referred to as
white FM jitter, as it can be produced
by feeding a white-noise signal into a

11

pdf 12

frequency-controlled clock generator.

(c) Correlated variations in the
clock edge events relative to an ideal
clock. By correlated we mean that
the instantaneous time displacement
measured on each clock edge is not
an independent event but is in some
way related to previous clock edge

times. For the technical reader: this
causes a jitter "spectrum" which is
nonwhite and may have spectral
peaks at particular frequencies. We
will refer to this type of jitter as cor­related jitter. If the variation in clock
frequency is "slow" compared with
audio frequencies, we will call this
low-frequency correlated jitter; if
these variations are fast compared
with the audio spectrum, we will call

this high-frequency correlated jitter.

2 The Pitfalls of Time-Domain
Measurements

It is common to estimate clock

jitter by using an oscilloscope with a

very accurate time base. This prac­tice is dangerous, as the results ob­tained depend on the type of jitter as
well as on the measurement tech­nique. It also is often the case that the
oscilloscope used will have more jit­ter in its time base than is present in
the clock itself. Advanced instru­ments are available to make accurate
measurements of jitter but are not
used enough.

Figure 1 shows one measurement
technique where an oscilloscope is set
to trigger on a clock edge and the time
base is set so that only the next edge
is visible on the scope. The variations
in the arrival time of the later edge

can be used as a measure of p-p jitter.
More sophisticated oscilloscopes can
plot a histogram of zero-crossings,
allowing a more accurate estimate of
the rms jitter without resorting to

"eyeball" measurements.

Since we are triggering on one

edge and measuring the arrival time
of the next, we are assuming that the
first edge (the one we are triggering
on) is in its "ideal" time position.
This technique is fine if we are
measuring white phase jitter as
defined above, where the errors in the
clock edge positions do not accumu-

late over time relative to an ideal

clock signal. But suppose that the fre­quency of the clock signal is slowly

wandering by a small amount (low-

12

frequency correlated jitter). This

slow wandering of the frequency
causes large displacements of the
clock edges relative to a stable clock,

but the edge-to-edge measurement
technique will not reveal this effect,
as each measurement is made in ref­erence to the last clock edge only.

Another common technique is to
trigger an oscilloscope on a clock
edge and, by using the delayed trig­ger feature, examine the edges that
occur at some later time (for exam­ple, 10 clock edges later). See Figure

2. If we assume that, again, we have
a clock signal with slowly varying
frequency, we can see that this mea­surement technique will start to re­veal this low-frequency jitter compo­nent as long as the trigger delay is
long enough for the frequency of the

clock signal to have changed substan­tially between the moment when the

trigger event occurred and when the

delayed edge is examined some time

later. But one danger of this tech-

nique is that it is possible the jitter
frequency is correlated in such a way
that at particular trigger-delay values
the delayed edge of the clock has re­turned to its correct position. This
technique therefore has periodic oc­currences of "blind spots" relative to
the modulating frequency of the
clock generator and is not to be trust­ed if the clock signal contains highly
correlated jitter components.

The predominant type of noise
mechanism present may be estimated
by examining a succession of de­layed edges and observing how the

jitter behaves as a function of delay

time. White FM jitter as defined
above will display a square-root rela­tionship between delay time and ob­served edge jitter, as each clock peri­od is an independent jitter event, and
therefore many such events add in
rms fashion. Jitter which contains
low-frequency modulation of its fre­quency will show a linear relationship
between delay time and observed jit­ter. White phase jitter shows no in­crease in observed jitter with trigger
delay time, as the errors do not accu­mulate over time. Correlated jitter
shows a more complicated relation­ship between edge-to-delayed-edge
delay times and observed jitter.

The discussion above indicates
that time-domain jitter measurements

are dangerous, although useful infor-

mation may still be obtained if one is
careful. It is preferable nonetheless to
use a high-quality FM or PM detec­tor in conjunction with a spectrum
analyzer, provided one knows how to
interpret the results [Robbins 1982].

3 Sources of Jitter in Practical

Clock Circuits

Consider the clock circuit of Fig­ure 3, which is a typical RC oscilla­tor, such as might be found in the
voltage-controlled oscillator used in
a PLL. The VCO works in the fol­lowing fashion. Assume at the outset
that the capacitor is charged to V

ref

Low and the logic has turned on the

switch which connects the current
source Iup to the capacitor. The volt-

age on the capacitor now rises with a
slope proportional to Iup. When the
voltage on the capacitor reaches V

ref

Hi, the upper comparator changes
state and the current I

down

is now
connected to the capacitor, causing
the capacitor to begin charging down
at a rate proportional to the current.
When the voltage reaches V

ref

Low,
the lower comparator changes state,
and the whole operation repeats it­self, forming an oscillator.

From the previous explanation,
we can see that the current I deter­mines the frequency of oscillation of
the oscillator.

There are at least three possible
sources of jitter in this circuit. Here

we are analyzing only the one

caused by thermal noise. Practical ef­fects such as correlated frequency

components on the power supply or
those picked up because of magnetic
coupling will of course be added.

The first source of error is noise

in the current that charges and dis­charges the timing capacitor C1.
Since this current directly controls
the frequency of the oscillator, noise
on this current source translates di­rectly into white FM clock jitter.

The second source of error is

thermal noise at each comparator in­put, which causes the comparator to
switch at the wrong time. Since each
pulse is referenced to the end of the
last pulse, any variation in clock ar­rival time will be "remembered" by
all subsequent pulses, and therefore
this mechanism must again produce
white FM jitter. This can be verified

THE AUDIO CRITIC

pdf 13

Jitter Sources:

1) Current source noise (makes FM jitter).

2) Comparator noise (makes FM jitter).

3) Buffer intercept noise (makes phase jitter).

Output

Figure 3: Jitter sources in a typical RC oscillator.

Vref HI

Vref LOW

Figure 2: Edge-to-delayed-edge jitter measurement.

Scope Display

Delayed sweep

p-p jitter

Scope triggers hers.

Scope Display

Jitter

Scope triggers here.

Ideal non-jittered
output

Jittered Output

Figure 1: Edge-to-edge jitter measurement.

ISSUE NO. 21 • SPRING 1994 13

pdf 14

by imagining that the comparator
noise is dc (equivalent to an offset).

It is easily seen that this causes a

shift in frequency.

The third source of error results
from thermal noise on logic gates
that are fed with finite rise-time clock
signals. Such noise on the inputs of
these gates is translated into jitter
from the delta-t to delta-v conversion
that occurs due to the finite rise time
of the input signal. Since this mecha­nism has no "memory," it results in
white phase jitter.

There are several other types of
oscillators that offer improved jitter
performance. Crystal-controlled os­cillators are best. Voltage-controlled

crystal oscillators are available, and
these are sometimes used to recover
the clock from the incoming serial
data stream (from the output of a CD
player, for example). While they
have low jitter, they suffer from a
limited frequency-adjustment range
(about 0.1% maximum). Varactor­tuned LC oscillators are better than
the RC oscillator described earlier,
but not nearly as good as a crystal os­cillator.

4 Sensitivity of D/A Converters
to Clock Jitter

The effect of clock jitter on vari-

ous types of converters is complex

and depends on many factors. Con­verter topologies may vary in their

sensitivity to jitter by several orders
of magnitude, depending on the na­ture of the jitter.

For the purposes of analyzing jit­ter sensitivity, D/A converter fall into
three classes.

(a) Conventional, resistive lad­der converters with or without inter­polation filters.

(b) Sigma-delta converters with
continuous-time output filters.

(c) Sigma-delta converters with
switched-capacitor output filters.

4(a) Conventional, resistive ladder

D/A converters:

The effect of jitter on the output
of a D/A converter can be analyzed
by subtracting the output of a D/A
converter that uses a jittered clock
from the output a theoretically per­fect converter that uses a nonjittered
clock, and then looking at this differ­ence in the time domain. Figure 4

14

shows this analysis technique for the
case of a conventional D/A converter
with two different input frequencies
and no interpolation filter.

Figure 4a shows a 1 kHz sine
wave sampled at 50 kHz. The differ­ence between jittered and nonjittered
D/A outputs is seen to be a series of
narrow spikes whose width is propor­tional to the instantaneous difference
between the arrival time of the ideal
clock and that of the actual jittered

clock, and whose height is propor­tional to the change in signal ampli­tude from the previous to the current

sample. Here we show the case for
white phase jitter, which does not ac-

cumulate over time. Note the modu­lation of the error spikes by the sig­nal slope, which causes the error to
become very small at the signal
peaks.

We are now in a position to ana­lyze the added noise and how it re­lates to the signal. The spectrum will
be white, as there is no statistical re­lationship between one error pulse
and the next. The rms amplitude of
the noise spectrum is related to the
average slope of the DAC output sig­nal, as large step sizes between adja­cent samples cause large error pulses
in the error signal. This fact can be
seen clearly in Figure 4b, where a
higher-frequency signal (6 kHz) has
been applied to the D/A converter,

resulting in larger step sizes between

adjacent samples and hence larger er-

ror pulses.

We can summarize by saying
that for white phase modulation of
the clock, the D/A output will be
corrupted by white noise whose rms
amplitude varies with the average
slope of the signal. The worst-case
signal for audio would therefore be a
full-scale 20 kHz sine wave at the
D/A output.

The situation is slightly different
when an interpolation filter is used in
front of the D/A converter. Analysis
of that is beyond the scope of this ar­ticle, but the results are simple: the
sensitivity to white phase jitter is re­duced in direct proportion to the
oversampling ratio. This means that a
D/A converter using a 16x interpola­tion chip will be four times less sen­sitive to jitter than one using a 4x
oversampling filter (assuming that
the absolute jitter in ps is the same

for both clocks).

If the jitter is not white phase jit­ter but rather a relatively slow varia­tion of the clock frequency (low­frequency correlated jitter), then the
situation is quite different. Assume
that we feed the DAC with a sine
wave. Spectrally speaking, a slowly
wandering clock signal will cause
narrowband noise "skirts" to appear
around the frequency of the sine
wave signal. Oversampling no longer

has much effect on the output spec­trum, as the errors introduced by the
clock modulation are all "inband"

(below 20 kHz).

Many types of jitter fall in be-

tween the pure white phase jitter and
slow frequency-variation type of jit­ter described above. In that case,
oversampling may improve the jitter
sensitivity to a certain degree, but not
as much as in the case of truly ran­dom white phase jitter. Jitter in
which the time base is sinusoidally
modulated will potentially produce
discrete frequency components spaced
around spectral sticks in the DAC

output signal.

For resistive ladder converters, it
is obvious that with no input signal
(or dc), jitter cannot have any effect
on the output. The output is not
changing, so it doesn't matter exactly
when it doesn't change! While this
observation may seem trivial, the
same statement cannot be made for
other types of converters, as we shall
see presently.

In summary, regarding resistive
ladder D/A converters, we can state
the following:

• For white phase clock jitter, the

jitter spectrum on the DAC output is

white and proportional to the aver­age of the absolute value of the sig­nal slope. Oversampling filters de­crease the jitter sensitivity in direct
proportion to the oversampling ratio.
With no input to the DAC, jitter has
no effect and does not raise the noise
floor.

• For "slow" variations in the
frequency of the clock signal, narrow
noise sidebands appear around sinu­soidal components in the D/A output
spectrum, again with an amplitude
proportional to the frequency and
amplitude of the sinusoid. Oversam­pling filters do not decrease the jitter
sensitivity in this case.

THE AUDIO CRITIC

pdf 15

Figure 4

Figure 4a:
Time-domain jitter error
for a 1 kHz signal.

Figure 4b:

Time-domain jitter error

for a 6 kHz signal.

ISSUE NO. 21 • SPRING 1994 15

pdf 16

Figure 5: Time-domain jitter waveforms for 1-bit D/A converters.

4(b) One-Bit Noise-Shaping D/A
Converters with No Switched-Cap
Output Filter

This type of converter has be­come very popular in recent years,
both because of its inherent linearity
and because it can be implemented in
an all-digital CMOS process.

One-bit noise-shaping converters
can be further divided into two class­es. The first is a single-loop modula­tor with 1-bit quantization, and this
1-bit signal is fed directly to a 1-bit
output stage at the modulator clock
rate. The second class of converters,
so-called MASH converters, involve
multiple loops with feedforward
noise cancellation. They internally

produce a multiple-level digital sig­nal, which is converted to a 1-bit sig­nal through digital pulse-width mod­ulation. To achieve the desired time
resolution for the digital pulse-width
modulator, a clock signal with a very
high frequency is typically used.

1

In the previous section, we saw
that the sensitivity to clock jitter was
proportional to the changes in the
DAC output from one sample to the
next. For 1-bit converters, this
"change" is always full-scale, regard­less of the actual input signal!

Figure 5 shows a 1-bit waveform

with and without clock jitter, and the

resulting error pulse, for the case of
uncorrelated jitter (white phase­modulation jitter). Differing from the
previous case, the effect of jitter is
largely signal-independent, and in
fact the input signal to the noise­shaper loop may be zero with no re­duction in jitter sensitivity. This is
because the modulator is still switch­ing vigorously even when the input
is zero. In some cases there may be a
slight reverse sensitivity to signal
amplitude, as the number of transi­tions per unit time in the output bit­stream generally decreases as the sig­nal approaches the maximum level
in either positive or negative direc­tions.

A rough estimate of the jitter

1

The new DAC devices of NPC use
single-loop modulators with multilevel
quantizers. As in the MASH devices, the
multilevel digital signal is converted to a
1-bit signal through digital pulse-width
modulation. The new Philips-designed
$400 Marantz CD-63 uses an NPC DAC.

—David Rich

16

sensitivity can be obtained simply by
taking the average of the absolute
value of the instantaneous jitter (in
ps, for example) and dividing by the
period of the high-frequency clock
that drives the 1-bit output stage, and
then dividing by the square root of

the oversampling ratio. This noise
power is then compared with the
maximum rms signal that can be pro­duced by the 1-bit output. A quick
calculation of a typical system indi­cates a jitter sensitivity on the order
of 20 ps rms for 16-bit performance!
This is more than an order of magni­tude more sensitive than for the case
of resistive ladder converters. It is
far from trivial to design an oscillator
this good; only crystal oscillators
have a chance of meeting the spec.
Elaborate test equipment is required
even to measure jitter as low as this.

Comparing two different sigma­delta D/A converters having different
clock rates by measuring the clock

jitter for each converter is not trivial.

For example, suppose that the output
stage of converter A runs at 12 MHz
while the output stage of converter B

runs at 6 MHz (half the rate of A).
Also assume that both clocks are di­vided down from a common 24 MHz
master clock signal with a certain
amount of white phase jitter. Note
that the absolute jitter in ps is the
same for both the 6 MHz clock and
the 12 MHz clock, as a digital divid­er will maintain the absolute edge po­sitions of the master clock on its di­vided outputs. The noise produced
by each converter as a result of jitter
is proportional not only to the
amount of white phase jitter but also
to how often it occurs. As a result,
the converter running at the 12 MHz
rate will produce 3 dB more total
noise than the one running at 6 MHz
(recall that each edge displacement is
an independent statistical event, and
therefore the rms noise increases by
3 dB for each doubling of the number
of error events). However, the noise
produced by converter B is spread
over twice the bandwidth as that pro­duced by converter A, and therefore
the two converters produce the same
amount of inband noise.

This analysis applies only to 1-

THE AUDIO CRITIC

pdf 17

Figure 6: Sigma-delta converter with switched-capacitor filter.

bit output DACs. If we apply the
same experimental setup as before to
a conventional interpolated R-2R
DAC, we do not get the same results.
This is because the DAC that oper­ates at the higher-frequency clock
rate is more highly interpolated,
causing the sample-to-sample differ­ences between one output sample and
the next to decrease in direct propor­tion to the oversampling ratio. As a
result, the DAC operating at the high­er clock rate would show 6 dB less
inband noise than the same DAC run­ning at half the clock frequency. This
result is different from the 1-bit re­sult because for a sigma-delta output
stage, the sample-to-sample transi­tions are always full-scale, regardless
of the clock rate or interpolation ratio.

It is clear from this analysis that

1-bit converters are very sensitive to
white phase modulation of the clock.
This can be explained in the frequen­cy domain by the observation that
phase modulation causes the out-of­band shaped noise to "fold" down
into the baseband. In the time do­main, it is intuitively clear that the
edge-to-edge jitter is what affects this
type of converter the most. Low­frequency correlated jitter of the
clock is not nearly as damaging to
the noise performance of the DAC,
although correlated sidebands around
the signal may become a problem.

One interesting aspect of this
high jitter sensitivity is that it is
difficult for manufacturers using this
type of D/A converter to meet the
high dynamic-range specs common
with today's equipment. For exam­ple, in a system that uses a conven­tional, resistive ladder D/A convert­er, the noise is measured with no
digital codes toggling, and therefore
is determined only by analog thermal
circuit noise, which may be much

ISSUE NO. 21 • SPRING 1994

lower than theoretical 16-bit quanti­zation noise (this figure of course is
not meaningful, as the noise will in­crease as soon as the music begins).
Clock jitter has absolutely no effect
under these conditions. But in a sys­tem with a 1-bit sigma-delta or
MASH converter, the 1-bit signal is
very "busy" even when the digital in­put signal is silent. Clock jitter will
most likely be the dominant noise
source, and therefore it will be
difficult for the manufacturer to
claim 110 dB of dynamic range. In
some cases an extra circuit is used to
detect "digital silence" and the 1-bit
D/A output stream is actually turned
off under these conditions to make
the numbers look good!

In conclusion, 1-bit D/A con­verters with no discrete-time output
filters are extraordinarily sensitive to
edge-to-edge jitter caused by white

phase modulation of the clock. As we
saw before, such phase modulation
often is caused by passing a clock
with finite rise time through a buffer.
One must be very careful to reduce
this edge-to-edge type of jitter to the
lowest levels possible.

4(c) One-Bit Noise-Shaping D/A
Converters with Switched-Cap Out-

put Filtering

Figure 6 shows a block diagram

of a D/A converter with switched-cap
output filter. The origin of the name
"switched-capacitor filter" is obvious
from the diagram. The switching of
the capacitor removes its memory
characteristic. A first-order analysis
of a switched capacitor shows that it
is equivalent to a resistor of value

1/fC, where f is the frequency at

which the switches change state. In
an integrated circuit, the exact value
of a resistor or capacitor can vary by
30% or more. This makes the design

of precision RC filters impossible.
The transfer function of a switched­cap filter is dependent on the ratio of
capacitors, which can be controlled
to 0.1% in an integrated circuit. In
Figure 6 we see two switched­capacitor circuits that replace resis­tors in a continuous-time circuit. If
we replace the switched-cap circuits
with resistors, we can see that the
filter in Figure 6 is a first-order low­pass.

A more exact analysis of a
switched-capacitor circuit is required
to take into account the sampling ef­fect of the switch. This more exact
analysis shows that a switched-cap
filter must be analyzed as a discrete­time system in much the same man­ner as a digital filter. It is often a

point of great confusion how a
switched-cap output filter is different
from a continuous-time active RC
filter when it comes to jitter. The an­swer is as follows. A switched-cap
filter will settle to a particular value
regardless of when the clock edge oc­curs (assuming the op-amp is fast
enough to settle completely within
one switching interval). It is a true
discrete-time system in that the dy­namic settling behavior of all circuit
voltages is not important as long as
the settled value is correct. This is
not true for the case of a 1-bit modu­lator feeding a continuous-time filter,
where an error in the switching time
of the 1-bit signal does have an effect
on the voltage at the end of the clock
period, and in fact will change the
output voltage for many hundreds of
cycles thereafter.

From this discussion we can
conclude that a 1-bit modulator feed­ing a switched-cap filter behaves the
same as a resistive-ladder converter
running at the same oversampling ra­tio, as both converters produce volt­ages that settle to the same value re­gardless of the timing of the clock
edge. This is true as long as the
switched-cap filter completely re­moves the out-of-band noise from the

1-bit modulator. If that's not the
case, then the jitter sensitivity de­pends on the sample-to-sample dif­ferential change. If the change is
dominated by the signal slope itself,
then the jitter sensitivity is un­changed from the case where the
out-of-band noise is completely re-

19

pdf 18

Figure 7: Time-domain view of sample-rate conversion.

20 THE AUDIO CRITIC

moved. If the sample-to-sample

change is dominated by unfiltered
out-of-band noise, then the jitter sen­sitivity will be in proportion to the
rms value of the sample-to-sample
changes. Note that in cases where a
switched-cap filter is followed by a
continuous-time analog filter to re­duce further the out-of-band noise,
only the switched-cap filter is useful
for reducing the sensitivity to jitter.
This fact implies that it is impossible
to predict the jitter sensitivity of a

chip-level D/A converter that has

both switched-cap and continuous­time RC filters onboard, as it is im­possible to tell how much of the
filtering is done in each section.

2

5 Asynchronous Sample-Rate
Converters (ASRC) and Jitter
Reduction

In a previous paper [Adams and
Kwan 1993] I described an algorithm
and VLSI implementation of it which
allow sample-rate conversion between

arbitrary asynchronous rates. Unlike
synchronous converters, the device
accepts external clocks at Fsin and
Fs

out

, and by performing various sig­nal-processing operations on those
clocks it is able to derive a high­accuracy estimate of the current sam-

ple-rate ratio, and this estimate is
continuously updated so as to track
real-time variations in the input or
output sample rates.

Figure 7 shows a time-domain
view of sample-rate conversion. Con­ceptually, asynchronous conversion
consists of interpolating the input
sequence to an extremely high fre­quency, which causes the amplitude
differences between adjacent interpo­lated samples to become very small.

The output resampling process then
consists of picking off the nearest in­terpolated point.

The ASRC chip described uses a
polyphase filter approach, with
65,536 unique polyphase filters of
length 64, each stored in compressed
form in ROM. This approach to rate

conversion is more efficient to imple­ment than the interpolation/decima­tion model, as unneeded interpolated
outputs are not computed. While poly­phase filtering sounds complicated, it
is actually quite a simple concept.

Every FIR filter has a particular

group delay, which defines how
much delay the filter introduces to
signals appearing on its input. A typ­ical interpolation filter might exhibit
about 600 µs of group delay, for ex­ample. Most FIR filter are designed

to be linear-phase, which means that
the delay introduced by the filter is
independent of frequency.

Normal linear-phase FIR filters
have a group delay that corresponds
to an integral number of clock cycles.
But it also is possible to design an
FIR filter which is linear-phase but
has a group delay of an integer plus
a fractional number of clock cycles.
For example, a normal FIR filter of
length 100 taps might have a group

delay of 50 samples (half its length).
But it is possible to design a linear­phase FIR filter with a group delay of

50.5 samples. Now suppose that we

had a large bank of FIR filters all
connected to the same input signal,
and each of these filters had the same
frequency response but a slightly dif­ferent group delay. When we change
the sample rate of a signal, we are ef­fectively attempting to resample the
signal at a point between the existing
sampled points of the input signal.

Using the filter bank described, we

could simply pick a filter output
whose delay matched most closely

the desired resampling point for that
particular output sample. For exam­ple, if the output clock signal (which
differs in frequency from the input
clock signal) were to fall halfway be­tween two edges of the input clock
signal, we would select the filter that
has a group delay of 50.5 input sam­ple periods.

To ensure that we have enough
possible group delays to select from,
the new AD 1890 chip uses a bank of

2

The Crystal CS4303, Burr-Brown
PCM67, all MASH chips, and all NPC
chips use continuous-time filters. An ap­plication note available from Crystal
Semiconductor on the CS4303 shows
how difficult it is to create the required
low-jitter clock. The Crystal CS4328 uses
a fourth-order switched-capacitor filter
for removal of most of the out-of-band
energy. It is followed by a second-order
continuous-time reconstruction filter which

removes most of the image signals that
arise from the switch sampling of the

switched-cap filter. The Philips 1-bit

DACs are a hybrid. They use an on-chip
first-order switched-cap filter (similar to
that in Figure 6), which removes some of
the out-of-band energy. This is followed
by an off-chip continuous-time filter.

—David Rich

pdf 19

Stable clock

Figure 8: Outboard D/A processor with sample-rate converter.

65,536 possible filters. Depending on
the internally calculated, desired out­put-resampling point, one of these
65,536 possible sets of coefficients is
selected to weight the surrounding in­put data values to produce an output
point.

By using such a large number of
possible group delays, the error intro­duced by the time quantization in the
resampling process is about at the

16-bit level for the "worst-case" sig­nal, which is a full-scale 20 kHz
input signal. For lower amplitudes
and/or frequencies, the error is sig­nificantly less and is ultimately lim­ited by the stopband rejection of the
polyphase filters.

The algorithm used in this chip
is quite complex, and interested read­ers are referred to AES Preprint 3712

for details. At the heart of the chip
lies a circuit that computes an inter-

nal estimate of the ratio of input to

output sample frequencies. This com­putation is done continuously, and in
fact will track real-time changes in

sample rates that are quite rapid. The

interesting part of this chip from the
perspective of rejecting jitter lies in
the fact that this internal estimate of
the frequency ratio is computed using
thousands of past input- and output­sample clock events, and is therefore
immune to small perturbations in the
arrival time of any individual clock
edge. This jitter-rejection capability
may be thought of as a filter where
the cutoff frequency is as low as 3
Hz. Any jitter components above 3
Hz are filtered with a lowpass charac­teristic of -6 dB/octave. For exam­ple, a 100 Hz jitter component is 5
octaves above 3 Hz, and will be at­tenuated by 30 dB. Higher-frequency

jitter, naturally, will be attenuated

even more.

This ability to reject jitter sug-

gests an interesting system architec­ture for outboard D/A converters.
Normally, they must recover a low-

jitter clock from the incoming serial

bitstream. Typical integrated S/PDIF
receivers have PLLs with relatively
poor jitter-rejection capability, and
even if the original S/PDIF signal is
of good quality, by the time it has
traveled through several feet of cable
it may have high jitter due to the in­tersymbol interference caused by the
finite bandwidth of the cable. That

ISSUE NO. 21 • SPRING 1994

makes it difficult for the circuit de-

signer to provide a stable low-jitter

clock to the D/A converter chip.

Figure 8 shows a system where a
sample-rate converter is used be­tween the clock-recovery and S/PDIF
receiver chip and the D/A converter.
Note that since the input and output
sample frequencies are decoupled,
the designer can use a crystal oscilla­tor to provide a low-jitter clock to
both the D/A converter and the out­put side of the sample-rate converter.
The sample-rate converter will pre­vent the input jitter from affecting
the output data, and the D/A convert­er is allowed to operate with a crystal
clock signal. Note also that the sam­ple-rate converter will reject jitter on
its output clock as well, in that the
output data will not be affected by
the jitter. But if this jittered clock is
used to clock the D/A chip, then er­rors will still arise in the D/A con­verter itself.

Evidently such ideas are confus­ing, as conceptual errors have al­ready appeared in print. Suppose that
one measures the jitter at the D/A
converter clock pin both with and
without the sample-rate converter be­ing used. Suppose that without the
sample-rate converter, the jitter is
measured to be 1 ns rms. In this case
we are measuring the jitter of the
PLL used in the clock-recovery cir­cuit of the S/PDIF receiver. Now we
measure the same D/A clock pin us­ing the sample-rate converter, and we
find that the jitter is measured to be

100 ps. One might be inclined to
state that the sample-rate converter
chip has reduced the jitter from 1 ns
to 100 ps. But this misses an impor­tant point: the sample-rate converter
does not produce an output clock;

rather, it accepts whatever clock the
user feeds to it. The reduction in jit­ter follows because the designer is
free to supply a crystal-generated
clock signal to the D/A converter,
and this clock can be as clean as he
or she can make it. In fact, even with
the world's worst sample-rate con­verter, as long as it was asynchro­nous the clock jitter measured at the
D/A converter would be that of the
crystal oscillator itself.

How, then, can we judge the
quality of the sample-rate conver­sion? There is no clock signal pro­duced by the sample-rate converter
chip that can be measured (it does
not even exist internally!). The only
way to judge the signal quality is to
look at the signal itself. Since the jit­ter is filtered to about 3 Hz with the
sample-rate converter chip, it should

be possible to "zoom in" on a sine­wave test tone using an extremely
long FFT (or a slow-sweeping analog
spectrum analyzer connected to the
D/A output) and see very narrow
noise "skirts" around it. These are in
fact visible if you have test equip­ment good enough to resolve such
narrow sidebands. Note that these
sidebands are much narrower and
lower in energy than the narrowband
noise-modulation products in such
bit-rate reduction schemes as DCC
and MiniDisc, so they will not be
audible.

To demonstrate the jitter-rejec-

tion capability of this chip, we have
constructed an artificially severe test
where both the input and output
clocks are jittered with 100 ns p-p bi­nomial-distributed jitter (produced
by feeding a random binomial bit­stream into an RC oscillator). Notice
that the degradation consists entirely

21

pdf 20

of noise tails around the fundamental
frequency (see Figure 9). This obvi­ously is an exaggerated case, de­signed to make the noise sidebands

readily visible. Normal amounts of

jitter cause much smaller amounts of

narrowband noise modulation, and in
most cases extremely long FFTs are
required to measure it with any accu-

racy.

6 Recommended Measurement

Practice

In view of the expertise required
to interpret D/A clock-jitter measure­ments, reviewers should stay focused
on the signal and not try to infer
signal degradation from jitter mea­surements. I propose the following
measurements, which would be par­ticularly sensitive.

