rdp@teddy.UUCP (10/03/85)
[]
This is real long, so bear with me...
The debate is flaming once again over the value of laboratory measurements
versus the "golden ear". Having had a lot of experience in lab measurements
of various hi-fi components (about 7 years worth) as well as having been
involved in research into audibility of distortions, my two cents worth is
being cast upon the waves.
First let me state, at the great risk of being flamed, that I find the
concept of the "golden ear" to be a complete crock. Some of the most highly
regarded "golden ears" are so inexperienced in listening to live music that
it is ludicrous the weight their opinioins are given.
Having said that, let's look at some summaries of the opinions expressed
on the network.
1. "poor frequency response is the major contributor to lack
of fidelity"
2. "Harmonic distortion does not contribute at all to the sound
of hi-fi components"
3. "The great difference in sound between amplifier A and amplifier
B can not be attributable to the differences of .01% and .1% in
distortion we measured"
4. "Since this set of measurements don't differ greatly, and the
components sound wildly different, then the audible differences
can NEVER be measured in the laboratory"
And so on, ad infinitum.
When I was in the audio business, the store I was the technical manager of
had the policy of measuring, as completely as possible, EVERY single piece
of equipment that even came near the store. That included every piece that
was sold. For electronics, as an example, we measured THD at full power,
1/3 power, 1 watt, 10 mW, and at 20, 200, 2000, and 20000 Hz. IM distortion
was measured at these same power levels. Transient response was evaluated
from input to output (including phono sections). Reactive load driving and
recovery from overload was tested. On some selected units, transient
modulation characteristics were tested. When THD was done, often a complete
narrow band spectrum analysis of the distortion components was done. The same
for noise, and on and on and on. I was able to collect a vast amount of
verifiable, correlatable data on the measured performance of commercially
available hi-fi components (probably totalling some 5000 to 6000 units).
Some immediate (and, hopefully, non-controversial) conclusions can be drawn
from this data:
1. MOst units barely squeeked by the manufacturers specs out of the box,
with the exception of tape recorders (of all types, cassete, reel-
to-reel, etc.), which failed, with few exceptions, miserably.
2. Most units could be made considerably better by simple tweeking.
Tape recorders are the biggest beneficiaries here, followed by
tuners.
3. Most units produced by the larger manufacturers were consistantly
better adjusted than those of small, esoteric, manufacturers. This
was most true of power amps. Those from many of the audiophile
companies were dismally adjusted, most never met specs. The reason
is probably that they don't know how do design amplifiers that could
be shipped properly.
Now, what has all this to do with the sound of various components? Well, as
presented, not a whole bunch. But it does establish an objective base from
which to proceed (before flaming me for pre-conceived notions, read on!)
Let's take an example of two amplifiers of similar characteristics that
sound decidedly different (I realize they are somewhat antiquated, but
I do have quite a bit of experience in this particular comparison, and it's
validity transcends the age of the experiment).
The crown DC-150 was a 75 watt/ch power amplifier with fairly wide bandwidth
and low distortion (measured at below .05% THD for most cases). The LUX
M1500 has very similar specs. Rather than deal with the similarities, let's
look at the differences. The Crown has a higher damping factor and a somewhat
wider bandwidth. The Harmonic distortion figures at full power are lower
for the Crown than for the Lux, but not by a lot.
In listening tests, about 85% of the people listening regarded the crown as
being "edgy" or "brittle" or "hot", or "very detailed" or "tiring", whereas
they regarded the Lux as "sweet", "reticent", "unagressive" and the like.
Simple THD, frequency response, IMD, etc. measurements revealed little
differences, especially when measured the way the manufacturers did.
