New Research on Audibility of Distortion in Headphones

[illusonic]

Sorry, but no. The whole basis of figuring audibility of low frequency distortion is the fact that the non-linearity is so strong, that it creates high order harmonics that land where our hearing is most sensitive: 3 to 5 kHz. High pass filter a speaker and you often hear increased clarity in those ranges.

IMD in higher frequencies will only have an effect if you don't have these bass harmonics. Even then, the music levels are going to be far lower than bass causing IMD components to be even lower than they already are. Artificial test tones at full amplitude do not represent actual music use case. I can easily make a case of a speaker being drive to maximum capability in bass. You can't do that with straight face for 1 kHz upper tone in IMD.

See the results of the research just posted where IMD had no correlation with listening results.

That is IMHO not a correct interpretation of distortion mechanisms. IMD products are not related to harmonics of the lower test frequency f1 and easily much higher in level than harmonics of f1. As was already pointed out above, f1 (high level bass signal as in music) drives the speaker through it's nonlinear BL range. The much lower level (again, like in music, for example 20dB below f1) midrange frequency f2 by itself would never excite the excursion that f1 causes. But since f1 is modulating BL, the strength of the motor is modulated and f2 is affected by that. The resulting modulation products are mainly f2+/-f1 and f2+/-2xf1. Harmonics of the bass caused by f1 play a minor role.

 

[amirm]

You are misunderstanding what I wrote...

 

[amirm]

That is IMHO not a correct interpretation of distortion mechanisms. IMD products are not related to harmonics of the lower test frequency f1 and easily much higher in level than harmonics of f1. As was already pointed out above, f1 (high level bass signal as in music) drives the speaker through it's nonlinear BL range. The much lower level (again, like in music, for example 20dB below f1) midrange frequency f2 by itself would never excite the excursion that f1 causes. But since f1 is modulating BL, the strength of the motor is modulated and f2 is affected by that. The resulting modulation products are mainly f2+/-f1 and f2+/-2xf1. Harmonics of the bass caused by f1 play a minor role.

Ok, let's step back as it seems that folks still think there is something magical happening when we use two more tones measuring a driver/speaker. Let's agree on the following:

1. If a system is 100% linear, then it does NOT have IMD or harmonic distortion. No new distortion products are created by definition and math.

2. Inversely, if a system is non-linear, it will create both IMD and harmonic distortion. It is never the case that it only creates IMD. Again, this is a matter of math.

Given this, what you have to show is that IMD distortion levels are higher than harmonic distortion for it to matter. And that such distortion is above threshold of hearing. Nothing stated in support of IMD demonstrates either.

Harmonic distortion as I have shown in every speaker review, is quite substantial. That by itself brings out issues that the designer needs to solve. Whether there is also some IMD, would only matter if it rises to good portion of harmonic distortion. No one has shown this to be case. And I post research that shows at low frequencies where nonlinearities dominate, IMD is below threshold of hearing whereas harmonic distortion is not.

We also have another proof point in the case of headphone research as shown:

Correlation was found with harmonic distortion but not so much with IMD, etc.

Further, what folks are asking out of the basket of Klippel tools is multitone. Multitone is a catch all measurement that includes many things including artifacts of the measurement system itself. It also has poor SNR. It also creates a noise+distortion spectrum that does not lend itself to psychoacoustics analysis.

If you are designing a speaker, you better aim to reduce harmonic distortion as you can as that also reduces causes of IMD.

 

[illusonic]

1. If a system is 100% linear, then it does NOT have IMD or harmonic distortion. No new distortion products are created by definition and math.

2. Inversely, if a system is non-linear, it will create both IMD and harmonic distortion. It is never the case that it only creates IMD. Again, this is a matter of math.

1. TRUE

2. DEPENDS ON DISTORTION MECHANISM AND FREQUENCY

Nonlinearities of the suspension (Kms) only cause low-frequency distortion, while nonlinearities of motor force (BL) cause distortion throughout the whole spectrum. The reason is that at higher frequencies inertia of the moving mass (Mms) largely outweighs suspension force (Kms), which neutralizes Kms nonlinearities. But motor force (BL) is required throughout the whole spectrum to move the membrane.

At medium to high frequencies, HD based on BL-nonlinearity is low because of low excursion. Once bass plays in parallel with midrange, excursion rises and midrange is affected by BL nonlinearity. The result is IMD in the midrange caused by bass tones. Midrange IMD rises with bass level.

Without knowing the nonlinearities of a mid-woofer (BL/Kms), the only way to know if it produces excessive midrange IMD is a measurement with more than one tone playing simultaneously. To create a meaningfull result at least one frequency must be in the bass and one in the midrange.

The Purifi tech note we discussed earlier, describes and demonstrates this in a very compelling way.