6(a) Noise Modulation with a Full­Scale 20 kHz Sine Wave

This test is easy. Using a spec­trum analyzer and a notch filter, ap­ply a full-scale 20 kHz test signal to
the D/A converter in question. Mea­sure the increase in the broadband
noise level, ignoring spectral distor­tion components. Note that sigma­delta or MASH converters with no
switched-cap output filters will not

be stressed by this test, as the slope
of the signal at the discrete-time/
continuous-time boundary is domi­nated by the shaped quantization
noise and not the signal. This test
will particularly stress conventional,
resistive ladder DACs that use low
oversampling ratios (4x, for exam-

ple).

6(b) Narrowband Noise Modulation

This test should reveal low­frequency jitter components. I would
define this as the difference between
a THD + N test made with a normal­width notch filter and the same test
done with a very narrow notch filter
(say, 50 Hz) using an input frequency
of 5 kHz or so. The problem with this
test is that it will not have any psy­choacoustic relevance, as narrow­band noise components will be
masked. It is of technical interest
only, as an aid to the design engineer.

6(c) Correlated Jitter

This test simply looks for fre-

quency components that seem to fol-

22

low the frequency of the input signal
with a particular offset. For example,
if distortion components were noted
at 800, 900, 1100, and 1200 Hz with
an input signal of 1 kHz, we could in­fer that 100 Hz FM was present in
the clock signal, especially if the

same relative spacing was observed
as the input signal was increased to,
say, 2 kHz. This can be a tricky mea­surement to interpret, as other prob­lems may produce the same spectrum
(for example, a small amount of 1
kHz and 2 kHz signal appearing on
the D/A reference pin, causing AM,
which is difficult to distinguish from

FM). It is also possible that some FM
components may be higher in fre­quency than the "carrier" (input sig­nal) and may therefore alias, causing
patterns that are difficult to interpret.
Still another potential problem is that
the jitter frequency may be related to
the signal itself (for example,

coupling of the MSB into oscillator
circuits), and therefore the fixed­offset relationship mentioned earlier
between the signal and its distortion
components may fail to appear. Note
that very low-frequency clock FM is
apt not to show up in traditional THD
+ N measurements, as the width of
the notch filter commonly employed
is usually large enough to attenuate

the sidebands as well as the main

signal. Again, one can point out that

low-frequency jitter is not important

when it falls within a critical band.

Comments

Traditional THD + N versus fre­quency tests and FFT spectrum plots
for input signals of various frequen-

cies are adequate to cover the effects
caused by jitter. There is no reason to
single out distortion components
caused by jitter as distinct from those
caused by such other effects as D/A
nonlinearity, op-amp distortion, etc.,
the only possible exception being the
broadband noise-modulation test men­tioned above. A full-scale 15 kHz or
20 kHz tone will have no masking
for frequencies in the 1-5 kHz range,
and for those very rare audio signals
with most of their energy in the last
half-octave, it is possible that, if ex-

cessive jitter is present, noise modu­lation might be audible at very high
playback levels. A high-frequency
THD + N measurement may not ade­quately show this noise modulation if
the residual distortion signal is domi­nated by discrete harmonic distor­tion. On the other hand, if the THD +

N at 20 kHz shows little or no rise
compared with the same measure­ment at 1 kHz, jitter is not a problem.

7 Conclusion

Any signal degradation caused
by using jittered clocks to drive D/A
converters is a complex combination

(continued on page 33)

Figure 9: FFT with 100 ns p-p binomial jitter on the L/R clock.

THE AUDIO CRITIC

pdf 21

Loudspeaker Systems Using

Forward-Firing Cones and Domes:

How Good Can They Get?

By Peter Aczel

Editor and Publisher

As drivers, designers, and system design software become more

and more sophisticated, the sound improves—but how much? We

tested a number of interesting new speakers to explore the answer.

It has become a truism—tweako cultists to the con­trary notwithstanding—that switching to a significantly
better loudspeaker will improve your sound more than
any change in electronic components, regardless of cost.
This publication is one of the most fervent advocates of
that tenet.

Even so, if you already own a fine forward-firing
dynamic speaker system—let us say one of the higher­priced models by Snell, B&W, Fried, or anything else of
that order—do not imagine that your audio life will dra­matically change if you switch to the latest and most

highly refined system of the same basic architecture.
There are no miracles in electroacoustics—if the music

comes out of 1-inch and 6-inch and 12-inch holes, it will
invariably sound more constricted than it would in
Carnegie Hall, even if those holes are exceedingly trans-

parent. If that sounds like an indirect endorsement of
large planar and line-source speakers, so be it—but that

approach has its own goblins and demons. If I were a

record producer and wanted to monitor a new recording
in progress, I think I would still opt for the most accurate
forward-firing dynamic speaker I could lay my hands on.
The truth—what do I really have on my tape?— is more

likely to be told by such a speaker. But for home listen-

ing, strictly for pleasure? I'm still not sure.

My vacillation is rather strikingly exemplified by

the DCM Time Window Seven, which I reviewed in the
last issue and which quite successfully splits the differ­ence between a foward-firing monitor and a multiple­wave-launch bi- or omnidirectional type of design. It

sounds bigger and more room-filling, though perhaps a

bit less precise, than the former but more focused and

sharply delineated than the latter, yet almost as expansive
and 3-D. I have become quite addicted to it and now rec­ommend it even more highly than I did before.

Anyway, perfecting the standard, conventional for­mats is always a most important part of technological

progress, and the speaker systems reviewed below are

ISSUE NO. 21 • SPRING 1994

some of the best recent examples of that endeavor. The
"monkey coffin, " as the old-time salesmen used to call
the classic forward-firing box speaker, has come a long
way.

Indeed, with each new generation of speakers and
each small improvement, we are probably not far from
the point where double-blind comparisons at matched
levels will be needed to evaluate and rank some of the
more similar designs. As amplifierlike accuracy is ap-

proached (I said approached, not achieved), amplifier­type test procedures will have to be used. I don't think
we're there yet; the differences are still not all that subtle;
but I am beginning to do some preliminary explorations
of speaker ABX-ing. (See also my comments on this sub-

ject in Issue No. 20, p. 39).

Note to new readers.

Background information on how we measure and

evaluate loudspeakers, why we do it that way, what we

don't do and why not, what we wish we could do, etc., is

scattered throughout Issues No. 10, 11, 14, 16, 17, 19,
and 20. There is no way to rehash all that material here
for new readers; all those issues are still available, how­ever, except No. 10.

B&W Matrix 803 Series 2

B&W Loudspeakers of America, P.O. Box 8, North Reading,
MA 01864-0008. Voice: [phone]. Fax: [phone].
Matrix 803 Series 2 floor-standing 3-way loudspeaker system,

$3000.00 the pair. Tested samples on loan from distributor.

This is not the flagship of the B&W line but it rep­resents the most recent and therefore probably the most
mature implementation of their technology and design
philosophy. It is a very well-designed, very accurate, one
could say almost flawless, monitor-type loudspeaker. If
my listening room were smaller, and if the fixed tweeter

25

pdf 22

level were set just a smidgen lower (maybe by only 1 or
2 dB), the 803 Series 2 would probably be a contender
for reference status in my ranking of speakers, at least as

a one-piece system of reasonable size—it's that good. As
it is, however, it sounds somewhat unassertive in size and
dynamics in the big room where I do most of my listen­ing and also a wee bit zingy on top. It sounds closer to

what I consider just right in a smaller room where I also
tried it.

The speaker is 40" high, 10" wide, and 13" deep (in
round numbers). The 1" metal-dome tweeter sits on top
of the cabinet in a separate streamlined pod, in the man­ner of all B&W 800 Series models. Three 6½" drivers
are mounted in a vertical line on the front side of the cab­inet. The topmost one is a Kevlar-cone bass/midrange
driver, the other two are Cobex-cone bass drivers. The
vented enclosure has its ducted port below the bottom
driver. The vertical edges of the cabinet are beveled, and
the grille is of a vaguely nondiffractive design (not as
clever as that of the Thiel grilles but of smaller impor­tance here because of the free-standing tweeter).

The bass system of the 803 Series 2 uses a vented

box with 4th-order Bessel alignment—more or less—

plus an optional filter that can be connected between
preamp and power amp, or alternately switched in out of
a tape loop, in case you prefer a 6th-order Butterworth
alignment. I was perfectly happy with the "Besselish,"
filterless alignment, which produced a nearfield response
curve that sloped gently downward from approximately
80 Hz to 20 Hz at an average rate of 2 dB per octave and
then broke sharply below 18 Hz at 12 dB per octave—an
overdamped profile. (I must add, however, that I always
have some minor reservations about the accuracy of a
bass response curve such as this, taken at the best appar­ent nearfield summing junction of the woofer and the
vent.) The box was tuned to 20 to 22 Hz; the maximum
output of the vent was at 19 Hz. The speaker put out a re­markable amount of tightly controlled, deep, accurate
bass for such a relatively slender, compact unit. The

filter, which B&W did not send me, would have flattened
the response in such a way that there would have been a
stonewall cutoff at 23 Hz and dead level output above
that, with much looser damping.

One widely advertised feature of all B&W 800
Series loudspeakers is their unique cabinet construction,
indicated by the Matrix designation before the model
number. The Matrix technique uses an elaborate honey­comb of internal braces to reduce cabinet resonances—
there is almost as much bracing as there is air inside the

cabinet. Here, in the 803 Series 2, the "next-generation"
Matrix 2 construction is introduced, with even more so­phisticated bracing of the upper (mid/bass) subenclosure.

When I tapped the center of the cabinet's side panel with
the little padded hammer I use for that purpose, it didn't
sound as dead as I had hoped; it appeared to be slightly

more pitched (tonal) in response to the tapping than some

26

cabinets without such high-tech credentials. I heard no
corresponding coloration, however, in the output of the
speaker.

The quasi-anechoic (MLS) frequency response of
the speaker from 300 Hz up proved to be extremely flat.
With the microphone on the axis of the mid/bass cone,
the response is within ±2 dB up to 20 kHz; furthermore,
the 300 Hz to 6 kHz segment is within ±1.25 dB, and the
6 kHz to 20 kHz segment is within ±0.75 dB. The overall

trend is ever so slightly upward, confirmed by the axial
response of the tweeter. The flattest response was ob­tained 30° off axis horizontally, with the microphone
aimed at the apex of the mid/bass cone: ±1.5 dB from
300 Hz to 16 kHz, down by 3 dB at 20 kHz. Maybe if the
tweeter were set a dB or two lower, as I suggested above,
the latter measurement would be unobtainable, but I

think I would still prefer the sound that way. Remarkable
results, in any event. (Incidentally, the phase curve was
also unproblematic, with only about 180° of rotation over
the measured range.)

In the time domain, tone bursts elicited no discern­ible ringing at any frequency—again remarkable. The
impedance curve, on the other hand, is a roller coaster,
fluctuating between 3.3 and 24 ohms (the latter at 1.5
kHz!) in magnitude and ±50° in phase. Not exactly an R­like load for the amplifier. (You can use two amplifiers,
however, as there are terminals provided for biamping
via the passive crossover or, needless to say, for tweako
biwiring, which is strongly recommended in the manual

—but who knows what marketing type wrote that.) The
nominal impedance of the speaker is given as 8 ohms,
which is the value at 400 Hz, as well as at 4 kHz. I did
not make extensive distortion measurements, only
enough to determine that the distortion is reasonably, but
not exceptionally, low. Efficiency is quite high but hard
to nail down because of the difficulty of measuring pow­er into that roller-coaster impedance.

In a medium-sized room the sound of the B&W
Matrix 803 Series 2 is that of a highly accurate forward­firing monitor and pretty much above criticism within
that genre. I have no reservations about it other than what
I have already noted. The speaker simply reproduces its
input without significant alterations and launches it for­ward into the room. If that's what you want, this is one of
the very few designs you need to check out. My plan is to
live with it a little longer, possibly do a followup report
with additional measurements, and use it as a convenient
norm in monitor-type speaker evaluations, since the
B&W 800 Series is widely recognized as a reference.

NHT Model 3.3

Now Hear This, Inc., 535 Getty Court, #A, Benicia, CA 94510.

Voice: [phone]. Fax: [phone]. Customer ser-

vice: (800) NHT-9993. Model 3.3 floor-standing 4-way loud-

THE AUDIO CRITIC

pdf 23

speaker system, $4000.00 the pair. Tested samples on loan from
manufacturer.

If you have read the interview with NHT's Ken
Kantor in our last issue, or have been exposed elsewhere
to his ideas, you already know that he is a formidable and
highly original thinker on the subject of sound reproduc­tion. This new loudspeaker system, his most ambitious
design to date, confirms and documents that perception.
In my experience as a reviewer, the NHT Model 3.3 is
the most unconventional implementation of basically
conventional transducer/enclosure technology and prob­ably the most nearly perfect. That doesn't necessarily
make it irresistible to me as a music lover, but it is cer­tainly a world-class design by any criterion.

The unconventionality of the speaker is immediate­ly apparent from its shape and driver deployment, both of
which are ingeniously conceived to compel correct room

placement—the user actually has no other practical
choice. The enclosure is a big, flat slab, 42" high by 31"
deep by only 7" wide, which must be placed with the nar­row front side facing the listening area and the opposite

narrow side against the back wall. (If you don't have an
available back wall, the speaker is not for you, at least in
my opinion.) The left and right units are mirror-imaged,

each with a 12" woofer mounted inboard in the "corner"
formed by the side of the slab and the wall. That provides
four times the radiation resistance of conventional floor
placement out in the room, with vastly increased drive
capability at the bottom frequencies. The narrow front
face of the slab is angled inboard (i.e., with built-in toe­in) and has three more drivers mounted on it, each in its
own subenclosure: a 6½ upper-bass/lower-midrange
unit, a 4" upper-midrange unit, and a 1" dome tweeter.

(See also the front cover of this issue.) Thus the woofer

is where it should be, in a corner, and the midrange/treble
part of the speaker is also where it should be, out in the
room like a typical audiophile minimonitor. Clever, these

Americans....

I developed a particular admiration for the woofer

design, which uses only 70 liters (2½ cubic feet) of
sealed space behind the 12" sideways-firing polymer­cone driver. The skinny cabinet allows no larger volume;
nevertheless, the bass is the finest I have ever heard out
of an unequalized and passively crossed-over sealed box
of less than giant size. It remains tightly controlled and
natural-sounding on the biggest bass-drum whacks and
most powerful organ pedal blasts. Ken Kantor credits
NHT's Bill Bush (an amateur bassist!) for the driver de­sign; the actual OEM for the 12" unit (as well as the 6½"
one) is Tonegen, a Japanese firm. My nearfield measure­ment of the woofer showed the characteristic 12-dB-per­octave bottom-end rolloff of a sealed box; the -3 dB
inflection was at 35 Hz; there is a tiny notch at 23 Hz,
which is probably the limit of useful response. The virtual
corner boosts the in-room response at the bottom end

ISSUE NO. 21 • SPRING 1994

considerably, of course. The fundamental resonance of
the sealed box appears to be 29 Hz from the impedance
curves, suggesting a very slightly overdamped condition,

which is desirable in this type of design.

The other three drivers are configured in accor­dance with NHT's so-called Focused Image Geometry.
That means the front baffle is angled inboard by 21°,
which in the Ken Kantor canon is the optimum angle for
minimizing interaural cross-correlation—as long as the
listener is equidistant from the two speakers—and there­by optimizing stereo separation and ambience retrieval.
(More about that below; see also the aforementioned in­terview in Issue No. 20.) A strip of absorptive foam on
the outboard side of the midrange and tweeter units helps
to direct the sound radiation toward the listening area and
aways from adjacent walls. The whole thrust of NHT's
approach is to bring the stereo information to a focus in
the listening area and minimize the effects of room
acoustics on the quality of the reproduced signal.

The drivers in the Model 3.3 are crossed over as
follows: 12" to 6½" at 100 Hz, 2nd-order slopes; 6½" to
4" at 320 Hz, 2nd-order slopes; 4" to 1" at 3.5 kHz, 3rd­order slopes. A positive-going pulse at the input of the
system causes a forward excursion of the 12" and 6½
drivers and a rearward excursion of the 4" and 1" drivers.

This is as it should be with the crossover orders used, ex-

cept for the 100 Hz crossover, where out-of-phase wiring
would be expected with the 2nd-order filters; the large
distance between the drivers probably accounts for the
decision to do otherwise.

The frequency response of the speaker is amazing-

ly flat; indeed, the on-axis response of the tweeter (a 25­mm aluminum-dome unit with magnetic fluid damping,
made by SEAS of Norway) is almost amplifier-flat—and
I mean ±0.5 dB—from about 8.5 kHz to 22 kHz. Way up

in the ultrasonic 25 kHz region the tweeter has a resonant

peak; below 8.5 kHz there is a tiny downward step; but
even so the 1-meter quasi-anechoic (MLS) response of
the total system on axis is flat within ±1.5 dB from 22
kHz on down to the woofer range—which is really some­thing and shows that the 4" SEAS upper-midrange unit is

no slouch, either, over its 3½-octave band. Off axis in

the horizontal plane one has to distinguish inboard vs.
outboard response, the latter being deliberately barricad­ed by the foam strip. Inboard there is very little off-axis
rolloff even at the highest frequencies, but remember that

"on axis" in this speaker means 21° inboard to begin
with, so that off-axis measurements beyond an additional
20° are hardly meaningful. In the vertical plane the off-

axis measurements show mainly a tendency toward a

suckout around the 3.5 kHz crossover point—not a major

fault in my opinion. The phase response of the system is

unproblematic and, given this particular configuration,
basically what I would expect. Overall, I haven't seen a
better passive 4-way solution by any other designer, or

even one as good.

27

pdf 24

The Model 3.3 also excels in THD + N. It is very
difficult to make it distort more than 1% at any frequency
above 50 Hz without pushing the output to nearly unbear­able levels, and that figure drops sharply as the drive is
reduced. At the lowest frequencies (20 to 40 Hz) all bets
are off at high SPLs, as is usually the case, but the high­er-order harmonics are still very low.

I must add that none of my measurements differed
significantly from the manufacturer's specs and curves,
which appear to be remarkably honest. I wish I could say

the same of all loudspeaker brands. One thing that NHT
does not talk about but I found to be outstanding is the
complete freedom from ringing on tone bursts, regardless
of frequency. The reproduction of gated square pulses is
far from coherent; the waveforms emerge somewhat dis­organized, but less so than with other 4-way speaker sys­tems known to me. Let us not forget about the impedance
curve: above the woofer range it varies only from 4.5
ohms to 7.5 ohms in magnitude (6 ohms nominal); at the
29 Hz box resonance it rises to 10.5 ohms (only!); the

phase is well within ±30° at all frequencies. A piece of

cake for any amplifier—or two amplifiers, as there are
terminals provided for biamplification via the passive
crossover. (Or biwiring if that harmless delusion makes
you feel better.)

The only thing I found objectionable about the
Model 3.3 had to do with the stabilizer bars which must
be screwed to the bottom to keep the heavy (123 lb.) but
wafer-thin cabinet from toppling over sideways. For
$4000 one has the right to expect high-quality machine
screws and threaded inserts for attaching the stabilizers;
instead I had to screw huge self-tapping wood screws
into hardwood and was rewarded with blistered palms for
my unskilled effort. I never affixed the pointy cones the
stabilizer bars are supposed to rest on; they are not for
the reviewer who constantly moves things around on the
floor.

Well, you probably thought I'd never get around to
the subjective listening quality, but here goes: Very accu-

rate, transparent, and neutral—like the best of Snell, but
even more so. Not very forgiving when the recording is
overbright. Perfectly balanced when the recording is. A
little more open and airy than the Waveform Mach 7 but
maybe not quite as dynamic. Medium-sized wave launch
into a large listening room—I wish it presented a larger
apparent source, but it certainly doesn't sound small.
Beautifully delineated, strong bass, as discussed above,
though not quite on a par with the most sophisticated
sub woofers.

As for the Focused Image Geometry, I am the
wrong person to ask. Amazing snap-into-focus 3-D imag­ing is not my Holy Grail because I do not hear it in the
concert hall; it seems to be achievable only with micro­phones. I am basically a tonality/balance/transparency
man. Ken Kantor wants the new owner of a Model 3.3 to
set it up with half-inch-by-half-inch trial-and-error

28

changes in position, separation, parallelism, etc., until the
focus snaps in like a seat-belt buckle. I have no patience
for that. After attaching the stabilizer bars I placed the
speakers perpendicularly against the back wall a little
over 8 feet apart, and they imaged/soundstaged just fine
without further ado. There are other speakers that will
not image one tenth as well, no matter what you do.

Bottom line: the NHT Model 3.3 is a marvelous
speaker and perhaps not even overpriced at $4000, but
for me it doesn't have that ultimate I-can't-live-without-it

quality because, ideally, I want a bigger apparent sound

source plus dead-flat low-frequency response down to the

subsonic region. Ken Kantor claims that this is best loud­speaker system he knows how to make, regardless of
price, but I don't for a moment believe that. It is probably
the best 3½-foot-high monolithic 9½-octave-range
speaker he knows how to make. Or anyone else knows
how to make, for that matter.

Sequerra Model NFM-PRO

R. Sequerra Associates, Ltd., 792 Pacific Street, Stamford, CT

06904. Voice: [phone]. Fax: [phone]. Model
NFM-PRO 2-way nearfield-monitor loudspeaker system,
$2000.00 the pair. Tested samples on loan from manufacturer.

Dick Sequerra clearly has nine lives as an audio en­trepreneur; this particular engineering/marketing effort
represents the sixth one I am aware of but I may have
overlooked one or two. Right now the NFM-PRO consti­tutes the sum total of his active product line, or close to
it, as I understand from my brief conversations with him.

This is a teeny-weeny speaker, 11" high and deep,
6¾" wide, with 10 of its 12 edges rounded, no grille,
black Nextel finish all around. The driver complement

consists of a 6" plastic-cone woofer and, believe it or not,

a 2¼" paper-cone tweeter with an aftermarket phasing

plug. Dick Sequerra hates dome tweeters and is the only
one in the hi-fi loudspeaker industry (to my knowledge)
to use this Japanese-made cone tweeter. His main reason,
I believe, is power handling. The crossover appears to be
first-order (Dick Sequerra also hates 4th-order Linkwitz­Riley, etc.), and the woofer protrudes l¾" from the box
by means of a short tube extension in pursuit of better
time alignment. The tweeter level can be adjusted up and
down over a narrow range with a 6-position rotary switch
in the back.

How can such a simple little speaker cost $1000
per side? For one thing, it is very solidly built and, in its
austere way, carefully finished. For another, it is made
primarily for the professional market, to sit on top of a
console for nearfield monitoring, and that market isn't
necessarily driven by apparent-hardware-value-per-dollar.
Furthermore, Dick's prices have always been somewhat
arbitrary, based on—shall we say?—audiophile-awe-per­dollar. The NFM-PRO is, in effect, a slicker and more

THE AUDIO CRITIC

pdf 25

highly refined version of the Sequerra Met 7 series of
minimonitors. I have no idea how many of the paper­cone tweeters, for example, Dick must test, select, and re-

ject before he picks one for this speaker, so I can't really

comment on the fairness of the price. Cheap it ain't.

My nearfield measurement of the woofer in its
sealed enclosure showed a typical "maximally flat" 2nd­order response profile, with the -3 dB point at 70 Hz and
a 12 dB per octave decline below that frequency. I can't
see anyone getting more bass out of that small box. Nor
can I see anyone using the speaker without a subwoofer
in a home music system. (A studio engineer who already
knows the bass response of the monitored loop is a dif­ferent story.)

The 1-meter quasi-anechoic (MLS) frequency re­sponse of the NFM-PRO is reasonably flat on axis—with
the tweeter level control in the 0 position—but falls apart
dramatically off axis. The axial response is flattest with
the microphone aimed slightly above the apex of the
woofer: ±3 dB from 300 Hz to 17 kHz. Even within that
tolerance the cone tweeter shows more roughness, more
small resonances, than typical dome tweeters. The phase
curve stays within a much narrower angular range than
that of more complex speakers. At 30° off axis horizon­tally, a huge suckout (-13 dB) occurs at 2.3 kHz and the
tweeter response remains flat only up to 6 kHz, dropping
to -6 dB at 9 kHz and showing no response at all above

14 kHz. Welcome to the world of first-order crossovers
and big cone tweeters. Vertically off axis the picture is
quite similar. The speaker must be placed at ear level and
toed in, and that of course is the usual position on top of
the studio console. (Think of a pair of huge headphones
and you'll know how to use the NFM-PRO.) I must ad­mit, though, that the speaker is efficient and will play
very loud. The 1-watt 1-meter SPL is at least 90 dB and
may be as high as 93 dB—I obtained different figures
with different methods of measurement. The impedance
curve is very nice, between 6.3 and 10.5 ohms in magni-

tude at all frequencies above the box range (i.e., 8 ohms

nominal), the phase remaining quite close to the 0° axis

throughout.

In the time domain, where the speaker is claimed to
excel, I observed good, but not amazing, reproduction of
gated square pulses. For amazing pulse-shape retention

you need great bandwidth, which is not the case here.
The leading edges looked nice, however, thanks to the
driver alignment and first-order network. The trouble is
that I no longer believe in coherence, per se, as an audi­ble benefit. Ringing, on the other hand, is unquestionably

audible, and the NFM-PRO reproduces tone bursts of all
frequencies quite cleanly, with only a small amount of

ringing here and there, nothing to fault seriously.

The sound of the speaker corresponds quite closely
to what one would predict from the measurements. Light
bass, no highs off axis, a bit of roughness on axis, very
good dynamics, no surprises. Imaging is good, as it is

ISSUE NO. 21 • SPRING 1994

with nearly all tiny loudspeakers. Everything considered,
in a world of good, bad, and mediocre speaker systems,
this one falls into the good category. I could live with it
(plus a subwoofer) if nothing else were available. But I
recall the Fried Q/4 at $498 the pair—returned long
ago—as being smoother and more transparent; David

Rich's Monitor Audio Studio 6's ($2499 the pair) proved

to be far superior on all counts in a side-by-side compari­son; and the new Velodyne DF-661 at $1695 the pair (in

black vinyl) is in an altogether different league—but read
the preview below. By top-of-the-console near-your-head
criteria, on the other hand (i.e., very small, very loud,
reasonably accurate), I would say the Sequerra Model
NFM-PRO delivers what it promises.

Velodyne DF-661

(quick preview)

Velodyne Acoustics, Inc., 1070 Commercial Street, Suite 101,
San Jose, CA 95112. Voice: [phone]. Fax: (408) 436-

7276. DF-661 "Distortion-Free Full-Range Loudspeaker,"
$1695.00 the pair in black vinyl, $2245.00 the pair in rose­wood. Tested samples on loan from manufacturer.

We were late, ridiculously late, and then this came
in. If we hadn't been late, I couldn't have inserted this
preview here, so every cloud has a silver lining—if
you'll pardon the cliche—because this speaker appears to
be a major development and you should be aware of it.
Please understand, however, that this not an actual test
report; that will be presented in the next issue, after
sufficient exposure to the speaker in the laboratory and
the listening room

The DF-661 is a bookshelf-sized 3-way speaker
system with a 6" metal-cone woofer in a vented enclo­sure, 6" metal-cone midrange driver, and 1" metal-dome
tweeter. The two 6" units are of a totally new and different
design, claimed to reduce distortion by a whole order of
magnitude (i.e., divide the distortion by 10) in compari­son with other good loudspeakers. That means distortion
on the scale of amplifiers rather than electroacoustic
transducers! Other than that, no special claims are made
for the DF-661—no claims regarding frequency response,
diffraction, coherence, cabinet construction, etc.

I have run only a very few quick-and-dirty tests on
the speaker, just to still my curiosity, and am far from
ready to discuss numbers here. I have listened to the unit
at some length, however, and I can report that the claims
for it appear to be true. Ultralow distortion in a speaker
does sound different, quite different indeed, and can easi­ly become addictive after several days of listening. My
initial impression is that as a system design the DF-661
could stand some improvement, but the texture of sound
that emerges from it—as distinct from structure and bal­ance—sets a new standard.

More than that I will not say at this point. To the

29

pdf 26

orthopedic surgeon with a couple of G's burning a hole
in his pocket my message is this: (1) my heart bleeds for
you, doctor, because your new toy, should you decide to
go ahead and buy it now, does not yet have my official

blessing, and that's that; (2) what the hell, doctor, throw
caution to the wind and get the DF-661 without waiting
for Issue No. 22—the risk is small.

...and, by contrast, a

dipole:

Magneplanar MG-1.5/QR

Magnepan Incoporated, 1645 Ninth Street, White Bear Lake,
MN 55110. Voice: [phone]. Fax: [phone]. Mag­neplanar MG-1.5/QR 2-way quasi-ribbon/planar-magnetic

loudspeaker system, $1350.00 the pair. Tested samples on loan

from manufacturer.

Yes, I have a soft spot for planar/dipole/line-source
speakers, mainly because of their "room-friendly" wave­launch characteristics; but no, I never had a big love af­fair with Magnepan's products, even though they have
been very influential in the market and have a number of
strong virtues to recommend them. My two main prob­lems with them have always been, as far back as the late
'70s, bass performance and ringing diaphragms. That

doesn't seem to have changed much to this day.

It is Magnepan's party line that Magneplanar bass

(remember, Magnepan is the company and Magneplanar

the brand) possesses quality, whereas the bass you get

with subwoofers and their ilk is characterized by mere

quantity. Well, this reminds me of locker-room arguments
about the mammary endowment of cheerleaders; the fact
is, however, that a peripherally clamped plastic sheet,
even if driven over its entire surface as in the Magnepla-

nars, is not really a quality solution for bass reproduction.

