The Missing Specification

In the 1952 edition of the Radiotron Designer's Handbook, long recognized as the "bible" of the industry, the permissible level of IM distortion for a high-fidelity amplifier was given as 3%, with the alternate figure of 2% being cited as a "rather extreme" specification. We wonder what the author of that statement would think of today's solid-state amplifiers with their measured IM of 0.01% and less. And we wonder what he would think about the fact that these super-amplifiers still have audible distortion.

The Radiotron Handbook was written during the heyday of the tube, but even in the latter days of that era there were people who maintained that triode output tubes sounded "better" than pentodes (or, rather, than tetrodes). Distortion measurements showed that neither was categorically better than the other, yet the triode-vs-tetrode controversy persisted until solid-state laid the whole issue to rest.

It was soon observed, though, that all solid-state components had their own characteristic sound. They were sharp, crisp, and quite brittle-sounding—fine for reproducing "hard" transients like triangles and xylophones, but not so fine for massed violins, which became annoyingly shrill. By comparison, most tube equipment was rather "sweet" and slightly veiled, which made strings sound fine but took the edge off harder sounds. After the initial infatuation with "solid-state" wore off, it became obvious that the "crispness" of transistors was just as much a form of distortion as the softness of tubes, only different. So, subsequent designs aimed at lower and lower distortion figures until these were in some cases significantly lower than in the best of the tube-type amplifiers. Yet the characteristic difference was still there: Transistors sounded harder, tubes sounded "sweeter." And the astounding thing about it was that the amount of hardness from one solid-state component to another was audibly different when measured distortion figures differed by as little as 0.002%!

The usual IM and harmonic distortion tests do not actually measure an amplifier's distortion, they measure the effects of it—the modulation of treble by bass, the production of sum-and-difference tones, and the introduction of spurious harmonics. But the fact that such minuscule measured amounts of these effects are nonetheless audible through loudspeakers producing hundreds of times as much distortion as the amplifiers raises the question of whether there might not be a different effect occurring simultaneously, one to which the ear is more sensitive than those that we measure. We don't believe, though, that the answer is quite that abstruse.

When an amplifier is adding harmonics, they are not likely to be equally strong. For example, the strongest harmonic from a single-tube amplifying stage will be the second. The third will be weaker, the fourth weaker still, and higher-order harmonics will be weaker still until they die out. (Push-pull operation cancels most of the second harmonic, making the third one the strongest.) In transistors, however, the distribution of harmonic energy is entirely different. Here, the harmonics are likely to be of fairly equal strength all the way out to and even beyond the 10th. There is little tapering off in the higher-order harmonics, as in a tube.

A total-harmonic-distortion test (THD) measures mainly the strongest harmonic. Thus, for a tube-type power amplifier (with push-pull outputs), the reading would be a measure of its third-harmonic distortion. The weaker higher-order harmonics would not register at all. On a solid-state power amp, though, the THD reading (which might be numerically identical to that of the tubed amp) may be that of the second, third, fourth, or fifth harmonic, or that of all the harmonics to out beyond the 10th. Obviously, identical THD measurements do not assure identical distortion characteristics. And since it is a proven fact that the ear is much more offended by the higher-order (fifth, seventh, etc.) harmonics than by the lower ones, it would seem equally obvious that the ear is responding to these little details that the THD measurement overlooks.

This is borne out by subjective observation. In tube equipment, a measured THD of 0.05% through most of the audio range (that is, using test frequencies of 40, 1000, and 5000Hz as the fundamentals) was almost an ironclad guarantee of rich, sweet sound, although there may have been a barely perceptible "veil" over the sound due to the audibility of the lower-order second and third harmonics. Things didn't start to sound hard until THD rose to above about 0.2%. In solid-state equipment, the sound is annoyingly hard at 0.2% THD, and some hardness is still perceptible at 0.05% THD (footnote 1). The "veil" rarely becomes audible, because it is covered up by the screech. Hence, the transistor's reputation for "clearer" sound.

Dynaco's Dave Hafler (who supplied much of the info for this piece) suggests that the industry devise a way of "weighting" the distortion readings to favor the higher-order harmonics, as we now weight signal/noise ratio readings to favor that part of the spectrum where the ear is most sensitive. This seems like a fine idea, but the industry seems in no hurry to consider it. Heaven forbid that the consumer should be given a better way of distinguishing the good from the bad!—J. Gordon Holt

Footnote 1: Stereophile's Test CD 2 has tracks where you can compare a pure tone with various amounts of second, third, and seventh harmonic distortion.—John Atkinson

tmsorosk's picture

What can I say after rereading The missing Spec, I miss JGH.

dce22's picture

J. Gordon Holt wrote this article in 1971 when audio test gear was just notch filter and a wideband volt meter, in this day and age of super sophisticated test gear and amp simulation coupled with experience and knowledge of yester year fixing hard sounding solid state amp or soft sounding valve amp are trivial things to a good amp designer.

dalethorn's picture

Bob Carver did an amazing amp challenge in the 1980's using "null difference" tests. It's a fantastic read:

John Atkinson's picture
dalethorn wrote:
Bob Carver did an amazing amp challenge in the 1980's using "null difference" tests. It's a fantastic read:

Good read, maybe, but it's also misleading. When I interviewed Bob in 1990, he admitted that the smallest null he had been able to achieve with production amplifiers was –36dB or 1.5% difference: see

And from my own measurements, that –36dB null was in the midrange; it was lower at the frequency extremes.

John Atkinson
Editor, Stereophile

dalethorn's picture

Yeah, not doubting the real differences then or now, but it is very interesting how much difference we can hear in premium amps today, with all of our advances in technology. In general, I'd prefer that amp designers aim for the best possible sound rather than compromise to match an arbitrary target.

otaku's picture

Funny: "some hardness is still perceptible at 0.05% THD"

I just checked, and my Bel Canto clocks in at 0.003% THD.
Quite a bit of progress.

JRT's picture

Just a few years later in 1975, Nelson Pass' Threshold 800A stereo amplifier was available at retail sale, and was providing 200W per channel with clean gain over an adequately wide bandwidth, exhibiting less than 0.01% IM distortion using audio test frequencies mixed with 80kHz, remaining very well behaved when fed the then sometimes excessive ultrasonic nonlinearities associated with playback of phonogragh records and analog tape.

Charles Hansen's picture

Gordon was far, far ahead of his time, and we all love him and miss him dearly. Yet it is easy to overestimate our knowledge. Nearly fifty years later I agree 100% with everything he wrote until the last sentence of the third paragraph, "We don't believe, though, that the answer is quite that abstruse."

Clearly it is far more abstruse, as otherwise all amplifiers would have sounded the same by now, nearly half a century later. David Hafler's idea of weighting distortion components is admirable, yet the evidence shows us there is far more to the story than that.

One fact remains unchanged - tubes still wear out.

Marc210's picture

There are differences between tubes and solid state, especially in the bass.
But between different transistors brands, they are negligible. In ABX of course.