DISTORTION is one of those specifications that every- one talks about but few really seem to understand.
For example, newcomers to the world of high fidelity frequently have the impression that distortion is something that bad amplifiers have and good ones don't. A manufacturer revises the distortion specification of one of his products from, say, 0.1 per cent to 0.05 per cent, and audiophiles immediately assume that it will now sound "cleaner" as a result. Or an engineer discovers that a product under test suffers from high "crossover" distortion at low power levels, and he immediately assumes that it will sound rotten in actual use.
Considering the efforts expended to reduce distortion to ever-lower levels, it seems curious that few, if any, engineers have ever stepped back from their test instruments long enough to ask themselves exactly what specific goal they are striving for. The difficulty, of course, is that the meters on distortion-test instruments have only numbers on them rather than such meaningful legends as HIGH, Low, and SUB-AUDIBLE. And in the absence of some definitive word from the psychoacoustics labs stating that "the trained listener is unable to hear distortion of complex program material when it is below x per cent," engineers will continue to strive onward—or is it "downward"? — for ever-lower distortion numbers.
At this point some readers may be thinking, "So what? Even if there is only a questionable advantage in lowering distortion far below the level at which it is audible, what's the harm?" Well, there is harm if the product thereby ends up costing more than it really has to, or if the money spent on lowering distortion could have been used elsewhere in the design to produce some more practical user benefits.
I have touched on these matters from time to time in my monthly Audio Questions and Answers column, and Julian Hirsch has also devoted at least one of his recent Technical Talk columns to the question of the audible effects of crossover distortion, but neither of us then had access to any hard experimental data to back up our suppositions and speculations. During a conversation with Robert Carver, designer of the Phase Linear equipment, I mentioned our interest in the subject and struck gold. It seems that he had been investigating the question of the audible threshold of distortion for years and had notes, laboratory data, and, what is more, firm opinions on the subject. But, since some of his findings seriously contradicted accepted audio dogma, he felt that, for credibility's sake, it would be best not simply to write an article (as I immediately urged him to do) telling what he had found, but rather to recapitulate, in condensed form, his original experiments for the eight "golden ears" attached to the technically oriented editors and contributing editors of STEREO REVIEW and allow us to come to our own conclusions.
And so, one Saturday morning early this year, Julian Hirsch, Ralph Hodges, Craig Stark, and I came together in Craig's basement laboratory/listening room and awaited the arrival of Robert Carver, his 200 pounds of test instruments, and what turned out to be some mind-blowing revelations. The following article, with some additional explanatory notes, is Mr. Carver's description of some of what we discovered during that fascinating weekend.—Larry Klein, Technical Editor
THE prospect of conducting a weekend of technical demonstrations for the assembled technical editors of STEREO REVIEW caused me a certain amount of trepidation. Was I being simply perverse in trying to prove that very low distortion really wasn't necessary—when I had already designed and was marketing two super-lowdistortion high-power amplifiers? I knew what my measurements had shown and what my ears had heard, but would I be able to convince the others? Would they find some flaw in my procedures, my instrumentation, logic—or even my ears?
After I arrived at Craig Stark's home and unpacked and set up the test equipment and a specially modified Phase Linear 400 amplifier, we measured the background ambient noise level of our temporary laboratory. It was found to be a mere 31 dB—about the same level of quietness as is found in the country, far from the city, in the early morning hours. That is very quiet indeed, so there was little or no chance of masking effects interfering with our ability to hear whatever distortion was present. To "calibrate" our ears we played an excellent wide-range recording (an early English pressing of a folk-rock album, "The Pentangle," Transatlantic Records TRA 162) with, among other instruments, drums, snares, and a solo female vocalist. One side of the record was played several times so that we all became very familiar with the overall character and sonic "flavor" of the material. (The very low tracking force we used precluded audible groove damage that might result from repeated plays.) After listening for perhaps thirty minutes, the group unanimously agreed that the overall fidelity and "sound" of our system was almost awesome. Transients were clean and clear, plucked strings had a transparency that was absolutely chilling, and the low-frequency detail was superb—nothing was missing.