A properly suspended and terminated stiff cone driven at
its apex and enclosed in a correctly designed box is a
much better-controlled and predictable low-frequency
transducer, though perhaps lacking in high-tech glitz by
comparison. As for ringing... but I'm getting ahead of
myself.

The Magneplanar MG-1.5/QR represents the top of
Magnepan's "cheaper" line, i.e., the line without genuine
ribbon tweeters. The tweeters in this not-quite-high-end
line are called "quasi-ribbon" by the manufacturer; less
scrupulous companies would call them ribbon, period,
which they aren't. In effect, these quasi-ribbon tweeters
are based on the same flat-Mylar-diaphragm-with­distributed-voice-coil principle as the low-frequency
transducers in the speakers. I have discussed this ap­proach to transducer design a number of times in the past
and do not intend to go over the same ground again (see,

30

for example, Issue No. 11). Magnepan's literature clearly

explains the design principle, but naturally only the pros

without the cons. Suffice it that the MG-1.5/QR is a 64"

high, 19" wide panel divided into a large low-frequency

section and a tall, narrow (i.e., ribbon-like) high­frequency section. The left and right units are mirror­imaged. The electrical crossover frequency appears to be
700 Hz, although the specs say the acoustical crossover
is at 1 kHz. The lowpass filter is 2nd-order; the highpass
filter is lst-order (series capacitor).

The measured impedance curve indicates that the
low-frequency and high-frequency transducers are both
essentially resistive (5.2 and 3.7 ohms respectively) but
that the crossover network synthesizes a fairly complex
load: magnitude, 3.7 to 29 ohms (peak at the crossover);

phase, -55° to +40°. The nominal impedance is 4 ohms.
Terminals are provided for biwiring (pure nonsense) or
biamping via the passive crossover (marginally advanta­geous). The low-frequency and high-frequency sections
are wired out of phase (probably because the overall
acoustical profile of the crossover is closer to 2nd-order
than anything else).

When the sound source is a tall structure like the
MG-1.5/QR, its frequency response is not as easily mea­surable as that of a minimonitor or even a fair-sized box
speaker. I opted for 2-meter MLS (i.e., quasi-anechoic)
measurements with the microphone aimed at the high­frequency strip from various heights and at various hori­zontal angles. I also took nearfield measurements of the
low-frequency panel. I would have preferred outdoor

measurements from a distance of 3 or 4 meters, but that

was not an available option. I am reasonably certain,

however, that I have a good handle on the speaker's over-

all response characteristic, even if the accuracy isn't the

highest possible.

The bass response of a speaker like the MG-1.5/QR
is hard to quantify with my usual nearfield technique be­cause the real-world bass output of a dipole exists only at
some distance from the launch point on account of the
back-to-front cancellation. I was able to determine that
the fundamental resonance of the "drumhead" (because
that's what the low-frequency panel really is) was at 44
Hz, with steeply falling output below that frequency, and
that there was another big bump at 75 Hz, the really
smooth response starting only at 125 Hz and continuing
(now on the basis of MLS) up to 1.3 kHz or so. Those
three-plus octaves look very nice and are probably the
long suit of the speaker. From there on up things get
rough again; the response up to 20 kHz is a series of

peaks and dips. The biggest dips are -7 dB on axis and

-10 dB off axis (30° inboard); the biggest peaks are
smaller, maybe +5 dB; the whole response profile is
much more ragged than anything we have seen in quality
speakers of conventional design. The high-frequency re­sponse above 11 kHz declines 6 dB per octave off axis.
Even allowing for some measurement errors, this is not

THE AUDIO CRITIC

pdf 27

an altogether happy picture.

In the time domain the picture is no happier. A tone
burst consisting of 10 cycles of 500 Hz makes the low­frequency panel produce 10 more cycles only about 14
dB down from the input. The same panel, when excited
with a single cycle of 20 Hz, produces another cycle at
the original level and then a decaying series of six more
cycles starting at only 9 dB down. That panel is alive!
The high-frequency strip also rings badly: a 4 kHz tone
burst of 10 cycles makes it produce 16 more decaying
cycles, starting at 6 dB down and persisting at 12 dB
down most of the way. As for coherence, I found a 1 kHz
square wave to be pretty badly mangled by the crossover,
but a 2 kHz square wave came off the high-frequency
strip looking quite nice.

How does all that affect the speaker in terms of
subjectively perceived sound? More or less as you would
expect: not enough bass, not enough highs (at least from
some listening angles), not enough focus, too much col­oration, an overall not-quite-neutral quality. Lest that
should appear like a totally negative reaction to the
speaker, I must add that I still liked the planar wave
launch—the height and width of the apparent sound
source were just right, and that made for reasonably hap-

py listening despite all of the above. Even a faulty planar

speaker has a certain dimensional Tightness about its
sound.

Since I had not looked at a Magnepan product for

years, I felt a little uneasy about the possibility that this

single unenthusiastic review would define my view of all

Magneplanars in the eyes of our readers. I therefore bor­rowed a pair of Magneplanar Tympani IVa's from a
friend and put it through its paces. The Tympany IVa,
with three screenlike hinged panels per side and a bona

fide ribbon tweeter, was until 1992 the top of the Mag­neplanar line at $3750.00 the pair. It is no longer made,
having been replaced by the single-panel MG-3.3/R
($3000.00 the pair) and single-panel MG-20/R ($8500.00

the pair). All I wanted, however, was a generic high-end
Magneplanar for reference and comparison, and the IVa
served that purpose very well. It is an incomparably bet­ter speaker than the MG-1.5/QR (at almost three times
the price it had better be!) but it still has some of the

same generic shortcomings. It rings all over the place in

much the same manner (not the ribbon, though), and the
bass is still far from superb, with a big bump at 35 Hz
and poor damping. A single Velodyne ULD-15 Series II
made the whole system sound significantly more authori­tative below 85 Hz. The frequency response of the IVa,
however, is much flatter and less ragged than that of the
MG-1.5/QR, and in the 2½ octaves from 100 to 600 Hz
(upper bass and lower midrange) the IVa is extremely
impressive, with tremendous impact and dynamics result­ing from the huge drive area. No conventional speaker I
know of is its equal in that respect. On the other hand,
imaging and directional cues are quite vague on the IVa,

ISSUE NO. 21 • SPRING 1994

and the overall sound has a subtly colored, zingy charac­ter. The apparent sound source is of course huge. Not a
negligible speaker, all in all. (Now it's history, in any
event.)

I am hoping that the Magneplanar MG-20/R has
the answers to all (or at least most) of the questions
raised by the intrinsically problematic nature of this kind
of planar-magnetic design and that Magnepan will not
punish us for our detached objectivity (read lack of
groupie adulation) by denying us test samples.

.. .plus, a very different

subwoofer:

Bag End ELF Systems
S10E-C and S18E-C

Modular Sound Systems, Inc., P.O. Box 488, Barring ton, IL

60011. Voice: [phone]. Fax: [phone]. ELF-1
two-channel dual integrator electronics, $2460.00. S10E-C
black-carpet enclosure with single 10" woofer, $234.00 each.
S18E-C black-carpet enclosure with single 18" woofer, $658.00
each. Tested samples on loan from manufacturer.

I was tempted to call this a preview because I am
still in the process of evaluating the ELF approach to
bass reproduction and far from finished, but then a pre­view is supposed to be much more tentative in its conclu­sions, if any, than mine already are about this remarkable

product. The truth is that I was just about finished with

speakers for this issue when the ELF system arrived, and
I wasn't going to get involved in it before press time.

Well, I did get involved in it and got hooked. So here

goes this first-look, more-than-just-a-preview early review:

I have never heard bass like this in my listening

room. I didn't even know what I was missing. How's that

for openers?

Now, in all fairness I must hasten to point out that I
never had in my system a pair of those floor-to-ceiling
multiple-woofer towers (as in the Infinity IRS, Martin­Logan Statement, Genesis I, etc.), nor a pair of Velo­dynes (only a single matrixed ULD-15 Series II), nor any
number of other highly regarded sub woofers. I have

heard these elsewhere but not in my familiar environ-

ment. To my ears none of them had the Bag End ELF
sound—but that, of course, is the kind of anecdotal report
you should never trust.

Bag End is the cutesy brand name chosen by Mod­ular Sound Systems, a Chicago firm, for their line of
speakers. (The name comes, I believe, from J. R. R.
Tolkien's The Hobbit.) Although virtually unknown in
the high-end audio community, Bag End is a familiar
brand in the field of professional sound (music stores,

31

pdf 28

rock groups, sound system specialists, etc.). What's good
for road shows, arenas, and churches isn't automatically
bad for high-end home music systems, as some snobby
tweaks will immediately think. Au contraire, the ELF
system was developed by heavy-duty technologists in
search of the most accurate bass, not by ex-salesmen who

learned their physics on the floor of an audio store and
talk about "fast" woofers. Our readers have probably
heard of one of the designers, Ed Long, who writes an
occasional technical review for Audio and owns various
audio patents. Ron Wickersham was the other half of the
ELF development team. Long and Wickersham are not
part of Modular Sound Systems, which is a licensee and
is run by Jim Wischmeyer and Henry Heine. ELF is a
cutesy acronym for Extended Low Frequencies—and
that means extended down to 8 Hz. Really. I measured it,
hobbits and elves notwithstanding.

The basic idea behind the ELF approach is one of

those hey-I-thought-everybody-knew-that insights, really

simple and obvious once it is stated. What is the main

problem to be solved by the designer of a superior speaker

system for bass reproduction? Answer: pushing the fun-

damental resonance as far down as possible and control­ling it down there. The ELF solution to the problem: let's
not solve it! That's always an excellent solution when
there exists a viable alternative, which in this case is
pushing the fundamental resonance as far up as possible,
let us say to 70 Hz. In a sealed box, the response below
that frequency will decline by 12 dB per octave, predict­ably, linearly, unproblematically, without resonances. All

you have to do is to boost the drive signal by 12 dB per

octave below 70 Hz and you can have flat, resonance-

free, perfectly controlled bass down to practically dc. Of
course, such a drastic boost takes a lot of amplifier pow­er, not to mention an extremely linear high-excursion
woofer, but watts are relatively cheap and a good driver
is needed with just about any design approach. All this is
a somewhat simplistic abstract of the ELF concept, but I
want all readers to grasp at once its simple and plausible
essence. (I do not guarantee, of course, that the idea had
never occurred to anyone before Long/Wickersham im­plemented it. One could even argue that Roy Allison's
Electronic Subwoofer from the late '70s bears some re­semblance to it, although Allison did not try to push the
resonance as high up as possible.)

There are other advantages that come automatically
with a higher fundamental resonance. The woofer cone
can be lighter. That means higher efficiency. The enclo­sure can be smaller. The Bag End S10E-C and S18E-C
systems aren't very much bigger overall than the carton
in which the 10" driver and, respectively, the 18" driver
came. One of the nontrivial design challenges posed by
the ELF concept is what circuitry to use before the power
amplifier input to boost the signal 12 dB per octave down
to 8 Hz (i.e., close enough to dc but still safely above it
for real-world amplifiers and woofers). There will be an

32

explanation in the next issue of the circuit details of the

unorthodox ELF solution, which is incorporated in the

$2460 ELF-1 electronic unit and which we haven't tested
yet as a separate component. The heart of the circuit is
described as a dual integrator (with the emphasis on
"dual"). This is rather different from the conventional
lowpass-filter type of bass equalizer and is claimed to
have many advantages. Very briefly, the main advantage,
in addition to inherently excellent linearity down to virtu­ally dc, is that the very low permissible low-frequency
cutoff (almost dc-ish at 8 Hz) results in significantly im­proved signal-delay characteristics in comparison with
conventional bass-response profiles. Time offset, which
in the case of low-frequency transducers is of more than

just tweako concern, thus becomes a relatively innocuous

issue with the ELF system.

The basic ELF electronics could be implemented
much more simply and cheaply than in the ELF-1, which
is an incredibly elaborate and versatile professional unit
with balanced inputs/outputs (only!) and a zillion possi­ble settings and adjustments. Its so-called concealment
circuit is a particularly sophisticated adjustable control
feature which is actually essential with the 8 Hz cutoff to
protect the typical amplifier from overload. The ELF-1
also provides electronic crossover and level setting facili­ties for the main amplifier/loudspeaker system. Details
about all that will be part of the more technical followup
review in the next issue; meanwhile I should note that

Modular Sound Systems is definitely interested in the
home audio market and new, more audiophile-oriented

versions of the Bag End ELF products can be expected.

The earliest sign of that is a substitute front panel for the
ELF-1 with a much more High End kind of look than the
very utilitarian decal-labeled original. As for the woofer
enclosures, I absolutely love the indestructible carpet­covered road-show versions they sent me, complete with
corner protectors and handles, but for the home market
they will have to make the entire line available in various
veneers, not just the one 18" model in oak veneer as at
present.

A single ELF-1 control unit can drive any number
of amplifier/subwoofer hookups, as long as it sees at least
600 ohms. One of the beauties of the system is that the
different subwoofer models, regardless of size, all have
essentially the same response characteristics, differing
only in efficiency and maximum SPL capability. Thus
multiple arrays can be used to increase subbass output
and reduce distortion ad libitum. I am not quite finished
yet with my measurements of the distortion characteris­tics of a single S10E-C and single S18E-C; the complexi­ty of the electronic signal path makes the process a bit
more than routine; I can confidently predict, however,
that the final results will be somewhere between very
good and excellent.

To get back to the subjectively perceived sound of
the Bag End ELF system, I find it more solid, clearer,

THE AUDIO CRITIC

pdf 29

more detailed, less nimbly, more revealing than that of any
other subwoofer known to me. It would seem that getting

rid of a very low fundamental resonance as the crutch for

low-frequency extension also gets rid of a certain vague-

ness of definition in that range. That of course is specula­tion without proof. If you want objectively verifiable and
repeatable observations, I have only gone as far as to
measure the nearfield frequency response. That goes down
flat well below the lowest audible tones, as I said, with
worst-case deviations of less than 1 dB. And that goes for
both models. So much for now—to be continued.

(Just before press time, a pair of Velodyne Servo

F-1500R's arrived. That will be some shootout!)

...lastly, an acoustical

accessory:

MSB Acoustic Screens

MSB Technology Corporation, P.O. Box 141, Moss Beach, CA

94038. Voice: [phone]. Fax: [phone]. MSB
Screen, 6 ft. high with three 2 ft. wide folding sections, $400.00
($800.00 the pair). Tested samples on loan from manufacturer.

I got into trouble with MSB Technology by being
disrespectful to their "EMA Isolation Plate" in an editori­al reply to a subscriber's letter in Issue No. 18. They
wrote a long letter, published and answered (with "case­closed" finality, in my opinion) in Issue No. 19; they

complained to us verbally as well; however, as of the

time this is being written, I still have no test data from
them proving that they are right and I am wrong, only
promises that such proof is coming. I shall now redeem
myself (in their eyes, I hope—in mine I need no secular
redemption) by warmly recommending their acoustic
screen product.

What we have here is a handsome and useful piece
of furniture, a three-section folding screen covered with
the same acoustical fabric on both sides. Two short,
rounded maple legs support the frame of each section and
keep it off the floor. The hinges work in either direction,
and the screen is self-supporting in just about any folded
or unfolded position except a straight line. Under the
acoustical cloth are three separate layers of special damp­ing material. I do not have an established test protocol for

quantifying sound absorption—one of these days, may­be—but I can tell you that the device works.

Two of these screens can make a serious difference

in stereo loudspeaker deployment. Putting a screen be­hind each speaker system in a shallow U shape will block
just about all backwall and nearby sidewall reflections,
yielding a strictly forward-firing launch. That can elimi­nate a lot of soundstaging confusion in certain setups.
You can also cover reflective surfaces anywhere else in
the room for improved acoustics. The screens can be
moved easily by one person and folded when not in use.

It would be an exaggeration to report that the MSB
Acoustic Screens changed my audio life, but I will say
that after you have used them for a few weeks you don't
want to be without them. They are a bit pricey, though,
for something that couldn't cost all that much to make. •

Jitter (continued from page 22)

of the jitter spectrum and the type of
converter used. One-bit noise-shaping
converters with no discrete-time out-

put filter are an order of magnitude

more sensitive to white phase jitter
on the clock than either resistive lad­der DACs or 1-bit converters that use
switched-capacitor output filters. Con­tinuous-time filtering of 1-bit outputs
does not reduce jitter sensitivity.

The use of oversampling filters
in conjunction with standard multibit
or 1-bit DACs with switched-cap
filters reduces jitter sensitivity in pro­portion to the oversampling ratio, as­suming white phase jitter. The effect
of low-frequency correlated phase jit­ter is not significantly reduced by
oversampling.

A new, properly designed asyn­chronous sample-rate converter al­lows jitter on either the input or the
output clock to be heavily filtered,

ISSUE NO. 21 • SPRING 1994

thereby dramatically reducing errors
caused by jitter. This filtering is ac­complished by computing an internal
estimate of the sample-rate ratio over
many thousands of clock cycles, so
that individual jittered clock edges
are prevented from affecting the
computation.

The use of this new IC in out­board D/A conversion applications
allows the designer to use a fixed
crystal clock circuit for the convert­er, while accepting a jittery input
clock (most often recovered from an
incoming AES/EBU or S/PDIF serial
stream).

It is recommended that jitter
measurement of internal clock sig­nals be used by equipment designers
as an aid to achieving good signal
quality, but that reviewers should not
attempt to assess signal quality based
on jitter measurements, since the

amount of signal degradation caused
by jitter is a complex combination of
many design factors. It is better to
measure the analog output signal it­self, as this is what is ultimately re­produced. Most conventional THD +
N and spectrum analyzer tests will
adequately expose any jitter-related
problems, especially those that sub-

ject the D/A system to high-

amplitude high-frequency signals. A
noise-modulation test may be added
to look for the effects of broadband
phase jitter.

8 References

Adams, R. W. and T. Kwan.
"Theory and VLSI Implementation
of Asynchronous Sample-Rate Con­verters." 94th Convention of the
AES, Berlin, Germany (16-19 March

1993): Preprint 3570.

(continued on page 44)

33

pdf 30

Analog Electronics: More

Power Amplifiers, Preamp/Control

Units, and Mild Surprises

By David A. Rich, Ph.D.

Contributing Technical Editor

This is the continuation of our ongoing preamplifier and power

amplifier surveys. The plot thickens but does not take altogether

new and different directions.

The reader is referred to Issues No. 18 and 20 for
introductory material, basic definitions, general engineer­ing concepts, circuit illustrations, and extensive reference

documentation on the subject of preamplifiers and power

amplifiers. Our first-time explanation of the PowerCube,
with five typical examples of measurement printouts, is

also in Issue No. 20. We cannot go over this groundwork
in each issue for new readers, but both No. 18 and No. 20
are still available as back issues for subscribers.

Stereo Power Amplifier

Harman Kardon PA2400

Harmon Kardon Incorporated, a Harman International Com-

pany, 8380 Balboa Boulevard, Northridge, CA 91325. Voice:

[phone]. Fax: [phone]. PA2400 stereo power

amplifier, $1199.00. Tested sample on loan from manufacturer.

Harman Kardon is not among the names that in­stantly come to the mind of an audiophile when he thinks
of the high end, but perhaps it should be, since this com-

pany attempts to design its circuits according to the same

engineering philosophy we see in the highest of high-end

products. Harman Kardon circuits are always discrete.
The circuit topologies use minimum feedback or no feed­back at all. What principally separates a Harman Kardon

audio component from a card-carrying high-end unit is
build quality and price.

The front end of the Harman Kardon PA2400 pow-

er amplifier is much more complex than is typical for an

amplifier in this price class. A floating current source is

formed with three bipolar transistors. The two outputs of
this current source are mirrored with a two-transistor
Wilson (very high output impedance) current mirror to
the complementary differential pairs. The supply rail for
the current sources is provided by a pair of regulated sup­plies. Open-loop emitter followers set by self-biased zen­er diodes form the regulators. The differential pairs are

34

formed by source-degenerated JFETs that are cascoded

by bipolar devices. The common-mode points of the two
differential pairs are connected together through a 2.2 µF
capacitor. This capacitor allows the amount of current

flowing through the differential pair to increase if the
amplifier approaches slew-rate limiting. The bases of the
cascodes are connected to a level-shifted version of the
common-mode voltage of the differential pair. This is
called a dynamic cascode, since the emitter of the cas­code device moves with the common-mode voltage of
the differential pair. The dynamic cascode keeps the
drain-to-source voltage of the differential pair constant in
the presence of a common-mode voltage. This prevents
the common-mode input signal from coupling to the out-

put of the differential pair through the gain devices. Only

one other maker of commercial amplifiers I have seen
uses a dynamic-cascode input stage, and that is PS Au­dio. Bob Odell worked for Harman Kardon before he de­signed these PS Audio amplifiers, so it is likely that the
technique originated at Harman Kardon.

The PA2400 is configured as a noninverting ampli­fier, as are nearly all other amps. Such an amplifier has a
common-mode swing which is equal to the input signal.

Recall that the differential input signal to the amplifier is

an error signal which the amplifier tries to keep as small
as possible. If the amplifier could keep the error signal at
zero, the output would contain no distortion products,
since the output/input transfer function would be con-

trolled by feedback resistors alone. In an ideal differen­tial amplifier stage, the common-mode signal is com­pletely rejected. Now, if the differential pair transmits

some of the common-mode signal to the second gain
stage, that stage will be unable to distinguish between a

signal resulting from the differential gain of the differen­tial pair and a signal resulting from the common-mode
gain of the differential amplifier. Consequently the out­put will be distorted because the amplifier no longer at­tempts to keep the differential input signal to a minimum.

THE AUDIO CRITIC

pdf 31

Figure 1: "Triple
compound" topology
of output stage.

OUT

The separation of the differential inputs occurs as the
amplifier responds to the common-mode signal. Note that
the small-signal common-mode gain of a differential pair
usually varies across the common-mode voltage range of
the amplifier, adding to the distortion effect. The dynam­ic biasing of the gain devices and the very high output
impedance of the current sources ensure that the PA2400
input stage has excellent common-mode rejection ratios.
A complementary differential input stage of a power amp
can be formed with as few as four transistors, but the ad­vanced stage developed for the PA2400 uses a total of 17.

The collectors of the cascode devices are terminat­ed to resistors. One side of the input stage is coupled to
an emitter follower, which then drives the second gain
stage. This isolates the first gain stage's output from the
nonlinear load of the second stage. The second gain stage
is a composite CE-CC device often called a Darlington
transistor [Gray and Meyer 1984]. This dual device can
source significantly higher current than a single transis-

tor. The balanced pair of Darlington devices are connect­ed together through a single-transistor VBE multiplier.

The output stage is novel, since it is in effect a
complete unity-gain feedback amplifier embedded inside
the complete amplifier. (See Figure 1, which shows a

simplified schematic of one side of the amplifier.) Input
signals to the first stage go through a common-emitter
amplifier with a resistive load connected to the respective

power rails. This is followed by an emitter-follower

stage, which then drives the base of a common-emitter
output stage. Connecting the collector of the follower
stage to the output increases the complete subblock's cur­rent gain. The unity-gain feedback is established by con­necting the output of the amplifier to the emitters of the
first stage of the output stage. This local feedback ar­rangement linearizes the output stage, but stabilizing it is
very difficult. Indeed, this stage may be thought of as a
three-stage composite output device, and even two-stage
composite devices are known to have stability problems

[Gray and Meyer 1984]. Bongiorno called this topology a
triple-compound device [Bongiorno 1984]. In his article
he notes that it has the lowest distortion of any three-

stage output device but calls it virtually impossible to sta­bilize. Harman Kardon engineers have been able to stabi­lize it, but no circuit trick is obvious from the schematic.

Harman Kardon's rationale for the output stage is that

linearizing it with the local feedback loop allows the

amount of global feedback to be reduced. Some propo-

ISSUE NO. 21 • SPRING 1994

nents of low-feedback design would argue that an inter­nal multistage subblock operated at a high feedback rate

constitutes cheating. Biasing the output stage is simple,

since it can be seen from Figure 1 that the total output

stage looks like a single npn transistor with very high
current gain supplied by the last two stages of the circuit.
The dominant pole of the complete amplifier is estab­lished by 47 pF capacitors across the BC junction of the
first stage of the output circuit. Numerous other passive
networks are also used to improve the amplifier's stabili­ty, among them an inductor in series with the output to
improve stability into capacitive loads. The amplifier is
completely direct-coupled, with a potentiometer in the
circuit to null out any dc offsets. The pot adjusts the rela­tive bias currents that flow in the two halves of the com­plementary amplifier. Unlike a dc servo, this arrangement
can drift with time. The advantage to the highflier—oops,
I mean the high-ender—is that the amplifier is flat to dc
and no additional electronics are connected to the sum­ming junction.

The large, single, toroidal power transformer has
dual secondaries, one for each channel. The secondaries
can be reconfigured for higher supply voltages by means
of a rear-panel switch, labeled 4 /8 . The higher volt­age in the 8 . position allows for higher power output
into load impedances of 8 ohms or more. With speakers
of lower than 8-ohm impedance this 8 option should
not be used; the amplifier's heat sinks and output stage
are not adequate to deal with the dissipated losses that re­sult when the amplifier delivers high power into the low­impedance loads. This is an interesting design compro­mise, since the advantages of a higher power rail can be
offered for 8-ohm speakers, but the amplifier cost can be
kept relatively low by not having to use a larger output
stage. Separate full-wave rectifiers and filter caps are
used for each channel; 12,000 µF of capacitance filters
each supply rail.

The layout of the PA2400 looks so strange at first,
it could be the audio equivalent of a Car Talk puzzler. A
huge heat sink for the output devices is at the center of
the chassis. RCA input plugs and a small PC board for
the bridging and autostart circuits are located at the far
side of the chassis, about as far from the low-level signal
board of the power amplifer as you can get. A good part
of the chassis is empty. The answer to the puzzle is that
this chassis is also used for an integrated amplifier
(HK6950R). Good-sized heat sinks are used on the pre­drivers and also on the full-wave rectifiers. A double­sided PC board mounted to the heat sink interconnects
the output devices. Offset from this PC board is another
PC board which incorporates the low-level electronics.
This one is, disappointingly, a single-sided board.

The protection circuitry of this amplifier activates
when an overcurrent or overtemperature condition oc­curs. The protection circuitry is also used during ampli­fier startup. When tripped, this circuitry shorts the ampli-

35

pdf 32

fier's input to ground and turns off the floating current
sources so that no current flows in the amplifier stages.
This is a good approach to protection against excess in­put voltage or output shorts, but it will not save your
speaker if an output transistor shorts to a supply rail.
(The protection circuitry in the Parasound amplifier—see

below—would save the speaker, since it has a relay in se­ries with the output terminals.) A unique auto-on feature
allows the amplifier to remain powered on at all times; in
this standby mode the floating current sources are turned
off so current flow from the amplifier is minimum. When
a signal is present at the amplifier's input, a circuit turns
on the floating current sources and thus activates the
amplifier.

Our laboratory measurements yielded the following
results: Into an 8-ohm resistive load via the high-voltage
secondaries, the PA2400 reaches a minimum THD + N
level of-83 dB at 170 watts with a 1 kHz input just be­fore clipping. No dynamic distortion is visible; i.e., the

THD + N curves are almost exactly the same for all fre­quencies. Elimination of dynamic distortion (TIM, SID,
etc.) is an important design goal for this amplifier and ap­pears to have been meet. Into a 4-ohm resistive load via
the low-voltage secondaries, the minimum THD + N lev­el at the onset of clipping (190 watts) is -74 dB. Again
no signs of dynamic distortion can be seen, but the over­all distortion level is relatively high; such static distortion
dominates any dynamic effects. The PowerCube system
measured a dynamic output voltage of 46.2 V (267 watts)
into 8 ohms in the high-voltage secondary mode. This

represents a dynamic headroom of 0.65 dB at 8 ohms rel-

ative to the continuous power output of this amplifier at

1% distortion (230 watts, significantly higher than at the

start of clipping, because clipping in this amplifier is rela-

tively soft). The PowerCube showed that in this mode the
maximum voltage output of the amplifier into nonreac­tive loads declined by 7% into 4 ohms, 19.5 % into 2
ohms, and 37.5% into 1 ohm. In the low-voltage secon­dary mode the PowerCube system measured a dynamic
output voltage of 36.3 V (165 watts) into 8 ohms and

34.3 V (294 watts) into 4 ohms. For the 4-ohm reading
this represents a dynamic headroom of 0.53 dB relative
to the continuous power output of the PA2400 at 1% dis­tortion (260 watts, again higher than at the onset of clip­ping). In this alternate mode the PowerCube showed that
the maximum voltage output of the amplifier into nonre­active loads declined by 5.5% into 4 ohms, 18% into 2
ohms, and 34.5 % into 1 ohm. Dynamic output voltage
into reactive loads was at all impedances and in both
modes equal to or higher than that into the resistive load,
confirming that the amplifier's protection circuitry does
not affect its performance and that the amp is very stable
into complex loads. In other words, the PA2400 drew an
excellent PowerCube.

