Why is distortion harmonic?

[AuroraB]

Originally posted by Wavebourn

I want to see facts that people hear instead of fact that people don't hear.

That's exactly where I think you err. It is a long established fact that our perception varies quite a lot, as our hearing varies from person to person. Now - I actually do believe that some people do hear sonic artefacts better than others, but this has a lot to do with training - very much like trained musicians can hear differences between other musicians, as well as different instrument makes - but a trumpet is still a trumpet? Right?

Several decades ago, there was a fairly broad concensus even in the audio industry, that THD measurements did not tell the whole story. This became quite obvious when the all novel high powered SS amps that measured so well, just sounded crap. Now enter Matti Otala and his TIM measurements and theories in the early 70s. Since then, several other techniques have evolved - FFT processors, cheap computing power and low noise line level amps , high speed ADs/DAs etc.
I do believe we have the tools we need - we may not fully understand the interpretation of things like distortion spectra, which I believe is a very important clue. I think most of us agree that even order harmonics is a more "pleasurable" form of distortion compared to odd orders, one of the long argued pro's of tubes vs. SS. But this no way implies there is another form of "math" for tubes as for SS. Your quoted statement seems to imply exactly this - the existence of a "tube amplifier math" being different from solid state.

EDIT:
Several posts ticked in while I was writing.. my crappy old laptop has a lot of contact bounce, so I use more time to edit out the errors than actually writing

Just for tthe record:
I AM an orthodox engineer WITH a physics background, and psychology is exactly one of the factors we like to keep out of the equations - it just clutters the mind and make you see things that don't exist.......maybe even hear them, too.........

( I used to operate FFT routines in the early 80s, when a simple 512 pt. took 12 mins to execute on a Commodore CPM...)

[Wavebourn]

Originally posted by EC8010
To get back (slightly) to the thread topic, you could measure distortion by plotting the transfer characteristic directly, or its differential. But in most cases, the distortion is sufficiently low that it's difficult to do, whereas applying a sine wave, filtering out the fundamental, then measuring the residual is easier.

The differential of a reasonable transfer characteristic should be a constant DC with deviations, so AC coupling it and amplifying would give just the deviations. I suspect that noise would be a problem...

The thing is, for all the criticisms levelled at it, measuring harmonic distortion is a good way of quantifying nonlinearities, and a harmonic weighting filter can always be added before the meter.

Exactly!

It was convenient many years ago, but such convenience leads to wrong directions when horrible sounding devices are considered better than better sounding ones. When transistor amplifiers took over tube amplifiers in 1970'th they were better according to such convenient measurements. Filters are good for measurement of frequency response, but for linearity measurements "windows" needs to be used instead.

[Wavebourn]

Originally posted by AuroraB



Just for tthe record:
I AM an orthodox engineer WITH a physics background, and psychology is exactly one of the factors we like to keep out of the equations - it just clutters the mind and make you see things that don't exist.......maybe even hear them, too.........

We agreed with John Curl in a phone discussion that best engineers have experimental physics background and they can't be called orthodox engineers, but heretics instead, may be.

Speaking of psychology, it may help to suggest hypnotically things that don't exist; but it helps also to understand perceptions better in order to satisfy our perceptions. And it tells that exponential signals are preferable, because the louder is the sound, the less nuances we hear, until a pain threshold is reached on level of 140 dB. T

hat's why I use an exponential signal instead of a sinusoidal one. But such a signal is hard to analyze using Fourier transformations. No big deal, since modern A/D and D/A devices have high enough resolution to compare signals point to point. When they generate steps in decibels instead of steps in microvolts such measurement would be more relevant because we less care about microvolts on 10 volt level than on a millivolt level.

I.e. I measure linearity directly weighting results dynamically, i.e. more steps closer to zero, less steps closer to maximal output.

[Miles Prower]

Originally posted by Wavebourn
If you disagree why are you still in vacuum tube's forum instead of SS one? Their designs are better according to "standard" measurements than designs on this one.

Wouldn't do any good. big Cat Fight going on over there about the same topic.

[Conrad Hoffman]

Maybe not relevant now, but some posts back I wanted to mention that THD analyzers don't actually single out HD. That is, they don't have any type of comb filters to limit them to HD. They null out the signal and whatever is left is, to the THD meter, HD. For all the reasons stated in this thread, that's almost entirely HD and noise, but whatever is there, gets read on the meter.

[xiphmont]

Originally posted by Wavebourn

You look from the wrong prospective. Non-linearity may be measured by angles. Any non-linearity. Without any complex transformations. Because a curve is direct graphical representation of a function.

I will point out that you're talking only about time-invariant nonlineariities. I know that's what you meant, but it's worth mentioning.

One or the things people are constantly mentioning (without providing any evidence) is non-time-invariant distortion as something that 'scientific measurements' miss and people hear.

Originally posted by Conrad Hoffman
Maybe not relevant now, but some posts back I wanted to mention that THD analyzers don't actually single out HD. That is, they don't have any type of comb filters to limit them to HD. They null out the signal and whatever is left is, to the THD meter, HD. For all the reasons stated in this thread, that's almost entirely HD and noise, but whatever is there, gets read on the meter.

That's THD+N and some analyzers certainly do notch out the nonharmonic data. My little spectrum analyzer app does in fact only inspect harmonics in THD mode, optionally ignoring the second. Of course, once you're in the digital domain, you have the full power of billions of cheap transistors to do your analysis bidding so adding many analysis variants is easy.

[Conrad Hoffman]

Don't waste your time, that digital stuff will never catch on.

[xiphmont]

Originally posted by Conrad Hoffman
Don't waste your time, that digital stuff will never catch on.

/me checks his wallet ;-)

[Wavebourn]

Originally posted by Miles Prower


I will point out that you're talking only about time-invariant nonlineariities. I know that's what you meant, but it's worth mentioning.

Right, when I trace a curve of one stage on very low frequency I may assume it as a time-invariant nonlinearity. But in the real world it is hard to imagine, especially when solid state devices are involved in the signal path.

Originally posted by Conrad Hoffman
Don't waste your time, that digital stuff will never catch on.

Modern 24 bit / 96 KHz converters are pretty good already for such measurements.