Skip to main content
Topic: Amplifier sound quality test - method and samples (Read 7808 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Amplifier sound quality test - method and samples

Reply #25
[quote name='Arnold B. Krueger' date='Mar 21 2015, 03:54' post='893071']
...
BTW  funny story - on the IM2 files I could clearly hear the 1 KHz spurious response, even with the .org file that obviously had none of any significance. The problem was with my monitoring system - it appears its the headphones.
..
/quote]

I hear the same, it is there on the original, then it is gone on the amp1 - some hf distortion added instead, and on the amp1 and C15 it is back again.
Weird, maybe the distortion gets nulled in amp1 because the transfer function is approximately a negative match.

I don't remember how this was in the loudspeakers, I will try it when I get back home later.

Amplifier sound quality test - method and samples

Reply #26
Kvalsvoll, one thing you might want to do is match the volume of the recorded files using something like the ReplayGain scanner in foobar2000. Otherwise you might hear volume differences in an ABX test.


I did not bother, because the abx tester does replay gain.
And replaygain does not work on all players, it does not work on Kodi on the big system.

Maybe I should run replaygain in foobar and then use the gain numbers to apply gain to the sample files.

Amplifier sound quality test - method and samples

Reply #27
I have uploaded additional files for the instrumentation test loop:

im2_test.flac: 19K+20K through playback and recorder, no amplifier.
im3_test.flac: 4K+10K+16K through playback and recorder, no amplifier.
im3_test_4x.flac: 4K+10K+16K through playback and recorder 4 times, no amplifier.


I am able to detect the test loop for all those samples.
To hear the distortion requires very high volume, and especially the im3_test is very loud and hard on the ears.
The 19K+20K required nearly maxing out the volume (this is on headphones), if you listen to music at this volume level this distortion is the least of your worries.
I did not notice noise in any of the test samples.
I conclude the test loop is significantly better than any of the amplifiers tested.

I have a nice spectrum plot from the test loop, showing the 19K+20k and the noise floor, unfortunately I did not upload it to the web server, and looks like you can not upload images to the forum directly.

Amplifier sound quality test - method and samples

Reply #28
I have uploaded additional files for the instrumentation test loop:

im2_test.flac: 19K+20K through playback and recorder, no amplifier.
im3_test.flac: 4K+10K+16K through playback and recorder, no amplifier.
im3_test_4x.flac: 4K+10K+16K through playback and recorder 4 times, no amplifier.


I am able to detect the test loop for all those samples.
To hear the distortion requires very high volume, and especially the im3_test is very loud and hard on the ears.
The 19K+20K required nearly maxing out the volume (this is on headphones), if you listen to music at this volume level this distortion is the least of your worries.
I did not notice noise in any of the test samples.
I conclude the test loop is significantly better than any of the amplifiers tested.

I have a nice spectrum plot from the test loop, showing the 19K+20k and the noise floor, unfortunately I did not upload it to the web server, and looks like you can not upload images to the forum directly.


Looks like your recording loop has as much as 1% nonlinear distortion.  Is it possible to run RMAA on it?

Link to Audio Rightmark tool free download

Amplifier sound quality test - method and samples

Reply #29
Looks like your recording loop has as much as 1% nonlinear distortion.  Is it possible to run RMAA on it?


That is not right, what did you use to find that?

In the spectrum plot of the 19k+20k all distortion products are below the -112dB line, distortion and noise is dominated by the 50hz and harmonics.

Note that the im2 and im3 samples are pulsed waveforms, the pulses are windowed (fade in/out), they are made like this to make it as easy as possible to hear the distortion.
If you analyse this signal to find distortion, that will not give a meaningful result.

I have measurements of all parts in the test loop, which consist of the computer player (Kodi set up properly for bit-perfect), DAC+pre (expensive and nice AV-receiver bought cheap on sale), ADC computer interface (USB audio interface, similar to any very good quality USB audio interface).
It is the noise from the avr that worried me the most, the DAC is pretty good, though not the very best.

Amplifier sound quality test - method and samples

Reply #30
Looks like your recording loop has as much as 1% nonlinear distortion.  Is it possible to run RMAA on it?


That is not right, what did you use to find that?

In the spectrum plot of the 19k+20k all distortion products are below the -112dB line, distortion and noise is dominated by the 50hz and harmonics.




I'm looking at  im2_test.flac and I'm seeing spikes that are probably representing spurious responses due to odd order nonlinear distortion at 15090 and 16090 KHz, amplitude about -48 dB, when the two test tones at 18990 and 19990 Hz are about 10 dB down.

There's an example of something similar in B&K paper about measuring distoriton figure 16 and further described under the heading "Two-Tone Interaction Distortion " on page 10.  Sorry about the ancient reference, but its the best I could find online in short order.

Twin tones don't just produce difference tones = those are mostly due to even order (mostly second) nonlineear distoriton, There's probably a matching set of sidebands > 20 KHz that got cut off by the 44.1 KHz sampling,  which is one reason  why do most of my testing with a 96 KHz sampling rate.

Amplifier sound quality test - method and samples

Reply #31
...
I'm looking at  im2_test.flac and I'm seeing spikes that are probably representing spurious responses due to odd order nonlinear distortion at 15090 and 16090 KHz, amplitude about -48 dB, when the two test tones at 18990 and 19990 Hz are about 10 dB down.


Yes, I see this, the 15k and 16k are present on all recorded samples of the pulsed 19k+20k, but not present in the original.

I can not say for sure why these tones are there, or where they originate from.
Since the 19k+20k continuous signal is clean, it is reasonable to believe this is not caused by distortion in the loop.
I may look at this further, and will report back here if I find the cause.

