New distortion measurement method for audio amplifiers

[PMA]

I have added measurements of the Sinclair-Z30 amp to the page

http://web.telecom.cz/macura/fmanalysis/freqdist.html

This is the class B, quasi-complementary amplifier with dif. input stage, VAS + CCS. This amp was my "reference" in the seventies , and it is one of those where you can hear distortion of the pure 1kHz sine (as higher harmonics from the tweeter), though THD is less than 0.5%. Please compare with results of the other two amps. In this case, there is a perfect correlation between measurement and listening tests .

 

[jan.didden]

Pavel,

Something still nags me. When I look to the results, I see a kind of "bunching" of the HD products at each harmonic order of the carrier.

As an example, the 10kHz with the 100 Hz deviation test for the Sinclair.

Suppose that I did a lot of THD tests with the following signals:
9.900Hz, 9.901 Hz, 9.902 Hz .... 10.099 Hz, 10.100 Hz.
Then I make a compound graph of all those results. I would expect to get exactly the graph you showed for the Sinclair with this test signal.

In other words, does the modulation really create extra test signals and extra DH products or do we see the results of a series of single-freq tests? And if so, what is the added value wrt single-freq HD tests? How do we know that this type of test signals really is worse for the amp than the traditional THD single freq test?

If for instance I take an IM test, the significance is that both (or more in a multi-freq IM test) signals are present at the same time. With the modulated test signal, at any instant only a single freq is present, or am I now simplyfying it too much?

Jan Didden

 

[DSP_Geek]

I have added measurements of the Sinclair-Z30 amp to the page

http://web.telecom.cz/macura/fmanalysis/freqdist.html

This is the class B, quasi-complementary amplifier with dif. input stage, VAS + CCS. This amp was my "reference" in the seventies , and it is one of those where you can hear distortion of the pure 1kHz sine (as higher harmonics from the tweeter), though THD is less than 0.5%. Please compare with results of the other two amps. In this case, there is a perfect correlation between measurement and listening tests .

Most intriguing. Have you tried applying an FM signal on top of a low frequency low duty cycle pulse train, just to see whether thermal effects could affect high frequency distortion?


Francois.

 

[anli]

Pavel,

I'm tracking this thread (and the thread concerning the error correction amplifier),
and all these days I have in my mind two questions:

1. If I understand well, the new method aim is to make a bridge between
human perception and measurements themselves. Till now I can not see
on your FFT diagrams any kind of such bridge. All diagrams seem to be a
linear superposition of ordinary harmonics measurements. Probably I have
missed something, probably these your investigations are at the beginning
only. Where must I see??

2. I have tried to understand internals of error correction without any
success. And, as I can see, I'm not alone here. Probably (between
measurements you can find a minute to publish some more words
about it. I don't mean an article as big as Hawksford's one (which is not
understandable for many DIYers too). One page may be sufficient

To conclude - please, do not treat my post like critiques. I just want to
understand. And I'm intrigued against your work!

Andrew

 

[PMA]

Andrew,

yes, it seems to be true for class A and error corection amp. But please have a more detailed look at Sinclair amp results, especially FM signal no. 4, FM signal no. 6 and 13+14kHz signal. I did not display this last one for class A and error correction amp, as they did not show anything unexpectable.

Pavel

 

[PMA]

In other words, does the modulation really create extra test signals and extra DH products or do we see the results of a series of single-freq tests? And if so, what is the added value wrt single-freq HD tests? How do we know that this type of test signals really is worse for the amp than the traditional THD single freq test?

If for instance I take an IM test, the significance is that both (or more in a multi-freq IM test) signals are present at the same time. With the modulated test signal, at any instant only a single freq is present, or am I now simplyfying it too much?

Jan Didden

Jan,

yes, I have the same doubts. But I do not think that the method is useless, especially for worse (i.e. quite usual) amps. IMHO it brings a combination of HD and IM method and one thing more - there are different amplitudes at different frequencies of test signals (esp. sig.4 and sig.6). This might show more of the non-linear behavior (related to different amplitudes) in one picture.

 

[jan.didden]

Jan,

yes, I have the same doubts. But I do not think that the method is useless, especially for worse (i.e. quite usual) amps. IMHO it brings a combination of HD and IM method and one thing more - there are different amplitudes at different frequencies of test signals (esp. sig.4 and sig.6). This might show more of the non-linear behavior (related to different amplitudes) in one picture.