So what we have here, on balance, is a power am-

plifier priced $99 higher than the Rotel RB-990BX re-

36

viewed in the last issue but less powerful. Build quality
in the two amplifiers is similar. The Harman Kardon
PA2400 does have better protection circuitry and has the
autostart feature. What it also gives you is classic high­end design, with a complex circuit topology that uses low
feedback and has no coupling caps. Dynamic distortion is

nonmeasurable but static distortion is up a little in com­parison with the Rotel. If you believe that feedback is the
causative agent of some kind of mysterious sonic pathol­ogy (I do not), then purchase the PA2400; it will be as
free from the mystery disease as amplifiers blessed by the
high-end crowd at more than twice the price. But if I had
to make the choice, I would pick the more powerful
Rotel.

Full-Function Preamplifier

Harman Kardon AP2500

Harman Kardon Incorporated, a Harman International Com-

pany, 8380 Balboa Boulevard, Northridge, CA 91325. Voice:

[phone]. Fax: [phone]. AP2500 stereo pream-

plifier, $599.00. Tested sample on loan from manufacturer.

Like the PA2400 power amp reviewed above, this
Harman Kardon preamp is conceptually a true high-end
design, even though it costs an order of magnitude less
than some audiophile preamps whose electronic design is
no more complex nor more faithful to the audiophile doc-

trine of discrete circuits and low feedback. How does
Harman Kardon do it? Build quality is similar to that of
other consumer products, for one thing, instead of Ml
tanks, and profit margins are set to be consistent with
those of other large consumer-electronics companies, not
fashion boutiques. The AP2500 is the successor to the
Citation 21, reviewed in Issue No. 18.

This is a full-function preamp incorporating a pho-

no preamp with optional gain for a moving-coil cartridge.
The unit has six high-level inputs. Two tape monitor out­puts are available. The AP2500 does not have tape moni­tor buffers, so the tape monitor outputs are connected di­rectly to the selected input. You should always turn on all
tape recorders even if they are not in use, since they may
present a very nonlinear load to the input if they are left
off. A separate three-position switch connects the tape
monitor outputs to the line outputs. Copying from Tape 1
to Tape 2 is possible but not the other way around. No
record function selector is included, so you must listen to
the same program you are taping. Putting a tape recorder
into a self-connected loop is not possible with the
AP2500; potential system-destroying oscillations are thus
prevented.

As there are a lot of circuits to go through, let us
start at the moving-coil stage, which is a minimal com­plementary differential amplifier with six transistors. No
cascodes, active current sources, or emitter followers
here. Because of its high output impedance, this stage is

THE AUDIO CRITIC

pdf 33

more accurately described as a transconductance (Gm)
amplifier. The MC amp is supplied by separate open-loop
emitter follower regulators. The zener diodes are biased
by JFETs configured as constant-current sources to en­sure the good power-supply regulation required for high­gain amplifiers. Other regulators in the amplifier only
have the diode biased by a resistor connected to the un­regulated supply. The MC stage is selected or bypassed
by a standard pushbutton switch, which may not be the
most reliable approach to handling low-level signals. The
RIAA amplifier is another six-transistor Gm stage. As
was the case with the Citation 21, the RIAA feedback
network loads the output stage of the amplifier and is
thus compensating it for stability by shaping the ampli­fier's open-loop gain. This approach keeps the return
feedback rate constant—well, at least at higher frequen­cies; at lower frequencies the amplifier's internal output
impedance starts to dominate and its open-loop gain be­comes a constant. Since the RIAA curve rises until 50
Hz, the return loop gain starts to decline at low frequen­cies. Problems caused by this decline will be seen in the
measurements discussed below. A more complex gain
element than the minimal six-transistor design is needed

to allow this circuit to work properly at low frequencies.
A passive RC network follows to cancel the added trans­mission that occurs when an noninverting amplifier is
used for the RIAA equalizer (see Issue No. 18). Follow­ing next is a composite two-stage bipolar amplifier
configured as a unity-gain buffer with a pullup resistor.
All three coupling capacitors are used in this stage. A bi­polar device shorts the output of the stage to ground on
power-up. The stage is powered by an open-loop regula­tor using a similar composite device.

The line-level preamp section starts out with a gain

stage which precedes the volume control. The gain of
this stage is switched by paralleling another resistor with
the feedback resistor, which is connected from the minus
input to ground. The gain switching is accomplished with
a JFET active switch. The front panel switch thus does
not switch the analog gain directly but only sets the dc
value to the JFET switch. No additional noise or cross­talk components occur with dc switching. Placing the
gain stage ahead of the volume control also improves the
amplifier's signal-to-noise performance, since at reduced

volume setting the signal level and the self-noise of the
amplifier are reduced. If the amplifier followed after the
volume control, its output noise would be a constant, in­dependent of the volume control setting. The problem
with the approach used in the AP2500 is that the signal
swing at the input to the volume control can be very
large, because the gain now precedes the volume control.
This can result in increased distortion. Being able to se­lect the gain of this stage (2.5 dB or 13.5 dB) helps mini-

mize this problem, as do the ±28 V regulated supply rails
delivered by another pair of open-loop emitter-follower

regulators. The active stage for this amplifier is a truly

ISSUE NO. 21 • SPRING 1994

minimal four-transistor single-ended Gm stage. If it were
a true voltage amplifier, the resistance of the volume con-

trol could be reduced to improve crosstalk and increase
bandwidth (as was done in the Boulder "Ultimate"
preamp), but the Gm stage can only drive a 30 k pot.
Other than the Boulder, the only preamp we have exam­ined with an active predriver for the volume control is
the Krell KRC-2, but the KRC-2's stage is unity-gain.
The coupling capacitors C1 and C3 are used in this stage,
but not C2. (See Issue No. 18, p. 18.)

The volume control is a high-quality sealed Alps
unit. The unit has no balance control. Following the vol­ume control is another four-transistor Gm cell set for a
gain of 2 dB. This is followed by another two-transistor,
open-loop, resistively biased buffer, which is connected
to the preamp's outputs. In the Citation 21 review, I com­plained that the Gm cell (which in that model was a much
more complex, fully balanced circuit) was driving the
load directly. The addition of the follower corrects this,
but its single-ended, resistive biasing will make it dif­ficult for this stage to drive low-impedance loads. Now,
Harman Kardon does know how to produce a good dis­crete amplifier that can drive a low-impedance load;
indeed, they have one in this preamp: the headphone
amplifier, which is an eight-transistor amplifier with a
push-pull Darlington output stage inside the feedback
loop. Why a similar circuit is not used for the line-output
amplifier is beyond me. The output is muted on power-up
by a bipolar pulldown. This is a feature that all preamps
should have but many four-figure models do not. Again,
the coupling capacitors C1 and C3 are used in the output
stage, but not C2. The headphone amplifier and line driv­er are powered by another set of open-loop regulators
with two-transistor composite drivers. That brings the to­tal number of regulators in this preamp to eight. Interest­ingly, the Citation 21, which also had eight regulators,
used a combination of four-transistor discrete closed­loop regulators and open-loop subregulators.

In the low-gain setting the line stage reaches a min­imum distortion of -90 dB at 3 V rms, rising to -78 dB at
8 V rms, where clipping starts. No dynamic distortion is
visible in these curves; the distortion figures are basically
frequency-independent. In the high-gain setting low­frequency distortion was similar, but the 20 kHz distor­tion curve started to deviate slightly from the low-

frequency curves at 2 V rms. Adding a 600-ohm load

shows that the output buffer has trouble, as expected,

driving a low impedance. Soft clipping starts at 1 V rms

with a distortion figure of -85 dB, and at 2 V rms output

the distortion rises to -40 dB. Again, no dynamic distor­tion was visible in the 600-ohm test. Channel separation

is exceptional, staying between 107 dB and 105 dB from
20 Hz up to 1 kHz and diminishing to 80 dB at 20 kHz.
The lack of a balance control helps in achieving this re­sult, but exceptionally careful layout of the single-sided
PC board is also required to obtain such numbers. Only 2

37

pdf 34

to 4 µV of 60 Hz hum and less than 2 µV of 180 Hz hum
were found at the line outputs (low-gain setting at full

gain), confirming that all the regulators are doing their

job. [The headphone amplifier was not measured, a truly

regrettable omission for which I'll have to take the
rap.—Ed.]

The phono stage does not perform as well. The
RIAA equalization error curve in the worse channel oc­cupied a strip almost 0.5 dB wide, and channel matching
was off by as much as 0.2 dB at some frequencies. A 0.3
dB rise in the response from 100 Hz down to 20 Hz is
probably due to inadequate loop gain at low frequencies,
a result that we expected from the circuit analysis. The
other systematic variations in the RIAA response are

probably due to nonoptimal selection of the RIAA pas-

sive components. Distortion tests of the phono stage
showed significantly higher distortion at 20 Hz than at 20

kHz, confirming the inadequacy of the loop gain at low
frequencies. Measured at the tape monitor output, the 20
Hz distortion reaches a minimum of -68 dB at 1 V rms,
rising to -45 dB at the 11 V clipping point. The 20 kHz
distortion curve is at a minimum of -82 dB at 5 V rms
and rises only slightly to -74 dB at 10 V clipping. These
are the figures for the moving-magnet stage; the moving­coil stage appeared to introduce no significant additional
distortion. The worst-case input-referred noise of the
MM stage was relatively high at 1.5 µV; the MC stage
had an input-referred noise of 0.38 µV.

Build quality is similar to that of other up-market

Japanese-made components. The PC board is single­sided. The PC-board-mounted RCA jacks are gold-plated
only on the ground side. The selector switch is mounted
near the rear of the unit. It is a linear switch, with a
smallish rotational-to-linear converter mounted in front.
The shielded transformer is relatively large for a preamp.
Regulator pass transistors are on good-sized heat sinks.

The phono performance of the AP2500 is sig-

nificantly worse than that of the 8% lower-priced Rotel
RC-980BX, but the Harman Kardon marginally outper­forms the Rotel at line level. The line-level channel sep­aration of the AP2500 is truly exceptional. Build quality
of the two units is similar. The AP2500 is an all-discrete
low-feedback design, whereas the Rotel uses op-amps. A
difficult choice, but I would save the $49 and go with the
Rotel. If a phono stage is not a requirement, then the Ac­urus L10 becomes the clear choice (among the preamps
we have tested so far), since it is much better built than
the other two units and is priced the same as the Harman
Kardon. More expensive preamps by B&K, Bryston, and
Sumo should also be considered (see Issue No. 18).

Line-Level Remote Preamplifier

Krell KRC-2

Krell Industries, 35 Higgins Drive, Milford, CT 06460. Voice:
[phone]. Fax: [phone]. KRC-2 remote-

38

controlled line-level preamplifier, $3700.00. Tested sample on
loan from manufacturer.

I have invoked the name of Krell in these pages
many times, but this is the first time since Issue No. 10
that we have had a new unit made available to us for test­ing. At $3700 the KRC-2 represents the middle price
point in Krell's $2700 to $6300 line of preamplifiers
(without phono). It is not a tweaky update with Wonder
Caps of a 1950s circuit design but a fully modern remote

preamp under microprocessor control. So it's an over­priced Sony? Far from it; the build quality is close to mil-

spec, and the analog circuits are all discrete with low or
no feedback. Indeed, the Krell topologies are unlike any-

thing else on the planet, as we shall see below. [On our

planet, yes, but what about the Forbidden Planet? (Inside
joke.)—Ed.] This is a line-level-only preamp with two

balanced and four unbalanced inputs, one balanced and
one unbalanced output. There is only one unbalanced
tape-monitor loop, and it can record only the source con­nected to the main outputs. Input and tape switching is by
relays controlled by a microprocessor. Selecting an un­balanced input causes the negative input of the preamp's
balanced input buffer to be grounded. No tape-monitor
buffer is included. So, if you want multiple tape loops, an
independent tape-function selector, and a tape-monitor
buffer, you will just have to save $3001 and purchase the
Sumo Athena II.

As I said, the circuits look to me as though they
might have come from a Martian E.E. They look so
strange because the designers are obviously trying to op­timize them for some performance attribute that is differ­ent from the standard goals of low static and dynamic
distortion. Note that they are not just trying to design a
high-performance preamplifier in the traditional sense
with low or no feedback. To do that, you use fully com-

plementary circuits, dynamic cascodes, nontraditional

methods of canceling distortion such as feedforward and
the Hawksford circuit, etc. An example of this approach

can be seen in the Tandberg TCA-3018A preamp.

Okay, so what's inside the KRC-2? The first sub­block is a buffer stage which uses no feedback. This de-

couples the input signal from the load presented by the
programmable volume control. First, the input signal sees

two source-degenerated JFETs with resistive loads (one

JFET for each polarity of the signal). This stage provides

a 10 dB loss (!!??) and rereferences the signal to the posi­tive supply rail. Following that is an emitter-degenerated
bipolar stage biased with a single transistor current
source. This stage is loaded with a resistor, with a resul­tant gain of 7 dB for the stage. The dc bias on the bases
of both transistors is several volts below the supply rails,
limiting the dynamic range of the stage. In series be­tween the gain element and current source is a stacked set
of diodes, which are used to bias a set of complementary
emitter followers. The dc offset of this stage is canceled

THE AUDIO CRITIC

pdf 35

by a dc servo. Identical circuits are used for the plus and
minus signals. No circuit is included to convert a single­ended signal to balanced.

The volume and balance controls are formed by a
resistor ladder and CMOS switches. The schematic of
this stage was not submitted to us, but with proper design
the CMOS switches can be arranged so that no static cur-

rent flows through them when terminated into a buffer

stage. But even under these conditions some distortion
can occur at high frequencies. This distortion is the result
of a voltage divider formed by the nonlinear channel re­sistance of the MOSFET and the nonlinear reverse­biased diode that is the drain of the MOSFET. The catch
is that if the width of the MOSFET is increased to lower
the channel resistance, the parasitic capacitance grows

proportionately. Using a device with a lower channel

length will reduce the on resistance without increasing
the parasitic capacitance, but the magnitude of the volt­age that can be passed through the device decreases be­cause small channel devices have lower breakdown volt­ages. The on resistance of the channel of the MOSFET
varies with the gate-to-source voltage and hence the sig­nal swing. For the case of an NMOS device, the channel
resistance increases with increasing signal voltage. Fortu­nately a PMOS device does just the opposite. By using
both types of CMOS devices in parallel, the variation in

channel resistance is minimized, especially when the in­put signal is kept well below the gate voltages on the
cmos switches. The disadvantage of CMOS is that if the

signal ever goes above the chip's supply voltage, the chip

can go into destructive latchup. It is therefore important
to clamp the input signal to the CMOS gates by means of

Schottky diodes. Without a schematic I cannot tell what
Krell does to prevent latchup.

The output of the volume-control circuit goes into
a unity-gain follower with no feedback. The follower is
required because the CMOS switches cannot be resistive­ly loaded, as explained above. You would logically ex-

pect this buffer to be similar in topology to the buffer on
the input, since they perform similar functions, but you

would be wrong. This buffer's first stage is a JFET

source follower biased by a single-transistor bipolar cur­rent source. Stacked diodes are in series between the fol­lower and the current source; the stage then drives a com­plementary emitter follower. A separate follower is used
for each of the balanced signal lines. The output of this
buffer stage has a significant dc offset, which is rejected

at the input of the next stage, since dc offset is a com-

mon-mode signal to this stage. The output of the buffer is
routed into a balanced amplifier using a classic all­bipolar complementary op-amp. The differential pairs are
biased by current sources, and cascaded emitter followers

are used for the output stage. A dc servo removes the dc
offset at the output. One op-amp is used as a differential­to-single-ended converter (see the Parasound power­amplifier review for more on this circuit) for the plus out-

ISSUE NO. 21 • SPRING 1994

put, and a second op-amp is used in an identical circuit

except for reversed inputs for the minus output. Why a
true balanced op-amp (two inputs and two outputs) was

not used here instead of the two single-ended op-amps is
unclear to me, but that is not the strangest thing about
this output stage. The strangest thing is the way its gain
is set.

Consider the feedback equation G = A/(l + AB),
where G is the closed-loop gain, A is the open-loop gain,
and B is the fraction of the output that is subtracted from
the input. The Krell's output amplifier has a low-gain
mode and a high-gain mode, switched by a relay. The
high-gain mode doubles the G. In the low-gain mode the
open-loop gain A is approximately 6 (the gain stages are
highly degenerated and heavily loaded), but the external
feedback resistors that determine B are set for a closed­loop gain G of 10. Clearly you are not going to get any
more gain out of the amplifier than its open-loop gain A,
although the presence of the feedback loop does have a
slight effect on the closed-loop gain G. Now, you are not
going to believe what that gain-switching relay changes.
No, it doesn't change B; it changes the open-loop gain A,
which no other designer would touch in this application.
It does this by reducing the emitter degeneration of the
first stage, so that the open-loop gain is now increased to
approximately 25. So what we have here, friends, is an
output stage that has just a little feedback in the high­gain mode and virtually no feedback in the low-gain
mode! (I told you the design came from another planet.)
Finally, the output stage is connected through a muting
relay to the output jack. This prevents turn-on and turnoff

pulses from getting to the output. This important feature

often does not show up in megabuck preamps. The mut­ing function is automatically engaged on power-up; you
have to disengage it before you can select an input.

The thick, high-quality sheet metal of the KRC-2's
chassis is what is expected at this price point; it is perfo-

rated with many closely spaced slits on top and on the

sides for both lightness and ventilation. Inside is a wall­to-wall 4-layer PC board stuffed with components. RCA

jacks are mounted directly to the rear chassis for added

structural integrity. Right-channel jacks are connected di­rectly to the main board. Left-channel jacks are connect­ed to an auxiliary PC board mounted above the main
board at the rear of the unit. This board contains the left­channel relay network.

The ac power line comes right onto the main board
and into the big (for a preamp) PC-board-mounted trans­former. All the circuitry shares the same power-supply
regulators. This is not a dual mono design!! Each half of
the power supply has its own bridge rectifier and 4700
µF filter cap. The voltage reference is a zener diode
biased from the unregulated rail by a resistor. This is then
filtered by an RC network. The regulator itself is formed
with an OP-27 op-amp which drives the base of a bipolar
pass transistor and receives an attenuated version of the

39

pdf 36

regulated rail on its negative input. The reference voltage
is connected to the plus input of the op-amp. The regula­tor is configured as a tracking regulator, so that the posi­tive power supply acts as the reference for the negative

supply. The TO-3 packaged pass transistor is on a big
heat sink. The supply for the microprocessor and other
circuitry used for the remote control is on a separate
transformer secondary and has 7812 and 7805 integrated
regulators. These regulators are also on good-sized heat
sinks, as are all the output devices.

The remote-control features of this preamp are no
gimmick. Once you have experienced the ability to adjust
signal levels at the listening position, you will never want
to go back to a preamp without a remote. The micropro­cessor-based circuitry for the remote control is essential­ly similar to that used in other remote-control products.
In less expensive products much of the signal switching
would be done with semiconductor devices, not the 18

relays used in this unit. A shaft encoder is used to simu-

late the action of a rotating volume control. In a high­volume product the selector pushbuttons and display de-

vices would be a custom implementation that would cost
significantly less per unit, but at very low volume the
tooling cost would make production of such units prohib­itive. In the KRC-2 these components are all on a PC
board mounted to the front panel. Even the handheld
pushbutton control unit for the KRC-2 is manufactured
by Krell in the USA. Unfortunately, without custom tool­ing the remote control is somewhat large and cumber­some. The code for the microprocessor still needs some
work. The balance control requires that the L or R push­button be pressed each time to advance it, but the volume
control advances only if the Level button is pressed con­tinuously. A pop could be heard through the speakers
when the Gain button was pressed to switch between low
and high gain. This indicated the intrusion of dc offset
when switching, a bug that should have been cleaned up

before the preamp was put into production. The gain set-

ting of the preamp would occasionally change arbitrarily

when the unit took a static discharge hit. (Sonys do not
do this.) The feel of the shaft-encoder-based volume con­trol, however, is excellent. The same cannot be said of
the power switch because this preamp hasn't got one!!
This is unacceptable, given the power dissipation of this
unit. [Come on, David, you know the rules of the high-

end game. Preamps must be powered up 24 hours a day,
365 days a year, otherwise they won't sound good. They
begin to sound really good only after several weeks, or is
it months? —Ed.]

The Krell KRC-2 showed no dynamic distortion in
our tests, but overall distortion was quite high for a
preamp, even if not nearly high enough to be audible.
With an unbalanced input and unbalanced output in the
low-gain mode, THD + N reached a minimum of -81 dB
at approximately 1 V rms output, then rose to -60 dB at 7
V rms, just before clipping. The Harman Kardon

40

AP2500, also reviewed in this issue, did significantly bet­ter in the THD + N test, even though it sells for less than
one sixth the price of the Krell. Since the AP2500 is also
a low-feedback design, Krell cannot use the old "you can
have very low numbers only if you pour on the feed­back" excuse. Distortion was very similar in both gain
modes and was unaffected by a 600-ohm load across the
output. (That is one test where the AP2500 does not do
nearly as well, since it lacks the big output stage of the
Krell.) Distortion with a balanced input and balanced out-

put was better, since even-order distortion products can-

cel in the balanced mode (provided the power-amplifier
input stage has as good a CMRR specification as our Au­dio Precision test unit). When the input was single-ended
but the output measured in the balanced mode, the distor­tion was very close to the measurements for unbalanced

in and out. That was predictable because with a single­ended input only one set of buffer stages sees a signal;
the other set is at ground, since this preamp does not
have a single-ended-to-balanced converter. Consequently
these stages operate single-ended and distortion cancella­tion does not occur. The THD + N in the low-gain mode
with a balanced input and balanced output reached a min­imum of -89 dB at 3 V rms differential (1.5 V on each
lead) and was -65 dB at 14 V rms differential, where the
stage clips. Running the test with a balanced input and
single-ended output yielded similar distortion results (the
output amplitude was of course reduced by a factor of 2).
This indicated that the principal distortion is not occur­ring in the output stage. Easily measurable hum compo­nents (7 µV at 180 Hz and 5.5 µV at 300 Hz in the low­gain mode with the level control at full gain) contaminate
the single-ended output. The lowly Harman Kardon

AP2500 has significantly smaller hum components and a

lower noise floor. Channel separation of the KRC-2 runs
from -123 dB at 20 Hz to -66 dB at 20 kHz. Again, the

Harman Kardon beats the Krell above 500 Hz. This de-

spite the Krell's four-layer boards, electronic attenuators,
and a fully balanced internal signal path. (When an adja-

cent channel couples into both the plus and minus signal
leads, it represents a common-mode signal and should be
rejected.)

Is the Krell KRC-2 a rip-off? No, it isn't, since its
price is consistent with other low-volume, retail­marketed audio products of similar build quality and
complexity. This should be the last preamp you will ever
have to buy, something you probably cannot say of the
Harman Kardon AP2500. It is interesting to point out
that just minor modifications to the KRC-2's circuits to
increase the feedback levels would reduce the distortion
significantly. But as I said above, it is clear that the engi­neers are designing to some set of parameters I do not un­derstand. One presumes that they think they are optimiz­ing the sound quality of the preamp, but we could hear
absolutely no difference in matched-level ABX testing
against other units. Krell cannot, however, make up any

THE AUDIO CRITIC

pdf 37

excuses for the significant hum components and less­than-outstanding channel separation in a $3700 preamp­lifier. At this point in time, the KRC-2 is the only remote­controlled preamp with balanced inputs/outputs and
built-for-life construction at or below this price (unless
others have appeared since this was written). If you need
those features, can overlook the occasional switching
pops, can live with the somewhat inconsistently operat­ing remote control, and have the funds—well, then go
ahead and buy it. I, from my perspective as a terrestrial
E.E. and audio consumer advocate, cannot endorse it
with any degree of enthusiasm.

Stereo Power Amplifier

Parasound HCA-2200II

Parasound Products, Inc., 950 Battery Street, San Francisco,
CA 94111. Voice: [phone] and [phone]. Fax:
[phone]. HCA-2200I1 Ultra High Current Power Am-

plifier, $1695.00. Tested sample on loan from manufacturer.

This is a big brute of an amplifier that is sold by
dealers as equivalent to a Krell or a Mark Levinson at a
fraction of the price. Before I analyze the truth of such a
statement I must digress with a story about how your
friendly audio dealer operates.

I occasionally hang out at audio stores to see how
the salespeople do their job. On one occasion a man
walked in and showed interest in the HCA-2200. The
dealer immediately went into the "this amp is just like a
Krell but costs a lot less because Krells are overpriced"
speech (he was not an authorized Krell dealer). The deal­er then hooked the unit up for a listen, and proceeded to

pour on the "sweet high end, great imaging" shtick, hop-

ing for a quick sale. "Did I mention it was designed by

John Curl, who used to work for Levinson?" he droned

on. The customer wanted to hear how the Parasound
sounded against a megabuck unit the dealer had on dis-

play. The dealer tried to discourage this, wanting to close

the sale quickly, but when it was clear the customer was
not going to produce his credit card instantly, the dealer
set up the demo (without level matching, of course). Dur-

ing the demo the dealer pointed out how similar the units

sounded, but the customer disagreed, preferring the high­er-priced unit. When it became plain that the customer

had the funds to buy this megabuck amp and had been

shopping for similarly priced units at other stores, every-

thing changed. Now the HCA-2200 had a sloppy bottom
end, a shrinkage of the soundstage width, lack of
definition on the highs...

Much of the 60 lb. weight of the HCA-220II is
from the thick sheet metal of its chassis, its large heat
sink, and its big power transformer. So from outer ap-

pearances this amp does appear competitive with higher-

priced units. But open the chassis and you find yourself
staring at a single-sided PC board, which contains most

ISSUE NO. 21 • SPRING 1994

of the amplifier's electronics. It is mounted on the rear.
Other, smaller boards, which contain power-supply com­ponents and which interconnect the output devices in the
unit, are double-sided. Wiring appears to be untidy.

Cheap push connectors without gold-plating were used in

many places to connect the wiring harness. A large num-

ber of capacitors were tack-soldered onto the backs of
other capacitors instead of being properly mounted di­rectly on the PC board. Many construction flaws can be

noted, including several components with bent leads.

This probably occurred when the PC board was quickly

squeezed into the tight space allotted for it. Next to an
underfilled soldered joint was another joint with a giant

solder blob (apparently a sloppy piece of rework). To

sum up, this unit is better built than the best Radio Shack
receiver but is not built to the standards of Krell or, for
that matter, Mondial and B&K.

Where value can be found is in the circuit design of
the HCA-2200II. The two voltage-gain stages are sup­plied by separate regulated power supplies. The huge
transformer has four separate secondaries. Two are for
the output stage, one per channel. The other two are at a
higher voltage for the voltage-gain stages of the ampli­fier. Again these two secondaries allow for dual-mono
configuration. Each secondary is followed by a full-wave
rectifier. Each unregulated output supply rail is filtered
by a 25,000 µF electrolytic capacitor plus three 0.1 µF
and six 0.01 µF capacitors. One or two small film capaci­tors are used as bypass caps to shunt the inductance of
the large electrolytic, but the nine (three plus six) small
capacitors must be used for some tweako reason that I
cannot fathom. To add insult to injury, all the film bypass
capacitors are connected before the rail fuses—but if they
are to have maximum effectiveness, they should be con­nected after the fuses. The transformer secondaries for
the voltage amplification stage must be at a higher poten­tial than the unregulated output stage, since the voltage
regulator will reduce the potential of the supply and the
regulated-supply output must at least be at the potential
of the unregulated output supply rail. Regulation on the
voltage rails is accomplished by an open-loop emitter
follower. Stacked zener diodes are connected to the base
of the follower. The zener diodes are biased by a resistor
connected to the unregulated supply. A more complex
regulator would offer better power-supply rejection and
lower output impedance, but such a circuit is not a trivial
or inexpensive design when 76 V regulated rails are re­quired.

The actual amplifier is a relatively straightforward
design. Complementary differential pairs with JFETs
form the first stage. The n-channel sources are connected
to the p-channel sources through a resistor network. Be­cause the JFETs have a negative threshold (for the n-
channel devices), this arrangement self-biases the differ­ential pairs. Unlike a constant-current biasing scheme,

such self-biasing allows the current in the differential

41

pdf 38

pair to increase when the differential pair is driven with a

large differential current. This improves large-signal dy­namic performance. The resistor network also allows for
some degeneration of the input stage. This arrangement,
also employed in the Hafler Transnova amplifier (see Is­sue No. 20), apparently was developed by John Curl
when he was working for the original Mark Levinson
company. The differential stages are cascoded with bipo-

lar devices, which in turn drive resistive loads.

An RC network across the collectors of the cascode
devices is part of the amplifier's compensation network.
The complementary outputs of the differential stage are
then passed to a second set of bipolar differential pairs,
which is the gain element of the second stage. One of the
collectors in each of the complementary differential pairs
in the second gain stage is connected to ground. The oth­er pair of collectors is connected together through a com­plementary two-transistor VBE multiplier. The use of a
differential pair in the second gain stage improves the
common-mode rejection ratio (CMRR) of the amplifier.
A capacitor from one end of the VBE multiplier back to
the negative input of the op-amp sets the dominant pole
of the amplifier. The output stage of the amplifier is a set
of six paralleled bipolar output devices. Driving six bipo­lar devices on a 70 V supply rail is a difficult task for the

predriver, since the base current requirements for the out­put stage can become very high when driving low-

impedance loads. In the HCA-2200II a single MOSFET

predriver in a source-follower configura-tion drives the
output stage. As discussed in Issue No. 20, common­emitter stages exhibit less distortion than a MOSFET
common-source stage when they are required to drive a
significant load. An additional problem caused by mixing
MOSFETs and bipolars occurs in biasing the bipolar out­put stage. Normally the temperature coefficient of the

VBE multiplier (which is mounted on the heat sink)

matches that of the output stage, and thus the quiescent
current of the output stage is independent of the output-

stage temperature. When the MOSFET is used as a pre­driver, its temperature coefficient, which is different from
that of the bipolar devices, will cause some variation in

the quiescent current of the output stage. Through careful
choice of bias currents, this effect can be minimized—

Figure 2:

Op-amp used as balanced-to-single-ended converter.

