Multi-tone intermodulation distortion testing

[sciguy]

Hello,

I have recently stumbled upon this forum and am absolutely baffled. This place is awesome and exactly what I've been looking for.

Now to the meat of my post, this question may largely be targeted at amirm, but I would like anyone's input.

What is the feasibility of running intermodulation distortion testing on multiple tones at once? I see you use the Audio Precision APX555. It seems to be a very capable unit. I have always had a vision of running a 20-tone intermodulation test. Injecting 20 tones separated by half octaves from 20hz to 20khz, then measuring distortion in a similar way to how it seems that you do SINAD. To me this seems like a more ideal stress test for audio equipment, as a true music signal could have hundreds or thousands of tones of varying level that must be reproduced. Is this a feasible test? Is there something I'm not thinking of that makes this test impossible or impractically difficult?

 

[andreasmaaan]

Certainly possible, in fact quite a standard test Here's an interview with Tom Kite of Audio Precision about it.

 

[SIY]

I use that test routinely in my reviews. I've posted several examples here. So far, I haven't seen anything unexpected compared with two-tone measurements, but it's quick and easy.

For those without AP units, they can also be done in ARTA and Virtins MI.

Scott Wurcer has demonstrated a reverse test of this sort- he puts multiple narrow notches into a white noise signal. Very clever, IMO.

 

[andreasmaaan]

Do you have a link to info about the Wurcer test, @SIY? A quick google didn't help me much.

 

[SIY]

Hmmm, no, but he might have put some info up at diyAudio. I'll drop him a note and ask if he's published it somewhere accessible.

 

[RayDunzl]

I have always had a vision of running a 20-tone intermodulation test. Injecting 20 tones separated by half octaves

What's a half octave in in terms of frequency or in terms of what you're trying to express?

Looking up "half octave"
I find double the frequency - which is an octave above
And half the frequency - which is an octave below

 

[andreasmaaan]

What's a half octave in in terms of frequency?

The ratio is approximately 1.414 to 1

E.g. 1,414 Hz is approximately half an octave higher in pitch than 1,000 Hz.

 

[RayDunzl]

To me this seems like a more ideal stress test for audio equipment, as a true music signal could have hundreds or thousands of tones of varying level that must be reproduced. Is this a feasible test? Is there something I'm not thinking of that makes this test impossible or impractically difficult?

That's true and seems terribly complex but it still boils down to a single voltage level at any particular time in a recording, and the reproduction of the recording.

Here's a visual (represents voltage). 3000Hz sine, 4242.64068712 Hz sine (3000 x square root of 2), those two combined, and all 88 piano fundamentals (happen to have that laying around).

 

[Blumlein 88]

People have the idea 20 tones is more stressful than a twin tone test. However, you have to reduce the level of each tone in order to have more tones. So the two tone test is more stressful or reveals non-linearity better than a multi-tone test. Typically distortion and non-linearity drop with the decreased level of the stimulating tone.

I suppose SY has a history of doing the tests to view. I've done several versions of multi-tone tests and find it not to uncover anything useful.

I've also done more of the testing where multi-tones or high level noise are used with an octave gapped out to see if any IMD products show up in the blanked out zone. That hasn't uncovered anything useful either. I've not tested the number of items SY probably has, but there is a reason those tests haven't been used very often.

Doing this digitally when I've found anything unexpected above the noise floor in such tests it normally was a result of aliasing/imaging at a low level. There are better ways to find aliasing/imaging artifacts than using these tests.

 

[SIY]

People have the idea 20 tones is more stressful than a twin tone test.

Of course, "people" also refer to single tone tests as "static," when in reality, they sweep through all possible voltage levels...

As you surmised, my main use of multitone testing is illustrative rather than revealing of hitherto unseen distortions. Like my advocacy of the use of power averaging for determining DAC linearity, it's an "as well as" rather than an "instead of," mostly because doing these tests with modern equipment is fast and painless, and it's useful for tamping down some of the objections of Fourier deniers.

 

[RayDunzl]

People have the idea 20 tones is more stressful than a twin tone test. However, you have to reduce the level of each tone in order to have more tones.

True on level reduction, ambivalent about "stress".

The "88 Piano Notes" above, I think I set the level of each tone to 1/88th full scale ( 0.01136363636 ), a level of -38.89dBfs, in case all the waves "lined up" at some point.



They didn't, and the resulting file peak was at -8.859dBfs, close to the beginning.

