Distortion perception - two studies

[andreasmaaan]

Over in the "Need for a general statement thread" a discussion sprang up about audible thresholds for non-linear distortion. I mentioned a couple of studies which appear to be very thoughtfully planned and well-executed, which examined aspects of this topic in different ways.

The first is by de Santis from 2007 and is called Perception & Thresholds of Nonlinear Distortion using Complex Signals.

Here, the author used software to introduce varying kinds and degrees of nonlinear distortion to music signals, and then had subjects undergo trials to try to determine which subjective reports correlated best with 4 different distortion metrics: the old and simple THD, IMD, and then the newer and more compex DS and Rnonlin metrics. Low correlations were found in the cases of THD and IMD, while impressively high correlations were found with the newer metrics. The speculated reasons for this were to do with THD and IMD's failures to take into account various aspects of human auditory processing, most importantly the relationship between critical bands in the auditory system and masking, but also other aspects. Although there is no discussion or experimentation in terms of absolute thresholds, it's a very interesting read.

The second study is by Robert-Eric Gaskell and is called Subjective and Objective Evaluation of Distortion in Analogue Electronics.

Here, the author digitally modelled non-linearities in capacitors and opamps and then set up controlled trials to determine audibility thresholds for the modelled distortions. There's a lot going on in this study but the most interesting conclusion from my point of view was this one:
"In the listening tests performed for this dissertation, participants were able to detect levels of capacitor distortion as low as 0.002% THD+N and levels of opamp distortion as low as 0.005% THD+N." Note, however, that the methodology used is complex and therefore this statement needs to be read in the context of the study as a whole.

Hope these provide food for thought.

Andreas

 

[Blumlein 88]

Your link to Gaskell doesn't work. Try this instead.

http://digitool.library.mcgill.ca/webclient/StreamGate?folder_id=0&dvs=1530790148842~680

 

[andreasmaaan]

Your link to Gaskell doesn't work.

http://digitool.library.mcgill.ca/webclient/StreamGate?folder_id=0&dvs=1530790148842~680

I see. A free membership may be required for the link I provided.

Strangely, the link you posted doesn't work for me!

Is it accessible for people from here perhaps?

 

[Wombat]

Yes.

http://digitool.library.mcgill.ca/w...s=1530790349417~418&usePid1=true&usePid2=true

249 pages. Summary?

 

[Krunok]

Is it accessible for people from here perhaps?

Yes, but only abstract.

 

[Krunok]

Yes.

http://digitool.library.mcgill.ca/w...s=1530790349417~418&usePid1=true&usePid2=true

249 pages. Summary?

This one doesn't work for me. ("Cannot process request: null")

 

[Wombat]

This one doesn't work for me. ("Cannot process request: null")

Same.

I clicked on the here link in post #3 and the page came up. Then I clicked on the small pdf icon, upper left, and the paper downloaded.

Obviously the address I posted is invalid. Try the process in the above sentence.

 

[andreasmaaan]

Yes.

http://digitool.library.mcgill.ca/w...s=1530790349417~418&usePid1=true&usePid2=true

249 pages. Summary?

None exists that I know of. The sections are well-organised and the headings give good indications as to which are most important.

 

[Krunok]

Summary?

Page 183, from "Overall, .."

 

[Blumlein 88]

Quick read of the de Santis paper. Interesting developments of new distortion metrics that track with subjective results. However, they used a Behringer HA903 headphone amp apparently for most of the testing and later found it was somewhat high in harmonic distortion and rather high in IMD. They substituted an Fostex device which was clean for some measurements, but it looks like the HA903 was used in the bulk of the listening tests. Which certainly confounds the correlation between IMD and THD metrics. I don't have a feel for what levels of distortion in a conventional sense were equivalent to the DS metric they mostly used. Some of them were hard clipping and the others may not have been subtle amounts. I wish they had fleshed out what levels of IMD and THD were involved even when not used as a useful predictive metric of expected subjective distortion.

Other than that, interesting, and don't use Behringer gear for scientific research.

