Aragon Answers Back
Dear Editor:
Hey, I've got an idea. Let's start a magazine where you review audio components with a faulty test setup and discontinued speakers engineered a quarter of a century ago. What speaker wire and interconnects should we use? Who cares, we'll just go to a hardware store and get some. Now comes the fun part:
Consumers will read the reviews and actually spend their money based on what we say.
What! Audio Magazine is already doing this? Boy, you no sooner come up with a good practical joke than you find out someone else is already doing it.
I felt compelled to write this letter to compliment you on your comic genius. Especially the subtlety of it. Some people would only do it in the April issue, but that's too obvious. You do it in the June issue, with the Aragon 8008ST amplifier review. I particularly liked where you mention the Allison One was the speaker used for the review but don't say how old they are. That will make your readers go into the Audio Annual Equipment Directory to try to find some thing about them, but they won't be in there. Even if they saved previous Directory issues for several years, they won't find the Allison One speakers listed because they were discontinued 10 years ago. It will make your readers go crazy trying to figure this out. I can hear them now, "Maybe it's a new speaker that's not out yet." Very few will suspect that it came out about a quarter century ago.
It was bad luck that the Aragon 8008ST is also reviewed in the June issue of Stereophile.
I think people may catch on to the joke. After all, Stereophile also used an Audio Precision measurement system to test the amplifier, except their Audio Precision was working properly. Readers will see that Stereophile didn't have a problem testing the Aragon and might figure out that your reviewer's Audio Precision test setup has a problem that the wide-bandwidth 8008ST makes evident.
Also, as luck would have it, Stereophile's reviewer didn't listen to the amp on a speaker designed 25 years ago but on the newly re leased Energy Veritas v2.8. Tom Norton found that on this current design the Aragon "produced one of the best sounding top ends overall." Because of this, people may start asking questions, so allow me to prepare you with some suggested answers.
Q. We commend you for finally revealing what Ed Foster listens to, but why is it that in previous reviews by Foster, you never gave what associated equipment he listened to during a product review?
A. We weren't trying to hide anything; we just don't think anyone else needed to know because it doesn't matter.
A. Hey, we're the experts, and we say there hasn't been any progress in speakers in the last 10 years. As a matter of fact, these speakers were engineered about 25 years ago, and there hasn't been any progress since then.
Q. Don't you think consumers who trusted Audio will be upset when they find out?
A. It doesn't matter, because we're changing the name of our magazine to Clueless.
Then we're getting all new test equipment and speakers and stuff, and our magazine will have the motto "Clueless, But With New Equipment."
Q. How could the editor let this go on?
A. Don't take things so seriously. Besides, I've been editor for only six months. I ordered new equipment from one of those 800 numbers in the back of Audio. They told me they could get me any brand I wanted. Don't blame me if six months later they still haven't delivered anything. They tell me the manufacturer is back-ordered.
Q. What is the date of the last receipt your reviewer has for sending his test equipment back to Audio Precision to be calibrated?
A. Sorry, editor, I couldn't come up with an answer for this question. You are on your own on this one.
Anthony Federici, Mondial Designs, Dobbs Ferry, N.Y.
Author's Reply:
That's funny. I don't remember Tony Federici disparaging my speakers, cables, or test setup when I reviewed the Mondial Acurus 200X3 in the February 1995 issue. But more about speakers and cable later. I'm more bothered by Federici's ground less (pun intended!) attack on my test setup.
When the first 8008ST blew a rail fuse, I called Mondial to discuss the amplifier's grounding arrangement because I suspected oscillation. Mondial questioned whether the outputs had been connected to a common ground, since that could cause the 8008ST to oscillate. That was not the case; the loads were independent and the levels were read by the balanced input amplifier of the Audio Precision System One. Mondial said it did not have AP test equipment in its design lab, but the factory used a System One for quality control and had no problem. I gave Mondial the specific settings I use; it was confirmed that the factory set tings were the same as mine.
When I remarked about the 8008ST's wide bandwidth and opined that it did not seem to have an input filter, I was told there was a filter and response should be-3 dB at 100 kHz. My response data indicated otherwise, and I suggested that perhaps the filter was missing on this sample, which might explain the oscillation. I shipped the amp back to Mondial, and, to check that nothing had gone awry with my setup, I put a different wideband, high-power amplifier on the bench; it behaved normally in all tests.
