So how much distortion is audible?

[user 37518]

I was wondering, how much distortion is considered to be audible, 0.01%? 0.1 %? 1%? 5%?
I listened to audio samples from Ethan Winer, both 5% distortion however one of them is due to clipping and the other is due to cross-over distortion, eventhough both audio samples have the same THD, the one with the cross over distortion is more evident.

Im also aware that IMD is probably more harmful than THD.

In your experience how much distortion starts to become audible?  With audio measurements its easy to loose track of how things actually sound

 

[ruffrecords]

Part 1

 

[ruffrecords]

Part 2.

Cheers

Ian

 

[emrr]

https://groupdiy.com/index.php?topic=65615.msg831345#msg831345

 

[JohnRoberts]

I was wondering, how much distortion is considered to be audible, 0.01%? 0.1 %? 1%? 5%?
I listened to audio samples from Ethan Winer, both 5% distortion however one of them is due to clipping and the other is due to cross-over distortion, eventhough both audio samples have the same THD, the one with the cross over distortion is more evident.

Im also aware that IMD is probably more harmful than THD.

In your experience how much distortion starts to become audible?  With audio measurements its easy to loose track of how things actually sound

As it appears you have already learned from Ethan Winer's examples, "it depends".  Any deviation from perfect linear reproduction is distortion, but the audibility of that distortion will vary with the nature of that nonlinearity, frequency of the input and frequency of the distortion components .  Human hearing peaks in the midrange so distortion there should be more audible, but a loud fundamental can mask audibility of nearby distortion.  A classic flaw related to interpreting THD (harmonic distortion) is that some higher harmonics can fall above human audio range so cannot be heard but are still there, while a LPF audio path can roll off the distortion too.

IMD (intermodulation distortion) can generate distortion product energy at lower frequencies so HF IMD can express as LF mud... Jangle a key ring close to mic with lots of HF boost and any extra LF grunge is IMD.

Back in the 80s when I killed way too many brain cells designing phono preamps, I realized that the falling RIAA transfer function at HF was understating the nonlinearity (distortion) in audio paths using simple THD metrics. I rolled my own two tone IMD test (using 19kHz:20kHz) that generated a very inband (1kHz) distortion product. This test was very revealing to show phono preamps that measured good for THD but weren't really that linear at HF.

There is no single answer to your question, but in general less is always better. In some cases it depends on the application. Back in the 80's I designed an audio test set that included a sine wave generator. There was a trade off between oscillator settling time and distortion (from AGC loop).  After a bunch of bench time I settled on a couple tenths % THD as a good compromise.  The distortion was low enough that you could use it to line up tape recorders that often made much higher distortion, and even use to listen for faulty drivers in loudspeakers, while settling time was fast enough to not irritate users.

On the bench I could hear the difference from even that couple tenths of a percent in close A/B testing, but without the close comparison to a cleaner reference it sounds like a clean sine wave.

JR

PS: Another conflating factor is clipping... Almost any amount of clipping on a (low-mid frequency) sine wave is pretty audible, while clipping complex waveforms can be far less audible, because it many affect short peaks. Further tubes clip dramatically differently than solid state stages, but that is a subject unto itself.

 

[user 37518]

Ian: Thanks for the article, its great, do you remember whats the name of the magazine

Doug: Do you know where to get the transcription?

JR: its funny that you mention that on some cases THD was ok but IMD was not, in my own experience, extremely good THD is rarely accompanied by high bad IMD, and viceversa.

 

[ruffrecords]

Ian: Thanks for the article, its great, do you remember what's the name of the magazine

Sorry, I don't and unfortunately it does not even have the date on the front but it is definitely vintage.

Cheers

Ian

 

[JohnRoberts]

JR: its funny that you mention that on some cases THD was ok but IMD was not, in my own experience, extremely good THD is rarely accompanied by high bad IMD, and viceversa.

