Test your ears in my new ABX test

[udok]

High Pavel,

Thanks for the THD diagrams of your amplifier!

I am pretty shure that 0.1% THD, which is shown in the diagram of your
amplifier can be detected in sine tests.

To test this i generated two intermodulation test files. The first one has two
sine frequencies at 10kHz and 12kHz.
The second one has 1% THD added.
Because of the nonlinearity a new frequency at 2kHz is created.

Even with my laptop soundcard it is easy to detect which one is distorted.

Please test it with foobar ABX. If there is interest and 5 people send me the ABX test results, i can generate files with 0.1% THD or with real music signals too.

Udo

Here the ABX result:

foo_abx 2.0.4 report
foobar2000 v1.3.16
2017-11-06 15:35:50

File A: a1.wav
SHA1: 8e69dd6c0beb8ac6694f6a9ed75c422015e99026
File B: b1.wav
SHA1: cc22f7688a6f89d3152e8f3dc4b81fa7fdd661f1

Output:
DS : Primary Sound Driver
Crossfading: NO

15:35:50 : Test started.
15:36:21 : 01/01
15:36:36 : 02/02
15:36:53 : 03/03
15:37:03 : 04/04
15:37:17 : 05/05
15:37:35 : 06/06
15:37:42 : 07/07
15:37:52 : 08/08
15:37:52 : Test finished.

----------
Total: 8/8
Probability that you were guessing: 0.4%

-- signature --
d886fa7be8e63320747b3b258e620f03d3a23dfc

 

[PMA]

To test this i generated two intermodulation test files. The first one has two
sine frequencies at 10kHz and 12kHz.
The second one has 1% THD added.
Because of the nonlinearity a new frequency at 2kHz is created.

Hi Udo,

thank you very much. First, I would like to make you sure that I understand what you are speaking about. I will try your files, but will not post 10+12kHz test, for the reason it is dangerous for hearing damage as I explained few hours ago. BTW, I can create it as well and I am doing the tests like this on a regular basis.

My suggestion is, do not guess for CCIF 10+12kHz distortion from 10kHz THD. This is at least misleading and depends on distortion generation process in the amp. I am sure that you have checked the THD vs. frequency plot with no load and 6R8 load that I posted. You have seen that it was a purely crossover distortion, so the 13+14kHz distortion, for example, is similar to 1kHz distortion and not to 10kHz distortion.

However, I will try your files and if you are interested, we may co-operate on some new test methods.

Last but not least, in the possible test, the whole frequency and amplitude distortion profile is to be added to make a comparable test, not just 1% THD only.

 

[Mooly]

I have tried Pavel's new 440 Hz files a few times on headphones and consistently get random results, a couple of 5/8's a 4/8 and a 2/8.

 

[Markw4]

Oh, I see. 16 bits? Because the files are much smaller and the audibility is exactly the same in the tests like this one. 16 bits have enough potential. Let's keep our Planet Earth less polluted!

My inclination would be to check that it works as theory predicts. Obviously, sometimes it doesn't as in the case of laptops and headphones that may be quirkily revealing of distortion as we saw here. Not that that was 16-bit vs 24-bit issue, but I would still be inclined to do some testing of the lower level distortion tests at 24-bits to see if other unexpected results occur and perhaps warrant investigation. In particular, because dither often sounds grainy to me rather than exactly like random noise, I don't know if there is any related masking or other adverse effect when listening to low level distortion. Would have to try it to see.

 

[PMA]

To test this i generated two intermodulation test files. The first one has two
sine frequencies at 10kHz and 12kHz.
The second one has 1% THD added.
Because of the nonlinearity a new frequency at 2kHz is created.

Udo,

I am posting one more comment on your files. I appreciate that you have created them, but the distortion used is completely different from the test amp here, so if you made test files, it might be interesting but there would be absolutely no common line with the test that I am doing here. I assume it would deserve a new thread for your test. Please see the spectrum, which you know well, I am sure.

BTW, are you using an Adddistortion SW by Keith Howard?

 

[PMA]

I have tried Pavel's new 440 Hz files a few times on headphones and consistently get random results, a couple of 5/8's a 4/8 and a 2/8.

Thank you for taking your time with the files, Karl. A positive result would indicate to some error or mistake.

Could you tell me if the switching between X and Y files in foobar abx is seamless, or is there an audible click? Because the fact that the waves are in phase says nothing, because some time is needed to switch between the files and the sound is definitely not continuous. Makes no difference if the tones are in phase or not. The best way would be to listen to complete files, to prevent possible differences in clicking between them.

 

[Pano]

Pavel, I have not yet tried the new 1K files or the 440Hz, but I don't think I'll hear any difference seeing how low the distortion is. What I really would like to know, is at what level do we start to hear this distortion spectrum? Is there any way to generate various levels of crossover distortion? Mooly did some of that in Spice, maybe that's a good way to test.

