New means of distortions' measurement (Weighted Distortions)

30 years ago I was young and stupid and invented Watt*Percent, i.e. power multiplied by THD to compare apples and oranges, now I don't think this measure is valid because harmonics need to be weighted, i.e. they need to have own coefficients of nastiness, and the higher is power the more excusable are distortions.

So, this will be better: WD = (L2*k2+L3*k3 + ...)/P

where L is power of the harmonic, k is coefficient of nastiness, WD is coefficient of Weighted Distortions. For the beginning, k may be equal to the number of the harmonic.
So, if to simplify, WD = ( L2*2 + L3*3 + L4*4...)/P

Curves for different amps would be interesting to see (power horizontally, WD vertically). The ideal amplifier should have always WD as close to 0 as possible.

I wouldn't base k simply on the number of the harmonic. Odd-order harmonics have a greater "nastiness coefficient", as you put it. In other words, 3rd harmonic is more objectionable than 4th.

Even with weighting of harmonics, steady-state THD measurements don't tell the whole story, anyway. (And I say that as someone who uses steady-state THD measurements because... let's face it, they're convenient).

I don't understand the point of this proposed "figure of merit." 1% harmonic distortion doesn't sound better at a higher output power, just louder.

IMO THD versus frequency tells you alot. especially if you plot the individual harmonics vs frequency. why try to sum it up into one number? if you can measure the relative amplitudes of each harmonic, show a family of curves.

if you can come up with a 3D plot that shows each harmonic vs frequency vs level, without getting messy and cluttered, id like to see it.

THD versus frequency tells you alot. especially if you plot the individual harmonics vs frequency.

Click to expand...

Absolutely.

This is too subjective for me.. All distortion is bad.

As I've probably mentioned before my go-to test was a modified two-tone IMD using 19:20Khz. Stressor was in band, distortion was very much in band.

Most NFB circuits are at their worst at the highest frequencies, so well exercised by 19+20kHz. Simple THD products are often out of band or sometimes LPF by circuitry being measured, but IM product at 1kHz is smack in middle of most passbands.

I found the two tone test pretty representative of real world performance, but YMMV.

JR

What ratio, John? 4:1 as in the usual 60Hz+7kHz test, or something else?

It's interesting to note that IMD was the preferred figure of merit in the '50s. The thinking was that a perfectly linear transfer function could not create intermodulation products--and that such products were more objectionable than simple harmonics. Sometimes THD wasn't even listed even while an IMD figure was quoted.

[quote author="NewYorkDave"]

It's interesting to note that IMD was the preferred figure of merit in the '50s. [/quote]

Because they measured mostly the 2'nd and 3'rd order harmonics, if to be honest. Push-pull amps looked better against SE amps, it was convenient. When people wanted to push transistors on market they measured total THD on 1 KHz frequency. Now, for OpAmp-like topologies couple of upper in the band frequencies is used (Bob Cordel, for example, likes similar approach, because he again and again develops all the same powerful OpAmp with FET output).

I want to push nothing on market, so I am interested mostly in some kind of aggregated parameter that shows a degree of sonical nastiness for absolutely subjective listener.

For example, weighted noise measurements reflect totally subjective, human-dependent parameters, why do you guys want some objective measurement of distortions?

Not everyone is in favor of weighted noise measurements. I think they're kind of a "cheat", myself.

Band-limited noise measurements are a different story; they concentrate on noise you can hear but don't confer more gravity on some frequencies over others.

[quote author="NewYorkDave"]Not everyone is in favor of weighted noise measurements. I think they're kind of a "cheat", myself.
[/quote]

You may try using this application: http://www.phys.unsw.edu.au/jw/hearing.html

However, your results may differ from official curves that I believe reflect acoustical parameters of the hall where Fletcher and Mudson experimented, also they had different amplifiers and speakers.

Band-limited noise measurements are a different story; they concentrate on noise you can hear but don't confer more gravity on some frequencies over others.

Click to expand...

Hmmm.... I don't understand what you mean. Can you explain please?

High-frequency IM tests are indeed useful for sleuthing out certain types of misbehavior. I use 19kHz + 19.5kHz, putting the first IM product (down-scale) at 500Hz, well away from the territory of hum. HF IM tests are typically done with a 1:1 ratio. A 3-tone test is also useful; the third tone can be 9.6kHz. That will generate 200Hz and 300Hz tones in addition to the 500Hz. All of these will also, if the DUT is bad enough, generate multiples: 1kHz, 2kHz, etc., along with 400Hz, 600Hz, etc..