I gave the example of RIAA phono preamps where the HF roll-off attenuates the HF THD components, most noticeable in top octave measurements (but RIAA starts rolling off at 2kHz). 

Measures better, doesn't mean it is better, actually bad if measurement does not reflect reality.

In fact THD and IMD are both caused by the same mechanism, nonlinear transfer function...  Back in the day I could predict SMPTE IMD from THD, and vice versa...but SMPTE IMD only used 7kHz for HF modulation. 19kHz:20kHz IMD can be predicted by 39kHz THD metric  but many audio paths have LPF rolling that off already in the octave above 20kHz, so you can't accurately measure THD at 39kHz (note THD @ 39kHz will occur at higher multiples of 39kHz). 

Of course different frequency stimulus will result in more or less distortion because of path behavior (HF is harder to reproduce cleanly).  Perhaps TMI but most (all) NF audio circuits use dominant pole compensation, so open loop gain, and therefore loop gain margin is falling with increasing frequency leading to higher distortion at higher frequency.

THD and IMD are measured differently, but pretty much measure the same thing (path linearity).  I stopped relying upon THD+N for higher end circuit design decades ago, but it can be useful for apples to apples rough benchmark comparisons if all else is equal.

Yes, high linearity paths should express low distortion for all tests, but you were asking about "audible" distortion which is generally fractions of a percent or higher (much higher in some cases).

JR

PS: Another conflating factor is the +N in THD+N. This plus noise term will double with every doubling of bandpass, so a better wideband circuit with wider HF response, can measure worse because of more N bandwidth included in the measurement.

 

[ruffrecords]

Here is a power point presentation of a more recent presentation on the subject. I have changed the file extension to .txt to upload it. You will need to change it to .ppt

Cheers

Ian

 

[emrr]

Doug: Do you know where to get the transcription?

I do not.

 

[Mbira]

I've been learning more about using clipshifters when mixing down drums and it is really shocking to me how much clipping you can do on percussive stuff and not even hear it. 

 

[JohnRoberts]

I've been learning more about using clipshifters when mixing down drums and it is really shocking to me how much clipping you can do on percussive stuff and not even hear it.

Not really if you think about it... clipping a narrow transient peak will not dramatically alter the waveform and can even add some HF edge that listeners may prefer (this ASSumes the clipping is clean and doesn't cause longer duration perturbations).  Clipping doesn't really become very audible until the LF energy in the signal envelope is involved.

JR

PS: Being a low distortion advocate, I was disappointed by listening tests where "unwashed" (inexperienced ) listeners preferred the sound of allowing power amps to clip on transients (like drum hits) vs. a clean fast limiter that prevented clipping. I suspect the extra loudness delivered by allowing the amps to clip also contributed to the listener preference for the clipped path. Louder often sounds better, even if the louder is a little dirty, as long as recovery from clipping is fast and the signal envelope is not stepped on. Of course excessive clipping sounds like crapo.

 

[Mbira]

Not really if you think about it... clipping a narrow transient peak will not dramatically alter the waveform and can even add some HF edge that listeners may prefer (this ASSumes the clipping is clean and doesn't cause longer duration perturbations).  Clipping doesn't really become very audible until the LF energy in the signal envelope is involved.

JR

PS: Being a low distortion advocate, I was disappointed by listening tests where "unwashed" (inexperienced ) listeners preferred the sound of allowing power amps to clip on transients (like drum hits) vs. a clean fast limiter that prevented clipping. I suspect the extra loudness delivered by allowing the amps to clip also contributed to the listener preference for the clipped path. Louder often sounds better, even if the louder is a little dirty, as long as recovery from clipping is fast and the signal envelope is not stepped on. Of course excessive clipping sounds like crapo.

I guess what I really mean is that I was surprised at how much louder you can (and people do) make dance music tracks by using that technique. 

 

[JohnRoberts]

I guess what I really mean is that I was surprised at how much louder you can (and people do) make dance music tracks by using that technique.

