Class D high frequency problems.

[Lars Risbo]

The main difference between AB and D is that the feedback loop bandwidth in a class D amp is limited by the switching frequency whereas a typical AB amp goes to the Mhz range (small signal at least). this requires class D to use more advanced higher order control loops to get enough feedback (and you want a lot of that). This again means that the loop gain drops off faster above 20kHz than for an AB amp that often uses a rudimentary 1st order loop. Luckily our human hearing is really poor above 20kHz. The Eigentakt amps are designed so that distortion products in the audible range are vanishingly low. For example, the CCIF 19+20k test even at full power yields extraordinary low level IMD products (actually better than for the AHB2 as the reviews here show). This CCIF is a very tough high frequency test since it’s pathological worst case audio signal with a high slew rate that tests the amp and it’s feedback control.

it’s only when the analyser is set to have a measurement bandwidth exceeding 20kHz that these modern class Ds shows rising high frequency distortion when the harmonics fall above 20kHz outside the audio band, the 3rd harmonic of a 6.7k sine starts being above 20k. The feedback loop gain drops above 20kHz and we see rising distortion readings.

Again, these amps are designed to suppress noise and distortion components in the audible range.

at 5W and below (the SINAD test here) is mostly just a noise level test since a good amp has harmonics in the noise floor at this low level.

 

[JRR]

Thank you for answering Lars. I understand that the integer harmonics, where class D distortion begins to rise, fall outside the audible range. Can you explain to me why it wouldn’t be relevant to run a 32 multi tone test between 5-20khz at 20w? Isn’t there potential for spurious products caused by IMD in the audible range as they do not appear as multiples of one fundamental?

Thanks for taking the time.

 

[Old_School_Brad]

The distortion I could hear was in the bass where we are less sensitive. I know what distortion sounds like in the midrange and up. Its a much more unpleasant sound.

View attachment 428025

Alright, cool. That’s pretty much what I expected from a well-trained ear and aligns with the Axiom Audio study. Your result is better than mine, by the way, especially with this track, since I always felt it sounded somewhat distorted even when it’s not, which resulted in some wrong choices.

 

[Lars Risbo]

Thank you for answering Lars. I understand that the integer harmonics, where class D distortion begins to rise, fall outside the audible range. Can you explain to me why it wouldn’t be relevant to run a 32 multi tone test between 5-20khz at 20w? Isn’t there potential for spurious products caused by IMD in the audible range as they do not appear as multiples of one fundamental?

Thanks for taking the time.

that is certainly a relevant test since the IMD products analysed are in the audible range. It can be hard to do at 20W though since such signal has a very high crest factor (peak to average power ratio)

 

[JRR]

that is certainly a relevant test since the IMD products analysed are in the audible range. It can be hard to do at 20W though since such signal has a very high crest factor (peak to average power ratio)

Thank you for answering and for your efforts.

 

[JRR]

Pretty good result - is that with headphone, or speakers?

Revel F226BE in my listening room.

 

[sigbergaudio]

Thank you for answering Lars. I understand that the integer harmonics, where class D distortion begins to rise, fall outside the audible range. Can you explain to me why it wouldn’t be relevant to run a 32 multi tone test between 5-20khz at 20w? Isn’t there potential for spurious products caused by IMD in the audible range as they do not appear as multiples of one fundamental?

Thanks for taking the time.

Another reason it may not super relevant is that 20w into a tweeter is just not a very realistic number in a real life scenario. Even 5w is quite a bit, as mentioned by others in the thread.

Simplified example: Relatively low sensitivity tweeter rated at 90dB/1w. It will do 93dB@2W, and 96dB@4w.

Assuming typical music content, you'll see maybe 12-15dB difference between 100hz and 5khz. So when you're at 96dB at 5khz, you're somewhere around 110dB at 100hz. So we're already in the territory where slightly lower SINAD is extremely unlikely to be audible, and you're only feeding 4W into the tweeter. And this is a pretty conservative example, most tweeters are more sensitive than this.

 

[wwenze]

Fortunately, take the above screenshot for example, the HF distortion is already there at low output voltage and the % stays rather constant vs output power

Well, that one fluctuates a bit, so here's another

And another. This one does increase at above 20W. But 5W and 20W are roughly the same, for 15kHz at least.

Assuming the HD% does not change with power, then the IMD implied as caused by that HD would not change with power either.

You can test at many power levels but ultimately it's balancing cost vs benefit (time of testing vs how much more useful data it gives)

 

[Lars Risbo]

Another reason it may not super relevant is that 20w into a tweeter is just not a very realistic number in a real life scenario. Even 5w is quite a bit, as mentioned by others in the thread.

Simplified example: Relatively low sensitivity tweeter rated at 90dB/1w. It will do 93dB@2W, and 96dB@4w.

Assuming typical music content, you'll see maybe 12-15dB difference between 100hz and 5khz. So when you're at 96dB at 5khz, you're somewhere around 110dB at 100hz. So we're already in the territory where slightly lower SINAD is extremely unlikely to be audible, and you're only feeding 4W into the tweeter. And this is a pretty conservative example, most tweeters are more sensitive than this.

very short term peak power into a tweeter can be surprisingly high: the crest factor pf high-passed filtered audio tends to be quite high. Average power must of course be quite limited in order not to melt the tweeter

 

[Matias]

Fortunately, take the above screenshot for example, the HF distortion is already there at low output voltage and the % stays rather constant vs output power

View attachment 428049
Well, that one fluctuates a bit, so here's another
View attachment 428050
And another. This one does increase at above 20W. But 5W and 20W are roughly the same, for 15kHz at least.
View attachment 428051


Assuming the HD% does not change with power, then the IMD implied as caused by that HD would not change with power either.