 

[Jaxx1138]

It would be nice to get some vindication finally, for a long time I have thought it is easy to hear how much punchier/faster low distortion planar magnetic bass sounds, especially in electronic music or signal closer to square waves. Even when comparing to other very high end dynamic headphones with level matched Harman EQ, the bass difference is very audible for me at normal listening levels. But a lot of people think it is nonsense.

All we need is some measurements to get started.

 

[amirm]

2. DEPENDS ON DISTORTION MECHANISM AND FREQUENCY

Which makes any standardized test for review purposes, impractical. A point I made earlier that this kind of testing is useful for drivers, or designers of speakers. But not for reviewers.

 

[amirm]

To create a meaningfull result at least one frequency must be in the bass and one in the midrange.

Proving what I said above. You can't pick any two tones for IMD test and expect that condition to happen in a single driver.

 

[Jaxx1138]

1. TRUE

2. DEPENDS ON DISTORTION MECHANISM AND FREQUENCY

Nonlinearities of the suspension (Kms) only cause low-frequency distortion, while nonlinearities of motor force (BL) cause distortion throughout the whole spectrum. The reason is that at higher frequencies inertia of the moving mass (Mms) largely outweighs suspension force (Kms), which neutralizes Kms nonlinearities. But motor force (BL) is required throughout the whole spectrum to move the membrane.

At medium to high frequencies, HD based on BL-nonlinearity is low because of low excursion. Once bass plays in parallel with midrange, excursion rises and midrange is affected by BL nonlinearity. The result is IMD in the midrange caused by bass tones. Midrange IMD rises with bass level.

Without knowing the nonlinearities of a mid-woofer (BL/Kms), the only way to know if it produces excessive midrange IMD is a measurement with more than one tone playing simultaneously. To create a meaningfull result at least one frequency must be in the bass and one in the midrange.

The Purifi tech note we discussed earlier, describes and demonstrates this in a very compelling way.

I am curious, why would it matter where the distortion is coming from as long as it shows up, or does not show up?

A transducer engineer has to find the source of distortion for a driver and where it is originating from.

If you evaluate a SYSTEM (completed active or passive unit) during the evaluation (like a reviewer) , you are no longer looking for sources of distortion but just a test of its performance and therefore you are looking at distortion across the board.

It appears as though you are looking at distortion characteristics from a designers point of view and Amir is arguing the more basic function of looking for distortion regardless of where it emanates from.

Isn't it obvious that IMD requires more than one tone in order to test it. Otherwise what are you modulating?

Maybe I am missing the point completely and am entirely lost here. Please help me understand this more clearly.

 

[Jaxx1138]

Proving what I said above. You can't pick any two tones for IMD test and expect that condition to happen in a single driver.

You cannot standardize that test for all conditions. But you can define parameters based on the passband. But above certain frequencies this test may not be very useful. It assumes a single driver of course.

 

[Jaxx1138]

You cannot standardize that test for all conditions. But you can define parameters based on the passband. But above certain frequencies this test may not be very useful. It assumes a single driver of course.

Nevermind, I believe you already stated this in a previous response.

 

[Chris A]

A few references that are apropos (enclosed) with respect to loudspeakers:

1) a short excerpt article on the effects of air-loading of drivers (whether or not that loading is good or poor and its effects on the output, including distortion)
2) effects of FMD (Doppler) in listeners, by Keith Howard, discussing its audibility and comparing FMD to jitter in electronics.
3) and a final reference to loudspeaker distortion vs. bass bin type

For your reading pleasure.

Chris

 

[Doltonius]

I do not feel the same at all, playing high dynamic range music with deep and strong bass. Use proper judgement and listen to your body and you will be fine in such usage in moderation.

However I think modern music recordings do not have a lot of dynamic range? Max 30db for classical recordings. If popular music then the dynamic range is extremely narrow. And so it is unlikely that bass alone is ~100db yet everything else is just 70db. In a usual classical recording, the brass instruments like trumpets and trombones will be as loud if not louder than the bass drums, and they of course produce a lot of high frequencies. And so unlikely you are listening in the way you are describing. You need to get specially mixed and produced music for that.

 

[amirm]

However I think modern music recordings do not have a lot of dynamic range? Max 30db for classical recordings. If popular music then the dynamic range is extremely narrow.

There is a huge difference between modern music and pop. Soundtracks for example are modern but can have incredible dynamics including deep bass. Even in pop there are well recorded tracks. Vast majority of my reference tracks are modern recordings.

In a usual classical recording, the brass instruments like trumpets and trombones will be as loud if not louder than the bass drums, and they of course produce a lot of high frequencies. And so unlikely you are listening in the way you are describing.

Maybe there are some classical tracks that are that way. And if so, I would not crank them up high as that spectrum is bothersome to the ear.

