On Class D Amplifiers Measurements

[pma]

Instead of inventing new tests, why not using the 32 tones already used with DACs? Even if not very useful for a designer, it shows strengths and weaknesses at a glance. That's personally the only one I am looking at in reviews.

Because average power of the multitone signal is low and the number of spectral lines masks the origin of the issue. Personally, I would prefer a tuned equidistant twin tone (sweep) or a set of triple tones, as suggested by Bob Cordell, Erno Borbely and others.

 

[Matias]

It's not low. Actually it is -12dBFS RMS like real music.

32 multitone would be perfect if it was tilted from bass to treble, again replicating real music distribution in frequency.

 

[Blumlein 88]

Because average power of the multitone signal is low and the number of spectral lines masks the origin of the issue. Personally, I would prefer a tuned equidistant twin tone (sweep) or a set of triple tones, as suggested by Bob Cordell, Erno Borbely and others.

View attachment 107759

View attachment 107760

What is the benefit of the triple tones over twin tones?

I've used swept tone pairs testing DACs or preamp, but not poweramps. Usually it hasn't shown up any surprises.

 

[pma]

It's not low. Actually it is -12dBFS RMS like real music.

32 multitone would be perfect if it was tilted from bass to treble, again replicating real music distribution in frequency.

The triple tone has DR=5, as most of the contemporary popular and streamed music.

As a circuit designer, I get almost nothing from the multitone, because the issue is masked by a gross number of spectral lines. The twin tone gives me clear hint where is the issue. If it is a triple tone with one frequency distant from the closely spaced twin, then I have another useful information.

 

[boXem]

The triple tone has DR=5, as most of the contemporary popular and streamed music.

As a circuit designer, I get almost nothing from the multitone, because the issue is masked by a gross number of spectral lines. The twin tone gives me clear hint where is the issue. If it is a triple tone with one frequency distant from the closely spaced twin, then I have another useful information.

While, like you, I like knowing what is the design issue behind an anomaly, some questions need to be raised:
1. What is the percentage of circuit designers in the audience of this site?
2. Do the guy running the tests on his free time agree to run a lot more of them?
3. How do you ensure you have covered all the possible issues with you set of 3 tones tests?

Maybe the multitone could be modified to have a spectrum closer to pink noise in order to increase the average power, but for assessing the validity of a design at one glance, it's quite efficient. Low and flat = good. Other combinations = not that good.

 

[SIY]

The triple tone has DR=5, as most of the contemporary popular and streamed music.

As a circuit designer, I get almost nothing from the multitone, because the issue is masked by a gross number of spectral lines. The twin tone gives me clear hint where is the issue. If it is a triple tone with one frequency distant from the closely spaced twin, then I have another useful information.

If you have Virtins, you can subtract out the multitone lines and measure the residual.

 

[sq225917]

Stupid question, if the input amplitude was set to a lower input level would the additive Imd raise it above the input level. Wouldnt that help show the addition instead of the even peaks we see in the current 32 tone?

 

[DonH56]

Multitone testing is very common in the RF/mW world, as is the NPR (noise power ratio) test. They offer different info, and can find an anomaly now and then that simple 2/3-tone IMD testing misses, but at some point you have to draw the line on what is practical and meaningful to test.

 

[boXem]

Stupid question, if the input amplitude was set to a lower input level would the additive Imd raise it above the input level. Wouldnt that help show the addition instead of the even peaks we see in the current 32 tone?

They look even, but the scale is logarithmic.

 

[pma]

As I already wrote in another thread, the sound of AIYIMA A07 with specifically JBL Control1 Pro speaker is awful, though class AB and class A amp sound "normal" with this speaker. The speaker is nothing to call home about, I agree. However, the harsh shhh sound with AIYIMA makes me curious.

So I made a measurement (at amplifier binding posts) of THD vs. amplitude at lower power into 4ohm load and into JBL speaker. I did not expect the result I measured ....
Yes, it is repeatable.

So, any class D expert here with an explanation? Please serious answers only, simple suggestions like banana contact will be ignored.
This suggests that measurements of distortion to resistive load only are pointless.

 

[boXem]

What is the EPDR of your speaker at 1 kHz?

 

[DDF]

As a circuit designer, I get almost nothing from the multitone, because the issue is masked by a gross number of spectral lines.

You might want to try John Risch's (Peavey) old Phi tones. They're mathematically spaced to best avoid harmonics and imd components from landing on the test frequencies. Different flavours, 3 tone, 6 tone etc. Drawback is you have to live with the crest factor, and no guarantee that really high harmonics aren't obscured by the test tones. I weight them to the IEC spectral average of music, found it useful.

