How much tweeter distortion is audible?

[scott wurcer]

Just to prevent more misinformation I am no expert at interpreting the "diffmaker" output, let it suffice to say what ever they did to the signal needs to be known before making a judgement. Letting the casual listener think that a simple compressive non-linearity which is common to many transducers is that audible is misinformation.

 

[mabat]

Well, I took the Keith Howard's stuff posted above and generated thee samples of music: one grabbed directly from CD, second with distortion added (2nd h.d. at -20dB and 3rd at -40dB, which gives rougly 10% THD), and the third sample without any distortion added (intentionally), just processed through the software.

I marked these samples A, B and C a posted these on local forum. Much of the folks couldn't tell any difference (at home, on their respective equipment), some prefered the distorted version, but many of them argued that the last sample ("without any distortion") sounded safely the worst of the three. How to explain this? I doubt if such a test has any validity at all... Which thw version should be compared then?

Marcel

 

[gedlee]

I did a similar test. I found that the incidence of clipping on CDs was quite disturbing. I wrote a program which could detect clipping and it would interpolate the clipped waveform and replace the missing data. It worked quite well in fact. However when I sent samples out to people they almost universally picked the clipped file as sounding better. People like distortion if its the right kind. We found that in our tests also. The only correlation of THD to perception was negative! It was weak, but if there was any trend at all it was that distortion was prefered.

 

[KSTR]

Last night I ran the two signals through "diffmaker" and the distortion looked more like harsh clipping than a smooth third order. I agree with the doctor, a simple out = in - A*in^3 where the peak distortion is 10% would not change the rms level much and I'd bet it was not very audible.

Scott, something seems to have gone wrong. I get a very neat xfer function, the typical (for electronics) expanding distortion(using LTspice's .WAV input).

And that "10% 3rd" is very well audible in that sample (lot's of IM grunge). The question is, is it really 10% THD peak distortion of a full-scale sine, or did they actually mean the 0.1 factor in an x-fer like y=x+0.1x^3?

- Klaus

 

[jeroen_d]

The 3dB RMS difference is not audible for me, I distinguished on sound quality. There is very noticable distortion added especially when a cord is played. A 3dB amplitude difference on files which are by no other means treated is a totally different sensation of audible difference.

http://web.archive.org/web/20070829215910/http://www.pcabx.com/technical/reference/trumpets1.wav
and
http://web.archive.org/web/20070829215910/http://www.pcabx.com/technical/levels/trumpets1-3dB.wav
differ 3dB according to the website.

Compare this with the piano pieces:
http://web.archive.org/web/20070829215910/http://www.pcabx.com/technical/nonlinear/piano_nlref.wav
and
http://web.archive.org/web/20070829...x.com/technical/nonlinear/piano_3rd_10pct.wav
They sound equally loud, if you want to distinguish you will have to listen to tone quality.

Ok, I didn't check if it is pure 3rd harmonic. I will ask a friend who is able to do that. The previous poster confirmed that it is indeed 3rd order, so who do we have to believe? Will have to check out myself I guess.

I am very curious about the 50% 3rd order sample that Earl could post. If he only would do that.

 

[mabat]

I mean, Dr. Geddes, how did you -in yours tests- assured that what is being compared is really only a HD, without any side effects of the digital processing alone?

I'm planning to test some more to see the statistics, but it seems really tough to make this right if it's only the harmonic distortion one is interested in (I asked for preference, actually).

Marcel

 

[Wodgy]

I could post a pure third order distortion that was almost 50% and wouldn't be audible.

This is bull. Sorry. I don't believe it, and I strongly doubt you can post such a file to back up your claim.

 

[jeroen_d]

I checked the RMS value of the 10% 3rd harmonic piano piece. Indeed it is of lower level. I again did the WinABX test, and this time I concentrated on volume, not on tone quality. Concentrating on volume it was also easy to make a distinction. Unfortunately this makes the WinABX test sample useless to make claims about audibility of 10% 3rd order harmonic, making it insignificant whether the 3rd harmonic is pure or not.

 

[DcibeL]

If you are unsure of the real difference between test files, I would suggest plugging them into some Spectrum Analysis software. There is some free Linux software called Baudline which should do the job rather well.

 

[KSTR]

Ok, I didn't check if it is pure 3rd harmonic. I will ask a friend who is able to do that. The previous poster confirmed that it is indeed 3rd order, so who do we have to believe? Will have to check out myself I guess.

I guess you mean me. I will not say that the transfer characteristic I plotted is that of a pure 3rd-order polynom x+ax^3, although it looks pretty much like that. More analysis is in order to confirm that.

However (and that's a big "however"), the order of a polynom has no correlation with what harmonics will be generated, I hope everyone is aware of that. It's just pure luck that the expansion of sin(x)^3 is (3sin(x)-sin(3x))/4, which has only the fundamental and 3rd harmonic in it. Similar with 2nd order polynoms, but above 3rd order the expansions gets more complex, e.g. sin(x)^4 gives the fundamental, 2nd and 4th harmonic.

- Klaus

 

[jeroen_d]

With the program AddDistortion v2.1 I have made some files by myself. I added 25% 3rd harmonic to the piano piece. First I processed this file to make it mono and then attenuated it 4dB to prevent clipping when adding distortion. It is nice that AddDistortion gives a warning if that happens.

I was surprised that the resulting file still sounded like music. If I added the 3rd harmonic without first attenuating the file, I got an audible result just like the sample provided by the PCABX website. Nonetheless, 25% is easily audible. 10% 3rd order, processed on an attenuated file, is not so easy to detect but it can be done.