Mondial sent a second sample. When it also acted strangely, I was back on the phone.
I advised Mondial that there was no 100-kHz filter in this one either and was told the cut off wasn't really at 100 kHz. So, where was it? Well, I was told my results were probably correct, whatever they were. Hmm! I recon firmed that the factory did use an AP System One, that it was set the same as mine, and that they did not have similar problems.
(This is significant because, if oscillation were induced by the test equipment, the factory should experience similar results.) I asked about the test cables used in the factory, since shield grounding can affect amplifier stability. My cables are designed to avoid such problems, but I suggested the factory send me the cables it uses so I could duplicate their setup precisely. I was told they didn't use anything fancy, "just a twisted pair." At this point, Mondial said it would get a System One on loan and try to duplicate my results in its lab. I put the 8008ST aside. A few weeks later Mondial called back saying it had the answer: The problem was Audio Precision's. Mondial had learned that Audio Precision uses small RF bypass capacitors from the balanced inputs to the chassis. In Mondial's opinion, this provided a feedback path from output to input and caused the 8008ST to oscillate. (I'll not recite the attack on AP in detail; suffice it to say that "stupid" was one of the milder epithets.) Mondial had not gotten a System One and tried the tests itself, however.
Well, admittedly, this is a possible explanation, but I don't buy it. RF bypassing both sides of a balanced input to chassis is good engineering practice to ensure high-frequency common-mode rejection. Furthermore, tying the AP chassis to the 8008ST chassis with a braided cable (as I did) grounds the presumed feedback path and should eliminate the feedback. Finally, if the oscillation were caused by the AP System One, why didn't the factory have a problem, since it uses the same equipment? I tried different connection arrangements between the second 8008ST and the test gear and came up with one (described in the re view) that gave reasonably consistent results.
I gave a few preliminary measurements to Mondial's designer, who agreed that they "seemed about right," so I completed the lab work and went to the listening tests. I approached the listening tests under the assumption that the problems were peculiar to testing and would not occur in a music sys tem, but I did not cotton to the sound of the 8008ST, as is evident in my review.
Federici was less than pleased. In an effort to convince Audio to print a retraction, he came up with another explanation of the bench-test problems. Again, it was Audio Precision's fault. The factory had found that some System Ones had a loose ground connection on the power cord and that if an 8008ST were hooked up to one of those it would do exactly what my sample had done when I first turned it on. Well, I opened up my System One, and the ground connection was solid as a rock. Not that a loose connection there would likely cause the problem anyway, since power-line ground is not effective at radio frequencies; it's there for safety reasons.
So Federici wrote a letter and forced this reply. I am reluctant to comment on another reviewer's measurements, but when Federici says "[Stereophile's] Audio Precision was working properly.. .[and that]...Stereophile didn't have a problem testing the Aragon," I'm left no choice. Obviously, I wasn't pre sent at Stereophile's test, so I can't be sure, but I find its published distortion curves on the Aragon 8008ST peculiar. Figure 9 (distortion versus output) shows abrupt shifts in "distortion," which suggests that Stereophile's System One was responding similarly to mine. Apparently their tester didn't question the validity of the measurement; I did.
Federici goes on to say that Stereophile re viewer Tom Norton found that the Aragon "produced one of the best sounding top ends overall." Federici does not tell you that this was Tom's "first reaction" and that Tom goes on to say he "would not describe the Aragon as sounding “sweet"' and that "the very top could turn slightly dry at times, with a trace of added zip." Other comments in the review could be interpreted as suggesting something was not quite right with the 8008ST's top end--Federici was particularly upset about my calling it "bright"-but I'll readily agree that Tom's review was far more positive than mine. I respect Tom Norton's integrity, but we do not listen to the same type of music nor do we necessarily listen for the same things or in the same way.
This brings us to listening, which, given the limitations imposed on reviewers, is necessarily subjective. If one is to impose any "science" on listening tests, one must try to apply the basic rules of scientific experimentation, one of which is to lock down all variables except the one under test. When re viewing a power amplifier, this means all other components in the system must re main the same and be well known to the ob server.
