It's interesting that a straight THD test was already recognized as inadequate in 1938 - over 80 years ago and yet we have still not developed an adequate test for amplifier distortion evaluation. There are probably a few reasons for this, but one must be a lack of clarity at a scientific and engineering level on exactly what makes an amplifier sound good, and the second is that it has only become feasible to precisely extract information from audio signals in the precision required with the advent of high-performance sound cards/test systems and DSP over the last decade or two.
Its not clear if the problem is that a test hasn't been developed. It may be more a question of whether the commercial market wants it, and or is willing to pay AP extra for it. I mean, if playing the low distortion game is selling product now, why complicate things with a better test?
As an aside, how do people buy computers? CPU speed, CPU model, number of cores? Or benchmark scores that aren't exactly like the buyer's use case? IOW, when the effects of the market are considered, what sells isn't necessarily what is better, even if someone does know how to measure whatever it is that's important.
As an aside, how do people buy computers? CPU speed, CPU model, number of cores? Or benchmark scores that aren't exactly like the buyer's use case? IOW, when the effects of the market are considered, what sells isn't necessarily what is better, even if someone does know how to measure whatever it is that's important.
Last edited:
Frank Massa actually wrote about intermodulation and multitone distortion. This is a summary of that article in Electronics 9/1938 by Frank Massa:
"Trigonometric power series analysis, with experimental measurements to show that distortion is greater when several frequencies are present than when only a single frequency is present."
Here is a picture from the article:
Note also that Sean Olive totally misses the text following Frank Massa's quote:
"When the class of the amplifier is specified (e.g. Class A1 single triode) the "total harmonic distortion" may be interpreted sufficiently, but it is always preferable to specify each harmonic separately."
So what is actually stated in that book is that THD as a single figure (like SINAD) is insufficient but THD presented as separate harmonics is ok. So no need to invent new distortion measurements.
EDIT: unfortunately the online version of Electronics 9/1938 does not include the 2nd page of Frank Massa's article. It would be an interesting read.
As I see it, the real problem is that speaker loads not only vary with frequency, but are also significantly non-linear, so the distortion of an amplifier is probably much more complicated when you consider an amp combined with a speaker. With some ill-defined length of wire between the two.
If speakers were perfect, any two amps with distortion >80dB down would probably sound very similar.
THD was probably a meaningful indicator of amplifiers when it was 1% or something?
This is probably why multi-tone tests are now becoming more popular. You have a lot of tones mixing to create IM products. My only complaint about some of these tests is they are done at very low power levels where the total tone power is low ie the amplifier is not operating in its large signal regime where large signal non-linearity may be contributing significant distortion.
Okay, but what about the compression effect PMA showed with an amplitude modulated test signal?
Also, part of the problem with FFTs is that is an integral transform. It means what you see is to some extent averaged. The peak of your spurs could be dancing up and down randomly due to noise problems and you might never know it. The spurs would just show as some averaged-out level. There would be no extra spurs due to the modulation because the modulation is by random noise (or approximately random, anyway -- it just needs to be spread out over enough bins to look like noise).
Also, part of the problem with FFTs is that is an integral transform. It means what you see is to some extent averaged. The peak of your spurs could be dancing up and down randomly due to noise problems and you might never know it. The spurs would just show as some averaged-out level. There would be no extra spurs due to the modulation because the modulation is by random noise (or approximately random, anyway -- it just needs to be spread out over enough bins to look like noise).
Last edited:
I have not seen a proposed test come out of academia or industry bodies and they’ve had 80 yrs to do it. This type of test hasn’t appeared for lack of opportunity. You could probably write something to run on a QA4xx.
Problem is we could go around in circles proposing new tests but without some solid evidence that it is rooting out the thing we think is a problem, we are just flying blind.
(Notice how this discussion has gravitated towards distortion again 🙂 )
And also the amplitude of individual harmonics is always low, because such multitones are usually constituted of 20 - 30 sine tones.
The advantage of FM modulated signal is that you work with full amplitude of the sine that is tuned/modulated in frequency. The stress sent to the amplifier is much higher.