Old-style SMPTE-IM (a low frequency plus a high frequency, typically 70Hz + 7kHz, in a ration of 4:1) is also useful; it shows up, among other things, poor power supply regulation in circuits which also have poor power supply rejection.

And I still look at 1kHz. Seeing how the harmonic spectrum varies is sometimes a good clue to what the device will actually sound like.

All of these are still models. Ideally you could simply subtract the input signal from the (attenuated) output signal and listen to the results, with music as the signal source. (Effectively a straight-wire bypass test.) David Hafler did that in the (I think) 1980s. The catch, though, is that this test gives high error signals unless the amplifier has very high bandwidth, because of phase shift at top and bottom of the audio band. Without even beginning to wade into the swamp of whether phase distortions are audible, this test privileges such distortions so powerfully that they swamp any other possible error mechanisms.

I believe Peter Walker at QUAD developed what he called a "crooked wire bypass test", where the amplifier's output signal was compared, not to its own input, but to the input signal which was processed through passive networks having the same phase response as the amplifier being tested. Those networks turned out to be quite complicated, however, and I don't thing anyone pursued this otherwise excellent idea.

In short, nobody has come up with a single "figure of merit" test for audio circuits. To properly characterize them still requires a battery of tests and good judgement.

Peace,
Paul

Oh yeah, and somebody (Walter Jung? Robert Cordell?) published an article in JAES comparing different methods of distortion measuremend, including 1kHz THD, 20kHz THD, SMPTE-IM, high-frequency IM, a mixture of sine and square waves (Otala's TIM test) and the "noise transfer test" (you high-pass filter white or pink noise, then run it through the DUT and low-pass filter the result. Anything remaining is distortion.) The conclusions were more or less inconclusive, finding at least one distortion mechanism that each measurement method utterly failed to measure. Nothing popped to the top of the heap, although I seem to remember that noise transfer was at or near the bottom.

Peace,
Paul

[quote author="NewYorkDave"]What ratio, John? 4:1 as in the usual 60Hz+7kHz test, or something else?

It's interesting to note that IMD was the preferred figure of merit in the '50s. The thinking was that a perfectly linear transfer function could not create intermodulation products--and that such products were more objectionable than simple harmonics. Sometimes THD wasn't even listed even while an IMD figure was quoted.[/quote]

IMD AFAIK was invented to look for problems with film projection and 7 kHz was considered HF for that application.

I used 19:20kHz at 1:1. That test waveform had a slew rate of a full amplitude 39kHz sine, so a significant HF test. I modified an old heathkit SMPTE analyzer since the archaic 60 Hz/7 kHz test was too easy for everything I was working on. My Heathkit THD analyzer was limited too, but I bought an old used spectrum analyzer and fed the distortion product from the heathkit into the SA for a respectable visibility into distortion products. The SA only had a 50 db range but added to the Heatkit's THD notch filter it got the job done well for 20 years ago and low budget.

I don't claim my two tone IMD was an all encompassing test, but I found it extremely useful. I was designing phono preamps at the time so conventional THD measurements would read better than the reality due to distortion products being reduced by the RIAA playback EQ. My little home brew tester instead had the the distortion products (1 kHz) boosted some 20 dB by the playback EQ and was rather instructive for parsing out weaknesses between sundry preamp design approaches.

On the subject of noise measurements, bandwidth limiting is absolutely important otherwise the wider response circuit will appear noisier than it sounds. Weighting curves can be useful to correlate with human hearing, but since our hearing curves are level dependent, it seems these weighting curves would also need to be anchored to a nominal level (in other words if the noise is loud enough you'll still hear it curve or no curve).

JR

Paul;
indeed many brilliant engineers used own tests in order to discover what still is wrong and why power amplification screws down sound clearness despise other tests proposed very clean reproduction; indeed some of them also invented new tests for marketing purposes to show own works in better light. I personally highly respect Peter Walker, but I doubt that he was satisfied by that test that demonstrated errors ignoring phase shifts only. His works demonstrate that he knew much more than explained publicly; his works are based on deep understanding of human perception. But his "Current Dumping" patent demonstrate that he wanted to hid his understanding from public. I agree that listening to errors ignoring phase shift helps to understand what is going on, and such test may be used to discover "coefficients of nastiness", but it does not show dependence of impact of errors on perception on power level. Walker's works show that he was aware of such dependence, but did not talk of it loudly.