You can test at many power levels but ultimately it's balancing cost vs benefit (time of testing vs how much more useful data it gives)

FAQ #7.

Class D FAQ

This topic is dedicated to Frequently Asked Questions about class D amplifiers. If you have been directed here the reason is that it is easier to concentrate the frequent answers here instead of repeating them in several topics. Q1: Class D power ratings cannot be true. How can these small and...

www.audiosciencereview.com

 

[JRR]

Fortunately, take the above screenshot for example, the HF distortion is already there at low output voltage and the % stays rather constant vs output power

View attachment 428049
Well, that one fluctuates a bit, so here's another
View attachment 428050
And another. This one does increase at above 20W. But 5W and 20W are roughly the same, for 15kHz at least.
View attachment 428051


Assuming the HD% does not change with power, then the IMD implied as caused by that HD would not change with power either.

You can test at many power levels but ultimately it's balancing cost vs benefit (time of testing vs how much more useful data it gives)

These are helpful actually. Thank you. Despite rising THD already appearing at 5w the worst case scenario for 32MT@5W is still only around -100db.

Is there a standard assumed when one references a 32 multi tone test? What array of frequencies is used?

 

[quattro98]

FAQ #7.

Class D FAQ

This topic is dedicated to Frequently Asked Questions about class D amplifiers. If you have been directed here the reason is that it is easier to concentrate the frequent answers here instead of repeating them in several topics. Q1: Class D power ratings cannot be true. How can these small and...

www.audiosciencereview.com

I was just about to post the FAQ. I'll link the post referenced in #7 here.

Review and Measurements of Purifi 1ET400A Amplifier

So what I'm wondering is how this will sound compared to the nc400. When is performance so good that it doesn't matter anymore?

audiosciencereview.com

 

[OldHvyMec]

Subjectively I do prefer the Topping A/B amplifiers.

When it's all said and done EVERYONE who is enjoying their system makes a subjective choice. I assume subjective means they/you listened to their choice
before they made a purchase. It's as simple as preferring Levi over Wrangler or Khakis over Denim or better yet, Redheads over Blonds, of course, it's all just a
preference. That preference and 5.00 dollars will still only get you a cup of Starbucks coffee. Just sayin'

Regards

 

[HalSF]

From the snake oil peddlers at Benchmark audio (sarcasm):

“Unlike class-D amplifiers, the AHB2 is noise free over its entire 0.1 Hz to 200 kHz frequency range. Don't settle for the limited performance of class-D amplification”

It’s funny you should say that because this sales copy from Benchmark actually does seem like an arguably and uncharacteristically snake-oily slip-up from an otherwise impeccably objective company dedicated to engineering excellence in audio playback.

I may have missed it, but I can’t find any follow-up text on Benchmark’s site where they explain how the *audible* performance of the AHB2 demonstrates its superiority over some inherent and *audible* “limited performance” of all Classic D amplification. Jokes about bats aside, a quotient of Benchmark’s hype seems pointed at its measured performance at crazy inhuman extremes like 0.1 Hz and 200 kHz.

My suspicion that Benchmark lapsed into subjective FUD-generation for a moment there may be wrong, but if so I’d like to know why.

 

[wwenze]

These are helpful actually. Thank you. Despite rising THD already appearing at 5w the worst case scenario for 32MT@5W is still only around -100db.

Is there a standard assumed when one references a 32 multi tone test? What array of frequencies is used?

There is no standard. Amirm started doing this test as a response to claims that music has many tones. So whatever he does is the standard.

 

[JRR]

There is no standard. Amirm started doing this test as a response to claims that music has many tones. So whatever he does is the standard.

Actually, my question was dumb. It’s pretty obvious which tones he used looking at the plot itself.

 

[Bjorn]

This is what we really want to see. Class D with no rising distortion in the highs with more power.

This is something that's not achievable with neither Hypex or Purifi modules. They all have raising distortion above 5 KHz area with increased output power. To avoid this with Class D it requires higher switching capabilities, i.e. what you get from GaN.

Check out Orchard Audio if you're looking for a class D amp with GaN and great measurements. This is the next step for class D, not simply newer Mosfet modules with lower SINAD.

 

[voodooless]

This is what we really want to see. Class D with no rising distortion in the highs with more power.

View attachment 428405

This is something that's not achievable with neither Hypex or Purifi modules. They all have raising distortion above 5 KHz area with increased output power. To avoid this with Class D it requires higher switching capabilities, i.e. what you get from GaN.

No, this is just the measurement bandwidth. The fact that the distortion goes down at the end is a tell tale sign.

What’s that amp anyway? Google knows nothing about it…?

 

[antcollinet]

This is what we really want to see. Class D with no rising distortion in the highs with more power.

View attachment 428405

This is something that's not achievable with neither Hypex or Purifi modules. They all have raising distortion above 5 KHz area with increased output power. To avoid this with Class D it requires higher switching capabilities, i.e. what you get from GaN.

Check out Orchard Audio if you're looking for a class D amp with GaN and great measurements. This is the next step for class D, not simply newer Mosfet modules with lower SINAD.

Purify/Hypex/Fosi etc will all measure the same with that measurement bandwidth - which looks to be about 15kHz.

Here is the Vera Audio P400 with 45kHz bandwidth measurement. (From Amir's test)

This is why it is important to have a good understanding of the measurement you are looking at before reaching conclusions.

 

[Lars Risbo]

This is something that's not achievable with neither Hypex or Purifi modules.

that is true: our modules measure much better than what the graph shows (0.002%). the measurement clearly uses a 20kHz bandwidth and yet you see distortion rising with frequency. The CCIF FFt plot would reveal even more. So this is a step forward?