Lately playing with REWs multitone, it attempts similar spacing idea but usefully computes the crest factor for the user.

 

[pma]

What is the EPDR of your speaker at 1 kHz?

 

[boXem]

Well, I have my idea. Could you do an amplifier frequency response with the speaker vs with 4 Ohm resistor?
Edit: and a 1kHz fft with speaker?

 

[pma]

Well, I have my idea. Could you do an amplifier frequency response with the speaker vs with 4 Ohm resistor?
Edit: and a 1kHz fft with speaker?

Sure. Below is the FR with 4ohm (green) and JBL (brown), then THD 1kHz and CCIF 18.5+19.5 kHz. Measurements are with R-C-R 40kHz LPF, to reduce HF pollution. Much more is here:

https://www.audiosciencereview.com/...-measurements-and-review-lm4562-option.19155/

I can say I have spent a week with in-deep measurements of this toy.

The only explanation I have is the interaction of AIYIMA output + LC filter impedance with non-linear speaker current, seen as a voltage drop at amplifier's binding posts. This is still a speculation, however. And I don't like the result. We are still at very low power. The issue is visible since some 100mW. Sonically it is like class B from sixties/seventies with almost zero idle current, in other words like crossover distortion.

 

[boXem]

Sure. Below is the FR with 4ohm (green) and JBL (brown), then THD 1kHz and CCIF 18.5+19.5 kHz. Measurements are with R-C-R 40kHz LPF, to reduce HF pollution. Much more is here:

https://www.audiosciencereview.com/...-measurements-and-review-lm4562-option.19155/

I can say I have spent a week with in-deep measurements of this toy.

View attachment 107934

View attachment 107935

View attachment 107936

View attachment 107937

View attachment 107938

The only explanation I have is the interaction of AIYIMA output + LC filter impedance with non-linear speaker current, seen as a voltage drop at amplifier's binding posts. This is still a speculation, however. And I don't like the result. We are still at very low power. The issue is visible since some 100mW. Sonically it is like class B from sixties/seventies with almost zero idle current, in other words like crossover distortion.

Sorry not to have followed the whole thread, it has been a tough week.
With my words (maybe we say the same thing), the high speaker impedance cause the output LC filter to have a very high Q. This cause the boost in high frequencies causing the crappy sound, but also boost the harmonics of the 1 kHz distortion, causing the huge degradation in THD you mentioned earlier.
Class D with outside of loop output filter has been, is, and will always be sonically messed up.

 

[pma]

You might want to try John Risch's (Peavey) old Phi tones. They're mathematically spaced to best avoid harmonics and imd components from landing on the test frequencies. Different flavours, 3 tone, 6 tone etc. Drawback is you have to live with the crest factor, and no guarantee that really high harmonics aren't obscured by the test tones. I weight them to the IEC spectral average of music, found it useful.

Lately playing with REWs multitone, it attempts similar spacing idea but usefully computes the crest factor for the user.

Thank you, @DDF , to turn my attention to Jon Risch. I have googled some of his papers and also his posts on audioreview forum. He knows what he speaks about and he is one of those from whom I can still learn something. Thanks again!

 

[DonH56]

The only explanation I have is the interaction of AIYIMA output + LC filter impedance with non-linear speaker current, seen as a voltage drop at amplifier's binding posts. This is still a speculation, however. And I don't like the result. We are still at very low power. The issue is visible since some 100mW. Sonically it is like class B from sixties/seventies with almost zero idle current, in other words like crossover distortion.

That is exactly what I started to post. The peaking may be due to interaction of the amplifier's output impedance into the speaker load, which both increases THD (since 2/3 HD will get emphasized with a 1 kHz fundamental) and produce a harsher sound due to the upper midrange peaking. That can happen at all power levels. Fundamentally it seems like the amp's output impedance is too high and/or not well controlled. This was one of the biggest problems of earlier class D designs, a problem I thought had been solved with modern designs, but apparently not...

As for power into purely resistive loads being pointless, I disagree, but certainly that is only the first step and does not tell the whole story.

 

[Matias]

a problem I thought had been solved with modern designs, but apparently not...

Depends on the modern design. I doubt this would happen with a modern high end design such as NCore, Eigentakt or ICEedge.

 

[boXem]

Depends on the modern design. I doubt this would happen with a modern high end design such as NCore, Eigentakt or ICEedge.

It would not. Not question of being high end, there are now integrated chips that control the output filter. Not the case of the TPA325x.