On Class D Amplifiers Measurements

[March Audio]

Unfortunately they don´t specify small signal BW, but slew rate was an incredible 200V/µs.

So here are the extremes in advance of power BW, which, after all has increased significantly over time...from my perspective class D still has to improve by a factor of 2-3, until they can really compete with state of the art class AB amps. This incorporates the associated impacts on distortion like the mentioned TIM, but the main focus is on large signal behaviour, close to the clipping point, at full musical spectrum...as said, this is my perspective on the subject.
My old Yamaha didn´t sound bad with its 50KHz, but listening to high quality dynamic material (via low efficiency speakers) at decent loudness, the amp´s clipping point is always near...

With respect I think you are just being dazzled by numbers without a proper grasp on the subject

Please explain what you mean by "but the main focus is on large signal behaviour, close to the clipping point, at full musical spectrum.“

IIf your amp was at clipping then I suggest any issues you hear have little to do with its 50k+z bandwidth

 

[eliash]

Purifi modules show 110 mhz gain bandwidth product. They are spec'd at 100 watts 20-20,000 at a distortion level with a couple extra zeroes vs the Yamaha. What is it you think is wrong with that or some sort of shortcoming that makes them unworthy? The 60 khz is set by a filter the underlying circuit is capable of much more, but with that filter they manage to be flat to within +/- .01 db 20hz to 20khz into any load within their power capabilities. Exactly under what conditions is the Yamaha better?

Maybe you can show us which part of their performance is the sub-standard part that needs 2-3 times the improvement before you'll consider them competitive with the class A or A/B SOTA designs.

https://purifi-audio.com/eigentakt/

If your still going on about 150 khz bandwidth, then I fear you still haven't understood how these things work.

Just to put it right, Yamaha was the starting point of the discussion, status from 1982. Same power bandwidth as Purifi et al today. Semiconductor (and circuit) advances made it possible to increase the power bandwidth above 150KHz in the following decade. This was considered as an achievement to be kept for the mentioned reasons.. So, nothing wrong with 50KHz power bandwidth amps, they probably sound at least as good as the Yamaha 40y ago. From a technology perspective, class D, at its current capabilities, is simply a step back in time, concerning power bandwidth and its possible obstacles.

Anyway, for me as single person, highend class D amps are problably already history (at home, in my listening environment), since I kept that Yamaha for some 35y and needed to upgrade to something for the next 30y, done!.


Besides, a suitable GBW is just a prerequisite for the whole discussion, no game changer...and yes, I consider Purifi amps as an impressive design, enabling TODAYS technology at its best.

 

[Blumlein 88]

Just to put it right, Yamaha was the starting point of the discussion, status from 1980. Same power bandwidth as Purifi et al today. Semiconductor (and circuit) advances made it possible to increase the power bandwidth above 150KHz in the following decade. This was considered as an achievement to be kept for the mentioned reasons.. So, nothing wrong with 50KHz power bandwidth amps, they probably sound at least as good as the Yamaha 40y ago. From a technology perspective class D, at its current capabilities, is simply a step back in time, concerning power bandwidth and its possible obstacles.

Anyway, for me as single person, highend class D amps are problably already history (at home, in my listening environment), since I kept that Yamaha for some 35y and needed to upgrade to something for the next 30y, done!.


Besides, a suitable GBW is just a prerequisite for the whole discussion, no game changer...and yes, I consider Purifi amps as an impressive design, enabling TODAYS technology at best.

You seem to have a hard time explaining the step back. And seem to wish avoiding straightforward answers. I think you don't have the beans to spill.

 

[March Audio]

Just to put it right, Yamaha was the starting point of the discussion, status from 1980. Same power bandwidth as Purifi et al today. Semiconductor (and circuit) advances made it possible to increase the power bandwidth above 150KHz in the following decade. This was considered as an achievement to be kept for the mentioned reasons.. So, nothing wrong with 50KHz power bandwidth amps, they probably sound at least as good as the Yamaha 40y ago. From a technology perspective class D, at its current capabilities, is simply a step back in time, concerning power bandwidth and its possible obstacles.