42

but get it wrong and the amplifier can go into thermal
runaway at the drop of a hat. Since the Parasound did not
blow up in our faces, I must conclude that they have the
problem under control. A total of 28 transistors are in the
active signal path of each gain stage. No capacitors are in
the signal path. A dc servo removes dc offset from the

output terminals.

A separate six-transistor circuit is used for the pro­tection of the amplifier. The amplifier's dc offset, output­stage current, and heatsink temperature are all monitored.
If any one of these exceeds preset levels, the protection
circuit opens a relay in series with the speaker terminals.
This forms a very effective protective circuit which does
not affect performance during normal operation. Tweaks

will roll their eyes into the back of their head at the
thought of relays in series with the speaker terminals,
but Parasound has wisely chosen to ignore the cultists
and produce an amplifier which will not be accumulating
frequent-flier miles between the owner and the factory
for repairs required after output terminal shorts. In addi­tion, if the amplifier should fail internally, the relay will
protect your speakers.

In Robert Harley's negative review of the original

HCA-2200 ("...slight grain overlaying midrange tex-

tures.. . The treble.. .a bit tizzy, with a dry forwardness...

soundstaging...flat and congested...," etc.—Stereophile,

April 1992), he complained about the presence of an
AD712 op-amp as the balanced-to-single-ended convert­er. Perhaps the op-amp, lacking the blessing of the High
End, predisposed him to write negatively about the amp­lifier's "sound." Such biases are unavoidable without
blind testing. Given the outrageous fact that a bad review
from Harley, no matter how unscientific, has a depressive
effect on sales, Parasound apparently felt obligated to re­vise the amplifier and remove the op-amp. Unfortunately
they replaced it with the simplest possible discrete circuit

in the II revision, a two-transistor open-loop JFET buffer.

* * *

Before we go any further, we need to review how

an op-amp can be used as a balanced-to-single-ended
converter. This is as good a time as any to discuss how
this circuit works. The circuit is shown in Figure 2. Its
operation is easy to understand. The voltage at the sum­ming junction of the op-amp is

From the virtual ground property of the summing

junction of the op-amp, the voltage Vx is also on the right

side of resistor R3. The current flow in R3, which is also

the current flow in R4, is then calculated as

The voltage at the output of the op-amp is then cal-

THE AUDIO CRITIC

pdf 39

with an op-amp, in front of one of the channels. The op­amps used in the original HCA-2200 could also have
done double duty for this function, in a simple re­configured circuit, but apparently this was not done. The
bridging circuit used instead has the output of the first
amplifier routed not only to the output terminal but also
to the second amplifier stage, which is reconfigured by
the switch that selects the bridged mode to be an invert­ing unity-gain amplifier. Now, a unity-gain amplifier is

harder to compensate than an amplifier with a gain of 20

dB, since the feedback factor is higher. The dominant
pole of the amplifier must be moved to a lower frequen­cy. The problem occurs when the amplifier is run in nor­mal mode and the output stage is set for 20 dB gain. In

such a case the amplifier is greatly overcompensated and
its open-loop bandwidth is significantly narrower. The re­duced open-loop bandwidth gives rise to significant dy­namic distortion, as documented below.

Into an 8-ohm load driven from the unbalanced in­put, the HCA-2200II reaches a minimum THD + N level
of -85 dB at 250 watts with a 1 kHz signal. Into 4 ohms
with a 1 kHz input the minimum THD + N level at the
onset of clipping (400 watts) is -80 dB. Above 1 watt the
20 kHz distortion curve flattens and then rises to -64 dB
at clipping into 8 ohms. Into a 4-ohm load the 20 kHz
distortion curve flattens out at half a watt at a level of

-75 dB and then rises to -59 dB at clipping. These sub­standard dynamic distortion results are probably due to
overcompensation of the amplifier for the bridged mode.
Another sign that the amplifier is overcompensated is
that, even though it has no inductor in series with the out­put, the damping factor starts to decline at 1 kHz. The
damping factor is proportional to the amplifier's closed­loop gain. Interestingly, the dynamic distortion of the
amplifier starts to increase above the static distortion at

the same point, 1kHz. A secondary possibility might be

that the MOSFET predriver is the source of the dynamic

distortion.

Switching into the balanced input mode, the distor-

tion figures degrade significantly as a result of the poor
performance of the input buffer. THD + N with an 8-ohm

load is the same regardless of frequency, reaching a mini-

mum of -74 dB at 5 watts and then rising to -59 dB at

clipping. The low-frequency distortion at clipping is thus

degraded by the buffer by more than an order of magni-

tude.

The PowerCube measurements turned out to be an-

other story. The PowerCube system measured a dynamic

output voltage of 53.8 V (362 watts) into 8 ohms. This

represents a dynamic headroom of 0.8 dB. The Power-

Cube showed that the maximum voltage output of the
amplifier declined by only 11.5% into 2 ohms with non-

reactive loads and by only 26% into 1 ohm. The dynamic

power into a 1-ohm resistive load measured 1595 watts.
The PowerCube measurements of voltage into reactive
loads were in all cases equal to or higher than those into

43

culated with R1/R2 = R3/R4 for simplicity.

The last equation is just what we want: the differ-

ence of the input signals is amplified and converted to a

single-ended output. Now we can build this circuit using
an op-amp and place it ahead of the single-ended power
amp. This is what was done in the original Parasound
HCA-2200. Indeed, Parasound used an even better cir­cuit, based on a two-op-amp balanced-to-single-ended
converter, which had improved distortion performance.

An alternative is to use the power amplifier itself as
the op-amp element. A problem with this second ap­proach is that the input impedance of the negative termi­nal can be very low. From the circuit above it is easy to
see that the input impedance of the positive terminal is
R1+ R2, which will be a high value, but the impedance
of the negative input terminal is much less. This impe­dance is calculated as

For a high-gain amplifier, this can be approximated
as R3/2. Now, if we want the input impedance of the
amplifier to be 50 k and the amplifier's gain to be 10,
then R4 is going to have to be 1 M , which is not a prac­tical value. The solution is to use a smaller value of R3 to
bring R4 down to a practical value. The low input impe-

dance is dealt with by adding a voltage buffer in front of
the negative input of the amplifier.

* * *

As I said, the two-transistor open-loop JFET buffer
regrettably used by Parasound in the II revision is the
simplest possible discrete circuit for this application. In
its simplest form a diode-connected JFET forms the cur­rent source that biases the single-transistor follower

stage. For matched devices the voltage drop across the
biasing device matches the voltage drop across the
source-follower device. Since the gate-to-source voltage
of the current-source transistor is zero, the drop across
the source-follower device is also zero, and the input and
output voltages of the buffer are matched. In the Para­sound implementation of the two-transistor buffer, a re­sistor is placed in series with each source of each JFET to
reduce static current flow and improve matching. This

simple circuit cannot follow the input signal accurately
enough to transfer it without distortion, as we will see be­low, because an open-loop JFET source follower in not
linear enough.

The bridged mode of the amplifier is also poorly
implemented, in both the original version and the II revi­sion. For the bridged mode to work, the input of one
channel of the amplifier must be inverted with respect to
the input of the other channel. Again, this is usually ac­complished by placing an inverting amplifier, formed

ISSUE NO. 21 • SPRING 1994

pdf 40

the resistive loads. This amp thus proved to be extraordi­narily stable into reactive loads. Peak current output was

198 amperes. The PowerCube therefore confirms that
this is the most powerful amp we have tested so far. As
for channel separation, we measured 90 dB or better be­low 1 kHz, gradually moving to 70 dB at 20 kHz.

To sum up, the Parasound HCA-2200II proved to
be a very powerful amplifier, with a power transformer,
output stage, heat sinks, and protection circuitry that are
not underdesigned for the task. The presence of regulated
supply rails and a dc servo indicates that the engineers
have not skimped on the design of the amplifier to meet a

price point. On the downside, distortion performance was

substandard as a result of some kludgy circuit design, al­though the distortion still is low enough to be inaudible.
Construction quality has to be judged acceptable given

the price and power output, but no more than that. Since
this unit is slightly more powerful than the Rotel RB­990BX and significantly better protected, it can be rec­ommended for special situations, but in most cases I
would pocket the $595 price difference and go with the
Rotel.

Now, since Parasound seems responsive to re­views, I look forward to a III revision of the HCA-2200
that will fix all my complaints.

Line-Level Preamplifier

Rotel RHA-10

(quick preview by the Editor)

Rotel of America, P.O. Box 8, North Reading, MA 01864-0008.

Voice: [phone]. Fax: [phone]. RHA-10 Stereo

Active Controller, $1800.00. Tested sample on loan from manu-

facturer.

Although introduced only recently to the U.S. mar­ket, this preamplifier was actually designed earlier than
the $550.00 Rotel RC-980BX reviewed favorably in Is­sue No. 19. I am told that the entire Rotel high-end line,
with model designations that take the form RH(X)-10,
was used to establish parameters and lay groundwork for

their more reasonably priced bread-and-butter line, which
we have found to yield such excellent performance per
dollar.

At more than three times the price and sans phono
stage, the RHA-10 had better offer something the cheap­er Rotel preamp doesn't. So far I can report that it looks a
lot snazzier, appears to be built to a higher quality stan­dard with better parts, and measures even lower in line­stage distortion—very low indeed. Whether it's worth

the $1250 price difference will be determined after a
more detailed analysis to be published in the next issue.

—Ed.

Stereo Power Amplifier

Rotel RHB-10

(quick preview by the Editor)

Rotel of America, P.O. Box 8, North Reading, MA 01864-0008.

Voice: [phone]. Fax: [phone]. RHB-10 stereo

power amplifier, $2700.00. Tested sample on loan from manu-

facturer.

The same applies here as above: this is the high-

priced predecessor and role model, so to speak, of the

$1100.00 Rotel RB-990BX, but a later arrival in the U.S.
Since the RB-990BX is a our favorite big amplifier on a
value-per-dollar basis, the question is again what you get
for 2½ times the price. Better protection, for one thing;

more glamorous looks, definitely; better sheet metal in
the chassis; better parts here and there (not everywhere);
but what about performance? I am not ready to tell you
because we ran into some minor technical problems,
probably unrelated to the design of the amplifier, that
make me question our measurements. I can report the ap­proximate clipping points: 210 watts into 8 ohms, 370
watts into 4 ohms. Our small problems will undoubtedly
be solved before long, and a detailed analysis of the unit
will published in the next issue.

I can tell you right now, however, that in the same

power category the Bryston 4B NRB, for $505 less, will

be a hard act for the Rotel RHB-10 to follow.

—Ed.

Jitter (continued from page 33)

"Interpolation and Decimation of

Digital Signals." Proceedings of the

IEEE 69 (March 1981): 300-31.

Koch, A., R. Lagadec, and D.
Pelloni. "Method and Apparatus for
Converting an Input Scanning Se­quence into an Output Scanning Se-

quence." U.S. Patent 4,825,398.

Lagadec, R. "Scanning Frequen-

44

cy Synchronization Method and Ap­paratus." U.S. Patent 4,780,892.

Lagadec, R. and H. Kunz. "Pro­cess and Apparatus for Translating
the Sampling Rate of a Sampling Se­quence." U.S. Patent 4,748,578.

McNally, G. W. "Variable-Speed
Replay of Digital Audio with Con­stant Output Sampling Rate." U.S.

Patent 4,716,472.

McNally, G. W., R. Lagadec,

and D. P. Pelloni. "Method and Ap-

paratus for Measuring the Time Dif-

ference Between Two Sampling
Times." U.S. Patent 4,564,918.

Robbins, W. P. Phase Noise in

Signal Sources. London: Peter Pereg-

rinus Ltd., 1982. •

THE AUDIO CRITIC

pdf 41

Digital Electronics:

More CD Players, D/A Processors,

Transports, and a First Look at the

Sony MiniDisc System

By Peter Aczel

Editor and Publisher

&

David A. Rich, Ph.D.

Contributing Technical Editor

Digital technology marches on and manifests some fine examples
of progress, but not every step is forward.

This is one of the busiest product categories in au­dio today, and we have tested more equipment since the
last issue than we can squeeze into this one. There will
be a spillover in the next issue, plus of course new items

just being tested. Among the reviews that should have

appeared below but are being postponed until next time
are those of the Cobalt 307 and Deltec Precision Audio

PDM 2 D/A converters (the latter with the Tl transport),

and of the Denon DCD-2700 and Marantz CD-63 CD

players.

Outboard D/A Converter

EAD (Enlightened Audio

Designs) DSP-9000 Pro

(Reviewed by Peter Aczel)

Enlightened Audio Designs Corp., 300 West Lowe, Fairfield, IA

52556. Voice: [phone]. Fax: [phone]. DSP-9000
Pro two-chassis digital processor with remote control,
$5500.00. Tested sample on loan from manufacturer.

In my review of the EAD DSP-7000 Series II and
DSP-1000 Series II in the last issue I may have left the
impression—or didn't I?—that this equipment leaves lit­tle or nothing to be desired and that further improvements
would be insignificant. Well, the EAD people obviously
don't agree with that because here they are with their
"statement" product at almost three times their previous
top price. EAD, as I have pointed out before, is a serious,
engineering-oriented company, almost entirely devoid of
the miasma of high-end gibberish—although they occa­sionally "hear" things that I and my associates don't—so
I have to take this all-out effort seriously. We aren't deal­ing with a bunch of tweaks here.

ISSUE NO. 21 • SPRING 1994

Conceptually, all three EAD outboard D/A proces­sors are the same, differing only in the elaborateness and
refinement of implementation. All three current models
use the 20-bit Burr-Brown PCM63P-K, the best multibit
DAC chip in the business (probably the best DAC chip in
the business, period). All three use EAD's proprietary
AccuLinear I-to-V converter. The DSP-9000 Pro puts a
lot of distance between itself and the other two, however,
by (1) having all the various analog and digital power
supply circuits on a second chassis; (2) providing eight
digital inputs, two of each type: professional AES/EBU
XLR, ST-type (AT&T) glass, coaxial, and Toslink; (3)
allowing all functions to be remote-controlled with great
sophistication; (4) reporting all ongoing electronic activi­ties within the processor on a front-panel display with—
count them—12 LEDs and 9 control buttons; (5) incorpo­rating all kinds of minor circuit refinements too numer­ous to list here.

Unfortunately, no circuit schematics were made
available to us (as a matter of company policy, I believe).
That's one reason why I am reviewing this unit, not Da­vid Rich ("no schematic, no review," quoth he). Worse
yet, not even a peek under the cover was possible, and
that brings me to an embarrassingly trivial but still very
real beef about this otherwise splendiferous equipment.
On each chassis, the sheet-metal cover is held in place
with 12 butter-soft, painted, Phillips-head machine
screws. The paint tends to cause some of the factory­torqued screws to seize. Then, when just a slight extra
force is exerted with the screwdriver to remove the

screw, metal shavings start to fly and the cross slot is

stripped beyond recognition in a single twist. Get me the
drill, Igor—but, of course, you can't drill into a $5500
masterpiece! I am told that this penny-pinching feature of

the DSP-9000 Pro was eliminated in later production; in

45

pdf 42

fact, I was sent a small bagful of "expensive" anodized
Allen-head machine screws, which are supposed to be
standard in currently sold units. That did not get my cov­ers off; one way or the other the problem will be solved
and I'll be able to look at the innards of the beast. "For
want of a nail...," etc.

If black-box measurements are the yardstick of
quality in digital audio and not the hardness of machine
screws, then the DSP-9000 Pro is arguably the best there

is, bar none. Even better than the other two EAD proces­sors? Very slightly, yes. I ran so many curves, in every
possible mode of operation, that it would be quite boring
and wasteful of space to discuss every one of them here.
The digital filter can be switched between 4x and 8x
oversampling; the analog output can be either single­ended or balanced; in each output mode three different
default output levels for a full-scale digital input are
available—the permutations and combinations are end­less. I don't believe in the look-Ma-I-have-an-Audio­Precision type of graph-cluttered review. (For that you go
to Stereophile, where the zillion graphs are purely cos­metic since their meaning is ignored in the evaluations.

My Ma doesn't even know I have an Audio Precision,

and I keep my graph printouts in the filing cabinet...)

In the digital domain, I found every measurement

to be textbook perfect, meaning full 16-bit resolution of a

16-bit input in all modes, no exceptions. With digital in­put levels in the 0 dB to -10 dB range, there was a very
slight amount of gain-related analog distortion, which di­minished to absolutely nothing at -20 dB. All digital-to­analog gear appears to have that small deficiency except,
mysteriously, the top-of-the-line Sony CD players. The
total absence of high-frequency distortion in the output
of the EAD at the -20 dB level also indicated that clock

jitter was of no consequence. Among the more spectacu-

lar measurement results obtained with the DSP-9000 Pro

were the 0 dB gain-linearity error all the way down to

-105 dB, stereo channel separation in the 112 dB to 104
dB range regardless of frequency, noise floor with digital
zero input in the -140 dB to -106 dB range up to 200
kHz (and below -130 dB at multiples of 60 Hz!)—I
could go on, but you get the picture: there is really noth­ing to report when everything is this good. (One hard-to­measure exception: EAD's long-standing radiated RFI
problem has not been solved; the unit must be unplugged
for clean FM and TV reception.)

It was interesting to see that 8x oversampling
yielded the flattest high-frequency response (-0.15 dB at
20 kHz) but 4x oversampling resulted in ever so slightly
better low-level linearity. The differences were minute,
in any event, but do you know of any other D/A proces­sor that allows you to make that 4x/8x change from your
listening chair by remote control? (What a totally nerdy
thing to do from your listening chair, when you think
about it!) The technologically most advanced feature of
the remote is actually the volume control (which is also

46

the feature most heavily used by those who bypass their
preamp). Various buttons permit analog volume steps of
6 dB, controlled by precision resistors, and digital vol­ume steps of 0.2 dB between those large steps. Thus the
maximum possible degradation of digital resolution is 1
bit (corresponding to 6 dB), eliminating the single disad­vantage of digital volume controls. Pretty neat trick. For

absolute-phase sticklers there is also a Phase button on
the remote. Now, these buttons here are for regular and
decaf... (just kidding, you deadly serious weenies).

Bottom line: should you buy the EAD DSP-9000
Pro? If $5500 means little or nothing to you and pride of
ownership means a lot, and if you need to switch be­tween a large number of digital sources, then by all
means go ahead, with my blessing. It's a superb piece of
equipment, able to stand up under the most exacting tech­nical scrutiny, unlike most of its high-end competition.

Does it sound different from the $999 DSP-1000 Series
II in an ABX comparison? Absolutely not, and I didn't
expect it to. Does an audiophile live by sound alone? I
don't have to answer that.

Compact Disc Player

Harman Kardon HD7725

(Reviewed by David Rich)

Harman Kardon Incorporated, a Harman International Com-

pany, 8380 Balboa Boulevard, Northridge, CA 91325. Voice:

[phone]. Fax: [phone]. HD7725 compact disc

player with remote control, $849.00. Tested sample on loan

from manufacturer.

One thing you can say about Harman Kardon is
that they do not make me-too products. The innovation in
the HD7725 CD player is their proprietary RLS ("Real­time Linear Smoothing") technology. The RLS circuit
performs linear interpolation between adjacent sample
points at the output of the DAC. Linear interpolation pro­vides more filtering than is obtainable by simply holding
the sampled signal's value for a sampling interval. The
calculation of the amount of filtering provided by linear
interpolation is relatively simple but requires a little
math. Since your Editor believes this would be enthusias­tically received by about 17 readers of The Audio Critic
[because it isn't just "a little" math—Ed.], I refer inter­ested readers to A. Papoulis's textbook Signal Analysis,
page 141.

The easily understood advantage of linear interpo­lation over sample-and-hold is that simpler, lower-order
analog reconstruction filters can be used. The hard part of
taking that route is the design of an analog circuit which
will perform the functions of linear interpolation and
filtering. In the HD7725, two DACs are used per chan­nel. The data into the second DAC is delayed in a digital
memory by one oversampling period relative to the data

THE AUDIO CRITIC

pdf 43

into the first DAC. The output of the undelayed DAC is
sent to the CD player's output. Added to this signal is the
interpolation signal, which is formed by the following in­teresting circuit. A signal is generated which represents
the voltage difference between the delayed and the unde­layed DAC just after a sampling interval. This signal

then drives a first-order RC network. When the differ­ence signal steps to a new value, the RC network starts
charging. The RC values are chosen so that the exponen-

tially rising signal at the output of the network approxi­mates a linear ramp during the time interval between
DAC output samples. Clearly, if the RC network's time
constant is not exactly right, the generated interpolation

signal can overshoot or undershoot the ideal interpolation

signal. The complexity of the whole circuit and its poten­tial to misbehave would seem to argue against its general
usage and point to other solutions, such as filtering the

signal in the analog domain or putting the entire interpo­lation circuit in the digital domain.

The CD mechanism of the HD7725 uses a number
of plastic gears and a small dc motor to articulate the la­ser pickup assembly. At this price point it is more typical
to use direct-drive linear motors, which should be in the­ory more reliable. (I must hasten to point out, however,
that I once had a problem with sticking rails on the linear
motor of my Pioneer Elite PD-71, resulting in random
skipping. This was easily fixed by cleaning and lubricat­ing the rails, so if you have such a problem don't let a
crooked repairman sell you a new laser!) The electronics
of the Harman Kardon CD transport mechanism consist
of three large-scale integrated circuits, two smaller-scale
circuits used to drive the motors, and about a hundred
discrete components. All the electronics associated with

remote sensing, keyboard decoder, fluorescent display,

and microprocessor management are in just one addition­al VLSI chip.

The power transformer of the HD7725 has four

secondaries. The analog supplies are filtered by 2200 µF

capacitors on the unregulated supply rails; the digital

supplies have larger 4700 µF capacitors. Five voltage

regulators are used in this CD player; all are discrete
open-loop devices. Zener-diode voltage references are
biased from the unregulated supply rails. The pass tran­sistors are a 2-transistor compound device. An additional
transistor forms a current limiter for the regulator. Regu­lated analog supply rails are a relatively low ±12 V, prob­ably because only 30 V rms is available on the transform­er secondary that powers the analog circuits. At this price
point ($849), I have come to expect a more robust power
supply. As for the PC board, it has a top-side ground
plane but does not have plated-through holes. The rela­tively thin sheet metal of the housing is reinforced at the
top and bottom of the unit with additional steel plates.

The digital filter chip is the NPC SM5840. No,
that isn't an advanced version of the SM5813. It is a a
cost-reduced version, with fewer than half the taps. As a

ISSUE NO. 21 • SPRING 1994

result, stopband rejection is 55 dB instead of 110 dB, and
the passband ripple is large enough to be observable (see
the measurements discussed below). This chip is not
what I would expect in an $849 CD player. The DACs
are Burr-Brown PCM61P-K multibit devices—good but
not the best. Even so, I find four of them in an $849 unit
to be quite generous. The DACs are enclosed in a shield­ed metal box; I am not sure what this accomplishes but
Krell also does it in the $3900 Studio. The I/V converters
use the internal op-amps in the PCM61P-K DACs. To
see high-feedback integrated op-amps used in a Harman
Kardon is equivalent to finding a rabbi eating a bacon
cheeseburger. I could never get a good answer from the
company on why the low-feedback, all-discrete­components approach the company expounds does not

apply to I/V converters. The fact that Krell does the same
thing in the Studio is not a defense that holds up in my
court. I thus sentence Harman Kardon to spend at least a
day evaluating the low-feedback Phototronics PA630
current-conveyor chip.

After the I/V converter, it's low feedback and dis­crete all the way. A 6-transistor complementary Gm cell,
similar to the circuit used in the phono stage of the Har­man Kardon AP2500 and in the Krell Studio (no cas­codes, only resistors to bias the differential pairs), fol­lows the I/V converter of the delayed DAC. It is
configured as an inverting amplifier. A passive RC filter
follows, and this in turn is connected to a 3-transistor
buffer stage identical to the one used in the AP2500

preamp (here Krell uses a complementary open-loop buf-

fer—the Harman Kardon approach is again surprisingly
close). Outputs are muted by two bipolar switches, some­thing that Krell would never do; at this price point I
would like to see a relay. Two sets of outputs are avail­able on the HD7725: the fixed-level output just discussed
and a variable output routed through a motorized volume

control. A separate headphone amplifier is connected to
the variable output, with an NJM4565 driving the head­phones.

As I have already said, realizing the linear interpo-

lation circuit is not trivial. Another open-loop buffer is
connected to the Gm cell's output. The buffer output is
connected to a 2-resistor summing circuit, which is also
connected to the undelayed DAC. The output of the sum­ming network is connected to the RC network. The out­put of the RC network is connected to the plus input of
the Gm cell. As the voltage on the plus input of the G

m

cell moves, its output moves, and thus the interpolation

signal is added to the CD player's output.

Frequency-response measurements showed some
minute ringing, of the order of ±0.025 dB (2.5 millibels),

due to the limited number of the taps in the digital filter;
otherwise the response was dead flat up to 20 kHz. De-

emphasis is done digitally, and no significant error was

measured. Channel separation was greater than 100 dB

below 4 kHz and was still 92 dB at 16 kHz. No hum

47

pdf 44

components could be identified in the noise spectrum, a
truly rare engineering achievement. Gain linearity error
with dither at -90 dB was +0.4 dB in the better channel
and a not-so-good +1.2 dB in the worse channel. The lat­ter showed +0.2 dB error even at -70 dB; the better chan­nel was off by only +0.1 dB even at -80 dB. The spec­trum of a 997 Hz tone with dither at -90 dB looked very
clean, however, in both channels. In the time domain we
also saw more ringing on low-level sine waves and on

the CBS test CD's monotonicity track than would be ex­pected from a good multibit DAC like the PCM61P-K.
The RLS circuit is the likely culprit for these strange ef­fects in both the frequency and time domains.

THD + N for a 997 Hz signal measured at the theo­retical limit at signal levels below -80 dB, then strangely
rose to 2 dB worse than the theoretical limit at -60 dB,

staying there at all levels up to -20 dB, then rising to 8
dB worse than the theoretical limit at a 0 dB signal level.
For a -24 dB input signal, the THD + N does not rise
with frequency. For a 0 dB input signal above 2 kHz, the
THD + N does rise, going to -73 dB at 16 kHz. I do not
think the distortion is coming from the analog section,
since the AP2500 preamp uses the identical circuit and
showed no significant distortion at 2 V rms. Nor is the
distortion coming from the K-grade DACs. So that
leaves—you guessed it—the RLS circuit as the likely
cause.

Since Harman Kardon claims considerable sonic
advantages for the RLS technology, we went out of our
way to give the HD7725 the opportunity to assert its su-

periority in listening tests. Three different experienced

listeners spent a good many hours ABX-ing the
HD7725's own fixed-level analog output against the lev-

el-matched output of the $999 EAD DSP-1000 Series II
processor, which was driven from the Harman Kardon's
digital output. That way the same CD could be played

with and without RLS at exactly the same volume level.

The double-blind identification scores were completely
random, no different from wild guessing. The two sys­tems could not be distinguished, even though the EAD
measures considerably better.

We all like to see audio manufacturers innovate,
but the innovation has to improve performance. RLS
does not offer improved performance; indeed, it seems to
degrade distortion and time-domain performance, even if
inaudibly. Because it requires an additional set of DACs,
it is expensive to implement. Perhaps this accounts for
the presence of a low-end transport, relatively wimpy
power supply, and cheap digital filter in an $849 CD
player. The use of an integrated, high-feedback current­to-voltage converter is also surprising, given the compa­ny's established design philosophy.

As it stands, then, the HD7725 is not the CD player
of choice in this price range. I suggest that Harman Kar­don start over, throw out the RLS circuit, and design a
CD player with the Burr-Brown PCM63P-K (or similar

48

PCM1702P-K) and the Phototronics PA630. The PA630
is a true low-feedback current-to-voltage converter and,
owing to its very advanced high-speed process technolo­gy, should outperform the simple discrete circuits used
by Harman Kardon in the HD7725's filter and output
stage.

Outboard D/A Converter

Krell Studio

(Reviewed by David Rich)

Krell Digital Inc., 35 Higgins Drive, Milford, CT 06460. Voice:
[phone]. Fax: [phone]. Studio software-based
D/A processor, $3900.00. Tested sample on loan from manufac­turer.

Yes, believe it or not, Krell sent us a review sample
of their $3900 midline D/A converter, the Studio (the top
of the line is a $14,000 unit called the Reference 64).
They even provided schematics. The Studio has six in­puts: three coax, one Toslink, one AES/EBU, and one
AT&T optical. A tape-monitor loop has coax and Toslink
outputs. The front panel is filled with LEDs to tell you
what you have selected and what the sampling rate is.
Both balanced and unbalanced analog outputs are avail­able. The most important feature of the unit is its 16x in­terpolation (oversampling) filter, made of two Motorola

DSP56000 devices and a handful of digital logic chips.