 

[DonH56]

Multitone testing covering a wide frequency range provides more opportunity to see things like the effects of band limiting and hysteresis in the circuits. It can expose resonances and signal interactions difficult to see in a single-tone test (or even swept sine test). Such testing, including things like the NPR test, are standard in the RF world. But, I am not sure how much added value they provide in this context (low-level audio components). A swept two-tone test might be useful, or just at a few select frequencies, but there is always the problem of determining what is an adequate test. It can quickly get unwieldy... I used to do 100 Hz, 1 kHz, 10 kHz using closely-spaced tones plus one test using 100 Hz, 1 kHz, and 10 kHz (three-tone test) to check nothing bad happened. It rarely did, but sometimes things happened... But, it was very rare that IMD testing did not track HD testing classically, obviating the need for a lot of two-tone tests IME/IMO. Now and then it would expose some unexpected behavior, but again it's hard to say if it is worth the time to run such testing on everything. My world involved a lot of high-rel and space-qual components so testing was very rigorous. Not sure we need to go there for most audio stuff.

Note a multitone test using the IEEE methodology or similar can ensure peak values are excited given the appropriate choice of target frequencies, sample rate, and record length, but it can be tricky getting all of that right. I usually checked to make sure I was hitting FS signal levels during a test.

 

[RayDunzl]

My 88 tone test was to see the in-room response...

I was unconcerned (read satisfied) with the expected/assumed electronic result (which I can't measure precisely anyway):

All the peaks should be the same height, but, that isn't going to happen in a room...

JBL LSR 308 and DRC

Martin Logan reQuest with cheeze subs and DRC

I think I could see a little higher HD/IMD/Noise, 10dB or so, particularly in the JBL higher frequencies, not unexpected, mildly amusing to see.

 

[sciguy]

Wow this is awesome. I'm impressed at how many of you have experimented with the idea of multi-tone testing, if not a little bummed that I wasn't the only one to think of it.

 

[Werner]

The root of it all:

Belcher (BBC), Wireless World, May 1978:

https://www.americanradiohistory.com/Archive-Wireless-World/70s/Wireless-World-1978-05.pdf

 

[SIY]

Do you have a link to info about the Wurcer test, @SIY? A quick google didn't help me much.

I just heard back from Scott. The files are on Jan Didden's site:

https://linearaudio.net/downloads

Scroll down to "Test Waveforms."

 

[andreasmaaan]

I just heard back from Scott. The files are on Jan Didden's site:

https://linearaudio.net/downloads

Scroll down to "Test Waveforms."

Thanks!

 

[Cosmik]

It seems to me that the reason for using tones in the first place is that, unlike the more intuitive time domain null or differential test, benign 'distortions' are ignored e.g. phase shifts, slight DC offsets and so on, and averaging allows random noise to be ignored.

But this advantage is also a disadvantage as it means that the nature of some distortions remain unknown: if there was a system that went slightly unstable when a castanet followed a decaying bass note while playing into a certain type of speaker - or whatever - it might show some little wrinkles in the standard distortion plot but give no meaningful clue as to where and when the distortion takes place. Or the instability may not be triggered by the test at all.

This leaves a gaping hole for the "Fourier deniers" to walk through, and no multi-tone test will satisfy them. What is really needed is a test that can use real music (to satisfy the audiophiles) - the sort that might do strange things to an amplifier's power supply, etc. - and can quantify the deviation from ideal in the time domain, homing in on that little burst of oscillation that occurs as the amp's power supply recovers from delivering a big transient of a certain type into a certain capacitive cable - or whatever.

 

[SIY]

This leaves a gaping hole for the "Fourier deniers" to walk through, and no multi-tone test will satisfy them. What is really needed is a test that can use real music (to satisfy the audiophiles) - the sort that might do strange things to an amplifier's power supply, etc. - and can quantify the deviation from ideal in the time domain, homing in on that little burst of oscillation that occurs as the amp's power supply recovers from delivering a big transient of a certain type into a certain capacitive cable - or whatever.

That won't satisfy them, either. There's always somewhere to move the goalposts. Note that a 19/20 kHz two tone IMD test is more rigorous than any acoustical recording.

Coincidentally, I've been working on a collaborative project to use music signals to characterize the efficiency and intelligibility indices of transducers (where highly nonlinear behavior is the norm); we're hoping to publish our first paper sometime early next year.

 

[restorer-john]

Coincidentally, I've been working on a collaborative project to use music signals to characterize the efficiency and intelligibility indices of transducers (where highly nonlinear behavior is the norm); we're hoping to publish our first paper sometime early next year.

I think we should be the first to see it. You know, a bit of peer reviewing, constructive criticism and general proof reading...