Anyway, for me as single person, highend class D amps are problably already history (at home, in my listening environment), since I kept that Yamaha for some 35y and needed to upgrade to something for the next 30y, done!.

Besides, a suitable GBW is just a prerequisite for the whole discussion, no game changer...and yes, I consider Purifi amps as an impressive design, enabling TODAYS technology at best.

But you havent actually talked about the reasons, you have just stated "someone said 40 years ago".

Power bandwidth does not dictate the sound quality in the way you seem to think.

 

[eliash]

You seem to have a hard time explaining the step back. And seem to wish avoiding straightforward answers. I think you don't have the beans to spill.

Yes, that´s true, I am not explaining the step backwards in terms of sound quality, I am stating a technological fact on which the audio and semiconductor industry has heavily worked upon, because they were convinced that this was for the good.

 

[boXem]

Purifi modules show 110 mhz gain bandwidth product. They are spec'd at 100 watts 20-20,000 at a distortion level with a couple extra zeroes vs the Yamaha. What is it you think is wrong with that or some sort of shortcoming that makes them unworthy? The 60 khz is set by a filter the underlying circuit is capable of much more, but with that filter they manage to be flat to within +/- .01 db 20hz to 20khz into any load within their power capabilities.

In the case of Hypex/Purifi amps, the output filter is part of the control loop. As a consequence, it is just 2 complex poles in the loop, it doesn't directly dictate the bandwidth. The bandwidth is decided by the poles and zeros placement. System order is decided by the difference between amount of poles / amount of zeros. Hypex amps have a first order behavior, Purifi ones a second order behavior, explaining the perfect flatness till 20 kHz.

 

[pedrob]

Another thing - is it OK if the power amp sends 460kHz rubbish like this into the tweeter? Are there any possible nonlinear effects in the tweeter driver? Or we just keep saying the only issue is negligible amount of added heat? Any serious explorations on this?

View attachment 39112

Did you ever get a response? Not a pretty sight. Now if the ripple showing on the 1kHz sine was added to music I'm fairly sure it would be audible.

 

[pma]

Did you ever get a response? Not a pretty sight. Now if the ripple showing on the 1kHz sine was added to music I'm fairly sure it would be audible.

I never received any valid response with a proof. This would be important, to get a proof. There may be thoughts and beliefs, but they do not count, to me.
I tried, myself, to make tests of speaker distortion when the ultrasonic tone was added to the audio tone. I tested frequencies between 30kHz and 440kHz. The biggest influence to tweeter distortion there was for the frequencies 30kHz - 40kHz. Nothing special between 100kHz - 440kHz. However, I suspect that more thorough study should be performed. Below you may see plots of tweeter distortion vs. frequency with and without the HF tone added.

HF sine turned off

HF sine 30kHz added

HF sine 40kHz added

 

[March Audio]

Another thing - is it OK if the power amp sends 460kHz rubbish like this into the tweeter? Are there any possible nonlinear effects in the tweeter driver? Or we just keep saying the only issue is negligible amount of added heat? Any serious explorations on this?

View attachment 39112

Yes it is ok because the tweeter cannot physically react to that high a frequency. Mechanically it just doesn't see it. It is both an electrical and mechanical filter.

A more appropriate experiment would be to test a class d amp and a similarly distirtion rated AB amp playing into a tweeter and measure the distortion acoustically from the tweeter. I might try and do this if I get time but I dont expect any difference.

I see potential with metal dome tweeters for issues where signals fall around their break up frequencies (maybe 25 to 35kHz) but to me that's just an argument against unnecessarily and useless high bandwidth recordings.

 

[SIY]

What’s the level of the added tone? And the effect of the added tone at the frequency you were fretting about (400khz)?