 

[andreasmaaan]

Quick read of the de Santis paper. Interesting developments of new distortion metrics that track with subjective results. However, they used a Behringer HA903 headphone amp apparently for most of the testing and later found it was somewhat high in harmonic distortion and rather high in IMD. They substituted an Fostex device which was clean for some measurements, but it looks like the HA903 was used in the bulk of the listening tests. Which certainly confounds the correlation between IMD and THD metrics. I don't have a feel for what levels of distortion in a conventional sense were equivalent to the DS metric they mostly used. Some of them were hard clipping and the others may not have been subtle amounts. I wish they had fleshed out what levels of IMD and THD were involved even when not used as a useful predictive metric of expected subjective distortion.

Other than that, interesting, and don't use Behringer gear for scientific research.

Agreed, this study is not much use if the focus is thresholds of audibility. And it's also a real pity that this error with the headphone amp undermines it somewhat.

My reading of it as a whole was that the study provides good evidence that the DS and R metrics correlate well, but weak evidence that the older metrics correlate poorly.

EDIT: @Blumlein 88 , on rereading the study, I can't seem to clarify at what point the Behringer was removed. One possible interpretation is that this was done prior to any testing, although it's hard to tell. From what in the study did you draw the conclusion that the Behringer seemed to be used for most of the listening tests? This is obviously a crucial point in terms of the validity of the study

 

[Blumlein 88]

Agreed, this study is not much use if the focus is thresholds of audibility. And it's also a real pity that this error with the headphone amp undermines it somewhat.

My reading of it as a whole was that the study provides good evidence that the DS and R metrics correlate well, but weak evidence that the older metrics correlate poorly.

EDIT: @Blumlein 88 , on rereading the study, I can't seem to clarify at what point the Behringer was removed. One possible interpretation is that this was done prior to any testing, although it's hard to tell. From what in the study did you draw the conclusion that the Behringer seemed to be used for most of the listening tests? This is obviously a crucial point in terms of the validity of the study

In the latter parts it lists equipment used in listening test, after the part where they measured the Behringer. In the appendix D page 79. It lists the HA903 with no mention of the Fostex.

 

[trl]

Just did a short REW test today at a pretty loud level to my ears, especially the 5KHz sinewave was really annoying and I'm glad I the tinnitus I got after the test disappeared in less than half of hour. Given the relatively high % of THD, DAC and amplifier is not important here, but speakers are 2 x CANTON GLE 496 + 2 x SVS SB1000.

Basically, from REW I've use the built-in signal generator and the option (checkbox) to add distortion manually. I've adjusted each harmonic while the tone was playing, so I don't hear the difference between the original sound vs. the distortion added one. It's pretty similar like an A/B test, especially if using the SPACE key to check/uncheck the distortion checkbox.

50Hz - below these numbers of 2nd...5th harmonics I can't hear any sound changes

50Hz - below these numbers of 2nd-3rd 90 degrees phase harmonics I can't hear any sound changes

1000Hz - below these numbers of 2nd...5th harmonics I can't hear any sound changes

5000Hz - below 0.05% on the 2nd harmonic I can't hear any sound changes

​

Seems that my CANTON speakers are able to reproduce more accurate trebles than bass, or perhaps for my ears is easier to differentiate fundamental sine from the added distortions on the 5 KHz sinewave vs. the 50 Hz one.

What I've noticed is that changing the phase of the added distortions is making a lot of a change in the output sound. In my case, the phase of the distortions was varying from -130 to about +180 degrees, depending on the harmonic's level (2nd, 3rd a.s.o.).

Just to be on the safe side, seems that speakers distortions in my home system needs to be lower than 0.05% (-66dB) for the 2nd harmonic and down to 0.005% (-86dB) for the 5th one, so I can't hear any change in the output sound.

 

[Blumlein 88]

Just did a short REW test today at a pretty loud level to my ears, especially the 5KHz sinewave was really annoying and I'm glad I the tinnitus I got after the test disappeared in less than half of hour. Given the relatively high % of THD, DAC and amplifier is not important here, but speakers are 2 x CANTON GLE 496 + 2 x SVS SB1000.

Basically, from REW I've use the built-in signal generator and the option (checkbox) to add distortion manually. I've adjusted each harmonic while the tone was playing, so I don't hear the difference between the original sound vs. the distortion added one. It's pretty similar like an A/B test, especially if using the SPACE key to check/uncheck the distortion checkbox.