Yes I think that is the case, but it is difficult to see in the curves what is what.

Since you do all this stuff in a virtual (computer) world, would it be possible to take the output spectrum and take out the spectral components of the test signal? That way we would see the IM products, and that *may* be more revealing.

This is the way Audio Precision does it with their FastTest. Since they know the test signal freq components, they remove those from the output spectrum, giving them the distortion.
Then they look at the harmonic components, and it gives them the THD.
Then they look at the "bins" with a sum- and difference value of the input components, and that gives them the IM products.
Then they look at the other (empty) bins and that gives them the noise under signal conditions.
Of course, the original spectrum gives them the freq response.
Edit: the phase response is also derived from the complex FFT of the test signal components of course.

FastTest is a very clever invention of Richard Cabot to fully characterise an amp with a single measurement.

Is that a challenge or what?


Jan Didden

 

[PMA]

Jan,

thank you for the hint, I have to think it over. And to find enough time, which is the most difficult.

Pavel

 

[jan.didden]

Jan,

thank you for the hint, I have to think it over. And to find enough time, which is the most difficult.

Pavel

If you want I can copy the relevant pages from the AP manual for you.

Jan Didden

 

[jwb]

I think it would be most interesting to test a class A amplifier with the normal kind of global feedback, taken from the output at the speaker terminals, rather then the feedback taken at the middle stage of the amp.

Also I'm curious about the ~ 50kHz artifact in the error-correcting amp. Is it some kind of damped oscillation?

 

[jan.didden]

Actually Pavel, something dawned just on me! If you generate a signal with multiple frequencies according to some specific spec, can you run that through your test system, and as output of the FFT get a list of frequencies and their levels? Then you can input that into the AP program and get your response curves and distortion graphs without the AP hardware!

Could you for instance handle a waveform as in the attached list?

Jan Didden

 

[PMA]

Also I'm curious about the ~ 50kHz artifact in the error-correcting amp. Is it some kind of damped oscillation?

I do not think so. Sometimes in this method you find some artifacts resulting from PC environment - single lines of small amplitude without any input signal etc. They move with different signals, sometimes dissapear, sometimes not. I assume this is one of them.

 

[PMA]

Jan,

I have to check it. I have read the file.

 

[PMA]

I do not know if this helps, but I have plotted 4 curves into 1 graph for FM signal no. 4. You can see envelope of the original spectrum and spectral envelopes for all of the amps under test. In this image I would like to demonstrate my usually used terms "mess" and "tail". Look what the amp with higher distortion (Sinclair) is doing with the signal. It produces wideband noise (mess) that makes the music nervous and unclear. This mess can also be seen at Sinclair 13+14 kHz IM measurement. Probably you will understand now what I mean when I am speaking about "silence under tones".

 

[jan.didden]

Pavel,

This is fascinating, insightful! Is that 'mess' from the Sinclair not due to noise? In other words, what are the relative noise levels of the amps without any signal? (I know, questions, questions, sorry...)

Jan Didden

 

[PMA]

Jan,

the output noise (i.e. without signal) is not very different for all of the amps tested. The "mess" is produced by distortion. In case you test by 1 sine signal you may not realize this fact. But multitone discovers the phenomenon.

I should add that I have shown envelopes of discrete lines (dots). I have also prepared data for comparison excel charts (output harmonic components with signal components subtracted).

 

[anli]

Pavel,

Indeed such diagram is more interesting. But I'm interested in difference
between Class A and Error Correction amps. Does it mean the last one
has much more clean spectrum (I mean higher harmonics lower faster)?
Or - what?

(as Jan have mentioned: "... questions, questions, ..."

Andrew

 

[PMA]

Pavel,

Indeed such diagram is more interesting. But I'm interested in difference
between Class A and Error Correction amps. Does it mean the last one
has much more clean spectrum

Andrew

Yes - exactly. For these 2 amplifiers tested it is true.

 

[Upupa Epops]

I mean, that very interesting should will be, if you compare this results, but by different output levels. I mean, that if is distortion small, ear or brain stop this perceive like distortion, but if this level of distortion is not constant, we perceive it like " tail " below music.

 

[anli]

Upupa Epops,

Do you mean 0.01/1W and 0.01/50W is better than 0.001/1W and 0.01/50W?
Or - are you saying about _absolute_ distortion values?