Class D high frequency problems.

[Bjorn]

You need do your own blind tests. No, it's not strange to mention the benefit of lower distorion if it's actually high. What's weird is marketing benefit of lower distortion in an area where it's already extremely low.

 

[Bjorn]

Feel free to set out to disprove if you'd like.

I have to myself. That's good enough for me.

GaN has its obvious advantages. But transparent is transparent. Whether Purifi, Hypex, or a properly implemented GaN amplifier. It just so happens one of these options costs considerably more. Maybe that's why you're pushing it?

I'm not selling GaN or have no financial interest in it either today. Quite the contrary, I'm selling amps with Hypex and Purifi.

 

[Buckeye Amps]

I have to myself. That's good enough for me.

I'm not selling GaN or have no financial interest in it either today. Quite the contrary, I'm selling amps with Hypex and Purifi.

I misspoke.

If you stand behind it so much, though, why not offer it?

 

[Sokel]

I have to myself. That's good enough for me.

I'm not selling GaN or have no financial interest in it either today. Quite the contrary, I'm selling amps with Hypex and Purifi.

If you want to see a direct apples to apples comparison for GaN against traditional MOSFETs there's a good one:

Is GaN worth it? - ICEpower

icepoweraudio.com

The benefits as it seems is efficiency and the potential of high power.Nothing else audio-wise.

 

[antcollinet]

But we don't see the same rise as we do with silicon MOSFET amps. And that's my point.

Because the limited bandwidth of the test can't see the rise. Not because the rise is not happening. Do the same test and test bandwidth as the amps you are claiming you can see the rise on and you will see it with this (fictitious) amp also. Or perhaps it is so amazing that it doesn't do that. But without a test done with a 45kHz measurement bandwidth, we'll never know.

And as I have pointed out, 3 out of 3 of the GAN amps that have been tested here all perform worse than the decent class D designs from Purify and Hypex. In fact they probebly perform worse than the good TPAxxxx chip designs (with post filter feedback) from the likes of FOSI.

Not much point having 1 out of 4 amps with (maybe but unlikely) better distortion results if it is never released to market. I mean, if it is so good why is the thing not being sold? They could clean up.

And 3/3 GAN amps showing poor performance is hardly a ringing endorsement of your theory that GAN is a superior technology.

 

[antcollinet]

The graph I showed earlier of a GaN amp (prototype that doesn't exist on the market and may perhaps never be released) was shown with 22 KHz, which is typically the accepted standard. Sure, with higher bandwidth the distortion would be higher. But we don't see the same rise as we do with silicon MOSFET amps. And that's my point.

We can look at a measurement conducted for both a Orchard Starkrimson GaN and Purifi 1ET400A by Audioexpress.

The Starkrimson GaN:
View attachment 428665

And the Purifi 1ET400A:
View attachment 428666

Source: https://audioxpress.com/article/fre...d-audio-bosc-and-purifi-audio-eigentakt-eval1

Notice that with the Purifi the distortion rises much quicker at higher frequencies. This would also be much worse for the Purifi above 2 KHz with a higher bandwidth. I don't know whether Audioexpress measured with 20 KHz or 22 KHz but it wasn't 45 KHz.

A poor implemented GaN amp obviously doesn't prove anything. There's no doubt that a well implemented GaN has advantages. Less rise in distortion at higher frequencies, close to a perfect squarewave and more efficient. What it comes down to if what's audible and that requires proper testing. People need to conduct that for themselves.

As mentioned before, I see no benefit with newer silicon MOSFET class D amps with simply lower SINAD numbers.

From the "wrap up" section of that very article you are taking those charts from (my bold):

For integrators or those who want to experiment with front-end design, the Purifi is the clear choice. For someone who just wants an amplifier to run the stereo, one of the many integrated versions of the Purifi or (if going for separates) the Orchard BOSC will do pretty much equally well — and by “well” here, I mean “superbly.” Special system requirements (e.g., common grounding between channels) or just a non-sonic esthetic choice (e.g., look, feel, reputation) might sway a prospective purchaser in one direction or another, but soundwise, either is essentially perfect.

Any measurement differences seen here fall orders of magnitude below human hearing thresholds.

So, again, you seem not to be really understanding what the measurements you keep quoting are telling you. Even when it is written in plain text in the article you are getting them from.

 

[Bjorn]

If you want to see a direct apples to apples comparison for GaN against traditional MOSFETs there's a good one:

Is GaN worth it? - ICEpower

icepoweraudio.com

The benefits as it seems is efficiency and the potential of high power.Nothing else audio-wise.