However, very large differences were measurable at realistic listening
conditions. The most complaints about the Crown occured when listening
to quite passages. Under these conditions, the Crown did have more THD
than the Lux, but it still was low (at 10 mW, the THD, excluding noise,
was approaching .2%). There was, however, a striking difference in the
makeup of the distortion components. In the case of the Lux, most of the
distortion products were confined to the second third and fourth harmonics,
totaling about .04%, whereas in the Crown, the distortion products consisted
mainly of third, fifth, seventh, ninth, etc, etc., and the spectrum continued
as far up as the fifteenth harmonic in some cases. It was noted that the vast
majority of the distortion occured as a result of dis-continuities at
the point where the output stages were switching from positive to negative
output and back again (That ol' devil, crossover distortion!) Additionally,
one of the transient tests we performed, which consisted of taking a 5Khz
sine wave at about 1 watt, and occasionally imposing a 10 watt (or greater)
10 mS wide cosine pulse, resulted in about .1% modulation from the Lux, but
20% (yes TWENTY) modulation in the Crown. Note that in both cases we were
nowhere near starining the amps power wise. Also note that the demodulated
waveform in the Crown did not closely resemble the pulse, rather it showed
a fast exponential rise, followed by a slow exponential decay, and all this
delayed by a few milliseconds (maybe heating of junction effects? ).
We did find that static damping factors don't mean s**t. We found, however,
that the crown, which had a damping factor of (as I recall) 150 at 20Hz, had
a damping factor of about 5 at 5 Khz, and under transient conditions, it was
lower still. The Lux managed to maintain it's factor of 50 out to 10 Khz.
It was interesting to see what happened to such things as THD and IMD as
a function of power. Both amps were quite low in the range of 3 watts to
near clipping. However, below that, the Crown's distortion rose monotonically
with decreasing power, whereas the Lux's remained fairly constant until the
point where it was totally swamped by noise (milliwat and lower levels).
What does this all prove? Well, to me it demonstrates that if we are hearing
differences that we can't measure, than we simply aren't measuring everything,
or we are not measuring them correctly. We simply can't stand back and declare
dogmatically that THD doesn't mean anything, or frequency response is
everything, or nothing. The ear, is indeed, a complex measuring instrument,
which reveals qualitative measurements. Laboratory instruments reveal, on
the other hand, qualitative figures. The two are certainly correlatable as
long as sufficiently precise, accurate, controlled, and unprejudiced
experiments are performed.
It has been my sad conclusion that the audio world is not interested in this
type of research, nor of the commercial application of the results. Too
many of the esoteric companies would fall flat on their faces in any fair
technical forum.
The point to hi-fi, in my view, is to, as accurately as possible, re-produce
the information provided to us by the media-mongers, however dismal that
may be. I have found, universally, that this goal, combined with good media,
most closely duplicates the musical experience of the live performance.
Then, if you accept this definition of accuracy, laboratory instruments are
admirably suited to measuring the deviation from this accuracy, because they
are consistant, untiring, and unprejudiced. They are not swayed by media
and guru hype, especially when a consistant measuring procedure is applied.
This, combined with a scientifically controlled, unbiased listening test,
results in a very good, predictable, correlation between sound and
measurements. In fact, what often happens is that the subtle audible
differences seem to magically vanish, never to return, when a controlled
experiment is performed.
About 12 years ago, an interesting measurement technique was described by
(I believe) Peter Walker of QUAD, U.K. in "Hi-Fi News and Record Review".
In the article, he described a technique of measuring amplifiers by what
he described as a bridge technique. Simply stated, the output is compared
to the input, via phase and delay compensating network, and the resultant
output is a direct measure of the inaccuracy of the unit. We applied the
same technique to the above experiment, and, lo and behold, the Crown would
reveal HUGE errors when people were bitching about it (low powers, etc).
The technique would not, unfortunately, reveal what kind of errors, just
their magnitude. (Note that in our case, we eliminated the phase and delay
compensation network, on the theory that any error is significant. Given the
bandwidth of the Crown and the Lux, it was deemed unnnecessary anyway). The
nice thing about this particular technique is that the signal can be anything,
including music.
Many research papers have been published in journals such as the Audio
Engineering Society, the Journal of the Acoustical Society, and so forth,
regarding the audibility of particular kinds of distortion. Most audiophiles
do not have either knowledge of or access to these articles, and many of them
would find (as I occasionally do) them to be quite incomprehensible. But
the fact remains that there has been much work done on the audibility of
these effects, and much of the denial of such stems from lack of knowledge
of these researches, or the inability or unwillingness to perform controlled
experiments without prejudice.