 

[pma]

Okay, so now I can bring my real contribution to class D amp measurements, as I bought, based on curiosity, the AIYIMA A07 TPA3255 amplifier. I have found other than expected basic issue, which can be generalized to all class D amplifiers with output LC filter out of the feedback loop. This filter has very limited choice of component values, if we want to keep some bandwidth and prevent response resonant peaking. This is also described in TI SLAA701A–October 2016–Revised November 2016 application note, LC filter design. And also TPA3255 datasheet and EVM. The usual LC values are 10uH + 1uF. This gives 50329 Hz filter cut-off frequency, no peaking with 4ohm resistive load and peaking with 8ohm load. There is no way to make it optimal for both resistances. Either you loose BW or get peaking. This would not be the worst - but - helas - the real world speaker has complex impedance, not the resistive only!! I measured the frequency response with 2 speakers, JBL Control1 Pro and Troels Gravesen CNO-T25, together with listening tests.

Measurement with JBL Control1 Pro

Please note the peaking between 3 and 10kHz and also fast decay above 10kHz. This is the result of capacitive impedance at high frequencies, see the impedance plot and read the phase as well

The impedance is capacitive from 3 to 10kHz and this changes the LC filter cut-off frequency.

Measurement with CNO-T25

Now there is no problem and the response is almost optimal. Why? The answer is in the impedance plot

The impedance is resistive from 4 to 8kHz and then is inductive. So there is no reduction of the filter cut-off frequency.

Listening
is in complete conformance with plots measured. The sound with CNO-T25 is fine, with JBL it is bright, harsh in upper mids and highs. This compared to well behaving class AB amplifier.

Conclusion

Testing of class D amplifiers into only resistive loads is absolutely insufficient and inadequate. At least complex dummy load is to be used as well. And it should be considered which speaker will be used with the amp and to test it with this speaker. This topology is not universal re load used. At least it is true for no PFFB designs.
It is also no problems to simulate the issues in MicroCap or LTSpice.

 

[Matias]

This topology is not universal re load used. At least it is true for no PFFB designs.

Yes, but not all class D amplifiers, as this should not happen with higher performance designs such as Hypex NCore or Purifi Eigentakt.

 

[Blumlein 88]

Yes, but not all class D amplifiers, as this should not happen with higher performance designs such as Hypex NCore or Purifi Eigentakt.

Wouldn't hurt to test those two. But some version of this result happens with all the other class D amps I am aware of.

Look at this Soundlab impedance plot.

 

[Matias]

I think it has to do with Bruno's designs taking feedback after the output filter, and therefore more imune to the load, something like that.
I agree it would be interesting to measure and check a popular module, say the NC252MP, driving a real speaker load.

As for lower priced designs, well, one gets what one pays for.

 

[DonH56]

The output impedance of virtually all amplifiers goes up with frequency as feedback (loop gain) falls. Many (most?) modern class a or AB designs have been able to forego the output filter so probably do better in that regard but I doubt it is a real-world problem for most modern class D designs.

The (or at least one of) the big breakthroughs in class D design was applying feedback after the output filter to minimize the variation across the audio band. Same thing happened with tube amplifiers when they got the output bandwidth high enough to support feedback after the output transformer.

Stereophile, Amir, and others have measured a number of class D amplifiers into emulated speaker loads. I don't recall any problems with mainstream designs of the past few years but not something I really follow.

 

[Matias]

The output impedance of virtually all amplifiers goes up with frequency as feedback (loop gain) falls.

Yes, and this is also one are where NCore is optimized.

NCore NC252MP below.

Purifi 1ET400A at first glance seems worst, but the Y axis scale shows that it is actually lower. And both lower than usual amplifiers.

 

[tmtomh]

Yes, and this is also one are where NCore is optimized.

NCore NC252MP below.

View attachment 105375

Purifi 1ET400A at first glance seems worst, but the Y axis scale shows that it is actually lower. And both lower than usual amplifiers.

View attachment 105374

So with regard to output impedance relative to frequency, Purifi varies less than Hypex and both vary less than a typical Class AB amp?