At this point we substituted an excellent 300-watt amplifier (made by a competing manufacturer) to convince ourselves that the Phase Linear 400 was indeed performing properly and that there was nothing in its sound quality that might somehow cloud our judgments during the experiments to follow. When we were finally able to get the output levels of the two power amplifiers exactly matched, there was absolutely no audible difference when switching between them while listening to either white noise or music. During the adjustments of the amplifiers, it was demonstrated dramatically that minute differences in volume level (sound quantity) that are too subtle to be heard as such are interpreted by the ear as "obvious" differences in sound quality. Everyone was startled by the effect — everyone, that is, except Larry Klein, who had touched upon the phenomenon some time ago in his Audio Questions and Answers column.
And so, psychologically and sonically prepared, we were all set, we felt, to deal with the big question: In what amounts and under what circumstances do the effects of distortion become audible? The additional signal sources we used for our tests included master tapes, a white-noise generator, and a variety of music with a wide dynamic range drawn from commercially available records. The speakers we used were AR-LSTs, chosen for their linearity, very wide frequency range, and high power-handling capability. An assortment of other lab instruments was employed to produce the scope photos and check the measurable data, but the basic point was not what could be measured, but rather what could be heard and how it correlated with what was measured.
OUR primary "test instrument" was the Phase Linear 400 stereo power amplifier that had been modified at the factory to have switch-selectable, pre-measured amounts of "crossover" distortion (see accompanying box). Klein, Hirsch, Hodges, and Stark assumed positions around the left-channel speaker. The amplifier, fed by a Hewlett-Packard generator, delivered a pure 60-Hz tone at a 1-watt level to the LST. The distortion switch on the Phase Linear was stepped upward:. 0.05 per cent, then 0.07, 0.1, 0.15, 0.2 — stop! There it is! 0.2 per cent. Wow!
Everyone was in agreement that there was some ever-so-slight change in the tone at 0.2 per cent distortion. The next increment was a large one — a jump to 0.75 per cent, at which point the distortion sounded like a separate buzz on top of the 60-Hz tone. It was audibly obvious, and all were in complete agreement as to its presence. Repeating the test at 4 watts into the speaker increased our perceptual sensitivity, and it was discovered that distortion of 0.15 per cent could then just barely be detected. (The STEREO REVIEW people all expressed surprise that such small amounts of distortion were audible.)
We next used two mixed tones: 60 and 7,000 Hz. At 0.15 per cent nobody heard the distortion. In fact, up to and including a level of 2 per cent, no one was certain that he was hearing distortion. However, the distortion suddenly became obvious to everyone at 2.5 per cent. It was heard as an "overlay" or harmonic tone added to the 60-Hz tone.
For the next test, three tones-60, 3,000, and 7,000 Hz—were mixed and fed to the speaker at a four-watt level. Even at 2 per cent no one heard the distortion. The lowest distortion perceptible to any of us was a whopping—and startling — 4 per cent!
Two more tests were performed, this time using music. The music was the recording of Pentangle used earlier for "ear calibration." The amplifier power output averaged about one watt. Distortion was stepped upwards as previously, and at 6 per cent we just began to detect a "strained" quality in the singer's voice. However, even at 6 per cent, distortion was not evident on the percussive instruments. It was only at 12 per cent that distortion began to affect the sound of the guitar and cymbals. Higher distortion levels caused an obvious "fuzziness" in the sound. To test the relationship of volume level to the ear's ability to detect distortion, we played the music at a very loud level with peaks exceeding 100 dB. There was no difference: distortion was again just barely audible at 12 per cent.
So far, our tests indicated that very small amounts of distortion (0.15 per cent) are perceptible if the program material is sufficiently simple—for example, a single pure, steady tone. Mixing two tones dramatically raised the threshold of perception to over 2 per cent. Three simultaneous tones, representing increasingly complex program material, resulted in a perception level of a surprising 4 per cent. With normally complex music, it was necessary to increase distortion to a full 6 per cent before it became just perceptible.