Amplifier sound quality test - method and samples

Reply #32
The distortion Arnold B. Krueger found in some of the samples is caused by an error in the soundcard driver on the recording computer.

The error occurs only on 44.1K sample rate, and at high frequencies with high levels.
It is therefore not likely that this could easily be detected on sound signals, and the distortion heard on the multitone samples I assumed were from the instrumentation loop.
The distortion is audible, there is a clearly audible difference for the 19K+20K when the sample is recorded free of this error.

If I select a different driver the problem does not occur.
It does not occur for 48K or 96K sample rates, which is what I normally use.
I chose 44.1 here so that the samples end up with the same sample rate as the original music samples, which are 44.1k.

I did not see this because the spectrum analyzer I use for monitoring is set up with 96K sample rate, due to the same reasons already described by Arny in this thread.

Since I can not say for sure which samples are affected, I will re-do samples and report back here when the updated samples are uploaded.

This was a nice catch, Arnold B. Krueger, and I believe this will have a significant effect on the results based on the quick testing I did on the 19K+20K error-free sample from the test loop.

Amplifier sound quality test - method and samples

Reply #33
The distortion Arnold B. Krueger found in some of the samples is caused by an error in the soundcard driver on the recording computer.

The error occurs only on 44.1K sample rate, and at high frequencies with high levels.
It is therefore not likely that this could easily be detected on sound signals, and the distortion heard on the multitone samples I assumed were from the instrumentation loop.
The distortion is audible, there is a clearly audible difference for the 19K+20K when the sample is recorded free of this error.


Its aliasing/imaging due to moderate quality resampling from 44.1->48k in your sound card.  This is very common.  Complete rejection of spurious frequencies close to Nyquist is computationally complex, and its usually not audible. 

If I select a different driver the problem does not occur.
It does not occur for 48K or 96K sample rates, which is what I normally use.
I chose 44.1 here so that the samples end up with the same sample rate as the original music samples, which are 44.1k.


If you want to test tones close to Nyquist, you must ensure that all hardware is running at the sampling rate you are using.  Compared to most equipment, even a high quality resampler will be fairly nonlinear within its transition band.  RMAA is a great tool to test for these problems. 

Amplifier sound quality test - method and samples

Reply #34
Yeah it's probably a good idea to use ASIO or WASAPI exclusive on Windows for recording and maybe even playback. That's why I was asking for the 19+20 kHz loopback test, because it's easy to introduce distortion somewhere in the chain that could be avoided.
Even if your test files are sampled at 44.1 kHz, maybe you should resample them (for example with the foobar2000 sox resampler plugin) to a higher rate and play and record at that rate. That way the anti-aliasing/imaging filters will work at a higher frequency and not mess as much with the 20 kHz tone, for example.
"I hear it when I see it."

Amplifier sound quality test - method and samples

Reply #35
...
Its aliasing/imaging due to moderate quality resampling from 44.1->48k in your sound card.  This is very common.  Complete rejection of spurious frequencies close to Nyquist is computationally complex, and its usually not audible. 
..


Indeed, it is the resampling, and this is done differently on different driver implementations.

Recording in 48K and then resample to 44.1K solves the problem.
Using the original 48K would be better, but some abx comparators require the samples to be equal in sample length, and the music is in 44.1K.
Accuracy is more than sufficient for this purpose, provided the resampler works properly.

Amplifier sound quality test - method and samples

Reply #36
I have fixed and uploaded new versions for all xxx_test-samples for the instrumentation test loop.

The old 5x music loop is removed, instead a 4x music loop is added.

I can not hear any difference now in distortion for the 19K+20K through the loop, it is masked by noise or simply not present.

I could verify a difference on the im3, for the 4x loop test.
I had to play very loud, and because of this listened to only one or two tone bursts for each iteration in the abx test, managed 9/10 this way.

There may or may not be a difference on the im3 (1x) loop test.
I would have to turn up the volume too loud to bear, I don't want to ruin what is left of my hearing just for this test.


I recorded the amp1 again today, have not uploaded the files yet.
Some very quick listening reveals that it is very likely I will be able to verify a difference on the im3 test.
The im2 test is dominated by the noise which masks all added distortion, part from the noise it is not possible to hear any difference.

Amplifier sound quality test - method and samples

Reply #37
Cool, I'll look at the files later.
"I hear it when I see it."

Amplifier sound quality test - method and samples

Reply #38
All files are now updated.

Multitone samples im2, im3 are noticeably different from the first uploads.
But it is still possible to identify the amplifiers.

Amp1 is very difficult to hear, the noise reveals the amplifier sample before you can hear the distortion on the 19K+20K.
On the im3 the loudness of the test tones gets too loud before I can tell for certain there is any added audible distortion.

Amp2 is clearly audible, and sounds like annoying distortion added to the im3 tones.

The C15 is also clearly audible, but it is interesting to note that the character of the distortion is subjectively much more pleasant.

This is when listening on the multitone test samples.
All the tested amplifiers have so low distortion levels that I think it will be very hard to verify a difference on the music samples.


Amplifier sound quality test - method and samples

Reply #40
Updated all music sample files with gain adjusted for level matching.

This required changing some of the original samples by level reduction amplification; -0.2dB or -0.4dB.

Levels were previously matched within approximately 0.4dB, that is not necessarily sufficient.
Now they are <0.1dB.
This is measured by using the foobar replaygain scan.

Amplifier sound quality test - method and samples

Reply #41
Some more test files and results of looping them through a mid-priced pro audio interface:

Post in uploads forum with file attachments


I will have look at your tests.
Such a "mid-priced" audio interface can be a very good instrumentation tool, as the specifications can be very good.
The recorder I use here in the amplifier test is a similar type.

 
SimplePortal 1.0.0 RC1 © 2008-2019