The current state of distortion measurements is bothering me

[anphex]

I might be wrong, but hear me out.

Whenever there is a distortion measurement in Audio Precision at the usual levels of 86 dBSPL and 96 dBSPL, in my understanding, this measurement is 100 % relative to the frequency response of that speaker.
Meaning that if the speaker already rolls off at around 120 Hz, of course it will have lower distortion at sub frequencies, artificially making it look more distortion-less than it really is.

I noticed this profoundly when I look at the KEF R11 Meta measurements. These speakers have a pretty strong bass-shelf reduction knee of about 6 db. This means that the 86 dbSPL of that distortion measurement are actually, for that range below 100 hz, only 80 dBSPL which is a range probably any semi decent speaker can handle. I am not making this example to pick on KEF but to show my point on this issue. (Measurement btw: https://www.audiosciencereview.com/forum/index.php?threads/kef-r11-meta-tower-speaker-review.53282/)

Isn't it a possibility during measurement to use the Audio Precision frequency response result to create a FIR correction to completely linearize the speaker from 20 to 20000 Hz and then make another distortion measurement normalized to 86 dbSPL (for me the level that is closest to real life usage) and the look for the "real" distortion behaviour? This way, all speakers would be actually comparable for once regarding their distortion behaviour.

Please correct me if I am taking a wrong turn with my thinking here.

 

[staticV3]

I believe what you're looking for is Klippel's equivalent input harmonic distortion.

 

[Sokel]

I just had a look at the Tekton to see what you mean.
There's two charts,one relative as you say but there's also the percentage THD chart where tilting does not matter as it seems cause the Tekton has kind of the opposite tilt.

(maybe you mean something else though,which I didn't got)

 

[AnalogSteph]

Isn't it a possibility during measurement to use the Audio Precision frequency response result to create a FIR correction to completely linearize the speaker from 20 to 20000 Hz and then make another distortion measurement normalized to 86 dbSPL (for me the level that is closest to real life usage) and the look for the "real" distortion behaviour?

Of course you could do this - but would this be representative of the speakers' actual use? After all, a bunch of speakers of the active persuasion specifically get a steep rolloff at the low end to improve level handling. And then the KEfs are explicitly shelved to make up for boundary/room gain.

Whatever you do, neither variant will be ideal.

One could still make a point for EQing flat from about 100 Hz upwards. That seems a lot more doable to me.

 

[No. 5]

Whenever there is a distortion measurement in Audio Precision at the usual levels of 86 dBSPL and 96 dBSPL, in my understanding, this measurement is 100 % relative to the frequency response of that speaker.
Meaning that if the speaker already rolls off at around 120 Hz, of course it will have lower distortion at sub frequencies, artificially making it look more distortion-less than it really is.

Non-linear distortion is always plotted relative to the level of the fundamental, not the frequency response. In other words, the measurement is showing the level of each added harmonic component relative to the signal that created them. So for the 86dB plot of the KEF R11 Meta, it can be seen that applying an input of a 30Hz tone into the speaker at a given voltage produces a 72dB 30Hz tone output (the fundamental), but it also produces a 25dB 60Hz tone (the second harmonic), and a 29dB 90Hz tone (the third harmonic), and a 32dB 150Hz tone (the fifth harmonic). So you end up with a speaker that has 1.7% THD at 30Hz (at that voltage level) because some additional sound pressure is coming from those harmonics. The SPL of other frequencies have no effect on that.

Note: some may have noticed that I didn't mention the forth harmonic. That's because its relative level is below the threshold Amir's measurements show.

This means that the 86 dbSPL of that distortion measurement are actually, for that range below 100 hz, only 80 dBSPL which is a range probably any semi decent speaker can handle.

What is more important for understanding distortion, is not the absolute SPL per se, but what the input voltage level is (and resulting current draw) because loudspeaker motors and suspensions become more non-linear (and produce more distortion) the greater the distance they have to move the diaphragm from its rest position, and the higher the input level, the further they move. Additionally, the lower the frequency that the speaker is reproducing, the greater the distance the diaphragm must move.

What this means is that even though less sound pressure is produced at lower frequency for any given input level, the speaker actually has to work a lot harder to do it. So the woofers may be moving twice as far at 50Hz as they are at 100Hz to produce half the sound pressure. And this can be seen in any speaker's distortion measurement: SPL goes down with frequency, but distortion goes up.

Isn't it a possibility during measurement to use the Audio Precision frequency response result to create a FIR correction to completely linearize the speaker from 20 to 20000 Hz and then make another distortion measurement normalized to 86 dbSPL (for me the level that is closest to real life usage) and the look for the "real" distortion behaviour? This way, all speakers would be actually comparable for once regarding their distortion behaviour.

You could use FIR correction to completely linearize the speaker from 20Hkz to 20kHz, but given the information above, it should be clear that the result would be waaaaay more distortion than the speakers would produce otherwise. Continuing to use the R11 Meta as an example, it would need almost 30dB of boost to be flat down to 20Hz. That's potentially a thirty-fold increase of the woofers excursion and with the added complication of being a ported speaker so the woofers would lose the protection of the box air spring below the tuning frequency.