So what's under the hood? First is a significant
number of components for routing the digital data lines,

interfacing with the front-panel switches, and displaying
the results on the LEDs. Digital data are decoded by an
encapsulated module identified in the schematic as the bi­phase decoder module. To find out more, I would have to
have destructively removed the module's cover, some­thing I am very reluctant to do with expensive equipment
on loan. Fortunately, Robert Harley has no such silly in­hibitions—see, I can say something nice about Bob—
and managed to open the module (in the Reference 64,
that is—see his review in the January 1994 issue of Ste-
reophile), finding only the Crystal CS8412 receiver chip
and a programmable array logic (PAL) for controlling
the chip and interfacing it with the DSP chips. No extra
PLLs, no proprietary decoding algorithm for data clock
recovery, no adaptive PLL loop filters, or anything else
special. Harley states that "Krell says this technique [pot­ting the decoder module] improves the processor's
sound." I think its purpose is to hide the fact that a state­of-the-art S/PDIF decoder is not used in that very expen­sive Reference 64 unit. I am assuming here that the Stu­dio uses the same potted module.

The decoder module is connected to the digital
filter. This is definitely the way to go when designing an
audio D/A processor of ultimate quality, since the digital
filter algorithm is under the control of the end product's
designer, not the chip designer. Another advantage is that

THE AUDIO CRITIC

pdf 45

the algorithm can be updated by changing the firmware
PROMs that contain the computer code which controls
the DSPs. I was given no information on the algorithm
used by Krell in the filter, so let's move on to the DAC.

The DAC is a Burr-Brown PCM63P (K?—the
grade is not given on the schematic), placed inside a
shielded box. Trim pots are included for adjusting the

MSBs of the colinear DACs. Adjusting these pots is very
difficult because they affect signals at half of full scale,
not around ground. I do not know what method Krell
uses to set them. The pedestrian 7805 and 7905 3­terminal IC regulators (what were you expecting, some
high-tech discrete regulators?) subregulate the ±15 V an­alog supply to ±5 V for the DACs. The current-to-voltage
converter is the SSM-2131 op-amp, which appears to be
identical to the PMI OP42G (the only specification dif­ferences are a slightly higher offset spec for the SSM­2131 and the addition of a typical THD spec). But wait!
Stop!! Hold the phone!!! What's a low-cost, high­feedback op-amp doing in a Krell? What happened to
discrete circuits, low feedback, and all that good high­end stuff? Now, don't get me wrong here; the OP42 is an
good op-amp for this stage, but it is by no means the low­est-noise, widest-bandwidth, lowest-distortion (see Ben
Duncan's article in the December 1993 issue of Audio
Amateur), and fastest-settling one around. It is not pro­cessed with a complementary bipolar process, so its prin­cipal attribute is relatively high speed at a low price. (See
Issue No. 15, Table 4 for a complete set of specs and
prices for this and other ICs used in CD players.) A good
price-performance ratio, however, is not exactly what I
had in mind as the principal attribute of a component in a
$3900 Krell. I wonder how many of Krell's customers re­alize that the designer could have used, for example, the
Burr-Brown OPA627, which outperforms the SSM-2131
in all respects by more than 2 to 1. The only disadvan­tage of the OPA627 is that it is three times the cost of the
SSM-2131. Note that the $14,000 Krell Reference 64
does use expensive complementary bipolar op-amps
(AD841 and AD846) for its I/V converter and deglitch
circuit.

The important point here is not the price of the op-

amp but the fact that using any op-amp breaks all the
supposed rules of "high-end" design. If it is okay to use
one op-amp in the signal path, why not make the whole
signal path out of op-amps? One would think that once a
signal has been "corrupted" by a high-feedback ampli­fier, no amount of discrete amplifiers downstream could
restore its purity. Once the signal has been negatively af­fected by passing through an on-chip diffused resistor,
why should it matter that other resistors in the signal path
are of the audiophile-approved variety or that audiophile­grade interconnect cable is used? Once the signal has

passed through the class AB (quasi-complementary in the

case of the OP42) IC output stage, why should it be im­portant to run the circuits downstream from the chip in

ISSUE NO. 21 • SPRING 1994

class A? In other words, why do I need to connect the
signal from the Studio into the line stage of the $3700
Krell KRC-2, instead of using an Adcom or Rotel?

In some places Krell does spend money on parts;
for example, the de-emphasis circuit for the I/V converter
is switched by a relay, not a cheap FET switch. After the
I/V converter comes a passive RC filter. (A first-order
filter is all you need with a 16x interpolator.) Then the
signal goes to a unity-gain inverter formed with another
SSM-2131. This stage is necessary in order to generate
the complementary signal required for the balanced out­put stage. The two signals at the input and output of the
inverter are each routed to a simple 8-transistor, discrete,
low-feedback output buffer. Now we are back in high­end country. The return-loop feedback resistor is con­nected to the second gain stage, and the complementary
emitter follower runs open loop. The dc offset is can­celed with a servo circuit. The power supplies of each
buffer are bypassed with 3300 µF capacitors. The circuit
looks remarkably similar to that of the Aragon Mark II
D2A (see Issues No. 15 and 16). The principal difference
is that the Aragon has the differential pairs biased by a
current source to improve the buffers' CMRR and thus
reduce distortion. The Krell uses only resistors to bias the
differential stages. You may also wish to recall that the
$1600 Aragon Mark II D2A did have a multiple-PLL
S/PDIF decoder and a fully discrete low-feedback I/V
converter, but it did not have balanced outputs and it did
not use a programmable filter chip. In the Krell, the sig­nals from the buffers' outputs are routed through a relay,
which is open on power-up, and finally are sent to the
output jacks.

The power supply of the Studio has separate trans­formers for the analog and digital sections. On the analog
side, separate bridge rectifiers are used for each supply
and 4700 µF of capacitance is on the unfiltered rails. Pri­mary regulation comes in the form of LM7818 and
LM7918 integrated regulators. This supply is down­regulated to ±15 V by MC1468 dual-tracking regulators
in conjunction with series pass gates driving each supply
rail. The analog supply is shared by both channels.

The mechanical build quality of the Studio is simi­lar to that of the Krell KRC-2 preamplifier. You do not
have to know anything about hi-fi to realize that the Stu­dio is an expensive unit. Look inside and you see three
separate multilayer PC boards for the power-supply, digi­tal, and analog sections. The boards are interconnected
with gas-tight mechanical connectors. The ac line comes
directly onto the power supply board. I/O jacks are di­rectly mounted with hardware to the rear panel of the
unit (not the main PC board) for added structural integri­ty. The boards are filled with near-mil-spec components,
as you would expect at this price point. What you do not
expect to find is that the two PC-board-mounted power
transformers and a filter cap overhang the edge of the
board. No standoffs secure this area of the PC board to

49

pdf 46

the chassis as would appear necessary with the overhung
components. It does not look very secure; the PC board
flexes significantly. I would complain about this if I saw
it in a Pioneer receiver. At the Studio's price point it is
appalling. Also appalling is the lack of a power switch on
a unit that consumes as much power as the Studio does.

Measurements showed that the frequency response

of the Krell Studio is down only 0.25 dB at 20 kHz, and
the square-wave response showed the typical overshoot
and wiggles that result from the truncation of Fourier
components above 20 kHz. This indicates that the digital
filter is designed to approximate an ideal brick-wall filter.
Krell does not play games with the frequency response of
the digital filter the way Wadia does. (Of course the Wa­dia sounds different—it's down 3 dB at 20 kHz!) The
Krell also rejects image tones almost completely, unlike
the Wadia. Crosstalk stays below -110 dB up to 5 kHz
and increases to -98 dB at 20 kHz. A 60 Hz component
was identified in the noise spectrum at a level of -108

dB. DAC gain nonlinearity was -0.2 dB at the -80 dB
level and -0.5 dB at the -100 dB level. THD + N for an
input of -20 dB was -76.5 dB to -77.3 dB (relative to
full scale) across the entire signal band, very close to the
theoretical limit of -78 dB. The hum components iden­tified above probably account for most of the small dif­ference.

For full-scale (0 dB) signals up to 2 kHz, the THD
+ N was -93 dB. This is 5 dB short of the theoretical lim­it of -98 dB. Reducing the signal level to -6 dB did not
improve the 5 dB differential. Furthermore, the THD + N
did not change when measurements were made in the
balanced mode, indicating that the distortion may not be

originating in the output stage. It is unlikely that the

SSM-2131 is the source of distortion, although the Ben

Duncan article I referred to above makes me wonder. My
best guess is that the DAC trim pots had not been adjust­ed properly. Burr-Brown says the adjustments are "in
practice.. .quite complex" and that "near optimum perfor­mance can be maintained at all signal levels without us­ing the optional MSB adjust circuitry." The EAD units
we have tested did not use the trim pots on the PCM63P­K and they performed better than the Studio in this re­spect. Full-scale THD + N rose at 4 dB per octave above
2 kHz, reaching a maximum of -83.5 dB at 11 kHz.
Again the results were the same for the balanced mode,
but in this range there is an improvement in either mode
when the level is reduced by 6 dB. It is not clear where
the rising high-frequency distortion is coming from; it is
even possible that it is a manifestation of high levels of
clock jitter.

On the positive side, radiated RFI was exceptional­ly low in the Studio. That is a truly remarkable engineer­ing accomplishment, given the multiple system clocks
and very extensive high-speed digital circuitry in this
unit.

In conclusion, the Krell Studio is clearly not the

50

D/A processor of choice, even for the audiophile who
can afford it, because it does not deliver the performance
that would be expected for the price. Yes, there is high­level engineering in the Studio, including the industry's
first properly performing DSP-based 16x digital interpo­lation circuit. Yes, the Studio is very well constructed
(except for that power-supply board!), with premium
parts. But the clock-recovery circuitry is identical to that
in units costing thousands less, and the analog stage is ac­tually less advanced than in units costing a small fraction

of the price. Indeed, the Sentec DiAna, which uses the
Phototronics PA630 analog IC, has all its analog stages
operating with little or no feedback and costs only $1150
for all its political correctness. Measured distortion in the
Krell Studio was disappointing, especially since it is a
high-feedback design. In view of all this, a high-ender
might be surprised that it did not sound a little worse than
some of the lower-priced competition, but a series of
ABX listening tests showed it sounded just like the rest
of the group here. So I am keeping my EAD DSP-1000
Series II and feel no loss now that the Krell is gone (ex­cept that I have to pull the EAD's power plug to avoid
the massive radiated RFI whenever I want to use the ra­dio or TV).

Integrated D/A Converter and Line Amplifier

Monarchy Audio Model 33

(Reviewed by Peter Aczel)

Monarchy Audio, 380 Swift Avenue, Unit 21, South San Fran­cisco, CA 94080. Voice: [phone]. Fax: [phone].
Model 33 Dual 20-Bit D/A Converter with Class A Line Ampli-

fier, $1199.00. Tested sample on loan from manufacturer.

The reader is referred to David Rich's reviews, in
Issue No. 19, of the Monarchy Audio Model 22A D/A
processor and Model 10 line-level preamp. The Model 33
combines the two at a saving of $779.00. So far, so good.

We have received no schematics, so I cannot report
to what extent, if any, the circuits have been modified.
The D/A performance is definitely improved; full-scale
THD + N is now -89 dB up to 2 kHz, rising to a maxi­mum of -83 dB at high frequencies, but this includes
gain-related analog distortion. The irreducible D/A dis­tortion due to other causes, measured with a -20 dB digi­tal input and normalized to full scale, is -94 dB, still not
doing full justice to the top-of-the-line Burr-Brown
DAC. The power-supply-related bumps are gone from
the noise floor with digital zero input, but the high­frequency noise level is too high (-104 dB at 20 kHz,

-79 dB at 200 kHz). Strangest of all, the D/A frequency
response is up one full dB at 20 kHz. All of the foregoing
was measured at D/A out, bypassing the line stage.

The line stage still measures the same as that of the
Model 10 (i.e., just fine), and the Model 10's loony-tune
source switching system is gone by default, since the

THE AUDIO CRITIC

pdf 47

Model 33 accepts only two external sources, switched by
a single toggle switch which can't be up and down at the
same time.

With just a little more engineering polish this could

become a highly recommendable unit for the money. I
know that Monarchy has access to excellent professional
engineering; unfortunately they also "hear" things with­out verifying them with double-blind ABX comparisons.

CD/Videodisc Transport

Monarchy Audio DT-40A

(Reviewed by Peter Aczel)

Monarchy Audio, 380 Swift Avenue, Unit 21, South San Fran-

cisco, CA 94080. Voice: [phone]. Fax: [phone].
Model DT-40A Audio-Video Laser Player, $1499.00. Tested
sample on loan from manufacturer.

We're in tweako territory here and I won't even
blame Monarchy because they're simply giving the
tweaks what they ask for, and after all the equipment
does work. The tweako belief is that a laser videodisc

transport is somehow better for playing CDs than an ordi­nary CD-only transport. More solid, heavier, more accu-

rate, lower in jitter, or something. I have never been

shown scientific evidence of that, nor have I ever met a
digital expert with serious credentials who actually be­lieved it, but there you are.

This is a late-model Japanese "combi" player chas-

sis dressed up by Monarchy to do battle as a super audio
component. It weighs 50 pounds, comes with special
chassis damping, has both S/PDIF and AES/EBU (pro­fessional) outputs for connection to a D/A processor,

and—let's not forget—will also play videodiscs when
plugged into a TV. It even has an S-video output but it
doesn't have all the sophisticated control facilities and
image-manipulation capabilities of some cheaper, more
video-oriented laser decks. It's definitely intended for the

audio tweak who wants more than "just a CD player."

I used the DT-40A for quite a few weeks—there's
really nothing meaningful audiowise to be measured on
such a machine—and I was definitely less happy with it
than with the Sony CDP-X707ES that's usually in my
system. It sounds the same in a double-blind comparison

(what did you imagine?) but it takes forever to reach
ready-to-play status on power-up and is slower in access,
more cumbersome to operate, and buggier. It has an auto­matic feature that puts it to sleep in standby mode when
not played for a while, and it that mode it has the habit of

suddenly sending loud, scary chirps and buzzes through
the system without the slightest provocation. I am told
that this bug has been fixed, but it sure makes for reduced
credibility.

I think Monarchy has some interesting and worth-

while ideas, but this isn't one of them. When it comes to
CD transports, I'll have vanilla.

ISSUE NO. 21 • SPRING 1994

Outboard D/A Converter

Sentec DiAna

(Reviewed by David Rich)

Phototronics Co. Regd, P.O. Box 977, Manotick, Ont., Canada
K4M 1A8. Voice: [phone]. Fax: [phone]. Sentec
DiAna Audio D/A Converter (made in Sweden by Antemi),
$1150.00. Tested sample on loan from distributor.

One of the persistent old wives' tales in the high­end community is that negative feedback causes audible
distortion even though the circuitry measures very well
on the test bench. These ideas were first proposed by
Matti Otala in the 1970s as part of his work on dynamic
distortion mechanisms. While his original work was
highly plausible, subsequent analysis has shown that dy­namic distortion can be prevented even in a high­feedback amplifier if some specific design rules are fol­lowed. That should have put the low-feedback argument
to bed, but it has not happened yet. If the proponents of
low feedback were only the people pushing Wonder
Caps, tubes, and $1000 speaker cables, we could dismiss
the case and move on to other things. It is more compli­cated than that, however. Some excellent engineers still
hold to the theory, such as Marty Zanfino of Harman
Kardon, Eric Lauchli of Coda, Jason Stoddard of Sumo,
my former partner Jon Schleisner of Precision Audio,
and Doug Wadsworth of Phototronics.

The last name may be unfamiliar to you, since Pho­totronics makes electronic components, not audio equip­ment. One of the former that the company makes is the
PA630, an IC specifically designed to form the analog
stages of a CD player that uses a current-mode DAC. The
current-to-voltage converter stage of the chip uses little
or no feedback, and the remaining stages use no feedback
at all. Wadsworth, the company's chief engineer, is a
highly respected expert in the field of current-mode ana­log design. He has authored a chapter in the principal
book on the subject (Analogue IC design: the current-
mode approach, edited by C. Toumazou, F. J. Lidgey,
and D. G. Haigh), in which he discusses the fundamental
block used in the PA630: the integrated current convey­or. Wadsworth has also been published by the IEE,
IEEE, and AES in a number of technical papers on the
subject. The Sentec DiAna D/A processor under review
here uses the PA630.

Whenever I encounter proponents of low-feedback
design, I ask them for one of three things: (1) a mathe­matical analysis showing the deleterious effects of feed­back, (2) an electrical test showing the advantages of low
feedback, or (3) double-blind listening test results show­ing that the low-feedback amplifier is audibly superior to
(or at least audibly different from) one with higher feed­back. So far I have not received any of the above, but I
have a lot of IOUs. So, at the moment, if you want to go

53

pdf 48

with the low-feedback crowd you have to do it on faith. I
am willing to admit that I used to believe in low feedback
myself until I started double-blind listening and could
hear no difference. Now, if anybody ever does satisfy
one of these IOUs, nobody will be allowed to forget that
I hedged a little bit in this review. If nothing ever comes
of it, nobody is going to remember the review. But if you
want to hedge your own bet, you might consider the Sen-

tec DiAna.

DiAna is the product of two Swedish companies:
Sentec (analog engineering, marketing) and Antemi (dig­ital engineering, manufacturing). The project engineer
was Svante Österberg at Antemi. The DiAna is enclosed
in a little extruded aluminum box about the size of a mo­dem. An external power transformer connects to the wall
outlet and supplies 15 V ac to the box. The unit has no
power switch; the manufacturer claims that since it draws
only 7 watts it can be left warmed up all the time. The lit­tle box is crammed with parts on a high-quality double­sided board. Extensive RFI filtering is supplied for both

the analog and digital power supplies. Four full-wave
rectifiers are used in the power supply. A pair of rectifiers

is for the ±5 V digital supplies, and the other pair is for
the ±12 V analog supplies. Each rectifier is ac-coupled to

the transformer. This allows each power supply to float.
The digital grounds of the two digital supplies are con-

nected to form the digital groundplane, and the analog

supplies are treated similarly. Digital regulators are
formed with 7805 and 7905 3-terminal integrated regula­tors. Analog regulators are open-loop regulators formed
with a voltage reference consisting of a zener diode
biased by a resistor from the unregulated supply rails.
The reference is filtered by an RC filter which then drives
the base of a bipolar pass device. The analog supplies are
further subregulated to ±5 V for the DAC power sup-

plies. Separate regulators are used for each DAC (again

the 7805 and 7905).

Data for the DiAna can come in through one of two
coax inputs or a Toslink. A front-panel switch selects the
input. Another switch on the panel is for inverting polari­ty. After the front-end circuitry selects which input is to
be processed, the signal goes into what is currently be­coming the standard chip set for high-end DSPs. First the
Crystal CS8412 chip decodes the signal and sends it to
the NPC SM5813 digital filter. Then the signal goes to
the Burr-Brown PCM63P-K DACs before it is routed to
the Phototronics PA630.

The Phototronics PA630 has been available since
July 1989 (the chip was discussed in Issue No. 15), but
DiAna is the first digital audio playback system to use it.
I do not understand why the chip has not found greater
acceptance among designers committed to low feedback.
These designers have instead turned to higher-feedback
circuits (see the Harman Kardon HD7725 and Krell Stu­dio reviews in this issue), apparently unaware of the
PA630's existence. The PA630 can also be used to make

54

low-feedback preamplifier line stages. Furthermore it can
be used in applications outside audio, where it does show
measurably better performance than a voltage-mode op­amp. The chip uses an advanced dielectrically isolated
complementary bipolar process.

The first stage of the PA630 is the current convey­or. This is the circuit block which is used for the current­to-voltage conversion. The current conveyor is essential­ly a current mirror with a unity current mirror ratio. The
voltage at the input of the current conveyor is set by a
second input of the circuit. This input is set at ground for
CD-player applications. The output of the current con­veyor leaves the chip to go to a load resistor, which in the
simplest case is connected to ground. The output voltage
is the DAC current times the value of the resistor. Add­ing a capacitor in parallel with the resistor forms a first­order lowpass filter, which is used as part of the antialias­ing filter. The de-emphasis network, which is switched
by a relay, is also connected from the output of the cur­rent conveyor to ground. To achieve good noise and dis­tortion performance, the input node of the current con­veyor must be close to 0 ohms, the output impedance
must be as close to an open circuit as possible, and the

current ratio of the current mirror must remain as close to

1:1 as possible over the entire current range. The com-

plete conveyor thus becomes a complex circuit. A discus-

sion of the complete circuit is given in the text cited earli-

er. In the actual implementation of the DiAna the gain­setting resistor is connected to the current conveyor's in­put. For a unity current mirror ratio this can be shown to
add 6 dB of feedback to the system.

The signal then returns to the chip to be buffered
by a 6-transistor open-loop complementary buffer, which
is very similar to monolithic buffer chips such as the Na­tional LM6321. After that the signal leaves the chip and
goes into the passive components of a Sallen-and-Key
filter, which generates the remaining two poles of the
3rd-order reconstruction filter. Another 6-transistor buf-

fer on the PA630 is used as the active part of the filter.
The output of the buffer is the output of the converter.
No muting relay is included, so any signal transients on
turn-on are passed directly out of the DiAna. I guess they
do not get power interruptions in Sweden.

In our sample, frequency response was essentially
dead flat (-0.05 dB at 10 kHz, -0.2 dB at 20 kHz). Cross­talk remained between -108 dB and -106 dB up to 2
kHz, then rose to -90 dB at 20 kHz. The noise spectrum
of digital silence showed negligible power-supply bumps
of -116 dB and -114 dB at 120 Hz and 240 Hz, respec­tively.

Gain linearity of the DAC in the less good channel
was off by +0.25 dB at a signal level of -80 dB, +0.6 dB
at -90 dB, +0.9 dB at -100 dB, and +2 dB at -110 dB.
The other channel was off by only -0.2 dB at -90 dB and
came back to zero error at -100 dB. Overall, this is a
slightly worse performance for a PCM63P-K than we

THE AUDIO CRITIC

pdf 49

have seen in the past but it is basically within the de­vice's specifications. THD + N at an input signal level of

-20 dB measured -77 dB across the entire band, just 1
dB higher than the theoretical limit for 16 bits, indicating
that the noise level of the PA630 is very low. This is of­ten not the case for current-mode and low-feedback de-

signs. At a -6 dB input signal level a THD + N of -85 dB
(the theoretical best is -92 dB) was measured up to 2
kHz, rising to a maximum of -81.5 dB at 10 kHz. For
full-scale input signals THD + N was -81 dB up to 2
kHz, rising to -76.5 dB at 10 kHz (the theoretical best
being -98 dB). These rather poor high-level THD results

are directly due the low feedback. The dynamic distor­tion evident above 2 kHz cannot, of course, be attributed
to feedback; instead, according to Phototronics, the rise
in THD at higher frequencies is the result of nonlinear re­actance terms associated with the second-order errors of
the current mirror. Viewed objectively, the distortion per­formance the DiAna is not particularly good compared
with units we have tested that rely on feedback. None of
the known measurement techniques would show that the
DiAna was at an advantage over high-feedback designs.

So what can we conclude? The DiAna is a well­built, well-engineered digital decoder. It uses a very
high-technology chip to allow it to work with little or no
feedback. Unfortunately, in objective laboratory tests it is
outperformed by competitive units using higher feedback
rates and in blind listening tests it does not appear to be
distinguishable from the latter. The EAD DSP-1000 Se­ries II thus remains my choice in competitive digital de­coders. For someone who believes that low feedback
rates are important and has already invested in a pream­plifier and power amplifier designed with low feedback,
the DiAna is clearly attractive; to me it makes more
sense, for example, to drive a Krell KRC-2 and KSA­200S with a DiAna than with a Krell Studio. But it
makes even better sense to take advantage of the virtues
of feedback, and so I would rather connect an EAD DSP-

1000 Series II into a B&K Sonata Series PRO-10MC and

then into a Bryston 4B NRB.

Compact Disc Player

Sony CDP-X707ES

(Reviewed by Peter Aczel)

Sony Electronics, Inc., 1 Sony Drive, Park Ridge, NJ 07656.

Voice: [phone]. Fax: [phone]. CDP-X707ES
compact disc player with remote control, $2000.00. Tested
sample on loan from manufacturer.

This is the supposedly updated successor to the

CDP-X779ES I endorsed so enthusiastically in Issue No.

18. I refer the reader to that review for all the details,
since the new unit is so similar that I don't even under­stand why Sony had to change the model number. The

heart of the design, the somewhat mysterious CXD2562

ISSUE NO. 21 • SPRING 1994

pulse D/A converter, is exactly the same. I had the oppor­tunity to have both players on the lab bench side by side;
David Rich analyzed the circuit schematics and service
manuals of both. Here are all the differences the two of
us could find:

Mechanically and cosmetically the two models are
identical, except for a seemingly more robust disc drawer
in the new model. This is just a small extra touch added
to what was already a state-of-the-art transport, with line­ar motor drive (as in the older CDP-X777ES), CXD2500
DSP chip, and CXD2501 digital servo chip. Tweako
high-end transports at higher prices are not as good!

Electrically there appear to be some minor changes
in component parts in the new model, plus a new audio
board that differs from the old one in only one important

respect, the digital filter IC. The new chip, making its de-

but here as far as I know, is the CXD2567, dubbed
"Score" by Sony. It replaces the CXD2560 used on the
old audio board. (Interestingly, the service manual does

not show this change, so I had to take Sony's word for it,
especially since the chip is mounted on the underside of
the audio board and is inaccessible without major sur-

gery.) In overall performance the CXD2567 appears to
be identical to the CXD1244 (see Issue No. 15). The
filter has a 20-bit data path, a 26-bit coefficient word

length, and it uses 213 taps. As can be seen from Table 3
in Issue No. 15, this is still a significant step down from
the mighty CXD1144 (22-bit data path, 28-bit coefficient

word length, 293 taps), giving up 20 dB in stopband at-

tenuation as well as a smidgen in passband ripple. The

accumulator output of the CXD2567 is very large at 45

bits. This is truncated to 20 bits with a quantizer block
which uses triangular dither—that's the major innovation
and the only possible justification of the model change.
The CXD2567 and the similar NPC SM5842AP are the
first monolithic digital filters to incorporate dither in or­der to prevent the truncation process from introducing
distortion.

In measured performance the two CD players differ

by only a fraction of a dB on all Audio Precision tests,

which is no more than what the expected production vari-

ation would be within a single model. The one exception
is the suppression of 60 Hz, 120 Hz, and 180 Hz power­supply spuriae in the noise floor of the CD player, which
is even better (by 7 to 10 dB) in the new model, although
the old one was certainly good enough in that respect.
The new digital filter chip does not appear to have made

a measurable difference, at least not on a CD-in/line-out
basis.

That's it; the old CDP-X779ES and the new CDP-

X707ES are otherwise indistinguishable. Consequently

this remains the CD player of choice for the well-heeled
audiophile who craves theoretical perfection whether or
not he can hear the difference. All measurements nudge
the theoretical limits of 16-bit digital audio (see Issue
No. 18 for the specific numbers). The only quibbles one

55

pdf 50

could possibly come up with have to do with the analog
circuitry following the DAC, and they are minor. Why
the slow-settling 5532's at this price instead of, say,
AD797's? (Not that the 5532's appear to cause any prob­lems.) Why is the single-ended output more sophisticated
than the balanced output? (The former uses a discrete
MOSFET output stage and a dc servo, the latter does
not.) The answer could be that Sony has altogether differ­ent design teams for digital and analog, the digital team

being more perfectionistic. Just a hypothesis.

By the way, if it was yet another Sony team that
designed the power supply circuits, they did the most in­credibly perfectionistic job of all, as David Rich pointed
out in a memo to me. The sophistication of the analog
and digital power supplies and the extent of regulation
are unequaled by much costlier high-end equipment,
such as Krell. This should have been mentioned in the
original review.

Those who have read my EAD DSP-9000 Pro re­view above and are into techie overkill regardless of cost
will now ask whether the Sony shouldn't be plugged into
the EAD for even "better" performance. The answer is

yes if you're planning to switch around among a whole

slew of other digital program sources, no if you wish to

play CDs only. Nothing will play a CD with greater accu­racy than the Sony CDP-X707ES.

Second-Generation MiniDisc Recorder

Sony MDS-501

(Reviewed by Peter Aczel)

Sony Electronics, Inc., 1 Sony Drive, Park Ridge, NJ 07656.

Voice: [phone]. Fax: [phone]. MDS-501 Mini-

Disc Recorder with remote control, $1000.00. Tested sample on
loan from manufacturer.

Originally I was going to review the MDS-101,
which was a rather kludgy embodiment of Sony's first­generation MiniDisc technology—a compact home deck
that betrayed its car-audio origins. It barely missed being
included in the last issue. Then, at the Winter CES, I
heard the exegesis of the second generation of MiniDisc

products by the brilliant Dr. Roger Lagadec, Sony's tech-

nical director in Europe. Shortly thereafter this slick new

machine arrived, and as far as I was concerned the MDS-

101 was history, especially since the price had not

changed.

By now the MiniDisc has been widely exposed and
explained; I refer the reader to David Ranada's "Inside
MiniDisc" (Stereo Review, March 1993), to the late
Leonard Feldman's "The Mechanics of Sony's MiniDisc:

Beyond the Caddy" (Audio, December 1992), to Sony's

own widely distributed brochure "MiniDisc: an overview
to the technology behind MiniDisc"—I see no need to go
over the same ground that others have covered so compe­tently. Nor do I wish to repeat here the comments I re-

56

cently made on digital technology vs. digital politics in
the wake of the R-DAT; see the DCC article in Issue No.
20 for that (p. 44).