If the speakers are changed from review to review, you don't know whether any difference you hear is caused by the amplifier, the speaker, or the interface. The most that can be said is that one combination sounds better (or worse) than another. Since the sound character of loudspeakers varies more than that of other components, I believe it is imperative that the speaker be maintained constant in the evaluation. That's why I use Allison Ones. They're not perfect, but I'm utterly familiar with their strengths and weaknesses. To me, on the types of music I use to judge other equipment, they reveal differences remarkably well.
I'm looking to replace the Ones-not so much because the design is old (they're still good speakers) but because, eventually, they're going to leave me for the great be yond. Then I'll need replacements, but until I've lived with a speaker for many months, I don't consider it fair to review other equipment with it. So I'm looking. When I find speakers as self-effacing yet revealing (to me!) as the Allison Ones, they'll go. So far, I haven't.
Regarding speaker cable, if the Aragon 8008ST requires magic wire to function well, may I respectfully suggest it be packed with the amplifier. Otherwise, I'll use low-resistance, low-inductance, low-capacitance copper wire through which electrons flow in both directions! Mine came from a major electronics distributor who sells it in bulk at a sensible price.
Finally, regarding calibration, as anyone familiar with the AP System One knows, it has a self-diagnostic routine that checks its performance.
The strangest thing about this whole affair is that nowhere does Federici question my numbers. In some respects, the 8008ST per formed better on my test bench than on Stereophile's. What really bothers him is my calling his amplifier "bright." And this with a speaker that, if anything, is rather "laid back" in the treble. Oh well!
-E.J.F.
Editor's Note:
Ed Foster was elected a Fellow of the Audio Engineering Society in 1980 "for contributions to audio measurement technology." He is a member of the Board of Editors of the Journal of the Audio Engineering Society and currently serves as the AES Vice President, Eastern United States/Canada. Foster was chairman of the IHF/EIA amplifier and tape recorder measurement standards committees and was principal author of both those standards. He serves on numerous other measurement standards committees and is Deputy Technical Advisor to the United States National Committee to the IEC, the international standards-setting organization. He holds advanced degrees in Physics and Business Administration.
-M.R.
Professin'
Dear Editor:
How can D. W. Fostle's "objective" comparisons of digital systems ("Digital Deliverance," April) have any meaning without listening to the analog signals from which the recordings were made? Certainly he should have discovered imperfections in his all-digital master "reference" system. Others, including me, who make recordings are very familiar with the sound of microphone signals and are all too aware of digital artifacts during playback. I can only presume from editorial comments and other statements that the dull and brittle timbre, up-front sound, and space-collapsing tendencies of digital systems must be a way of life. While focusing on small artifacts, the author, not having a suitable reference, appears to be unaware that the HDCD system has removed major annoyances. He then criticizes performances closer to analog feeds as sounding different from another digital system and labels these differences incompatible.
The late Richard Heyser clearly described self-centered objectivity when he wrote: "It is a foolish person who will draw conclusions about the `audibility' of certain technical flaws in the physical reproduction based on limited `listening' tests and ignorance of the possible differences in the frames of reference." This statement appeared in Audio's March 1978 issue. The article, "Hearing vs. Measurement," should be read by all equipment reviewers.
Important engineers now operate HDCD systems for productions by top artists. These people know, use, and live by equipment chosen from exhaustive comparisons. Their observations are like my experience at recording sessions. Timbre neutrality, harmonic integrity, and balance-as well as re production of minute details within complex sound-are all second to none. Someone at Audio should have carefully listened to analog signals from tape recordings made for comparisons. He would have experienced richer, more mellow sound that is "wetter than conventional [digital] recordings." With live analog comparison, the "aggressive and obvious" limitations and sonic character of your digital "reference" system should be revealed.
A major part of HDCD system technology corrects sampling-rate consequence and achieves timbre neutrality where "traditional" systems fail badly. The author is unaware. It is a shame that the EAD and Proceed players were criticized for synthesizing mellow analog-like sounds, moving instruments back, and presenting hall ambience. Fostle's preferences may have been for dry, spoon-feed multi-mono with centered, close-miked, chronic mezzo-forte solos; however, in the "Moonglow" example from Reference Recordings, the trumpet plays loud be hind instruments and the recording was made in an auditorium. Hence, the much-criticized three-dimensional staging presented by HDCD players is correct. It matches the analog signals making the recording.