It is a modern curse (IMO) but I have a personal bias against distortion. 

The pursuit of loudness has been with us for several decades. 

JR

 

[scott2000]

I always try to match the rms levels more or less when comparing the quality of making things louder..... I read that from Bob Katz info somewhere....

It can be a very sobering experience when the loudness perception is taken out of the equation..... If you are going for transparency that is.....

But it is surprising how much you can get away with subjectively, especially in the material with less low frequency content....

 

[user 37518]

  I stopped relying upon THD+N for higher end circuit design decades ago, but it can be useful for apples to apples rough benchmark comparisons if all else is equal.

On what do you rely on? IMD?

 

[benb]

Here's an interesting discussion:

Basically through an ellaborate test of some 25 college students we were able to show that THD and IMD are meaningless measurements of distortion as far as perception is concerned. Basically one cannot say that something does or does not sound good based on these measurements. .01% can sound outrageous in some cases and 25% can be inaudible in others. The numbers are meaningless.

http://www.diyaudio.com/forums/multi-way/121253-geddes-distortion-perception.html

 

[user 37518]

Here's an interesting discussion:http://www.diyaudio.com/forums/multi-way/121253-geddes-distortion-perception.html

Thats a great find! thanks!

 

[JohnRoberts]

On what do you rely on? IMD?

I thought I already shared this, but back in the 70's I modified my Heathkit SMPTE IMD analyser (7kHz and 60Hz) to work with 19kHz and 20kHz. By the '70s SMPTE (only 7kHz) was already too easy for modern high performance op amps, but 19kHz:20Khz still weeded out a few weak sisters.  The thing I liked about my 19kHz:20kHz IMD is that the test signals are nominally in band (<20kHz) and the distortion product (at 1kHz) was in the middle of peak audibility.

A friend and fellow engineer at Peavey (with a case of golden ear syndrome) Jon Risch wrote an AES paper about an improved multi-tone IMD  http://www.geocities.ws/jonrisch/PhiSpectral1.htm  Apparently the simple two tone IMD does not reveal all flaws in path performance, so using even more tones are even better for finding nonlinearity.

JR

PS: FWIW back in the late 70's the improved crop of ICs were better than my Heathkit THD analyser. I discovered I could extend the measurement range by feeding the product output into a used (very old Singer) spectrum analyser I bought second hand. Adding a few tens of dB to the resolution range of the the heathkit, revealed its own path distortion, so I had to run signals into the heathkit at -10dB below the 0dB nominal to not see distortion added by my test equipment.  Back then test measurements were not as cheap and easy as it is now.

 

[user 37518]

I thought I already shared this, but back in the 70's I modified my Heathkit SMPTE IMD analyser (7kHz and 60Hz) to work with 19kHz and 20kHz. By the '70s SMPTE (only 7kHz) was already too easy for modern high performance op amps, but 19kHz:20Khz still weeded out a few weak sisters.  The thing I liked about my 19kHz:20kHz IMD is that the test signals are nominally in band (<20kHz) and the distortion product (at 1kHz) was in the middle of peak audibility.

A friend and fellow engineer at Peavey (with a case of golden ear syndrome) Jon Risch wrote an AES paper about an improved multi-tone IMD  http://www.geocities.ws/jonrisch/PhiSpectral1.htm  Apparently the simple two tone IMD does not reveal all flaws in path performance, so using even more tones are even better for finding nonlinearity.

JR

PS: FWIW back in the late 70's the improved crop of ICs were better than my Heathkit THD analyser. I discovered I could extend the measurement range by feeding the product output into a used (very old Singer) spectrum analyser I bought second hand. Adding a few tens of dB to the resolution range of the the heathkit, revealed its own path distortion, so I had to run signals into the heathkit at -10dB below the 0dB nominal to not see distortion added by my test equipment.  Back then test measurements were not as cheap and easy as it is now.

Thanks for the link JR, that Heathkit mod sounds great, do you still have it or did you dump it?