A testament to the audibility of distortion, or rather lack thereof

[AnalogSteph]

You would think that something measuring like this...

or like this 12 dB lower...

ought to sound obviously bad. Right?
Right?

Well, when I saw the measurements I nearly fell of my chair. (It was even worse initially.) I had not picked up on it during listening. To be fair, this is not the kind of device you typically use with your own playback material - it's a tuner, and this one normally feeds some Edifier MR3s in the kitchen with less than ideal acoustics.

(Never mind the 50/150 Hz, I think I haven't quite gotten the ground loopage out of my test setup or something. I do think there's some power supply issues in this tuner, the tuning voltage supply may not be as quiet as it should be. It's a silly '90s design that keeps everything running in standby, so finding some tired electrolytics is kind of expected.)

They can, incidentally, also measure more like this:

 

[boxerfan88]

Yeah, looks similar to the harmonic spray that is often seen with R2R DAC 1kHz measurement.

Unless the harmonic spray levels are too high, it's quite challenging to pick out while listening.

/

 

[pma]

As long as it is masked by music content, speaker distortion and ambient noise, it is all inaudible. I am repeating over and over that harmonic distortion effect is overemphasized.

 

[Leeken]

As long as it is masked by music content, speaker distortion and ambient noise, it is all inaudible. I am repeating over and over that harmonic distortion effect is overemphasized.

When I done one of those tests which show when can you hear distortion,I realised it was so ridiculously high that any electronics put on the market in the last 50 odd years if operating under their original specs are more than good enough for me.

 

[KSTR]

Post the distortion residual, not the magnitude spectrum.
One and the same spectrum can vary fom totally audible to inaudible, depending on what the nature of the distortion is, notably vs. level.
Think clipping vs zero-crossing misbehavior.

 

[respice finem]

One of the factors may IMHO is the background noise level.
If listening at 65 dB peak, with 30 dB background noise...

 

[AnalogSteph]

Post the distortion residual, not the magnitude spectrum.
One and the same spectrum can vary fom totally audible to inaudible, depending on what the nature of the distortion is, notably vs. level.

Any good way of measuring that? I suppose I could rig up the unused output channel as a direct loopback and turn on preemphasis in the generator, then I should theoretically get a good phase match. Then I could subtract the channels, either in software or by routing both unbalanced outputs onto a stereo TRS, making use of the fact that I have balanced inputs.

The spectra do kind of remind me of crossover distortion.

Given the way that output level varied by tenths of dBs as I was going through the filter passband with the generator at 10 kHz deviation, I bet it'll look rather adventurous.

One of the factors may IMHO is the background noise level.
If listening at 65 dB peak, with 30 dB background noise...

65 dB peak would be pretty quiet even by my standards. Also, I tend to value my peace and quiet, and ambient SPL readings can be misleading (they might show 30-40 dB and yet you may be able to hear white noise all the way down to 4 dB SPL). That being said, my kitchen isn't exactly low on reverb, which could absolutely be a factor when it comes to masking. While I did some initial listening comparisons vs. my trusty Kenwood KT-80 with headphones (the latter gives about 0.03% mono (dom H3), 0.1% L-only at 40 kHz deviation, with near ruler-flat FR, so well in the green generally speaking), I haven't done anything like that since.

Now, speaking of which, where would you guys be posting tuner measurements in the confusing vast expanse that is ASR?
https://www.audiosciencereview.com/forum/index.php?forums/audio-electronics-and-hardware.26/
? Or
https://www.audiosciencereview.com/...iers-phono-preamp-and-analog-audio-review.13/
?
This one's been bugging me for days.

 

[KSTR]

Any good way of measuring that? I suppose I could rig up the unused output channel as a direct loopback and turn on preemphasis in the generator, then I should theoretically get a good phase match. Then I could subtract the channels, either in software or by routing both unbalanced outputs onto a stereo TRS, making use of the fact that I have balanced inputs.

The simplest way would be to record the output and apply a notch filter with an audio editor. Applied twice, with time reversal of the file between passes, creates a linear phase filter which best preserves the residual when the filter isn't narrow-band enough to have negligible phase shift at the lowest harmonics, and it doubles notch depth.

Given the strong distortion direct or indirect waveform subtraction would certainly work as well.

 

[pma]

IMO the distortion residual shape in time domain may be well predicted from the frequency spectrum as well, no special need to make possibly imperfect notch procedure. Both (frequency or time domain evaluation) depends on experience of the observer.

 

[DVDdoug]

I had not picked up on it during listening.

I've NEVER heard distortion from anything that wasn't broken, defective, or over-driven. It's not something I worry about. Except maybe on a few vinyl records in the analog days where maybe it was just a bad recording/pressing or maybe my cartridge just wasn't able to track the particular record, so maybe nothing was 'broken". A lot of 45 RPM singles sounded pretty bad too.

And I have something on a CD that I think is over-compressed/limited to the point that it sounds distorted to me. And I've heard some "bad sounding" low-bitrate MP3s which is a different kind of distortion that may not show-up in regular distortion measurements. The original cell phones were pretty bad too, and CB radio or other communication radios...

 

[AnalogSteph]

The simplest way would be to record the output and apply a notch filter with an audio editor. Applied twice, with time reversal of the file between passes, creates a linear phase filter which best preserves the residual when the filter isn't narrow-band enough to have negligible phase shift at the lowest harmonics, and it doubles notch depth.

Here you go (upsampled 8x for more accurate display):

Nice and ugly, eh?

Also, here are the results for 75 kHz deviation:

That's 1.6%.

UPDATE:
I noticed that the IF seems to increasingly go haywire at higher RF input levels.