Many times at sessions and trade-show playbacks, we have interchanged conversion methods and have observed listener anxiety, attention loss, and the sense of anguish that occur from compromised digital sound.
During compromised playback, people talk more, feel head pressure, and sometimes ob serve harder attack transients and faster rhythm. Some call this crisp digital sound; we label such discordant suborder distortions "head drill." To assure complex sounds remain full-bodied and defined, Pacific Microsonics has developed unprecedented part-per-million conversion accuracy. Performance from transient attack and complex signals found in music is far better than any thing done previously. Whether decoded or not, HDCD recordings are virtually free of added artifacts causing hard sound, agitation, and incorrect rhythm perception. Pace is not exaggerated by hard brittle sound, as the author and other reviewers not using analog references have observed and some times prefer. HDCD playback, decoded or not, matches sound and rhythmic time experienced from signals making the recording. It is correct and compatible.
Technical misconceptions can be misleading. Clearly, the author has narrowed his thoughts and field of observation to limitations of companders and other noise-reduction devices that operate by frequently altering levels and equalization. HDCD A/D converters have no programming to change equalization, and they will not operate on signal levels unless program dynamics prevent an accurate fit to CD standards. Without intervention, sonic problems are virtually guar anteed when dynamics won't fit 16-bit coding. Then the only option other than catastrophic overload is to reduce recording levels, thereby subjecting more of the entire program to dither limitations. High frequencies become lost as signals drop below LSB (least significant bit) resolution limits. This fundamental shortcoming of dithering and 16-bit coding dulls the decay of room sounds and instrument harmonics.
Noise shaping, supersonic dithering, and after-the-fact processes are not going to recover lost information.
To overcome this well-known sonic short coming, HDCD systems can provide a few inaudible top peak truncations, as well as boosts to smallest signals jeopardized by dither compromise. These processes are turned on only when dynamics will not fit the CD and are never used for middle levels, as erroneously described and presented many times in the article. When HDCD processes are used, higher program levels and a boost to smallest signals restore timbre neutrality of subtle information for both de coded and compatible playback. Much literature supports inaudibility of the sparingly used HDCD high level process. Smallest-signal LSB restoration can make a more reverberant playback from very good equipment, as the author has observed. However, the less sophisticated filtering, D/A conversion, and time-base performance always found in practical equipment will truncate smallest signals and frequently add substantial hash and spike noises. The fragile live-sound illusion is damaged by such losses and distractions, which combine to reduce "wetness." A situation like this is very clearly shown in the author's three-dimensional spectrograms, Figs. 10 and 11 (on page 32 of the April issue). Unless something has been doctored, an unprocessed recording described as having correct sound, but hopefully not the "reference" one, is full of mid-scale hash and noise during final decay. 1 would be very concerned, as these distractions are many times larger than the kinds of things criticized and are probably strong enough to create speaker awareness to pull the recording forward. Add to this an inevitable dither compromise reducing ambience and decay harmonics, and you have the sound preferred by the author. (A dull, uninteresting recording comes out better.) The HDCD ex ample is virtually distortionless and will obviously sound ("aggressively"?) different.
Someone should have observed that many non-decoding machines playing back HDCD recordings sound closer to the originating analog signal than they have any right to do.
Instead, the author made a seriously wrong assumption that HDCD is a compander.
Prejudice dominates, as Fostle invalidates his technical assessment by manipulating data to show companding action where none can possibly exist. Other errors abound, ranging from misinformation and sonic speculation to other examples of flawed measurements with preconceived interpretations. Without analog comparisons, the author has focused on preserving an up-front, bright reproduction stripped of spatial cues. (Audiophiles and recording engineers usually complain about such occurrences.) Infrequent events in the musical program are given undue emphasis. Real audibility of such isolated events is unlikely, because program conditions rarely need intervention and HDCD processing's select logic prevents such rapid-fire activity.
The technological witch hunt begins by comparing room and microphone noise in a 16-bit HDCD recording with noise from a 20-bit A/D converter without input signals.
What does this show? Certainly a reader quickly glancing through the article will get a bad impression. Incorrect statements create more confusion. (Reference Recordings does not use analog tape recorders for HDCD production, and sampling rates are much, much higher than 88.2 kHz.) Clearly no inference of dither spectrum, noise, or any consequence to music reproduction is possible from this misinformation.