[quote author="JohnRoberts"]

On the subject of noise measurements, bandwidth limiting is absolutely important otherwise the wider response circuit will appear noisier than it sounds. Weighting curves can be useful to correlate with human hearing, but since our hearing curves are level dependent, it seems these weighting curves would also need to be anchored to a nominal level (in other words if the noise is loud enough you'll still hear it curve or no curve).
[/quote]

May be we'll hear the part of it that is over the threshold? People say that transistors "hiss" (high end of specter!), while measurements show 1/f that is contrary to audio perception?

There's an IM test with carefully considered numerology that Gerald Stanley, the prime mover at Crown, considers the most effective for evaluating audio equipment. In particular it avoids certain fortuitous and misleading cancellation effects that can occur with other combinations of stimuli.

Unfortunately I don't have the reference handy, but I'll bring it up when I find it.

[quote author="bcarso"]There's an IM test with carefully considered numerology that Gerald Stanley, the prime mover at Crown, considers the most effective for evaluating audio equipment. In particular it avoids certain fortuitous and misleading cancellation effects that can occur with other combinations of stimuli.

Unfortunately I don't have the reference handy, but I'll bring it up when I find it.[/quote]

An engineer I worked with at Peavey wrote a paper on the subject. He proposed a 3 tone test with specific ratios to overcome limitations of simpler tests.

Preprint #4803, "A NEW CLASS OF IN-BAND MULTITONE TEST SIGNALS" by Jon M. Risch, Presented at the 105th Convention of the Audio Engineering Society 1998 September 26-29th, San Francisco, CA.

JR

There are a million ways to measure all kinds of different distortions. Audio Precision will let you test IMD about 5 or 6 different "basic" ways with many possible variations. In the end there is no better tool than your ears. (Though some kinds of distortions are actually enjoyed by some listeners). I personally think that IMD correlates pretty closely to THD for modern designs.

I have seen some people use A-wtd filter for THD measurements on switch-mode power amps. I think that is a bogus methodology.

>>>However, your results may differ from official curves that I believe reflect acoustical parameters of the hall where Fletcher and Munson experimented, also they had different amplifiers and speakers.

Funny thing I found a couple years ago while looking at loudness curves. The actual number of test subjects that F&M used to derive their curves. It was only 11 subjects. Wow, talk about poor statistical sampling... But some more recent studies show that they were pretty close.

Check this article for more info on loudness curves:
http://www.nedo.go.jp/itd/grant-e/report/00pdf/is-01e.pdf

HTH!
Charlie

[quote author="bcarso"]There's an IM test with carefully considered numerology that Gerald Stanley, the prime mover at Crown, considers the most effective for evaluating audio equipment. In particular it avoids certain fortuitous and misleading cancellation effects that can occur with other combinations of stimuli.
[/quote]

There is another set of test signals based on the mathematical Fibonacci sequence. The reason to use this kind of signal is to avoid the masking of generated harmonics (distortion) by the test signal. Also there is an advantage that there is a way to determine which components are at the origin of the resulted harmonics. Also because of the complex nature of the signal it has a very similar crest factor like real music has (20dB between RMS and peak)

read here:
"A NEW CLASS OF IN-BAND MULTITONE TEST SIGNALS
Jon M. Risch, Peavey Electronics Corp., Meridian, MS. USA Copyright 1998
Presented at the 105th Convention of the Audio Engineering Society
1998 September 26-29th, San Francisco, California as preprint #4803"
http://www.geocities.com/jonrisch/PhiSpectral1.htm

and here:
http://www.geocities.com/jonrisch/page10.htm

I generated myself these signals digitally at 24bit/96K resolution and use them for testing.

chrissugar

IMO ears are OK for final QA but make poor design aids.

I concur that A weighting may not be optimal for THD measurements but it seems unfair to lump switching amp artifacts above our hearing in with total nonlinearities. This strikes me as yet another candidate for multi-tone IMD as the stress and distortion products are mostly in band and unaffected by output reconstruction filtering.

JR

[quote author="JohnRoberts"]IMO ears are OK for final QA but make poor design aids.
[/quote]

It greatly depend on experience of the owner of this brilliant tool.
Also, psychological organization impacts: bright visual for sure will trust curves on a screen and paper more than own ears and will not hear sounds looking at curves.

I concur that A weighting may not be optimal for THD measurements but it seems unfair to lump switching amp artifacts above our hearing in with total nonlinearities. This strikes me as yet another candidate for multi-tone IMD as the stress and distortion products are mostly in band and unaffected by output reconstruction filtering.

Click to expand...

I do not know what is really above our perception, though I believed before in accepted boundaries of hearing ranges... It does not work, and huge money audiophiles pay for audiophoolery prove that... I know many of them as smart, critically thinking, and well hearing people...