I'm well aware of that article. Did you notice that they don't look at THD+N vs frequency? And which is what I have primarily pointed out.
Interestingly enough, the higher frequencies has actually been icePowers biggest weakness for many years. For instance was one of their ASX modules very load dependent in the higher frequencies, and could almost sound like a tube amplifier with a lot of roll off.

They do however, point out the benefit of the efficiency.

when using GaN but most importantly the idle consumption is 20% lower and when this is combined with the efficiency of a switch mode power supply it will be more like 40%.

Another benefit is close to perfect squarewave response (some mild over and under shooting in the corners with the GaN amp below). I'll take that over lower SINAD when it's already super low as well.

 

[Sokel]

Did you notice that they don't look at THD+N vs frequency?

You can surely tell by the 6.67kHz measurement though.I measures straight as the 1kHz one even though I suspect 20kHz BW only as usual.
And the days of the dependency are far back,maybe 10 years ago before they made the Edge but they may still have some of these modules on line,probably for repairing reasons as drop-ins.

I'm waiting for a high-power version of the Conductor though,I'm curious about it as the ones in line now are simply unbeatable down low and the rest Edge versions are decent (100dB THD+N or so at classic 1kHz) .

 

[voodooless]

Another benefit is close to perfect squarewave response (some mild over and under shooting in the corners with the GaN amp below). I'll take that over lower SINAD when it's already super low as well.

This is again clearly bandwidth limited. I don’t see anything special here.

So you’re claiming to hear differences in distortion beyond 5 kHz between 90 and 100 SINAD with (some) music? I very much doubt that.

 

[HalSF]

My conclusion based on tests today is that neither Hypex or Purifi are transparent under all conditions. There's potential audible distortion that can lead to some harshness with the right material with high enough volume (doesn't need to be super high though)…For me this comes down to actual listening tests and not simply consider it from a theoretical standpoint.

This argument seems slippery and weak to me.

Advanced Class D amps are not “transparent under all conditions.”

Audibly not transparent? Under exactly what conditions?

The issue is “potential audible distortion” but then “actual listening tests” (no details offered about these tests) seemingly reveal actual audible distortion — “some harshness” — confidently attributed to Class D non-transparency. But again, the conditions that produce these alleged findings are withheld.

Following this is like watching 3 card monte or a shell game — a combination of fast moves and burred steps and hidden fundamentals.

There’s a current thread elsewhere devoted to a similar, mostly subjective argument that current Class D amps feature some inherent flaw that makes music unpleasantly “synthetic” and not suited to true hi-fi sound.

Why might the sound quality from Class D amplifiers sound somewhat synthetic?

I've heard some Class D amplifiers from time to time over the past several years. My latest experience was listening to a new amplifier designed around...

forums.stevehoffman.tv

I can’t tell, as a non-engineer, what this hammering at the integrity and viability of Class D is actually proving beyond offering a reflexive rear-guard action in defense of traditional amp designs, or perhaps a pitch for some finer, flawless emerging Class D. But it seems half-assed and sneaky.

 

[Bjorn]

This is again clearly bandwidth limited. I don’t see anything special here.

So you’re claiming to hear differences in distortion beyond 5 kHz between 90 and 100 SINAD with (some) music? I very much doubt that.

How do you know for sure it's between 90-100 SINAD around 5 KHz? Crest factor can still be high here, which will draw considerable power at some distance.

 

[voodooless]

How do you know for sure it's between 90-100 SINAD around 5 KHz? Crest factor can still be high here, which will draw considerable power at some distance.

Well, given that the average spl at those frequencies will be some 30 dB down from the bass notes, it will be about a factor 1000 less power at those frequencies as well. There is no considerable power at those frequencies…

 

[Sokel]

Well, given that the average spl at those frequencies will be some 30 dB down from the bass notes, it will be about a factor 1000 less power at those frequencies as well. There is no considerable power at those frequencies…

To tell the truth here 5kHz is still low to have such a roll-off with new stuff.
Works like Billie Eilish - Ocean Eyes for example have the same energy there as at the 800Hz ballpark and there's even way worst examples.
It's different after 10kHz though,the roll off there is larger,yes.

 

[Lars Risbo]

Another benefit is close to perfect squarewave response (some mild over and under shooting in the corners with the GaN amp below). I'll take that over lower SINAD when it's already super low as well.
View attachment 428701

This waveform is solely a function of the small signal transfer function of the amp and this is determined by the control loop design and has virtually nothing to do with the type of transistors used in the power stage.

Any limitation of the bandwidth gives to varying degree overshoot and ringing in the time domain for square waves.

At high signal amplitudes, there will be limitations set by the LC filter of the class amp. The control loop servos the LC filter output to be controlled. This may require more voltage on the LC input than the power supply voltage and we get clipping.