Spectrum plots showing variances between Sony and HDCD recordings are wrongfully interpreted as large differences in frequency response. Uncorrelated spectral shapes at upper frequencies of the "Moonglow" samples can be observed. High-frequency portions won't merge when placed on top of one another, suggesting distortions or information differences are present. One tries to avoid these uncertainties by calibrating on the most identifiable, infrequent, highest peaks from the two recordings. This works when the systems being compared have similar waveshape responses and distortion is low. Since HDCD recordings were not decoded for these plots, the tops of infrequent, largest peaks sought out for calibration purposes may be compressed. Since these events seldom occur, sound is not changed. However, reference conditions have changed, and the remaining analysis will be unpredictable. This is evident from text descriptions and observing varying high-frequency differences from one plot to another.
Clearly one event causing small changes be tween two systems is not a different frequency response throughout the recording.
Other factors concerning spectrograms should have been discussed. Large plots of dynamic range, like those shown in Fostle's article, are very sensitive to distortions and hash-like quantization noises found in older digital systems, such as the Sony 1630. In creased high-frequency energy from these distortions is revealed when strong signals have few upper harmonics or decay quickly.
It should be obvious that the HDCD system, showing less high frequencies from added distortion, should sound better. However, from the author's frame of reference the soundstage exhibited "warpage." Also, wave form response differences of anti-aliasing filters can become part of a spectrum plot.
Quick-settling HDCD filters can put more of the event-related waveform in the analysis window, while the smallest responses from other filter designs can fall outside that window and become lost. These and other small-signal changes with tape-motion variations from one setup to another might account for heavily criticized measurement variations.
Very misleading interpretations are given to high-treble piano spectra. The highest frequency, smallest-signal portion shows dither noise at the same frequencies but lower amplitude than from a converter that the author likes and documents earlier. The prototype HDCD processor making the recording didn't have gain controls, so to be safe for the live recording, I set levels approximately 5 dB below optimum. Piano plots show this, but in no way do they show changing equalization, peak limiting, or the like. One can also see in the unprocessed, 16-bit conventional recording a spike anomaly only 30 dB down from signal. Hopefully this is not a recording used for sonic comparisons.
Energy-versus-time plots might show added "wetness" for compatible playback and its removal upon decoding. HDCD and 16-bit renderings in the article look the same to me. Had someone from Audio asked us or communicated with us before this article was published, we would have suggested using logarithmic coordinates to reveal these smallest signals before and after sub-LSB preservation. A front-panel control on the HDCD converter allows the recording engineer to make that choice. I think most recordists and engineers prefer the full process.
How can peak levels from two Jimi Hendrix CD releases that were mastered by different engineers, with different equipment, with different equalization, and possibly with different analog tapes be compared? How can a peak variation of 1 or 2 dB be tween these recordings have relevance? Maybe one could find out by examining the analog signals making the recording.
Maybe in an "alternative future," sound from every serious monitoring facility will reveal a recording system whose playback will be "perfect" enough that everyone of all ages, experience, and sonic preference can forgo live analog-signal evaluation. HDCD is a step closer to this ideal. Those proposing audio DVD, and most of the professional audio industry, also know that conventionally recorded CDs are not reference-capable. The only conclusion I can draw from Fostle's article is that there was a lack of scientific method and an unwillingness to open the mind to many things others hear: Subtlety, nuance, involvement, more sense of the artist, and possibly an experience. These things are bankable. They are also very important.
--Keith O. Johnson, Technical Director, Reference Recordings, San Francisco, Cal.
Author's Reply:
Mr. Johnson entirely misses the point in talking about "analog feeds." The telling reference is not the "analog feed," but the actual sound of the instrument. Having written a 703-page book on Steinway, the family, and the piano's history and design with a specific chapter on its acoustical behavior in comparison to that of other pianos, I believe I am qualified in the matter of piano sound.
On this basis, it is my view-one substantiated by the data presented in the article and by listening tests with multiple persons skilled in the art-that HDCD, particularly undecoded, has pronounced negative effects on piano sound.
If, as Johnson claims, HDCD removes "major annoyances," it introduces others.