I tried out a political zinger on Dr. Lagadec at his
press briefing, and he handled it candidly and illumina­tively. I asked him if Sony would ever have come out

with the MD if the R-DAT hadn't run into heavy politi­cal opposition. He replied that they probably wouldn't
have (!), not in the MD's present form, but that the idea
of a wafer-thin, 2¾ square, hard-shell, highly efficient,
recordable/erasable data-storage medium was too attrac­tive not to be pursued by the Sony technical teams, and
therefore some sort of product would have come of it. I
was satisfied with that answer.

The fact is that the MD, when you hold it in your

hand, is smaller, cuter, cuddlier, more lovable than the

DAT cassette, let alone the relatively massive DCC. You

want it to be the winner, even though in purely sonic

terms it isn't. The Sony party line, in all their advertising

and PR, is that the MD is not intended to compete against

the CD or the DAT—heaven forbid!—but that it is the
modern replacement of the dinosaurian analog cassette.
Time will tell whether such positioning is viable; right
now a recordable MD costs six to eight times as much as
a high-quality blank analog cassette—not very tempting
to defectors. (I said recordable MD because prerecorded
MDs are actually a different medium in the same format,

just a tiny CD inside the plastic cartridge.)

The principal audiophile reservation about MD has
to do with the 5-to-l data reduction via the ATRAC
(Adaptive TRansform Acoustic Coding) compression al­gorithm. As our readers know, I and other accountable
reviewers (as distinct from out-of-control tweaks) have
found the 4-to-l PASC compression in DCC to be trans­parent to music, at least the music tried so far. The addi­tional compression with ATRAC poses a problem. The
second-generation improvements have made a difference,
but on the basis of my very limited double-blind CD vs.
MD listening comparisons I am not ready to declare MD
transparent. In all fairness to Sony, they do not claim
such transparency. I plan to do more testing because the
sonic degradation, if any, caused by the current version
of ATRAC is quite subtle, and the prerecorded MDs
sound pretty much like acceptable, but not great, CDs.

For example, if I walked in while the Dvorak cello con­certo with Yo-Yo Ma, Lorin Maazel and the Berlin Phil-

harmonic (Sony Classical SM 42 206) was being played,
it would never occur to me to say, "Hey, what's that? It's
not a CD." It sounds like many middling CDs in my col­lection (fine performance, though). It should be noted
that German researchers using the NMR (Noise/Mask
Ratio) technique have found that ATRAC does not com­pletely satisfy the mathematical/psychoacoustical model
of a presumably transparent perceptual coder. That re­search, however, comes from the first-generation era of
the system.

THE AUDIO CRITIC

pdf 51

Editor's Note: Tom Nousaine is tac-

itly assuming here—a bit optimisti­cally, or even parochially, I think—
that all readers will remember from
his last column what Clark Johnsen,

the medicine man of Absolute Polari­ty, had called him—yes, "a weasel"—
and that Tom now wears the label as
a badge of his audio philosophy. (See
also Johnsen's latest protestation
anent Nousaine in this issue's "Box

978.")

* * *

The Editor's comments about
the usage of geek and geak in the last
issue made me realize that our feel-

ings about audio can obscure a clear
view of the best means for rational
pursuit of what I call High Definition
entertainment. High Definition to me
means a maximum "sense" of accura­cy, realness, or involvement for a
given set of resources. The realiza­tion that resources—time and mon­ey—are always limited means a
High Definition Geak (or HD Weasel
for you English majors) needs to pur­sue a strategy that will maximize ex­pected gains for a given set of
financial inputs. We can't afford to
waste energy chasing our tails, spit­ting in the wind, tilting at windmills,
or tugging at Superman's cape.

The basic issues can easily be di­vided into four categories. First, there
is little argument from anyone­consumer or pro, golden ear or engi­neer, scientist or mystic—that loud­speakers, microphones, artistic per­formance, and recording techniques

can radically affect the quality of the

listening experience. I call these Giv-

ens. Notice that they affect each

"end" of an audio system.

Next we have the Inevitables. It
doesn't much matter whether you
think LPs are better than CDs. The

former are on their way out. So are

compact cassettes. CDs and laser

discs are here to stay. Data-reduced
formats (DCC, MD, and others) are
coming. And so are new digitally
based video formats. These are sub-

ject to availability and only partially

subject to personal preference.

There are few arguments con-

testing that additional processing of

signals as they travel through the sys­tem alters the way things sound.
Equalization, noise reduction, video
and surround processing are common
examples of Additives. Some, like
me, find them useful. Others hate
them, but we don't argue about
whether they make a difference.

Finally, there aren't enough
hours in the day for people to argue
about whether switching, gain, and
transmission devices degrade or im­prove sonics. Basically there is
debate about whether amplifiers,
switchers, converters, cables, line
cords, and so forth have special audio
qualities other than moving a signal
from one point to another or making
it bigger. These are the Question­ables.

No one disputes that Givens and

As for the MDS-501 itself, it may not be an entire­ly representative example of the second-generation MD
line because it is a 17" wide home deck, whereas most
MD products are designed to be portable. I wanted a
sample of the MD technology at its most advanced be-

fore I looked at the more plebeian versions. An MD
Walkman review is in the pipeline.

If I had never seen a high-end CD player or DAT
deck, I'd be totally blown away by the MDS-501. What a
nifty machine! The sophistication of the microprocessor­controlled status display, the editing facilities, the speed
of track access, the look and feel of the controls, includ­ing the remote, are all on the order of sci-fi spaceship
gear. I'm not kidding. Don't play with this deck in the
store because you'll take it home and you probably don't

really need it.

As in the case of the DCC, I was not interested in
measuring with the Audio Precison just how drastically
the perceptual coder alters the signal; that's Sony's psy­choacoustic lookout. I was again interested, however, in
what is supposed to remain unaltered, i.e., in the accura­cy of D/A conversion and in purely analog distortion. To
my surprise I found that in these respects the MDS-501 is
a match for all but the absolute best CD players, D/A
processors, and line-level preamplifiers. At line out, the
full-scale THD + N of the delta-sigma DAC at most fre-

ISSUE NO. 21 • SPRING 1994

quencies is -89 dB, kicking up to -83 dB at 20 kHz.
Some of this is gain-related analog distortion; the irre­ducible D/A distortion due to other causes was found to
be -92 dB at all frequencies. Few DACs are better. The
gain-linearity error is 0 dB down to the -90 dB level and
only 1.6 dB at -100 dB. Highly respectable. The purely
analog THD + N (line-in/line-out at full gain, a worst­case setting) is noise-dominated and reaches a minimum
of -74 dB (0.02%) at 2 V out, with virtually no dynamic
distortion apparent. Could be better, but I've seen a lot
worse. The line-in/line-out crosstalk at 2 V out is sensa­tionally low in one channel (in the -113 dB to -90 dB
range, depending on frequency) and ordinarily low in the
other channel (steadily rising with frequency from -107
dB to -57 dB); this must be due to parts layout. The digi­tal-in/line-out crosstalk ranges from -124 dB to -83 dB,
depending on frequency. There are no frequency­response deviations of more than 0.05 dB at any frequen­cy in any mode.

We are in the earliest stages of the perceptual cod­ing era; the plot will surely thicken and various solutions
will come and go before we see any permanence. Even
though I have some apprehensions about the possible
threat to linear PCM technology as the gold standard, for
the moment I am not discouraged by the MD solution. A
lot of very good engineering went into it. •

57

pdf 52

Additives affect the sound. It is obvi­ous to me that their sonic impact has
to be orders of magnitude greater
than those of the Questionables,
where there is so much argument
(even if you ignore deciding who's
right). Debates over the sonic attrib-

utes of cables are a laughable tem­pest in a teapot compared with the
questions raised by differences in
loudspeakers and recordings.

Therefore, the clever HD Weasel

adopts a policy of spending his mon­ey and time where they are most

likely to have maximum positive ef-

fect. His strategy will devote the
most energy to selecting better loud-

speakers and recordings. He will
make informed decisions about for­mats which may later affect his
choices in loudspeakers. He knows
amplifiers, preamps, and transmis­sion devices will greatly enhance the
usability and style of the system but
have minimal sonic impact, and he
makes his choices accordingly. He
designs systems as a whole and im-

proves them from the outside to-

wards the middle.

For example: The list below
shows my perspective on the most
appropriate order of resource expen-

diture in terms of maximum sonic
impact for a given expenditure. Of
course, I don't always spend my
money in exactly the proper order,
but I am aware that while maximiz­ing personal satisfaction I might also
be suboptimizing sonic performance
by ignoring the rules.

But it is a choice...and I can't
change reality with my feelings. It
doesn't matter what I would like to
believe; no one has been able to tell
competently designed amplifiers apart
with the logos removed. Buying

products with my heart and not my

head may compromise sound quality.

To maximize sonic impact,

spend money in the following order:

Givens.

1st—Better recordings. You can
use your heart here as well. Plus, on
an individual basis, they are relative­ly inexpensive.

2nd—Better loudspeakers.

Transducers with flat on-axis re-

sponse, smoothly controlled direc­tionality, extended bandwidth, and
wide dynamic range yield the next­biggest bang for the buck.

3rd—More loudspeakers. Add a

subwoofer to extend the bandwidth
of your system. Employ surround/cen­ter systems to solidify the image,

widen the listening area, and improve
dynamics. See the 4th step.

Inevitables.

4th—Give your system an im­age. Video and surround processing
is on the way. Don't fight it. The best
way to improve the imaging of your
system is to literally give it an image.
You might argue that a surround

processor is really an Additive. May­be you're right.

Additives.

5th—Test equipment. Want to

spend a few grand and learn the
truth? Get a Techron TEF, MLSSA,
or LMS measurement system. You
will find ways to improve the work­ings of your system by finding out
lots of stuff you probably didn't
want to know when you started.

6th—Go ahead and get that

equalizer. Use it to touch up your
main speakers. Important to use the
test equipment to adjust. Don't over­do. You can't make a 40 Hz speaker
do 20 Hz with an equalizer.

Questionables.

7th—Buy amplifiers with enough

power. Buy preamplifiers with enough

inputs and adequate controls. If you
want to spend a few extra dollars,
make certain you get better build
quality with it. Get good remotes.
It's far easier to tune your system
from the listening position than
standing at the equipment rack. Buy

CD players, laser disc players, tape
recorders and decks, and radios with
useful features and good build quality.

Also, buy cables that are long
enough, with decent connectors and
adequate shielding. Buy stands that
position the speakers correctly and
look good. Minimize expenditures on
accessories. If you're tempted to
spend a lot of money on a Question­able, buy some new recordings in­stead—while you reconsider.

Remember: exotic cables, fancy
tubed amplifiers, expensive A/D and

D/A converters, and almost anything
that exudes weirdness are mostly per­sonal image options. They may im-

prove your self-image or make your

system look cool, but they are not
likely to improve the sound.

Summary.

Therefore, spending the bulk of
your money and time in areas where
everybody agrees differences can be
heard will improve the probability of
maximizing the definition, and hence
the long-run enjoyment, of your au­dio and/or video system for a given

dollar expenditure. That's High Defi-

nition thinking.

High-Def Weasels take this one
step further. If a piece of rusty baling
wire rescued from a ditch cannot be

distinguished from an expensive
speaker cable under blind conditions,
the Weasel considers the baling wire
superior because it frees cable mon­ey for more recordings, or concert
tickets, or something. That's the trou­ble with Weasels—you just can't
trust them!

Coming:

The long-delayed first review of the new Win SM-8 studio monitor speaker and a

great deal more on the Velodyne DF-661 low-distortion speaker system.

A shootout between the Bag End ELF S18E-C and Velodyne Servo F-1500R subs.
All the reviews of CD players and D/A processors squeezed out of this issue (see

page 45) plus more digital equipment tests.

The beginning of the promised but slow-to-jell survey of FM components.
Continued reviews in depth of preamplifiers and power amplifiers.

58 THE AUDIO CRITIC

pdf 53

Hip Boots

Wading through the Mire of Misinformation in the Audio Press

Editor's Note: Many of our readers confess that this is the first page they turn to when they look at a
new issue. A few others refuse to understand the purpose of this column. They think we just want to

put down our competition. Not so. If Hustler, which is definitely not our competition, wrote about

the "pace" or "ambience" of a speaker cable, we would castigate them for it as much as we do Ste-
reophile and The Absolute Sound (and as we did Business Week and Fanfare, also not our competi­tion, when they printed deleterious misinformation about audio). Naturally, the most frequent of-

fenders are castigated most frequently. By the way, this time I am the sole author of all the items here.

Aftershave.

Ken Nelson, Stereophile's chief advertising getter

and business booster, has confessed that he was the writer
of the "Larry Archibald smiles as he shaves" promotional
letter that tickled us so much here (see the first item in

"Hip Boots" in Issue No. 20). I have known Ken Nelson
for approximately 30 years and have the highest respect
for his professional competence and personal integrity.
Indeed, I suspect that Ken was the undeserved lightning
that struck Larry Archibald and made his magazine the
commercial success it is today. I know that Ken is smart
enough and honest enough to admit (to himself, privately,
when he shaves) that everything David Rich and I have
written in this column about Stereophile is 100% true.
The very fact that he parries our thrust ever so allusively
indicates what he must think.

Therefore, in reply to his advertising hyperbole that
"Stereophile is written to the highest technical and literary
standards anywhere in the publishing world," I make my
standard offer: Produce three electronics/electroacoustics
experts, Ken, with university graduate degrees in engi­neering or physics who are not commercially linked to
your magazine, or to the ultrahigh-end audio manufactur­ing/retailing/publishing business in general, and who will
confirm in writing that your quoted statement regarding
Stereophile's technical standards is correct—after having
examined a number of recent issues to be pointed out by
me. If you can do that, I shall admit in print that we have
been wrong all along, reprint the statements of your three
experts, and mail the issue of The Audio Critic containing
my admission and the reprints to our entire subscription
list as a free extension of each subscription. Fair enough?

60

Harry Pearson goes off the deep end.

This time HP really did it. No, I'm not talking about
the latest bit of HP lifestyle gossip. I'm talking about his
utterly swinish insinuation in Issue 89 of The Absolute
Sound that I may have paid Larry Klein—"hired him" is
how HP puts it—to comment enthusiastically about The

Audio Critic in the "Audio Update" column Larry writes

for Electronics Now. HP actually calls the latter magazine
Popular Photography (twice!), that's how conscientiously
he researches his shots from the hip. The man is out of
control and unaccountable. Most of his remarks on the
subject are self-parodying, but a few comments from me
are in order.

Larry Klein and I have known each other for well
over 20 years, going all the way back to his tenure at Ste-
reo Review as technical editor; we have nearly always
agreed on audio matters, so I don't have to "hire" him to
editorialize to the effect that mine is "a thinking audio-

phile's magazine" and that HP's is "a witches' brew of
pseudoscience and unabashed subjectivism.. .influenced...

by price, fads, technical ignorance, and unfettered egos."
That's what Larry thinks free of charge, that's what he
has always thought, and he would have to be totally in­sincere to write anything else. HP turns around and calls
him "ignorant and wrong" on the subject; this is like, say,

David Koresh calling Pierre Teilhard de Chardin ignorant

of theology. (It was HP himself who made me turn to

theology for my example; he writes that "born-again audio

fundamentalists now rule the roost" at The Audio Critic.
Apparently 2 + 2 = 4 is fundamentalist in HP's view and
2 + 2 = 5 is enlightened and liberal. Anyway, HP has no
technical credentials and Larry Klein does.)

THE AUDIO CRITIC

pdf 54

The funniest part of HP's tantrum is his assertion
that both Stereophile and The Absolute Sound "are dra­matically better in technical accuracy and in solid judge­ments than they once were." Isn't that delightful? Exactly
when did you stop being incompetent, Harry? When did
you stop beating your.. .uh... wife?

A Carver ribbon by any other name...

Remember how much Stereophile disliked Bob
Carver's "Amazing Loudspeaker"? This goes back to

1990; they actually managed to wreck the sales of the
speaker for quite a while. (See also the sidebar I inserted
in Issue No. 16, page 9, on the subsequent lawsuit.)

Now, in Stereophile's report on the 1993 Summer
CES (in their August 1993 issue) Contributing Editor
Robert Deutsch, one of their leading tweaks, goes gaga
over the $55,000 Genesis I loudspeaker by Arnie Nudell
and Paul McGowan. "How did it sound? Absolutely ef­fortless, with far better coherence than I would have
thought, given the multiplicity of drivers." In a footnote,
RD adds: "Arnie Nudell says that the resemblance to the
IRS is in form only. He says the drivers and other aspects
of design are quite different." They sure are. Take a look
at the picture of the Genesis I on the following page, all
you golden-eared Stereophile gurus. See that five-foot
ribbon smack in the middle? You know what that is? It's
the ribbon from the Carver "Amazing," amigos. Supplied
to Genesis Technologies by Carver Corporation—as is,
no frills, totally unmodified! Are your faces red?

Logical conclusion: When a novel transducer is

part of a speaker system introduced at $1576 the pair, it

is baaaad. When exactly the same transducer is part of a
$55,000 tweako rip-off, it is gooood. That's credibility,

High End style.

And now, the obligatory Harley Howler.

No "Hip Boots" column would be complete with­out a technical foot-in-mouth item from the bounteous
pen of Robert Harley, patron saint of the electronically
semieducated. In the August 1993 issue of Stereophile
(yes, the same issue as I cited above—no need to go to
another one to find more blunders), he reviews the
$6000-plus Genesis III five-way speaker system (the one

without Bob Carver technology in it) and makes the fol­lowing fascinating statement: "The crossover slopes vary
between second-order (12dB/octave) and third-order

(18dB/octave), with increments in between."

He sent me scurrying to all the standard texts on
analog filter theory, where I was hoping to learn more
about filters of the 2.1st order (maybe 13 dB per octave?),
the 2.2nd order (14 dB per octave?), the 2.3rd order (15
dB?), and so forth. To my dismay, I found that such
"incremental" orders do not exist! If you add another
pole to a lumped-parameter second-order filter—just a
teensy-weensy pole, Bob—you jump straight to third­order. Son of a gun! The laws of physics are so frustrat-

ISSUE NO. 21 • SPRING 1994

ingly unforgiving, aren't they?

Unfortunately, my altruistic appeal in the last issue
for contributions to the SHEESH (Send Harley to E.E.
School in a Hurry) Fund fell on deaf ears out there in
golden-ear country. Only the irrepressible Joe Cierniak
(Editor/Publisher of Sound Off in APO land) sent in $10
to be added on top of my pledged $50, and that was it. As
I wrote to Joe when I regretfully returned his check, $60
isn't enough for Bob's supply of nerdy plastic pocket

protectors for four years, let alone E.E. school tuition.
There is no charity left in the audio world...

(Not all misinformation from Harley is due to lack

of knowledge, though. Some of it is just dirty politics. In
the March 1994 issue he rhapsodizes that the new Adcom
GDA-600 is the only D/A converter under $1500 to use
the 20-bit Burr-Brown PCM63 DAC chip. He conven­iently forgets—because EAD is on Stereophile's baaaad
list—that 11 months earlier he himself reported the debut
of the EAD DSP-1000, a $999 unit that uses the PCM63,
in the K grade as against the lower J grade in the Adcom.)

Skull and crossbones on the crossover.

Speaking of Joe Cierniak, it was he who called my

attention to this one. Aaron M. Shatzman, archly listed
on the masthead of The Absolute Sound as Philosopher­in-Residence, wrote a review of the Paragon Acoustics
"Jubilee" speaker (yet another small two-way system
with big claims, made in Minnesota) in their Issue 92.
The review is the usual soundstage-obsessed subjectivis­tic gush fest, but I wouldn't bring it up for just that.

In a separate little box captioned "Speaker Specifi­cations," the resident philosopher lists crossover slopes
of "12,000 volts/octave." Holy Moses, Aaron, that's dan­gerous! The unit couldn't possibly be UL listed under the

circumstances but it should at least have a highly visible
death's-head warning label on it, for heaven's sake.
Please be careful! Electrocution is such an undignified
and historically unsuitable demise for a philosopher.
Hemlock, okay, or even crucifixion, perhaps an auto-da-

fé—but 12,000 volts...

Epilogue, on a theme by Larry Archibald.

In the January 1994 issue of his magazine, Larry

Archibald writes: "...I believe that Stereophile offers the
most well researched and authoritative reviews available,
and that other magazines are, well, basically unnecessary."

If you will accept a bouquet, Larry, from the most

unnecessarily critical and factual of the unnecessaries (as
you must see us), I consider that quotation to be an
immortal classic. I rank it with Herbert Hoover's gem
announcing prosperity to be just around the corner, with
George Bush's "read my lips" pearl, and with Hermann
Göring's declaration that the Allies will never bomb Ger­many or his name is Schultz (or was it Meyer?). I think
the best-educated members of the audio community—
and you know who they are—will share my sentiment. •

61

pdf 55

Recorded Music

Editor's Note: My massive backlog of CDs to be capsule-reviewed (as in Issue No. 19) had to yield
once again to David Ranada's more interesting and equally opportune reviews, but be prepared for
a major catch-up sequence in the next issue (No. 22). Meanwhile I'm appending just a few can't­wait-to-tell-you items of my own to the end of Davids column. Don't confuse my opinions with his!

A Miscellany of CDs and

Musical Videodiscs

Broadway Musical

Richard Rodgers and Oscar Hammer­stein: Opening Night, The Complete

Overtures. Hollywood Bowl Symphony
Orchestra, John Mauceri, conductor.

Philips D 100190.
Leonard Bernstein: On the Town. Lon-

don Symphony Orchestra, Michael

Tilson-Thomas, conductor. Deutsche

Grammophon 437 516-2.

No musical idiom needs rescu­ing from the depredations of record­ing producers and engineers more
than the Broadway musical (though
big-band jazz comes in a close sec­ond). Original-cast recordings of
Broadway material only rarely pro­duce the sonic effect of a live pit or­chestra in a real Broadway theater.
All of these buildings are within a

10-minute walk from where I work,
and I've attended performances in a
great many of them (probably all
those still suitable for musicals).
None of them is as large or reverber-

ant as the typical original-cast album

will have you believe. The sound is
instead hard-hitting, vivid, direct,
rather dry, and not very rich.

62

By David Ranada

Contributing Editor at Large

Not that you can tell from either
of these CDs. Mauceri's recording of
Rodgers overtures is severely com­pressed and has a reverb characteris­tic seemingly tacked on from another
space altogether. The reverb machine
should have been have been turned
off. The percussion in particular
lacks bite and impact. But the sound
is rich and luxurious—and altogether
inaccurate, unrealistic, and unauthen­tic. Unfortunately, due to the corrupt­ing influence of recordings like this
one, live Broadway productions now
are overmiked, overamplified, and
disgustingly reverbed to sound like—
recordings!

Musically, I liked the perfor­mances, though an hour of Richard
Rodgers's instrumental music, even
in its original orchestrations (none by
him), is a bit too much of a good
thing—I wanted to hear some sing­ing. But if this is what you want, this
is the only game in town.

There's singing in abundance in
Tilson-Thomas's rendition of Bern­stein's 1945 On the Town, which re­ceives its most complete recording to
date with Deutsche Grammophon
production, one recorded using

DG's proprietary "4D" technique,
about which I will have more to say
in a future issue. Suffice it to say that
despite the use of "21-bit" A/D con­verters, the recording's waveform is
clipped during many loud portions
of the music. It's not audible per se
and it occurs at very slightly below
CD-maximum level, indicating either
that the signal passed through a digi­tal console or other processor that
clips below unity gain, or that the
signal was clipped earlier in the
chain (perhaps at the mike preamp or
initial A/D conversion). In any case,

sloppy production is indicated—
there's no good reason for any digital
recording ostensibly made with reso-

lution greater than 18 bits ever to

clip. (A conspiracy theorist will say

that it is deliberately clipped in order

to obtain thereby some degree of

compression, to produce a "louder,"

more impactful disc, as is typically

done with FM radio signals.) While

we're speaking of slop, there's also

some low-level grungy electronic-

sounding noise in the right channel at

the end of Track 4.

The orchestra here—the London

Symphony—is quite a bit larger than

THE AUDIO CRITIC

pdf 56

your typical pit band. It was also re­corded live in a hall (London's Bar­bican Center) that I recall not being
as reverberant as it sounds on this
disc. The dead giveaway is the obvi­ous artificial reverb tacked onto the
end of Tracks 2 and 13. This record­ing is, like the Rodgers disc, a multi­miked multiple-mono extravaganza.

I liked the performance here too,
though Tilson-Thomas's conducting
often sounds manic compared to the
composer's own excellent 1960 re­cording (nicely remastered—and
with a very theaterlike ambience—
on Sony SM3K 47154). The over­wrought feeling is abetted by the size
of the performing forces: you can't
move that many people that quickly
without making it sound frantic. The
singing is more operatic than in
Bernstein's recording (except for the
always-yelling Tyne Daly) but still
within the bounds of Broadway
style. The disc includes, in their
proper sequence, all the extended
dance episodes and several vocal
numbers never before recorded. Get
both: Tilson-Thomas for singing and
completeness, Bernstein for interpre­tive and sonic authenticity.

Opera

"The Golden Ring: The Making of Solti's
'Ring'." A BBC film directed by Hum-

phrey Burton. London videodisc 440

071 235-1.

Anybody with an interest in clas­sical music recording techniques and
in the role of the recording producer
must see this program, which is also
available on VHS tape (440 071 235-

2). The film, by Humphrey Burton, is
of the 1964 recording sessions for
Georg Solti's London/Decca perfor­mance of Wagner's Die Götter-
dämmerung (London 414 115-2).
Unlike more recent recording-session
programs (which are really promo
videos that have a suspicious habit of
appearing on PBS at the time the as­sociated CD is released), this one has
a point of view: to explore the rela­tionship of a recording's creators, es­pecially the ones on the technical
side—"high priests of flawlessness,"
Burton calls them—to the work be­ing recorded.

We see the recording producer,

ISSUE NO. 21 • SPRING 1994

the late, great John Culshaw, in his
element: deciding orchestral balances,
which takes to use, where the takes
need to stop and start to ease the task
of splicing, what the musical defi­ciencies of the actual performance
are. We hear him agonizing over
"what the composer wanted" and
how to "do the best by Wagner." In
short, we watch him doing what a re­cording producer should always do.

Students of recording techniques
will enjoy seeing the massive
amounts of technical nitty-gritty ex­posed to the camera: mike placement
(including the famed Decca "tree"),
the mikes (mostly Neumann, as far
as I could tell), the tape recorders
(½-inch open reel), the tube electron­ics, how instruments like the harps
are moved from their customary

placement instead of being given

spotlight mikes, and how the singers
are scrupulously directed by a "stereo
man" to chessboard-like numbered
squares on the stage to achieve spe­cific—then revolutionary—stereo ef­fects. We are told by an engineer that
12 orchestra microphones are "suffi­cient" for this massive work; twice
that number, and more, is now com­mon. There are also 3 stage mikes
for the singers and 5 mikes for spe-

cific offstage effects, such as the

stierhorns called for by Wagner's
score.

Careful viewers will notice that

the stierhorn pickup is mono and

sounds artificially panned-in during

the finished production (one of the
recording's few technical miscalcula­tions). Wagnerites will undoubtedly

note how the actual last sung words

of the entire Ring cycle (Hagen's

"Zurück vom Ring!") are not per-

formed during the take of the "immo-

lation" scene we see being recorded.
Perhaps it, like the sound of the col­lapsing hall of the Gibichungs, was
dubbed in later. Musicians will enjoy
actually seeing a bass trumpet in ac­tion, the noble-sounding "Wagner"
tubas, and other brass instruments

with the uncommon valve mecha-

nisms used by the Vienna Philhar-

monic.

You'll also see some strange
singer behavior, like the way they
lunge toward the microphone at the
end of some phrases (perhaps to em­phasize a weak note) or slightly back

off on high or loud notes (perhaps to
prevent overload somewhere in the
chain). It's interesting to go back to
the finished recording to see how
these actions have affected the
sound. At the very least, the backing­off motion increases the vocalist's
"leakage" into the orchestra mikes.

The rare person who likes to
make measurements of recordings
will be fascinated to find that those
musical segments of the finished re­cording that in the movie obviously
originated from video still carry the
near-ultrasonic horizontal scan fre-

quency that leaks out of video came­ras and monitors and is easily picked

up by microphones. This can be easi-

ly seen with an FFT-based spectrum

analysis, which shows a distinct

spectral component between 15.75
kHz and 15.78 kHz about 80 to 90
dB below full level during those seg­ments. Even more fascinating is that
this spectral (and spect[e]ral) compo­nent is slightly higher than the stan­dard European horizontal scan fre­quency (which has long been 15.625
kHz). This possibly indicates that the
CD's analog master tapes were run­ning slightly faster (by about 17/100
of a musical semitone) during the
mastering process than they did dur­ing the sessions. Rosetta stone in­deed!

Other technical notes: The sound
is very good mono, the picture black­and-white with variable quality de-

pending on whether the shots are

from film or video. Although the nar­ration is in English, as are most of
the on-camera comments, it is a boon
to know German, the language Solti
usually uses with the orchestra. The
camerawork is good—meaning con­servative—with few zooms and no
dental-exam ultraclose-ups of spit­spewing and sweating vocalists.
There are a few misdirected shots in
some of the extended musical por­tions (such as Siegfried's funeral pro­cession), but they serve only to lend
a video-vérité feel to the proceed­ings. Again, a must-see.