The ET amps are designed to have a 2nd order Butterworth low pass response. Butterworth gives the flattest possible amplitude response. Other choice could be made and again, this is totally independent of the transistor type.

 

[Snickers-is]

@Bjorn have you considered the difference in rail voltage between the amps you compare? It trends to have a huge effect on the driver stage as well as the dead time. As packaging is also a factor there does not seem to be a huge difference between silicone and gan on the metrics you compare of the two amps are actually comparable.

 

[Bjorn]

Well, given that the average spl at those frequencies will be some 30 dB down from the bass notes, it will be about a factor 1000 less power at those frequencies as well. There is no considerable power at those frequencies…

Based on the measurements I've seen and considering crest factor for this area as well as speaker sensitivity and distance, SINAD can be as low as 70 dB at areas between 5 KHz and 8 KHz.

 

[voodooless]

To tell the truth here 5kHz is still low to have such a roll-off with new stuff.
Works like Billie Eilish - Ocean Eyes for example have the same energy there as at the 800Hz ballpark and there's even way worst examples.

Really?

There are some blips past 8 kHz (good luck hearing 16kHz distortion residue about 100 dB lower than the loudest parts). Otherwise, there isn't very much going on past 4 kHz. Never mind the dropoff past 17 kHz, I grabbed this from the official YouTube video of the song, so bandwidth is a bit limited. Overall, though, I'm still right, it's about 30 dB down from the bass and 20 dB down from 100 Hz to 1 kHz. Even if there is a higher crest factor (let's say it's 5 dB more), it's still between 15 and 25 dB less power, so between ~30x and ~300x, give or take. That means that when your bass is pumping at 200W, the tweeter will do somewhere between 0.7 and 7W beyond 5 kHz.

And yes, there will be exceptions to this, as there always will be. That still doesn't mean you'll be able to hear the distortion.

 

[Sokel]

Really?
View attachment 429290

There are some blips past 8 kHz (good luck hearing 16kHz distortion residue about 100 dB lower than the loudest parts). Otherwise, there isn't very much going on past 4 kHz. Never mind the dropoff past 17 kHz, I grabbed this from the official YouTube video of the song, so bandwidth is a bit limited. Overall, though, I'm still right, it's about 30 dB down from the bass and 20 dB down from 100 Hz to 1 kHz. Even if there is a higher crest factor (let's say it's 5 dB more), it's still between 15 and 25 dB less power, so between ~30x and ~300x, give or take. That means that when your bass is pumping at 200W, the tweeter will do somewhere between 0.7 and 7W beyond 5 kHz.

And yes, there will be exceptions to this, as there always will be. That still doesn't mean you'll be able to hear the distortion.

That's how it looks like with peaks:

Now,I admit it's not my tool but I can retest to see.
The way it shows it is in the same 1kHz park up to 8kHz or so.

I can find a lot of examples (some oddities too).But I still think 5kHz is too low,the rapid fall starts at 10kHz usually as much as I have seen in Deltawave and REW.

 

[voodooless]

That's how it looks like with peaks:

I don't think it's that different, It seems like mostly a display thing. Showing 160 dB of dynamic range isn't exactly ideal. Here are the two roughly overlayed:

It's by no means perfect, but is gives enough of an impression.

 

[Frank Dernie]

Do you think an IMD product at 7khz 70db down could be audible?

FWIW I did a test on me to get an idea of what level of didtortion I was likely to be able to hear or at what level it seems to me it would be masked.

What I did was to listen to music at my normal level then lower the level until I could barely hear it, at which point I reasoned that if there was distortion sound at that level when I was listening at my normal level it would be drowned out or masked by the actual music I was listening to.

For me that was at 60dB down, I could still hear music, just, but am satisfied that it was quiet enough to be masked if listening at normal levels.

That was wide band, ie no separate tones and given both the lack of high frequency loudness in real music and how loud HF needs to be to be audible at all I would be astonished if anybody could hear 7kHz distortion 70dB down, I can only hear 7kHz if it is pretty loud tone never mind a low level distortion product.

My advice is to give this sort of experiment a try on yourself since your hearing will differ from mine.

Also try an app like SPLnFFT and watch the fft whilst listening to your favourite type of music to get an idea of its frequency content and form an opinion on what levels of HF are important to you. You may be surprised, I was.

My first job was in noise and vibration research so I am familiar with measuring stuff but find this sort of "hairy arsed" experiment on myself very informative.
OTOH I am just a music lover and apparently trained listeners can hear stuff I can't, particularly those who studied lossy compression (I have a friend who did) so just valid for me.