So, I have more understanding now for the claims made by Earl, however, 50% 3rd order not being audible is a statement of which I cannot find prove, to put it mildly....

Here are the files:
http://home.kabelfoon.nl/~dezaire/piano_nlref_left.wav
http://home.kabelfoon.nl/~dezaire/piano_nlref_left distorted.wav

In the picture you see the wav before and after processing.

 

[KSTR]

If you are unsure of the real difference between test files, I would suggest plugging them into some Spectrum Analysis software.

That's not very conclusive, unless you can isolate something like a clean single note and then check how the harmonics weigh in. And even then it's hard to tell what really happens. Been there, done that. Transfer function analysis is much more suitable.

- Klaus

 

[jeroen_d]

Here you can see the transfer function of the 25% 3rd order distortion as provided by AddDistortion v1.2. Now it is explained that, as long as the volume of the piano is low, the distortion is non-existent. Therefore 25% 3rd order harmonic distortion is meaningless if you want to make comparisons unless the full dynamic range is used.

 

[jeroen_d]

To conclude, although Earl makes bold statements, I have to agree that distortion numbers as being provided by modern drivers are an insignificant factor in achieving good audio quality. In any case, as long as the higher order harmonics are low in level which is true for most of the drivers.

 

[scott wurcer]

Scott, something seems to have gone wrong. I get a very neat xfer function, the typical (for electronics) expanding distortion(using LTspice's .WAV input).

And that "10% 3rd" is very well audible in that sample (lot's of IM grunge). The question is, is it really 10% THD peak distortion of a full-scale sine, or did they actually mean the 0.1 factor in an x-fer like y=x+0.1x^3?

- Klaus

That would explain a lot. Crossover is expansive, speakers, microphones, and phono cartridges are compressive. I return to my vinyl example. The distortion on an LP at highish level is horrible but it decreases to nil at low level, this is the opposite of poorly designed electronics. I am generalizing here there are many ways to make poor electronics.

I came to sort of the same conclusions as Dr. Geddes 20yrs. ago (without his rigor of course). How could an LP be even listenable? The THD by any normal metric was awful. It still is BTW "line contact" stylii make little difference.

 

[erjee]

Here you can see the transfer function of the 25% 3rd order distortion as provided by AddDistortion v1.2. Now it is explained that, as long as the volume of the piano is low, the distortion is non-existent. Therefore 25% 3rd order harmonic distortion is meaningless if you want to make comparisons unless the full dynamic range is used.

What kind of transfer characteristic would one expect for a real driver? I would expect a compression curve, instead of an expansion curve as you show.

 

[gedlee]

Interesting discussion, I somehow missed most of it.

The % THD from a transfer characteristic is indeed ambiguous because it depends on the level of the signal along the transfer characteristic. It also depends on the type of music with complex music being the most benign and piuano being an almost worst case because of its dynamics. So it is difficult to say that 50% is inaudble or that 10% is. What I said was that I could generate a 50% THD sample, but I would of course use benign music, not a piano.

The point that absolutely has to be stressed here is that no matter what number we throw arround its NOT 1% or 2% that is audible for low orders of nonlinearity. And lets remember that this is all playback level dependent because masking goes up with level.

Bottom line here is that it is impossible to say anything based on THD level and that is really the point. Its just not a reliable metric of perception and all the discussion here points out why that is.

As to the statement

"However (and that's a big "however"), the order of a polynom has no correlation with what harmonics will be generated"

is completely false because there is a perfect correlation between the harmonics and the polynomial, but I think what was meant is that a nth order nonlinearity does not generate only an nth order harmonic. It generates all orders from N down, by two, until zero. For example an 8th order nonlinearity will generate a harmonic at 8th, 6th, 4th, second and zero or DC. A 7th at 7th, 5th, 3rd and first, or a change in gain. It is possible to create a tranfer characteristic that will generate a pure nth order harmonic but it will be a multiorder polynomial.

"Dr. Geddes, how did you -in yours tests- assured that what is being compared is really only a HD, without any side effects of the digital processing alone? "

This is a major problem because the harmonics generated by digitally processing a file will alias if they go above the Nyquist Frequency. And since they alias durring the calculations you can't just LP filter after the calculations. So you have to make sure that the bandwidths of the signals and the sample rates are such that this doesn't happen. I got bit by this in the earliest tests, but cleaned it up in later ones. Some of the very high order nonlinearities were aliasing and causing the examples to sound worse than they really were because of aliasing.

Maybe I'll do that demo file if I get the time.

 

[gedlee]

What kind of transfer characteristic would one expect for a real driver? I would expect a compression curve, instead of an expansion curve as you show.

Real drivers have a very large third order nonlinearity which compresses the signal. There can also be some second because the magnet structure is never completely symmetric, but the presnce of higher orders is rare. A sharp VC BL(x) shape can have Higher orders, which is why I look for more Gaussian BL(x) curves than flat ones.

 

[gainphile]

If drivers distortion are indeed inaudible, can a blanket statement such as "just use any drivers" correct? Assuming the FR is good.

I am using no-name tweeters at the moment and to my surprise they did not generate complaints. Currently waiting shipment for a well Zaph-measured drivers (HiVi K1). Keen to know the changes later...

 

[pjpoes]

well the name or brand of a driver doesn't automatically equate to good or bad sound. If a driver measures well in the space/time domain for your needs, and it's dynamic limits exceed your needs, I don't see why that driver would be bad. I've always interpreted Dr. Geddes statements there to basically mean that.