After lowering the level from my previous 70 to 50 dBµV and realigning the discriminator, distortion at 40 kHz mono looks like this now:

0.165% still isn't that much to write home about (a +20 kHz offset gives 0.091%), and stereo L-only distortion remains fairly high at 0.56% (0.38%), but it's a whole lot better than it was.
Those figures all pretty much double at 75 kHz, only mono distortion at +20 kHz doesn't change much and remains at little over 0.3%.

I am using an adjustable RF attenuator to keep RF intermod in check (the frontend isn't the last word in IM3 performance to put it mildly), so that may well have been helping things in practice.

 

[AnalogSteph]

Umm... folks? I can't hear y'all over the sound of those darn crickets...

 

[KSTR]

@AnalogSteph, thanks for showing the residuals, and nicely done with the original waveform as phase reference.
It looks like the strongest error actually is near zero-crossings but the nature of it still is quite benign, no hard edges. For isolated low frequency signals around 100Hz pure sine or so this could still be audible (with low-distortion headpones or speakers) as some sort of sizzling/buzzing character (now that you have the residual, you can try listen to it -- without make-up gain).
With "normal" music content it will be likely masked off.

As mentioned, it all depends on the behavior vs level. If the zero-crossing error basically remains the same when signal level is reduced (until the signal is small enough enough to fully live within the crossover region) distortion would go up and could reach ~10% range.

 

[Frank Dernie]

Whilst I accept that the nature of distortion may influence its audibility my tests on me, and what I could hear, a few years ago were crude but satisfied me.

I listened to some music at my normal playback level then listened again to the same music at progressively lower levels.

At -60db from my normal level I could only just hear some of the louder bits.

I reasoned that if music had been playing at the same time I would not notice that so concluded that, for me, 0.1% distortion must almost certainly be un-noticable.

I also notice that the annoying hum from my cinema screen motor becomes completely inaudible when watching a film even though it is intolerable in a quiet room and I have to cut power to it all the time it is not in use.

OTOH I could be a cloth eared music lover.

I have written many times that IME the sound quality of the many recordings I have bought often varies more than the plethora of equipment I have used for listening to them over the decades.

 

[KSTR]

I listened to some music at my normal playback level then listened again to the same music at progressively lower levels.

At -60db from my normal level I could only just hear some of the louder bits.

Listening to a simple linear 1:1 copy of the music just at lower levels like -60dB is not representative of distortion of -60dB. Distortion is basically strongly correlated noise.
For the same reason, testing with a different track at -60dB, or simply using random noise, is not adequate.

0.1% of distortion may or may not be audible, depending on the type. I had an eye-opening moment when I assessed one of the earlier/cheaper class-D chips, TDA8953 vs TDA8920, when the latter went obsolete. On paper, the 8953 had similar or even better specs in -60dB category, yet it was unusable because of a nasty low-level distortion that made decaying piano notes sound like a digeridoo....

 

[Frank Dernie]

Listening to a simple linear 1:1 copy of the music just at lower levels like -60dB is not representative of distortion of -60dB. Distortion is basically strongly correlated noise.
For the same reason, testing with a different track at -60dB, or simply using random noise, is not adequate.

0.1% of distortion may or may not be audible, depending on the type. I had an eye-opening moment when I assessed one of the earlier/cheaper class-D chips, TDA8953 vs TDA8920, when the latter went obsolete. On paper, the 8953 had similar or even better specs in -60dB category, yet it was unusable because of a nasty low-level distortion that made decaying piano notes sound like a digeridoo....

Maybe you are right, as I said in the first line of my comment, but I am personally satisfied that a sound that quiet, whatever its nature, added to the music I listen to at my normal level would be unlikely to be noticed by me.

I can't speak for anybody else, obviously, and it is just me listening to sound 60dB quieter than I normally listen and deciding I could barely hear it in a quiet (26dBA) room.

I can't imagine any sound at such a quiet level being audible over my music, and frankly don't want my musical enjoyment ruined by training my ears to hear that sort of thing

 

[AnalogSteph]

@AnalogSteph, thanks for showing the residuals, and nicely done with the original waveform as phase reference.
It looks like the strongest error actually is near zero-crossings but the nature of it still is quite benign, no hard edges. For isolated low frequency signals around 100Hz pure sine or so this could still be audible (with low-distortion headpones or speakers) as some sort of sizzling/buzzing character (now that you have the residual, you can try listen to it -- without make-up gain).
With "normal" music content it will be likely masked off.

Now everyone can listen to it. Audio file of residual attached. (original 1 kHz tone @-3.8 dBFS approx)

As mentioned, it all depends on the behavior vs level. If the zero-crossing error basically remains the same when signal level is reduced (until the signal is small enough enough to fully live within the crossover region) distortion would go up and could reach ~10% range.

It never gets that bad. Here is a little series I took at a generator output level of 80 dBµV:

Deviation - THD

80 kHz - 0.505%
(Hey, that even meets spec, but man is it ever ugly.)

40 kHz - 1.261%

20 kHz - 1.286%

10 kHz - 1.486%

5 kHz - 0.629%

2.5 kHz - 0.585%

1.25 kHz (= 2.5 kHz - 6 dB) - 0.437%

0.625 kHz (= 2.5 kHz -12 dB) - 0.263%

Things go fully monotonic eventually around the 2.5 kHz mark (24 dB down), but it sure takes a while.

 

[AnalogSteph]

By contrast, here is a similar series at a 30 dB lower (50 dBµV) RF level:

Things are always monotonic and go fully linear around 5 kHz dev. By 2.5 kHz, all harmonics have disappeared below the noise floor. Also note the general absence of +/-100 Hz sidebands. Noise isn't even any worse.

80 kHz - 0.406%

40 kHz - 0.202%

20 kHz - 0.147%

10 kHz - 0.090%

5 kHz - 0.048%