Specifically, HDCD falsifies the bass of a Steinway grand and, through its manipulation of low-level system gain, distorts the natural sustain and modifies the piano's tone throughout its compass. At best, the process exchanges one form of offensive small-signal behavior for another. This might be called the "grunge-for-goo-tradeoff." I suggest that anyone interested compare an HDCD recording of a Steinway to the readily available real thing and any conventional digital or analog recording. The HDCD timbral alterations, both bass and treble, as well as the amplified sustain, will be apparent, as documented in the spectrograms on page 32 of the April Audio.
The same effects are also audible on the Testmasters CD, (see May Audio, page 75), which provides a convenient comparison of several digital processes, two versions of analog, and HDCD as recorded by Johnson. In sum, the reference was not digital, but the real piano, other recording processes, and HDCD with measurements.
There was no adverse criticism of Proceed or EAD converters. Their presentation in comparison to the respected and widely-used "pro" Apogee DA-1000 converter was simply reported. The makers of the Proceed, with whom I have spoken, expressed no objection to the report, and the EAD was favor ably reviewed in the March issue of Audio, an assessment with which I generally concur.
For the record, I will state my belief that /he use of other "pro" converters, such as those from Prism, DCS, or dB Technologies, would not have materially altered the report. This is not to say these devices sound the same.
Notwithstanding any of this, those concerned should listen for themselves and, having done so, make up their own minds. The cautious will undoubtedly want to wear a Kevlar helmet to fend off the dreaded "head drill" of which Johnson warns.
Moving briefly to the concrete, HDCD reduces the gain on peaks during recording and increases it on playback. That's companding, no matter what the process's promoters say. HDCD raises low levels on recording and leaves them there, unless one of the miniscule number of expensive de coders (compared to tens of millions of conventional CD players) is available. That's envelope distortion and is exactly what was heard, measured, and reported in the article.
Johnson tells us that "program conditions rarely need intervention." His patent states that the process is doing something several times per second, "at most." Which of these statements is true? The word "equalization" does not appear in the article. Also, as noted last month, the lower, reference noise floor (red curve) shown in Fig. 5 on page 30 of the April issue is not the raw output of a 20-bit A/D converter, but rather the output of such a converter dithered down to 16-bit via a Meridian 618 processor.
There is no "mid-scale hash" shown in the spectrograms, just the standard depiction of a whitish noise floor. As to the "sound preferred by the author," I have not expressed to Johnson-or anyone else-what my preferences in recorded sound are. And while maligning my taste, he fails to note my favorable comments on his own Mike Garson recording. Or is that disc an example of "dry, spoon-feed multi-mono with centered, close miked, chronic mezzo-forte solos"? As to HDCD's sampling rate and the statement on analog sources, both these representations were made by the president of Pacific Microsonics, Michael Ritter. Likewise, the comparison of HDCD and "20-bit" Hendrix recordings was suggested by Ritter, who provided the HDCD version and stated the master tapes were the same for both. Mike and Keith, you need to talk.
Regarding Johnson's comments about "a peak variation of 1 or 2 dB" on the Hendrix comparison, as Fig. 8 on page 31 clearly shows, the maximum variation is 4 dB. He also refers to a spike that is "30 dB down from signal." The only spike is plainly visible in Fig. 14 in the noise floor and is cited in the text. The same is barely perceptible in Fig. 10 where, by referring to the key on page 28, it can be seen to be roughly 50 to 60 dB below the leading edge of the signal.
In fairness, graphs like Figs. 10 and 11 are unfamiliar to most people in the audio industry; Pacific Microsonics, again according to Ritter, was entirely unaware of Maximum Entropy Spectrographic Analysis before I brought the method to their attention. The software, originally developed by Bell Labs, with contributions by Massachusetts Institute of technology and others, is in use at hundreds of research facilities world-wide.
Pacific Microsonics' response to the results may stem, in part, from the fact that they had not made measurements with actual music but instead used "complex tones." And is anyone else bothered that nowhere in the Johnson letter does a single specification appear, nor is any clear statement made about the HDCD process, other than what it is not? Regardless of what one may think of conventional digital processes, at least we know how they work.
-D.W.F.
(adapted from Audio magazine, Aug. 1996)
= = = =