There is research going back about 90 years that examines harmonic masking. I have not been able to find any recent research. A topic for related thread, perhaps.

 

[PMA]

Udo,

I am posting one more comment on your files. I appreciate that you have created them, but the distortion used is completely different from the test amp here,

Udo,

the CCIF IMD 2 tone distortion, 13+14kHz, of the amplifier under test, looks like this. It is ugly, but the distortion components are deep below the fundamentals, 70dB and more. That's why I spoke about similar distortion as for 1kHz THD. 2nd harmonic component (1kHz) is quite small, about 90dB below fundamental, odd harmonics (those near 13 and 14kHz) are higher. But again, almost sure below masking curve, with music.

 

[udok]

Hi Pavel,

You are right with the warning regarding the dangerous level - i hope that 10kHz is hearable by anone.

Yes, i am referring to your plot with no load and 6R8 load. Your amp has about 0.1% THD at 10kHz.

I am pretty shure that for the IMD/CCIF test the THD value at the frequency of the strong 10kHz sine signal is the important number
(there is no strong signal at 2kHz which could generate harmonics).

This is possibly the most hardest test as the new frequency is very good separated
from the test frequencies (no MP3 kind of masking) and falls in the frequency range the ear is most sensitive.
And the new frequency is not always harmonic related to the test frequenies.

I have not investigate how exactly the THD and the CCIF/IMD values relate, but i know that they can be converted quite easily.

My "amplifier" is a very simple simulated one with only 2nd order nonlinearities. The files are generated in Matlab.

At the moment i am interested in the detectable level of distortion,
and i like your ABX test idea!

I would be very delighted if the outcome of this investigation is a zip archive with useful amplifier test files.

Udo

 

[PMA]

Udo,

have youu seen the real CCIF plot of the amplifier under test that I have shown just above your post?

Re THD and CCIF, they do not correlate directly, but depend on distortion generation mechanism. I will try to find AP paper for you that reveals something on this.

 

[udok]

Pavel,

My test file is there to show if 1% of THD could be audible. It has nothing to do with
your amp. The THD is generated by sending the two sine waves through
a nonlinearity y(x) = 1*x + 0.01*x*x

If you wan't to limit the discussion to your amp, i can create a new thread.

Udo

 

[PMA]

This is the file, I hope it is OK .....

 

[udok]

Ok, i am lagging behind your posts. I will read the paper you posted.

 

[PMA]

If you wan't to limit the discussion to your amp, i can create a new thread.

Udo

No, I did not want to limit the discussion, I just was under impression that you are trying to simulate the distortion of the amp under test with the 1% of 2nd in CCIF. This is fine. But for 1% real test I would prefer for you to open the new thread. Because I would not like to mismatch the crossover distortion of -70dB with 2nd harmonic distortion of -40dB. And I assume that such test, with 1% of 2nd added mathematically by adddistortion, was here several years ago, shall I search?

 

[udok]

Yes, please search
If no such test are here i can generate some testfiles for testing purpose.

 

[scott wurcer]

The THD is generated by sending the two sine waves through
a nonlinearity y(x) = 1*x + 0.01*x*x

Don't forget cos**2(ωt)=1/2[1+cos(2ωt)]

 

[PMA]

Back to CCIF IMD and crossover distortion, the important part of the paper is this one:

As expected from the difference in feedback, the 20
kHz THD+N is substantially larger than the other results.
The DIM30 and CCIF second order results are significantly
less sensitive than the others, approximately a factor of 30.
As seen earlier, the CCIF 2nd order test gives little indication
of this distortionmechanism because it measures even order,
not odd order nonlinearities.

It (with image plots in the paper) explains why CCIF IMD difference tone is not a good indicator of the crossover distortion and that's why I raised my eyebrows when I saw your test file, and documented it by my CCIF measurement. One cannot simply tell from THD at 13kHz to 13+14kHz IMD without knowledge of the distortion creation mechanism in the amp under test. I hope this helps, because this is a very interesting subject and I can see that you are interested in it.

 

[udok]

I have done some searchind and have found threads which deal with distortion and blind testing.
Thanks,
Udo

 

[udok]

Yes i know that IMD test is not very sensitive to 3rd order distortion.
Therefore i have used the 0.01*x*x second order term.

But the idea was not to model your amp (should be possible by including higher order
terms in the polynomial), but to do ABX testing of the audibility of distortion.

The nice idea is behind the polynomial nonlinearity is that i can use real music signals and
hear if the distorion is audible.

 

[udok]

Don't forget cos**2(ωt)=1/2[1+cos(2ωt)]

I have filtered the DC and the 2*wt term because this would introduce aliasing..