 

[MRC01]

However I think modern music recordings do not have a lot of dynamic range?

Depends on the genre. Mostly true for pop/rock. Modern classical and some jazz is largely free of the nonsense of heavy-handed post-processing and dynamic range compression.

... Max 30db for classical recordings. ...

Max? Nope. 30 dB is typical for classical, and can be 60 dB for some large ensemble symphonic music.

 

[PierreRtings]

Then you're clearly not familiar with these areas at all.
It doesn't matter in which areas studies, research, measurements, etc., take place—for example, food, health, environmental pollution, and so on.
And there have always been attempts to manipulate the parameters to favor the desired outcome.
This isn't always malicious; often it's simply about obtaining more or more accurate results, but this can also distort everything.
There are hundreds, if not thousands, of examples.
Just Google it; it will shock you.

Therefore, one must always question the results of such research and the methods used to obtain them, examining and scrutinizing them from all sides.

Are you familiar with the quote, "Never trust a statistic you haven't falsified yourself"? There's more truth to that than you seem to realize.

Addendum due to current events:
I have only made a general statement here regarding studies and statistics.
At no time was it my opinion, assumption, or statement that the study ‘The Perception and Measurement of Nonlinear Distortion in Headphones’ by Dr. Sean Olive and co-written with Pierre Lelièvre, or its results, were in any way flawed or manipulated.
The same applies to ‘Measuring Distortion the way we Hear It’ by Steve Temme.

I conducted the listening tests. I there had to be a Bias, it would have to be the other way around. Me and Sean where both stoked by the results. I was concerned there could be falses positive, but false negatives, no. In the sense that people could hear it, but the test would make them fail, no. Don't forget they where listening a song recorded trough a headphone, to the same recording untouched. The participants where given any possible ways that if there was something tpo be heard, we've let them try very hard. They could set a loop on very short passage, if they feel they where sensing some small differences on a word or a transient or a drum hit, They could try for as they wanted. To us it would have been cool to have listener able to easily hear the distortion. There would have been more to report and to bring to the science.

 

[PierreRtings]

Thanks for linking to the report. They did a lot of work but I am uneasy with the experimental designs. Even the terminology seems odd to me. Using Klippel tool for distortion training is also not what I would have used. The test itself should have been used for training. Lack of control was unfortunate as well. As is blending the average outcome for a group as the basis of the conclusion. Even use of 5128 is odd as it has higher distortion then my GRAS 45CA due to smaller microphone size. Music selection raises an eyebrow for me as well. Fast Cars is good for tonality testing but would not be my test track for distortion. Using one of the microphone with all of its own distortions as the baseline is also a conflating factor.

I was also surprised that they did not ask the listeners what was too loud as another side experiment. Why just assert that 104+ dBSPL is too loud? And why use an average number like that rather than peak SPL?

Fortunately some of the outcome is as expected in that few people have higher sensitivity than the general population.

Some of those are valid concerns, I guess, but may I ask what you mean by lack of control? There is indeed various way to report SPL, and indeed it can induce confusion. but the way we reported it, when we say too loud, there is absolutely no one that would ever listen to these song at the audibility tresholds that we got. To clarify, we have assesed these Level, By normalizing noise at the same LUFS as the music extract, and measuring a weighted SPL. Think about it for a second, where the peaks would be. Or where the Bass SPL would be. 104+ dB SPL measured that way, is stupidly loud. Asking the listeners if they find this loud and not just assert it would have got me sued for damaging their hearing.

 

[amirm]

Some of those are valid concerns, I guess, but may I ask what you mean by lack of control?

Objectively and subjectively audible distortion. This would verify the overall protocol and weed out listeners who can't detect such impairments at all.

 

[amirm]

but the way we reported it, when we say too loud, there is absolutely no one that would ever listen to these song at the audibility tresholds that we got.

The song choices highly impact this assessment. Why did you use preference test tracks for this use???

 

[PierreRtings]

Objectively and subjectively audible distortion. This would verify the overall protocol and weed out listeners who can't detect such impairments at all.

Sorry I don't fully understand what is the distinction between subjectively audible and objectively audible. ABX assess obviously objectively audible, if you can't differenciate an audio clip till the other, you objectively don't hear it. The similarity test is a "subjective test" We are asking how similar this recording is to the clean file. Keep it mind, that was our goal, to determine the impact on preferences. Subjectivity was really on our work table, but you can't prefer what you don't hear. but if you cannot objectively detect it as you cannot even find which is the hidenn reference, what can we do?

 

[amirm]

Sorry I don't fully understand what is the distinction between subjectively audible and objectively audible.

You objectively and hence provably generate a distorted clip. You then listen to it with your trained listeners to make sure impairments are audible. This is sometimes called a low anchor.