The one article that comes to mind that I found most amusing is the one
person who over a period of a half an hour substituted of few conponents
in his CD player, and then noticed a "dramatic" improvement in the sound
of the unit. Two serious challenges arise against the validity of his
claim. First, it has been more than adequately demonstrated that the acoustical
memory can be quite poor over the long term (more than a few seconds).
Secondly, having invested in a CD, these ridiculously expensive replacement
components, a lot of time, effort, sweat, burned PC board traces and
fingertips, and lots of weeping wailing and gnashing of teeth, I would not
expect any rational person to be willing to admit that there was no difference,
or even a negative improvement. Damned if I would be so honest as to admit
how stupid I was for screwing something up (although, now that you confront
me with it, I have screwed up quite a few things in my time!).
So folks, let's be more scientific about things. Every "golden ear" who has
tried to demonstrate, under fair, controlled conditions, that such and such
a unit sounds such and such a way, has failed to duplicate the results.
The comment one gentleman made about "the differences vanishing under
double-blind testing" is well taken, and quite accurate.
Dick Piercerdp@teddy.UUCP (10/08/85)
[] The response to my article about correlations of hearing vs. measurements has been universally quite favorable. I have received quite a bit of mail about it, and I thank the respondants for their letters. I have yet to receive the (I thought) inevitable flames from those among us who often flame only for flaming's sake, but, rest assured, I am hunkering down, reinforcing my bunker against the expected onslaught of "Oh Yeah?" replies. Two recurrent themes are apparent in many of the replies. The first is "where do I find a journal with this sort of information?" Well, I am not aware of any such publication devoted to either mainstream consumer audio products or the audiophile (nutcase?) market. My departure from the business is not totally coincidental with the rise in popularity of such "journals" as Absolute Sound and the like. In the face of such drivel, it is hard for a person with any reasonable sense of integrity to stay involved in the audio business. I fear that, having reviewed the currently available journals, nothing strikes me as being worth the postage. I might be accused as being either too hard or too snooty, but when there is much objective information to be had, there is little place for the witch-doctor PHD or the "golden-ear" mentality. I especially found the reference to joining a local audio club for "ear training" to most pretentious. Secondly, several people asked me questions like, "Well, if these journals don't exist, then why don't YOU write one?" Well, my answer is simply a correlary of Suffolk Audio's First Law of Acoustical Physics, to wit, "Any idiot can write an audiophile magazine, and, unfortunately, many do." Why, if there are people interested inn objective measurments and listening data about equipment, do the magic rags flourish so well. Are there 2500 people out there who would be willing to pay $25 for such a journal? I have seen no evidence of it, but then, I haven't been looking very hard. There are other reasons why I don't do thism sort of work. Back when I was involved, my equipment inventory was such that it embarrased the research labs for the likes of companies like Phase Linear, JBL, Advent, AR, Harmon Kardon, and so forth. I had GR graphic level recorders, 1/3 and 1/10 octave spectrum analyzers, high precision THD and IMD analyzers, narrow band spectrum analyzers, scopes, precision, low distortion (<.001%) signal sources, etc., etc. For insurance purposes, I once had to come up with replacement costs for all this stuff, and I came up with a conservative figure of something like $600,000!. Also, it is something that cannot be done by a single person. Inorder to ensure impartiality, it was necessary to have one person conduct the measurements, and another to conduct the listening tests. Where, in addition, does one find critical, non-partial guinea pigs (by critical, I don't mean "golden ear", I mean questioning, skeptical, etc.)? I figure it might cost $25,000 to equip a lab with the minimal amount of instrumentation necessary. One piece that I briefly explored was a programmable FFT analyzer. (I actually am much closer to that now, owning, as I do, a PDP-11/23 with 16 channels of 100KHz A/D. But I don't have Unix, thank God!) Lastly, when does one do all of the work? I am casually interested in such a venture, more as a diversion than anything else. I refuse to be the crusader for truth and justice I once was, it is too frustrating to deal with the AbNonsense out there. If anyone has any comments or suggestions, or might be willing to participate, let me know. The worst that could possibly happen is I actually do it! Dick Pierce