View attachment 29760
50Hz - below these numbers of 2nd...5th harmonics I can't hear any sound changes


View attachment 29759
50Hz - below these numbers of 2nd-3rd 90 degrees phase harmonics I can't hear any sound changes



View attachment 29758
1000Hz - below these numbers of 2nd...5th harmonics I can't hear any sound changes


View attachment 29757
5000Hz - below 0.05% on the 2nd harmonic I can't hear any sound changes

​

Seems that my CANTON speakers are able to reproduce more accurate trebles than bass, or perhaps for my ears is easier to differentiate fundamental sine from the added distortions on the 5 KHz sinewave vs. the 50 Hz one.

What I've noticed is that changing the phase of the added distortions is making a lot of a change in the output sound. In my case, the phase of the distortions was varying from -130 to about +180 degrees, depending on the harmonic's level (2nd, 3rd a.s.o.).

Just to be on the safe side, seems that speakers distortions in my home system needs to be lower than 0.05% (-66dB) for the 2nd harmonic and down to 0.005% (-86dB) for the 5th one, so I can't hear any change in the output sound.

Did you look at the distortion of just your speakers without adding your own distortion? Just as a for instance, if your speaker is generating .08 % distortion and your threshold is .1% then you only need add .02% for the threshold to be exceeded.

 

[RayDunzl]

Did you look at the distortion of just your speakers without adding your own distortion? Just as a for instance, if your speaker is generating .08 % distortion and your threshold is .1% then you only need add .02% for the threshold to be exceeded.

Depending on the phase of the speaker's distortion, and the phase of the added distortion, it could add or subtract.

Remember this?

Using REW, send a 450Hz tone at -20dB to the system (JBL LSR 308 speakers), take an RTA with UMIK-1.

Note high level of 3rd harmonic distortion in the speaker output.

It is clearly audible.

Using the Distortion controls, add third harmonic to the signal at -40 degree phase angle.

 

[gene_stl]

Jonathan Novick of Audio Precision on audibility of distortion

 

[pkane]

Just did a short REW test today at a pretty loud level to my ears, especially the 5KHz sinewave was really annoying and I'm glad I the tinnitus I got after the test disappeared in less than half of hour. Given the relatively high % of THD, DAC and amplifier is not important here, but speakers are 2 x CANTON GLE 496 + 2 x SVS SB1000.

Basically, from REW I've use the built-in signal generator and the option (checkbox) to add distortion manually. I've adjusted each harmonic while the tone was playing, so I don't hear the difference between the original sound vs. the distortion added one. It's pretty similar like an A/B test, especially if using the SPACE key to check/uncheck the distortion checkbox.

View attachment 29760
50Hz - below these numbers of 2nd...5th harmonics I can't hear any sound changes


View attachment 29759
50Hz - below these numbers of 2nd-3rd 90 degrees phase harmonics I can't hear any sound changes



View attachment 29758
1000Hz - below these numbers of 2nd...5th harmonics I can't hear any sound changes


View attachment 29757
5000Hz - below 0.05% on the 2nd harmonic I can't hear any sound changes

​

Seems that my CANTON speakers are able to reproduce more accurate trebles than bass, or perhaps for my ears is easier to differentiate fundamental sine from the added distortions on the 5 KHz sinewave vs. the 50 Hz one.

What I've noticed is that changing the phase of the added distortions is making a lot of a change in the output sound. In my case, the phase of the distortions was varying from -130 to about +180 degrees, depending on the harmonic's level (2nd, 3rd a.s.o.).

Just to be on the safe side, seems that speakers distortions in my home system needs to be lower than 0.05% (-66dB) for the 2nd harmonic and down to 0.005% (-86dB) for the 5th one, so I can't hear any change in the output sound.

I've been planning to write some software to allow me to apply some distortions to music, with the ability to control the amount. Mostly to hear what it sounds like, and at what point I can hear it. Much better than listening to tones, I would think

I've not yet seriously started work on this, but a prototype version of a harmonic generator is almost working. It applies a variable non-linear transform to a music file of my choice, producing various combinations of harmonics, from all even, to all odd, to some mix. The graphs show the spectrum of a 1kHz sinewave and the non-linear transfer function. Here it is with the two controls set to no distortion:

Mostly second harmonic, third missing, others at decreasing levels:

Odd harmonics:

Just moving the sliders on the right one can generate a large number of combinations of various harmonics and hear them applied in real time, as the music is playing.

To tell you the truth, I can hardly tell the difference when playing with this over headphones. Maybe I'm just deaf