Wolfgang Amadeus Mozart: Don Gio­vanni. The London Classical Players,
Roger Norrington, conductor. EMI
CDCC754859 2.

Don Giovanni, the greatest of

63

pdf 57

Mozart's operas, receives here its
most "authentic" recording to date,
in musicological rectitude a veritable
doctoral dissertation compared with
the term-paper quality of the only
other original-instrument recording
(Östman's on L'Oiseau Lyre 425
943-2) and with the high-school es­say that was Richard Bonynge's ear­nest early attempt at correct perfor­mance practice (London/Decca, not
yet on CD).

The authenticity in this case ex-

tends to the ambience (a rather dry-

sounding Abbey Road Studio 1), the

employment of unwritten, but stylis­tically mandatory, appoggiaturas by
the singers, and a true 18th-century
operatic seating arrangement for the
orchestra (and not just a version of
the 19th-century/divided-violins lay­out). A seating diagram is helpfully
given in the accompanying booklet,
which also describes the simple CD-

player programming to use to make

the discs play either the original
(Prague) version or Mozart's own re­vised (Vienna) version. Music is pro­vided for both.

Most surprising to most listeners

will again be Norrington's tempos,

which are usually—but not al­ways—faster than normal. The most
notable exception is the so-called
Champagne Aria ("Finch'han dal

vino," which mentions only wine).
[Champagne is a wine but it did not
exist in Don Juan's early-17th­century Seville. Amontillado Aria
would probably be more appropri­ate.—Ed.] For once it is taken at a

tempo that doesn't force the singer to
bark out his notes and gains thereby
an extra dimension of snarling lasciv­iousness.

Otherwise, the drama sweeps
along at a theatrical—not "operat­ic"—pace, with possibly too few
moments of repose. There certainly
is little very soft singing, even when
it might be appropriate dramatically
and sanctioned by the score. While
vocally there is no weak cast mem-

ber, characterization during the arias

and ensembles is shortchanged by all
the singers. A little more personality
from them would have helped inte-

grate these pieces more fully into the
rapidly moving recitatives.

Whatever one may think of these

points, there can be little disagree-

64

ment that Norrington is extraordinar-

ily successful at choosing tempos and
manipulating dynamic levels during
the condemnation scene, the most
hair-raising music of the entire 18th
century. The only problem here is
the sound, which for once fails to de­liver: the chorus of demons sounds
like panned-in mono and is not well
integrated into the sonic picture
(maybe because they're supposed to
sound "from below"), and there is a
noticeable lack of bass here. And I

do think that the last scene (the post-

mortem ensemble) follows on too
quickly. Not even the quick-acting
stage machinery of a Baroque theater

could have changed scenery so rapid­ly, nor could the singers get onstage

so fast.

These points are minor. Because
of the opera's status as a cultural icon
and the overall quality of the perfor­mance, this recording is required lis­tening for all Mozart lovers. It will
be a cleansing earwash to those
brought up on a "traditional" Don

Giovanni.

Orchestral

Felix Mendelssohn: Symphony No. 3 in
A Minor, Op. 56 ("Scottish"); Symphony
No. 4 in A Major, Op. 90 ( "Italian "). The

Chamber Orchestra of Europe, Nikolaus

Harnoncourt, conductor. Teldec 9031­72308-2.

This 1992 Harnoncourt disc

again features the iconoclastic con­ductor leading The Chamber Orches­tra of Europe. The ensemble uses
modern instruments but produces a
very unmodern sound. The strings
are lean and sinewy rather than lush
and "romantic," a consequence of
vibratoless, and probably gut­stringed, playing. The woodwinds
are also more prominent than usual,
as can be heard at the very opening
of the "Scottish." Harnoncourt has
throughout replaced traditional ro­mantic soupiness with meticulous ob­servation of Mendelssohn's specific
dynamic inflections (such as swells
and various types of accents).

But do not think of these perfor-

mances as attempts at a historical re­construction of a mid-Romantic or­chestra. At the very least, the modern
string seating is demonstrably incor-

rect for Mendelssohn (Norrington on
EMI 7 54000 2 is more authentic in
this and many other areas). These
performances are instead typical idio­syncratic Harnoncourt, who is turn­ing into sort of a modern-day Men­gelberg replete with arbitrary and
bizarre tempo changes. Cases in
point are his distinct slowdown in
the first movement of the "Scottish"
at the final subject of the exposition
([1] 5:48, less marked at 8:52), and
his near cessation of the pulse at the
end of the exposition ([1] 6:30) be­fore he takes the repeat. Neither of
these Wagnerian "nuances" are spe­cified in the score; Mendelssohn
would have been horrified. Harnon-

court has allowed a few moments of

ugly playing to stand as well, like the
screeching entrance of the violins in
the fourth movement ([4] 0:03). But,
also like Mengelberg, he can produce
some wonderful effects, like the un­usually tense introduction of the
"Scottish" and the glorious cello line
during that symphony's first­movement coda ([1] 12:03).

The sound quality for the "Ital­ian" seems brighter than that for the
"Scottish," a difference suiting the
two works' contrasting characters.

However, Harnoncourt's sometimes
exaggerated dynamic inflections of
the melodic line tend to lose the last
notes of phrases. And despite the
clearer sound quality of the "Ital­ian," the inner lines are often ob­scured, even on headphones. I also
found the second movement a bit too
slow and the finale rather lackluster,
even though Harnoncourt's attention
to Mendelssohn's dynamics shifts the
musical emphases to unusual places.
On the whole, this release is a Har­noncourt-style disappointment: "al­ternative" interpretations worthy of
hearing if not treasuring.

Joseph Haydn: Paris Symphonies (Nos.

82-87). Sinfonietta de Montreal, Charles

Dutoit, conductor. London 436 739-2.

What a difference a year makes!
The first disc of this two-disc set,
containing Nos. 82-84, was record­ed in Montreal's St. Eustache in May
of 1990. A year later, the remaining
symphonies were recorded in the
same church, with the same producer
and engineer, but with a totally dif-

THE AUDIO CRITIC

pdf 58

ferent musical/sonic effect.

The earlier sessions suffer from
being recorded at medium distance in
what is obviously a very reverberant
acoustic that has been used with rath­er consistent success for Dutoit's
London recordings with the larger
Orchestre symphonique de Montréal.
While the reverb promotes a certain
lushness of string tone, it also tends
to obscure many details, such as the
woodwinds, and to smooth out the
nuances of phrasing that are essen­tial for Classical-period music. Large
contrasts in dynamics are preserved,
smaller ones—like the many sforzan­di ("forced," heavily emphasized
notes) sprinkled throughout these
scores—are not. There's also a slight
bass heaviness.

These effects have affected Du­toit's interpretations, which on the
first disc are underarticulated and
slightly slack. The best performance
on the first disc is of No. 84, which
features some marvelous woodwind
solos in the second movement and,
for a change, some verve in the min­uet.

Disc 2 is a different story: the
sonic distance to the orchestra has
been reduced and the musical lines
have gained thereby in clarity and
vividness. The orchestra sounds more
alert to Dutoit's—and Haydn's—
dynamic nuances. Even the tempos
seem more spirited. No. 87 in partic­ular is distinguished by its woodwind
solos, even though the sound is now
close enough so that you can hear
Dutoit's grunts in the first movement.
This second disc makes the entire set
of these marvelous pieces worth con­sidering.

Miscellaneous performance notes:
Modern instruments are used in the
modern string-seating plan (all vio­lins left, cellos and violas right).
Only the first half (the exposition) of
each first movement is repeated, al­though Haydn has indicated that the
second half (the development and re­capitulation) of those movements
should be repeated as well (except
for No. 86). The same goes for those
symphonies whose last movements
also have a second-half repeat (i.e.,
all except for Nos. 85 and 86). I
know of no two-disc sets of these
symphonies that take these rarely ob­served repeats. If they were all taken,

ISSUE NO. 21 • SPRING 1994

the playing-time limits of the CD for­mat would be exceeded. There's a
tentative-sounding restart after a
grand pause in No. 87's first move­ment ([9] 4:40) and, finally, Dutoit or
his tape editors missed the fermata
(a hold) during the pause separating
the exposition from the development
in the last movement ([12] 1:19 and
2:29). Haydn intended this and simi­lar pauses to desynchronize the lis­tener's internal metronome, a trick he
liked to pull.

Hector Berlioz: Symphonie fantastique.
Orchestre révolutionnaire et romantique,
John Eliot Gardiner, conductor. Philips
videodisc 440 070 254-1

Hector Berlioz: Symphonie fantastique;

Overture to Le Corsaire; Overture to Le
Carnaval romain; "Danse des sylphes"
and "Marche hongroise" (from La Dam­nation de Faust). L'Orchestre de la
Suisse Romande, Ernest Ansermet, con­ductor. London 433 713-2.

The Philips live-performance
video fascinates mainly by its depic­tion of a reconstruction of an early­19th-century orchestra playing in the
very hall where this groundbreaking
work had its first performances (the
old hall at the Paris Conservatoire).
Talk about authentic ambience! Only
the 1830 costumes are lacking. But
that would have prevented the video
director (Barrie Gavin) from featur­ing, as he apparently does, his favor­ite sleeveless-dress bass player. (One

does get tired of seeing the same fac­es, many of them not very interest-

ing, over and over again, as if the

cameras were at locations with re-

stricted views.) There are a few shots
of the wrong instruments at the
wrong time, and the harps appear
out of nowhere for the second move­ment. Gardiner uses six of them,
amply fulfilling the composer's re­quirement of "at least two" per part.

It makes a substantial difference

compared with the paltry pair you
get in most performances.

By far the most important direc­torial fault is the lack of early "estab­lishing" shots of the orchestra's
overall placement within the auditori­um, and of the size and shape of the
auditorium. Although one could
figure it out from the short reverb
time, especially in the bass, the

smallness of the hall doesn't become
visually apparent until the final cred­its are rolling, when the cameras
swing around to show the audience.
The video does show a Berliozian or­chestral layout, which itself is as fas­cinating as the use of a serpent­shaped ophicleide in the last two
movements and of an angled-tube
cor anglais (usually translated into
"English" horn) in the third. The vis­ual direction of the final pages of the
work is very well handled, too.

The size of the hall is an impor­tant acoustic factor in the success of
Gardiner's performance, which takes
tempos very close to Berlioz's metro­nome markings and in which, thanks

to the use of period instruments,
Berlioz's written-in orchestral bal-

ances are inherently produced. The
"eagle-eared" will also note with

pleasure how easily the old wind in-

struments seem to perform Berlioz's
spooky portamentos (pitch-to-pitch

glides) shortly after the beginning of
the last movement. Nice, portentous,
offstage bells in the finale.

Philips also has a CD version of
this performance (434 402-2). It uses
different takes—it can't be synchron­ized with the videodisc—and there's
no audience applause at the end. The
sound and performance quality re­main the basically unaltered.

As an example of how this work
used to be played, there are few ver­sions better than the newly reissued
Ansermet performance with its mod­ern instruments, modern seating, two
harps, and post-Romantic conduct­ing. You get to hear the modern
French woodwinds and their deli­ciously saxlike bassoons, as well as
some unusual-sounding bells custom­made for this recording's 5th move­ment. The sound—discreetly multi­miked—is excellent for its time
(1967), as it is in the filler works re­corded in 1964. You can even hear a
vehicle outside the hall at [3] 0:30­0:40. Recommended as a midpriced
second performance.

Gustav Mahler: Symphony No. 5. Dallas
Symphony Orchestra, Andrew Litton,
conductor. Dorian DOR-90193.

Andrew Litton's "interpretation"
of this work—I use the word only for
convenience—is here undone by the

65

pdf 59

recording quality, which is inferior to
that of most of the Mahler Fifths re-

leased over the last, say, 15 years.
Even some recent bargain-basement
readings of the piece have served the

music better. The principal fault is a

lack of "heft" to the sound—it is re-

corded at a far-too-distant perspec-

tive. There's low end in abundance,
but no body to the strings, no pres­ence to the woodwinds, and no spar­kle to the percussion. All you seem
to hear are laser-beam trumpets
piercing through the echoey haze.
Turning the volume some 10 dB
higher than where you would normal­ly have it helps, but not much. I can
only hope the acclaimed McDermott
Hall where this was recorded doesn't
actually sound this bad in person.

[My observation has been that

Craig Dory usually gets it right the

first time in a new recording venue,

as he did in this hall with the Stravin­sky Sacre, and then becomes bored
or restless and starts experimenting
with more and more ambience. Even
that first recording was miked about
as distantly as you would ever want,
and now he has gone too far.—Ed.]

Young conductor Litton doesn't
yet seem to have a total grasp of the
Mahlerian idiom. He almost brings
the middle of the third movement (a
scherzo) to a dead stop, and there is
no underlying rhythmic pulse (slow
as it could be) to the fourth-

movement Adagietto, which also fea-

tures some steely-sounding strings.

Among the nuances of tempo that he
simply gets wrong is the detectable
speedup following measure 338 in
the last movement ([5] 6:19), in di­rect contradiction to the composer's
explicit instruction not to hurry (nicht
eilen), a sin I consider quite serious.
But by this point it hardly mattered,
since the recording quality lost me at
the opening of the first movement.
Although this is billed as a live re­cording, the audience is very quiet
until the applause at the end. Maybe
they were all asleep.

Ralph Vaughan Williams: Symphony

No. 7 ("Sinfonia antartica"); Fantasia
on a Theme by Thomas Tallis. Indianapo­lis Symphony Orchestra, Raymond Lep-

pard, conductor. Koss Classics KC-2214.

Following the dubious lead of

66

André Previn's RCA/BMG recording
of the Antartica (the composer's
spelling), in which Sir Ralph Rich­ardson read the motto-epigrams with
which the composer prefaced each
movement in the score, this record­ing uses passages from Captain Rob­ert Scott's Antarctic journals and in­serts them into the music. (In the
published score, the texts actually all

appear on a single page before the
music, implying that they are best
used in mood-setting program notes
or only as interpretive guides to the
performers.) The narrated Scott texts
(as well as a contribution from Cardi­nal Newman, "Lead, kindly light,"
that appears out of nowhere), while
appropriate for the general mood of
the piece, always interrupt the musi­cal argument, ruining any coherence
the performance may have had. This
is particularly noticeable when the
music is allowed to come to a com­plete stop, without even a held back­ground chord. The Antartica is not a
suite of film music, although some of
its themes derive from the compos-

er's score to the 1948 docudrama

Scott of the Antarctic. It is a sympho­ny. It is also not Peter and the Wolf,

nor Tubby the Tuba, two works that

unfortunately sprang immediately to
mind when listening to this record­ing. I'm amazed that the whole mis­guided venture was the conductor's
idea!

Too bad, because the sound
quality (excluding the lack of lower
frequencies in the dubbed-in narra­tion) is very good and quite probably
the best this piece has ever received.
The occasional huge outbursts come
off particularly well. I hope that Koss
Classics saves what might actually be
an excellent performance by editing
out the narrator's contributions alto­gether. They can then also help the
composer by performing a slow elec­tronic fade-to-zero at the end of the
work where Vaughan-Williams wrote
a decrescendo to niente. As it stands,
you can still hear the performers ac­tually stop.

The famous Fantasia receives
no extramusical accretions, though it
would have been thoughtful for Koss
to have had the choir used during the
symphony perform the original Tallis
theme as a preface. But here the con­cert-hall recording quality is less suit-

ed to the music, which was written
for Gloucester cathedral. The work
divides a string orchestra into three

sections: a large body of strings, a
string quartet drawn from that body,

and a small orchestra meant to be

spatially separated from the other
two sections. Here that separation
takes the form only of an increased
right-channel presence with no addi­tional distancing via ambience. The
whole ensemble is a bit too closely
recorded for my taste (you can hear
somebody's almost asthmatic breath-

ing at [6] 6:54-7:03), though this
does suit Leppard's passionate, but
unrelaxed and nonmystical, interpre­tation. Best used a massed-strings
test track.

Ottorino Respighi: Roman Festivals;
Brazilian Impressions; Pines of Rome.
Dallas Symphony Orchestra, Eduardo

Mata, conductor. Dorian DOR-90I82.

Ottorino Respighi: Church Windows;

Brazilian Impressions; Roman Festivals.

Cincinnati Symphony Orchestra, Jesús

López-Cobos, conductor. Telarc CD-80356.

My principal interest here was in
Roman Festivals, the last-composed
of Respighi's Roman trilogy. It's too
bad that the other two pieces of the
trilogy (Pines of Rome and Fountains
of Rome) are more frequently record­ed because Festivals is by far the
most sonically spectacular of the
three. As music, it could well be the
soundtrack for a Fellini film of Ben
Hur Meets Petrushka. But as a dem­onstration piece, few works can
match its orchestral forces or the
range of massive and delicate effects
they can produce.

For the record—and since nei­ther CD's booklet prints a list (which
should be required for all orchestral

recordings)—Respighi's score spe­cifies: 2 flutes, 1 flute doubling on
piccolo, 2 oboes, English horn, pic­colo clarinet, 2 "regular" clarinets,
bass clarinet, 2 bassoons, contrabas-

soon, 4 horns, 4 trumpets, 2 trom-

bones, bass trombone, tuba, timpani,
tambourine, ratchet, sleigh bells, side
drum, tenor drum, triangle, cymbals,

bass drum with attached cymbals,
tam-tam, glockenspiel, 2 bells, xylo-

phone, 2 tavolettas (defined by the

Grove dictionary as "a board or table
struck with a hammer," but the Telarc

THE AUDIO CRITIC

pdf 60

disc seems to use standard wood
blocks, soon after the beginning of
the last movement), piano (2 players),
pipe organ, mandolin, first and sec­ond violins, violas, cellos, and dou­ble basses. (And you wonder why
multimiking has such appeal for
some record producers!) Three bucci­nas—ancient Roman brass instru­ments—are also called for at the be­ginning of the piece. But three
trumpets, the composer-suggested sub-

stitutions, are usually used instead,
probably because nobody seems to
know exactly what a buccina was.
Valveless baroque trumpets may ac­tually make a better substitute.

A good recording would enable
you to hear every one of these instru­ments' contributions when they can
reasonably be expected to be heard.
That doesn't happen with the Dorian
disc, which is saddled with virtually
the same "heftless" sound quality as
their recent Mahler recording (see re­view above). On the Respighi disc,
the measured reverb time is nearly 3
seconds, which is excessive for virtu­ally all orchestral music except possi­bly Bruckner's. It certainly does no
good to Festivals, which is lost in
the no-impact haze. [Clean, natural
haze, though.—Ed.] All that sticks
out is the brass, including those trum­pets-cum-buccinas, and some incred­ibly scrawny violins ([2] 2:45 and
4:14). That's just as well, since none
of Mata's sleepy performances would
otherwise recommend this disc. Look
to Telarc instead.

Their disc is yet another bass­drum special and easily one of Tel­arc's most spectacular recordings in
their Grammy-studded history of en­gineering achievements. The sound
scores highly in heft, impact, power,
clarity, tonal balance, instrumental

balance, realistic imaging in depth,

and an appropriate ambience (2­second reverb time). There are solid
musical values here also, López­Cobos managing the many tempo
changes in Festivals as smoothly as
did Toscanini (the work's first con­ductor). Cue to Track 8 and let 'er
rip. Ave Telarc, those whose ampli­fiers are about to go into clipping te

salutant...

Peter Ilyich Tchaikovsky: Symphonies

No. 4, 5, and 6. St. Petersburg Philhar-

ISSUE NO. 21 • SPRING 1994

monic, Yuri Temirkanov, conductor.
RCA/BMG 090226-61377.

Think of the claims for authen­ticity that can be made for this re­cording: Russian music, Russian con­ductor, Russian hall (possibly with
"original" acoustics), and a Russian
orchestra using—wonder of won­ders—the "old" 19th-century seating
plan for which these works were
written. That plan has the first vio­lins, cellos, and basses on the left,
and the second violins and violas on
the right. It results in an "opening" of
the string texture, in addition to re-

producing any antiphonal left-right

effects the composer may have in­tended. Luckily, this has all been pre­served in completely low-noise,
low-distortion Western sound that

does not artificially fill in the rather

short reverb time.

The performances are excellent,
overall. Temirkanov often makes ex­aggerated and/or unmarked tempo

changes—as is customary but not

necessarily authentic Late Romantic
style—that became exasperating dur­ing the first two movements of the
5th and the first movement of the 6th.

He is also a conductor who prefers
the last movement of the 6th to be
played in slow motion. He takes 12

minutes for a movement that usually

takes around 9 to 10 minutes if Tchai-

kovsky's metronome markings are

even vaguely followed. (But that's

nothing compared with Bernstein's
achingly glacial, 17-minute-plus read-

ing in his Deutsche Grammophon re­cording.) Even so, close as he might
come, Temirkanov never completely
drops the "line," a feeling of rhyth­mic continuity that is difficult to con-

vey without being able to see his
"beat." And the Russianness of the

horns and brass, the sensuousness of
the woodwinds, the clarity of the

string writing, and Temirkanov's ul-

trascrupulous observance of Tchai-

kovsky's often-ignored accentuation
and phrasing instructions were more

than enough to sustain my interest.

Highly recommended.

Instrumental

The Renaissance Lute [music from
around 1500 to 1600]. Ronn McFarlane,
lute. Dorian DOR-90186.

This is the first solo album of
McFarlane's that I've heard, and if
his other solo discs are half as well
done as this one, I'd have no hesita­tion at recommending the lot. Here,
the excellent sound quality is of sec­ondary consideration (just don't play
it too loud, lutes are inherently soft).
More important is how McFarlane's
musicianship and virtuosity show up:
impeccable phrasing, effortless orna­mentation, spot-on tempos, and in­credibly variegated tonal colors
coaxed out of temperamental instru­ments. That musicianship also shows
in the sequencing of the disc, which
maintains interest despite having lots
of short pieces that weren't intended
for such massive doses. Even McFar­lane's program notes leave us want­ing more. Extraordinary.

Britteniana

Benjamin Britten: Peter Grimes. Chorus

& Orchestra of the Royal Opera House,
Covent Garden; Bernard Haitink, con-

ductor. EMI CDCB 54832.
Benjamin Britten: A Midsummer Night's
Dream. City of London Sinfonia, Richard
Hickox, conductor. Virgin 7 59305-2.
Benjamin Britten: Gloriana. Orchestra
and Chorus of the Welsh National Opera,
Sir Charles Mackerras, conductor. Argo
440 213-2.

Peter Grimes has fared well in

performance and sound on its three
recordings, starting from the first
with the composer conducting (on
London 414 577-2). All have been
models of state-of-the-art sound for

when they were recorded. This EMI

production—one of the best-sounding
opera recordings ever—is no excep­tion. The clarity of the orchestral
pickup is of demonstration quality,
yet the voices are well integrated at
an appropriate distance into the image.
Obtaining this type of voice/orches­tra balance, where neither obscures
or overpowers the other, is perhaps
the most difficult task of its kind in

all of classical music.

The performance has much to

recommend it as well, starting with

Haitink's surefire pacing and master­ly control of the large orchestra. Vo­cally, this is the best-sung of all the
Grimes recordings. The only problem
is with the title role, sung by Antho-

67

pdf 61

ny Rolfe-Johnson. His voice is clear-

ly more beautiful than that of either
of his predecessors (Peter Pears and
Jon Vickers) and he sings accurately

enough. But he doesn't convey, in to-

nal shading or vocal pitch inflections,

quite the intense anguish that Pears

and especially Vickers brought to the
role. A little ugliness may have
helped. Highly recommended none­theless.

Also recommendable, though
less sonically faultless, is the second­ever recording of A Midsummer

Night's Dream. While the pickup of
voices and chamber orchestra (with
expanded percussion section) is gen­erally excellent, the engineers seemed
to have had trouble balancing the
countertenor (James Bowman, who
plays Oberon). This seems to be a
voice type that is unusually difficult
to balance, as this problem occurs on
other countertenor discs. (Or maybe
he just wanted to be too unatmo­spherically loud.) Also, when Puck
rushes in, he sometimes comes on
stomping, distractingly. But the piece

is marvelous, particularly the varied
colors Britten obtains from the small
orchestra.

Returning to a more massive

scale, Britten's opera written for the
coronation of Elizabeth II is about
Elizabeth I (Gloriana). The work re­ceives its first audio recording with
the Argo release. I've seen it onstage
(the production I attended, by the En­glish National Opera, is available on
videotape), and it is a grand piece of
musical theater with a controversial
ending that has Elizabeth speaking—
instead of singing—what I believe
are words of the historical Elizabeth
I. As staged by the ENO, this pro­duced a thrilling effect as the Queen
stepped out of her "artificial" operat­ic world and, as a "real" person, re­gally yet lovingly addressed us di­rectly, we who had now become her
subjects. Unforgettable.

Argo's recording mishandles
this theatrical coup. It breaks the in­theater sonic illusion it had main­tained up to that point, by having

Josephine Barstow speak Elizabeth's

words in a close-up conversational

tone. Instead of grand rhetoric we get
a Barbara Walters interview. It is too
intimate and incapable of creating the
grand, tragic effect of a more theatri-

68

cal delivery. Forewarned, you can
now safely explore the beauties of
the rest of the music, which are con­siderable (especially the writing for
the chorus).

Unfortunately, the recording qual-

ity does not come up to the very high

standards set by the other two sets.
The hall is somewhat boxy-sounding,
the important choral pickup has indi­vidual voices sticking disharmoni­ously out, and an onstage procession
in Act HI is virtually panned mono.
While excellent actors, both the main
characters, Elizabeth and the Earl of
Essex (Philip Langridge), have pitch

problems with the inner notes of
phrases, and Essex's voice often los-

es the sensation of specific pitch (his

overtones don't line up with his fun­damentals). But this opera is likely

not to be recorded again for decades
and is therefore self-recommending

to anyone interested in Britten's mu-

sic.

* * *

Editor's choice of recent releases:

Béla Bartók: Sonata for 2 Pianos & Per­cussion (also works by Ernst von

Dohnányi and Zoltán Kodály). Artists

and guests of the Chamber Music Society
of Lincoln Center. Delos DE 3151.

One of the great Bartók master-

pieces, very effectively performed

and sensationally recorded. Did
someone say bass drum? You just
wait! The Dohnányi and Kodály

pieces are also far from negligible.

•

Gustav Hoist: The Planets. Royal Phil­harmonic Orchestra, James Judd, con­ductor. Denon CO-75076.

Possibly the best-phrased, most
musical performance of The Planets
in the digital era. The sound is more
loudspeaker-sensitive than usual; on
a super system it's quite wonderful.

•

W. A. Mozart: Symphonies No. 35, 36,

38, 39, 40, and 41. Anton Webern:

Works for orchestra. The Cleveland Or-

chestra, Christoph von Dohnányi, con­ductor. London 436 421-2.

Intermingling the greatest sym­phonies of Mozart with the orchestral
cameos of Webern in a three-CD set
may be an off-the-wall idea, but I
don't care. The Cleveland is my fa­vorite orchestra, arguably the most
virtuosic of them all; Dohnányi's per-

formances combine heart and intel­lect in equal proportions; the John

Pellowe recordings possess both

warmth and transparency. It hardly
ever gets much better than this.

•

Igor Stravinsky: The Composer, Vol. Ill,
TV, and V. The Orchestra of St. Luke's &

The Gregg Smith Singers, Robert Craft,

conductor. MusicMasters Classics.

Robert Craft continues his high­ly authoritative traversal of the com­plete works of Stravinsky with
world-class orchestra players, good
singers, and flawless recording. Ev­ery Stravinskyite needs the whole set.

Ludwig van Beethoven: "The Complete
Sonatas." Richard Goode, piano. Elektra
Nonesuch 9 79328-2.

A monumental ten-CD set by the

Artur Schnabel of our time (in my
humble opinion, at any rate). Pro­found, spellbinding performances of
the utmost musicality, recorded by
Max Wilcox in unexaggerated, natu­ral, just-right sound. What a treasure!

•

Giuseppe Verdi: La Traviata. Edita Gru­berova, Neil Shicoff, Giorgio Zancanaro;
London Symphony Orchestra, Carlo Riz­zi, conductor. Teldec 9031-76348-2.

A sleeper. Beautifully intimate

performance that suits the character
of this essentially three-singer opera.
Highly rehearsed, polished studio re­cording, fine singing, excellent sound.

•

Richard Wagner: Das Rheingold; Die

Walküre. Recorded live at the Bayreuth
Festival. Bayreuth Festival Orchestra,

Daniel Barenboim, conductor. Teldec
4509-91185-2 and 91186-2; Teldec Vid-
eo 4509-91122-6 and 91123-6.

The first two sets of CDs and

videodiscs of the complete Baren­boim/Bayreuth Ring. The coming
Siegfried and Götterdämmerung need
to be no more than equally good to
make this one of the truly important
cycles in the recorded Wagner canon.
Barenboim has seldom been this
good; he works with excellent sing­ing actors though no great voices; the
orchestra is fabulous; and the very
accurately recorded Bayreuth acous­tic is clearly the only 100% right one
for the Ring. The Unitel video pro­duction (same sound track) proves
Harry Kupfer's staging to be a gro­tesque, self-indulgent travesty. Ugh!•

THE AUDIO CRITIC

pdf 62