Although there were five different listeners of varying age involved in the tests, whatever distortion was present was usually heard by everyone—or no one. There was no disagreement. For example, on the single-tone test, 0.1 per cent was not perceptible to anyone, but 0.2 per cent was immediately evident to everyone.
It seems clear that, since crossover distortion levels below 6 per cent are not audible when listening to complex musical material, an amplifier whose measured crossover distortion is below 6 per cent will probably sound fine most of the time. But since distortion as low as 0.15 per cent was audible with a single test tone. we therefore set out to determine to what degree our test tone resembled anything that might be encountered in normal program material. In other words, could we find a recording that contained musical material "simple" enough to allow us to perceive distortion at a level below 6 per cent? We found one: a recording of Mozart's four horn concertos (Vanguard S-173) in which a single French horn is featured in several solo passages.
During these sections, we found that some change of tonal quality could be heard with distortion at a very low 0.35 per cent. With the distortion switched in, the horn acquired a "richer" quality, presumably because the harmonic distortion generated by the crossover notch added a little harmony (no pun). As higher levels of distortion were switched in, the added harmonics subtly changed the character of the horn, each time making it sound as if it had been exchanged for a different—though equally good — instrument. Only when the crossover distortion climbed above 12 per cent did the horn begin to sound fuzzy or "bad."
The objection might be raised that when distortion was not perceptible it was because the total "normal" distortion in the complete recording-playback chain was great enough to mask the distortion added during the experiment. However, our high sensitivity to added distortion on a single test tone and our low sensitivity to distortion with three test tones suggest that entirely different psycho-acoustic mechanisms were operating.
THE human ear has a loudness-sensitivity response that is roughly logarithmic — which means that for every doubling of subjective loudness the objective power level of the sound has to be boosted ten times. It is this logarithmic characteristic that permits the ear to hear sounds over a loudness range of about ten trillion to one. In addition, the sensitivity of the ear varies not only with volume: its sensitivity to any given frequency may also be affected by the presence of other adjacent frequencies. This would explain why.increasing the number of test tones resulted in a decrease in our ability to perceive distortion. When a single 60-Hz tone was used. our ears were operating at maximum sensitivity for frequencies far away from the fundamental. Hence, it was relatively easy to hear the higher harmonics produced by the distortion components added to the fundamental. However, when the second tone was added, enough "masking" energy was present at the upper frequencies to cause the sensitivity of our ears to decrease significantly. The reduction in sensitivity was sufficient to render the distortion components inaudible until they were boosted by a factor greater than ten. This psychoacoustic masking phenomenon is well known today, and is made use of in the Dolby and some other noise-reduction systems. However, it has not generally been thought of as having any bearing on distortion perception.
In our final discussions, the editors admitted to some shock and surprise at both their sensitivity and lack of sensitivity to distortion under the various test conditions. And I pointed out, perhaps unnecessarily, that we had investigated only the most common and "worst-case" type of distortion encountered in amplifiers, that there are many other design factors that affect the sound of today's transistor equipment. However, that must await another time, another discussion, one which will include some of the other data and insights that have emerged on the question of what makes transistor amplifiers sound bad — or good.
Robert Carver has taught physics at California State University. Designer of the Phase Linear equipment, he's just completed a preamp with noise reduction based on psychoacoustic masking.
From FrankieD's lips to your ears: Sunfire - a quiet box of endless power.
Sunfire TG-IV/400~7 Amp
Carver SD/A-360 CDP
Active bi-amp: Ashly XR-1001 & 2 Rane PEQ-15s
Main: HotRodded AL-IIIs
Sub: Klipsch RT-12d
Center: Sunfire CRS-3c
Surround: Sunfire CRS-3 (x 2)
OconeeOrange wrote:"Gary likes to play it 'loud' as do I. His system begs you turn it up until you die"
RIP WIlliam B. Dibble, 1948-2012. I'll miss you my friend.