 

[Matias]

NC252MP is below 0.004 ohm, and 1ET400A, although it varies more than the NC252MP, is below 0.001 ohm, both through the audio range.

Putting them into context, for example:

Hegel H160: below 0.1 ohm.
"The output impedance was very low over most of the audioband, at 0.07 ohm, rising to 0.1 ohm at 20kHz. (Both figures include 6' of speaker cable.) As a result, the modulation of the H160's frequency response with our standard simulated loudspeaker (fig.1, gray trace) was minimal."
https://www.stereophile.com/content/hegel-music-systems-h160-integrated-amplifier-measurements

Krell KSA-50S: below 0.28 ohm.
"its output impedance was highish for a solid-state design at a calculated 0.28 ohms. This was the same at 20kHz as it was at 20Hz, however. "
https://www.stereophile.com/content/krell-ksa-50s-power-amplifier-measurements#

CH Precision M1.1: below 0.1 ohm best case.
"The output impedance with 100% feedback, including a 6'-long, spaced-pair speaker cable, was a low 0.1 ohms across the audioband. With 20% feedback, it rose to 0.24 ohm at low and middle frequencies, and 0.27 ohm at the top of the audioband."
https://www.stereophile.com/content/ch-precision-m11-power-amplifier-measurements#

McIntosh MC275: below 0.38 ohm best case.
"The McIntosh's output impedances were on the low side for a transformer-coupled tube design, at 0.33 ohm from the 4 ohm tap, 0.57 ohm from the 8 ohm tap, and 0.87 ohm from the 16 ohm tap. These figures applied at low and midrange frequencies; the impedance at 20kHz was a little higher, at 0.38 ohm (4 ohm tap), 0.7 ohm (8 ohm tap), and 1.1 ohms (16 ohm tap)."
https://www.stereophile.com/content/mcintosh-mc275-power-amplifier-measurements

 

[tmtomh]

NC252MP is below 0.004 ohm, and 1ET400A, although it varies more than the NC252MP, is below 0.001 ohm, both through the audio range.

Putting them into context, for example:

Hegel H160: below 0.1 ohm.
"The output impedance was very low over most of the audioband, at 0.07 ohm, rising to 0.1 ohm at 20kHz. (Both figures include 6' of speaker cable.) As a result, the modulation of the H160's frequency response with our standard simulated loudspeaker (fig.1, gray trace) was minimal."
https://www.stereophile.com/content/hegel-music-systems-h160-integrated-amplifier-measurements

Krell KSA-50S: below 0.28 ohm.
"its output impedance was highish for a solid-state design at a calculated 0.28 ohms. This was the same at 20kHz as it was at 20Hz, however. "
https://www.stereophile.com/content/krell-ksa-50s-power-amplifier-measurements#

CH Precision M1.1: below 0.1 ohm best case.
"The output impedance with 100% feedback, including a 6'-long, spaced-pair speaker cable, was a low 0.1 ohms across the audioband. With 20% feedback, it rose to 0.24 ohm at low and middle frequencies, and 0.27 ohm at the top of the audioband."
https://www.stereophile.com/content/ch-precision-m11-power-amplifier-measurements#

McIntosh MC275: below 0.38 ohm best case.
"The McIntosh's output impedances were on the low side for a transformer-coupled tube design, at 0.33 ohm from the 4 ohm tap, 0.57 ohm from the 8 ohm tap, and 0.87 ohm from the 16 ohm tap. These figures applied at low and midrange frequencies; the impedance at 20kHz was a little higher, at 0.38 ohm (4 ohm tap), 0.7 ohm (8 ohm tap), and 1.1 ohms (16 ohm tap)."
https://www.stereophile.com/content/mcintosh-mc275-power-amplifier-measurements

So this is a non-issue then, in properly designed/filtered Class D amps, yes?

 

[RichB]

So this is a non-issue then, in properly designed/filtered Class D amps, yes?

I don't think so until amps are measured into stressful reactive loads.
I have read that some negative feedback are sensitive to load as well a frequency where they become less effective.

- Rich