During a discussion on the Joenet someone asked Lynn:

This is quite possible, it just takes better equipment than a simple THD meter. See my article in Glass Audio, also the comments of Frank Deutschmann are very much to the point. Complex topologies with multiple forward paths and feedback do indeed generate a larger number of high-order spectra than simpler topologies. The high-order terms are at low levels (at least 10x to 100x below 2nd harmonic), but are there nevertheless. Some engineers claim that anything below 0.1% is inaudible, others say this depends very much on the order of the harmonic. Myself, I side with D.E.L. Shorter of the BBC and Norman Crowhurst in suggesting that harmonics should be weighted to the square or even cube of the order. If you do that, transistor amps with 20 to 30dB of feedback don't measure very well compared to zero-feedback amps.

True in part. Some "mainstream-audiophile" SE amps are intentionally designed for lots of "euphonic" distortion. I won't name names, but I think we all know who makes them ... they run the tubes at suboptimal operating points, choose audiophile-favorite driver tubes instead of low-distortion alternates, use cheapo output trans with fancy covers, etc etc. The ones mentioned in this forum, SP, and VALVE are another story ... they do indeed have quite low distortion once you look beyond the large 2nd harmonic.

As I mention in my Glass Audio article, 2nd harmonic dominates the THD measurement, yet has the lowest audibility of any harmonic. This encourages the construction of amps with low 2nd harmonic and abundant upper harmonics, since the THD measurements look nicer, and reviewers and salespeople are taken in by the magic of a low THD number. After all, if the goal is merely low THD, any rack-stereo receiver can easily meet that, with hundreds of watts to spare.

Several reasons, all measurable with the right equipment:

• Cleaner spectral response, due to an all-triode signal path with no loop feedback. Triodes have the cleanest spectra of any amplifying device ever made, especially if you weight the harmonics as mentioned above. Feedback multiplies low-order harmonics into smaller but more numerous quantities of high-order harmonics. See articles by Norm Crowhurst in Glass Audio for the mathematical derivation of this.
• Crossmodulation. As a long-time speaker designer, I know that speaker drivers are wicked devices, cheerfully storing resonant signals for many milliseconds and then sending them right back to the amplifier. If there is zero loop feedback the trouble stops right at the plates or emitters, otherwise it goes to the input stage and cross-modulates the desired input signal. The usual engineering assumption that the feedback summing node has zero distortion is lazy thinking and is simply not true. If the summing node has non-zero distortion the driver resonances will cross-modulate with the amplifier distortion. This cross-modulation is partly responsible for the notorious matching problems between amps and speakers ... worse, some of the very best speakers are the most reactive loads. If you want reactive loads, check out ESL's or horns. If your amp can't handle reactance gracefully, then you are stuck with Magneplanars, which are largely resistive loads.
• Greatly reduced problems with TIM/slewing distortion, which is an especially unpleasant form of signal-dependent time dispersion. Feedback generates large error overshoots when the amp nears clipping, slewing, or Class AB transitions. These overshoots can saturate low-current input stages, which then lose the ability to drive the Miller C's of the following stage. See articles by Matti Otala and many other authors in the AES Journal on this topic.
• Feedback greatly worsens the audibility of clipping, which is a serious problems with amps that don't have a 200 to 300 watt output capability. Music typically has a 10 to 20dB peak-to-average ratio, so 90dB average levels require peaks of 100 to 110dB. If the amp/speaker system is not capable of that, it is highly desirable the peak is gently compressed, not hard-clipped. The worst case is a feedback amplifier with output devices that saturate or become misbiased during clipping; not only does feedback hard-clip the amplifier, it may take a long time for the output devices to sweep out the charge, cool down the silicon die, regain normal operation, and return to a low-distiortion regime. This stretches out clipping and makes it extremely offensive, in the manner of a transistor radio.
• The effects discussed above interact in complex ways. For example, a very common problem with many transistor amps (nearly all, actually) is momentarly loss of phase margin stability when the amp clips. In the severe case, there are actually quick bursts of 2 to 20MHz oscillation when the amp clips into a speaker + cable load. (If this goes on long enough, the output devices and the tweeters will fail. This is one of the most common reasons that solid-state amps blow up.) Assuming the above problem is addressed by the designer, when the feedback amp clips, the TIM distortion will still greatly increase until the output transistors leave saturation, cool down, and re-enter their linear region. As long as the output device is outside the linear region, overshoots in the feedback network are very large and easily overload the input section, which then in turn has to recover from saturation. What starts as a simple overload problem in one part of the amp spreads to the whole amp thanks to feedback ... and none of it shows up in a THD measurement due to the transient nature of the problem. The amp can be very unreliable, sound terrible, yet measure very well in a THD measurement.

There is a common attitude in the audio industry that the critical listener is easily deluded, and the "proof" of this is that most music lovers don't care for the sound of modern low-THD electronics, from digital audio to op-amps to high-feedback amps. I take the opposite stance, and say that mainstream engineers should promptly discard outdated measurement protocols that have proven useless for assessing subjective quality.

Unfortunately, the mainstream audio-engineering community has little interest in doing so, and is instead vigorously pursuing techniques to discard 80 to 90% of the digital data in a medium that was grossly flawed even back in 1980. The 44.1kHz sampling rate was chosen by Philips since it would accomodate Beethoven's 9th Symphony in its entirety, fit in a standard IEC aperture for a car-stereo receiver, and be a submultiple of standard television scanning rates. In doing so, Sony and Philips ditched a decade of research by Tom Stockham, Denon, and the engineers at Decca Records, who were using 50kHz and had proposed a 60kHz rate to the AES Standards committee in 1979.

In the subjective world where we all live, if THD has any meaning at all, a mass-market receiver connected to $100 CD player should sound 20 to 100 times more realistic and lifelike than a Westrex 300B amplifier connected to a phonograph. I imagine there is no shortage of well-paid marketers and slick-magazine writers who would tell us that is exactly the case. And if Lucasfilm/Dolby/Microsoft changes their story next week, last week's party line will go into the memory hole, and it will be AC-5/Win99/DVD-II that will be "Perfect Sound and Vision."

I say let's just ignore these hucksters, and do our own research on what is truly audible and important. Switching moods from curmudgeon to Santa, I want to say a hearty "Congratulations" to the Joelist for doing exactly that for the last 2 years! Thanks to all of you for the sheer enthusiasm and hard-core R&D, and a Merry Christmas and Happy New Year to all!!!

Lynn T. Olson

E-mail: lynno@teleport.com

Web: http://www.teleport.com/~lynno/Ariel.htm

Editor, Valve & Tube News

Transformer coupled PP triode Illusion Engine by Lynn T. Olson.