It's been known for quite a while that intermodulation matters more than harmonic distortion and standardized intermodulation tests have existed for many decades - unfortunately many different ones.
https://www.ap.com/technical-library/more-about-imd/
I have a lot of trouble with the "sound quality" discussions. I need some sort of reference for what people are hearing. My hearing has changed dramatically over time and so has my judgement of sound and what I like. Today, I have pretty severe tinnitus and notable age related loss of high frequencies. Ok so I am a pathologically impaired listener, but I did not get here over night it has been a slow and bumpy process for many years. So when did I become "disqualified as critical listener?
What I do care about is that I can go to a concert and enjoy the performance and then listen to it at home and still enjoy it. I can still tell the difference between an Issac Stern recording and a Yehudi Menuhin or Jascha Heifetz recording. Something I have been able to do since my non-pathological 20's. As long as the playback aberrations are masked by the remembered and perceived real sound I am happy. I think Siegfried Linkwitz expressed it well ..... "true to the original". From my perspective many of the "sound quality" issues being discussed are dwarfed by the music itself, even if we cannot reproduce the concert hall. In a way, some (not all) are a bit like talking about 0.00001% THD. I need a reference, perspective, context or something to compare to for relevance.
Bill
What I do care about is that I can go to a concert and enjoy the performance and then listen to it at home and still enjoy it. I can still tell the difference between an Issac Stern recording and a Yehudi Menuhin or Jascha Heifetz recording. Something I have been able to do since my non-pathological 20's. As long as the playback aberrations are masked by the remembered and perceived real sound I am happy. I think Siegfried Linkwitz expressed it well ..... "true to the original". From my perspective many of the "sound quality" issues being discussed are dwarfed by the music itself, even if we cannot reproduce the concert hall. In a way, some (not all) are a bit like talking about 0.00001% THD. I need a reference, perspective, context or something to compare to for relevance.
Bill
Time for a "formal" definition of a Power Amplifier:
A function definition + defining characteristics.
As a bullet list. Black box.
Any takers?
I think it's better to work from this perspective and once a good description is at hand and agreed, tests will not be o hard to define. Talking test is kind of backward engineering.
//
A function definition + defining characteristics.
As a bullet list. Black box.
Any takers?
I think it's better to work from this perspective and once a good description is at hand and agreed, tests will not be o hard to define. Talking test is kind of backward engineering.
//
Okay, but is frequency modulation sufficient by itself, or does amplitude modulation testing have any additional value? One thing it has for me is accurate reproduction of beat notes, which are not frequencies; they are modulated envelopes produced by merely summing signals. I suppose a multitone test may do the same thing, at least dependent on the relative phases and amplitudes of the tones. So, should we use a Hilbert transform to measure the envelope fidelity?
If they are randomly phased, the probability distribution of a sum of sines quickly approaches a Gaussian curve, see Bennett's 1948 article Distribution of the sum of randomly phased components, https://www.ams.org/journals/qam/1948-05-04/S0033-569X-1948-24592-9/S0033-569X-1948-24592-9.pdf
Not surprising given what they knew about statistics at the time. How about for non-gaussian bell curves as seen in real world data? Even for white noise, there are multiple types as shown in the graphic:
https://www.edge.org/response-detail/11715
IOW, does "random phasing" mean random according to a presumed gaussian distribution of randomness?
Why don't you read the article?
If you would read the article, you would see that the phases are assumed to be uniformly distributed over 2 π radians and that Mr. W. R. Bennett and Mrs. A. J. Shanklin also calculated the probability distribution you get with small numbers of equal-amplitude sine waves. The fact that the sum converges to a Gaussian distribution was no surprise (central limit theorem), the question was how quickly this would happen and what the probability of excess would be with smaller numbers of sine waves.
If you would read the article, you would see that the phases are assumed to be uniformly distributed over 2 π radians and that Mr. W. R. Bennett and Mrs. A. J. Shanklin also calculated the probability distribution you get with small numbers of equal-amplitude sine waves. The fact that the sum converges to a Gaussian distribution was no surprise (central limit theorem), the question was how quickly this would happen and what the probability of excess would be with smaller numbers of sine waves.
Last edited: