On Class D Amplifiers Measurements

[LTig]

A wide bandwidth is required for maximum flatness and minimum phase shift within the audible range. A wide bandwidth gives a faster rise time.

Is a faster rise time required for accurate reproduction of audio? I thought a 20 kHz sine at full power would be the steepest signal one can think of.

To prevent TIM in the power amp a fast rise time is one possible solution, the other one is a lowpass filter at the input which has other advantages as well.

 

[March Audio]

A wide bandwidth is required for maximum flatness and minimum phase shift within the audible range. A wide bandwidth gives a faster rise time.

...and yet flat amplitude and phase.

 

[March Audio]

Is a faster rise time required for accurate reproduction of audio? I thought a 20 kHz sine at full power would be the steepest signal one can think of.

To prevent TIM in the power amp a fast rise time is one possible solution, the other one is a lowpass filter at the input which has other advantages as well.

indeed.

However this article will be interesting re TIM

https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf

Don is absolutely correct. Square waves are of interest as they can provide information about amp stability,

 

[restorer-john]

Yes. I already told you I misread your post.

Where did you say that? I respond to what is posted, not what is edited later. That's a set of moving goalposts nobody can deal with.

I only see one reference to misread in this thread and that is the one above...

In the interests of peace and goodwill to all men, I'll let you have the last word for today as we've both got work to do- go assemble some more of those amplifiers!

J

 

[tensor9]

Why doesn’t someone hook a class D amp to to typical speaker crossover and measure the output from the crossover to see if the high switching frequency is filtered before it ever makes it to the speaker?
I’m sure the behavior of the crossover could be predicted without actually doing the measurement, but it’d be a nail in the coffin at least.

 

[March Audio]

John I think everyone here acknowdges that you have a bit of an issue with class d, shown by your long term posting of that nature.

All the technical complaints you have are either simply not correct or demonstrably irrelevant.

 

[SIY]

Why doesn’t someone hook a class D amp to to typical speaker crossover and measure the output from the crossover to see if the high switching frequency is filtered before it ever makes it to the speaker?
I’m sure the behavior of the crossover could be predicted without actually doing the measurement, but it’d be a nail in the coffin at least.

I measured the current going to an unfiltered tweeter and posted the results earlier in this thread. It was something less than negligible.

Re: loads, the last Class D amp I reviewed (shipped it back last week, sadly) had the lowest source impedance through the audio band that I've ever measured, and may very well have been lower than my measurement limit.

 

[March Audio]

Why doesn’t someone hook a class D amp to to typical speaker crossover and measure the output from the crossover to see if the high switching frequency is filtered before it ever makes it to the speaker?
I’m sure the behavior of the crossover could be predicted without actually doing the measurement, but it’d be a nail in the coffin at least.

The residual switching frequency is around 350 mV in Hypex designs. That will be further reduced through the crossover but some signal will make it through to the tweeter.

The tweeters impedance at 450kHz is much higher than the in band nominal 8 ohms, might be 50 ohms or more. This means that the current flowing and power dissapated in the tweeter is truly tiny. The tweeter has no response at 450kHz so the switching signal is a total non issue.

 

[ahofer]

You propose a ridiculous and expensive challenge and then wonder why nobody is interested in it. It's like me saying I'll give you a free house if you can statistically tell the difference between a $2 RCA lead and a $2000 cable in double blind testing. But you have to fly over here at your expense and give me $1000 if you fail.

Huh. I’d be happy to do this wherever there is a competent and neutral engineer to run the test. I have no idea where pma is located. But I can only run it where there is someone to execute.

Wagers are useful because the stakes tend to cause people to react and reveal what they really think as opposed to what they simply say. Calling a simple challenge that many in audio have posed and executed “ridiculous” and objecting to a small relative payment on failure suggests to me that you don’t think he could tell the difference either (remember, I don’t know you, I can only draw some inferences from your words here).

And that’s what we want to know and is at issue in the discussion above-are these effects audible? There’s rather a dearth of blind tests with class D, so it would be good to have some.

My stated offer seems rather gentle relative to much of the dialogue on this site. I’m sorry I have offended you.

 

[SIY]

Huh. I’d be happy to do this wherever there is a competent and neutral engineer to run the test. I have no idea where pma is located. But I can only run it where there is someone to execute.

He's in the Czech Republic. An engineer may be useful in the setup, but any testing should be done by someone competent in sensory science. PMA has done a lot of excellent work preparing test files for people to play with, and the results are generally what you'd expect, so I'm honestly surprised at some of the assertions.

 

[tmtomh]

A wide bandwidth is required for maximum flatness and minimum phase shift within the audible range. A wide bandwidth gives a faster rise time.

Thank you for this additional information. I understand the importance of sufficient bandwidth for flat frequency response and minimal phase shifts in the 20Hz-20kHz range. But again, it seems to me that you were using older Class AB amps' 100kHz max bandwidth as a way to denigrate Class D amps' 40kHz (give or take - again, sorry if I'm misremembering) max bandwidth, and to cast the Class D amps' specs as less robust or even borderline cheating compared to more rigorous, standardized measurements. So with respect, is there any empirical evidence that having "only" 20kHz worth of bandwidth above the top of the audible range is insufficient to ensure flat frequency response and proper phase behavior? Do you really need 80kHz worth of extra bandwidth instead?

 

[March Audio]

I measured the current going to an unfiltered tweeter and posted the results earlier in this thread. It was something less than negligible.

Re: loads, the last Class D amp I reviewed (shipped it back last week, sadly) had the lowest source impedance through the audio band that I've ever measured, and may very well have been lower than my measurement limit.

Yes, on this point about output impedance class d is typically very low, and as a result will have minimal deviation in frequency response. From the plots above it certainly is as good or better than those class A and A/B amps.

 

[restorer-john]

and to cast the Class D amps' specs as less robust or even borderline cheating compared to more rigorous, standardized measurements.

No, not cheating per-Se, just being economical with the truth, particularly with specifications in relation to continuous power and THD in the traditional sense. More investigation on this will reveal itself I guess, as time goes on.

As for bandwidth, I'd rather be able to hit a LPF/BW limit button on my test gear, than worry about not being able to measure something, or worse, getting erroneous results because I'm seeing something I can't hear and doesn't seem to make sense. (that happens with oscilloscopes/analyzers sometimes). But I want that choice. Same with preamplification and power amplification. The same choice people seem to want with so-called HiRes recordings. But let's face it, the benefits (!) of HiRes are more likely the greater bit depth than the sampling rate and upper frequency extension.

The RIAA eq deviation on some of my preamplifiers is specified to 100KHz. I don't need that, but I like the ultra wide bandwidth designed carefully and thoughtfully. There's a lot more effort to produce a product that tests flat to 100KHz than 20KHz in the case of amplifiers and preamps. I like that aspect, knowing my gear can hit x10 times the highest musical frequency I might throw at it.

So, I'm not at war with Class D, far from it, I want to have some decent ones here to play with (and test). Maybe some March Audio gear. All I attempt to do is put things into perspective both from a historical sense and also to remind people that much of what is presented here as "game changing" and the "latest in science" may well be debunked and rapidly obsoleted down the track, when Class D takes another speed step, and then all of a sudden, 200KHz is not only achievable but eminently desirable in the future audio landscape.

Watch the sales hype kick in when that happens. Remember, many of us have seen a lot in this hobby and industry over many decades.

 

[pma]

For those who might be interested how the alias of 900kHz input frequency, described in the 1st post of the thread sounds, here it is in the zip file.

 

[pma]

As for bandwidth, I'd rather be able to hit a LPF/BW limit button on my test gear, than worry about not being able to measure something, or worse, getting erroneous results because I'm seeing something I can't hear and doesn't seem to make sense. (that happens with oscilloscopes/analyzers sometimes). But I want that choice. Same with preamplification and power amplification. The same choice people seem to want with so-called HiRes recordings. But let's face it, the benefits (!) of HiRes are more likely the greater bit depth than the sampling rate and upper frequency extension.

Exactly. We were doing A-B tests with the additional 100kHz/-3dB RC filter added in front of the power amplifier/preamplifier input and our result was that it made a difference. One of the reasons why I want wide BW and flat amplitude response and small or negligible change of phase response in the audio band, which is directly connected to flat amplitude response. Rule of 10x works here as well. >=200kHz/-3dB is the required parameter.

 

[Juhazi]

ClassD amps having poor performace in treble is a myth to me, considering audibility.

Yes, there is rising distortion and amplitude/phase variation in measurements, but we must remember where the output goes to (crossover/loudpeaker) and how it's converted to audible or even measurable acoustic energy. Same kind of "bad" measurements from a dac or preamplifier might lead to alias effects, IMD or oscillation etc. real problems in the power amplifier.

Eg. distortion of frequencies above 10kHz are not heard by humans, even if a loudspeaker can reproduce it. Dome tweeters and compression drivers have resonance peak at 25-40kHz that amplifies distortion of treble too, but is inaudible as well. Inaudibility applies to phase shift in treble too!

 

[March Audio]

Exactly. We were doing A-B tests with the additional 100kHz/-3dB RC filter added in front of the power amplifier/preamplifier input and our result was that it made a difference. One of the reasons why I want wide BW and flat amplitude response and small or negligible change of phase response in the audio band, which is directly connected to flat amplitude response. Rule of 10x works here as well. >=200kHz/-3dB is the required parameter.

Purifi manages flat fr and phase.

 

[pma]

Eg. distortion of frequencies above 10kHz are not heard by humans, even if a loudspeaker can reproduce it. Dome tweeters and compression drivers have resonance peak at 25-40kHz that amplifies distortion of treble too, but is inaudible as well. Inaudibility applies to phase shift in treble too!

As long as the worse linearity at high frequencies does not result in intermodulation in audio band. Nonlinear transfer function, in our case with frequency dependent non-linearity, applies for all kinds of distortion visualization. IMD, CCIF, DIM are just another ways how to look at the same nonlinearity.

Cabot, Hofer, Metzle: Nonlinear Audio Distortion, chapter 13.3

 

[pma]

Purifi manages flat fr and phase.

View attachment 40460

I can only see 2 plots of amplitude response (Purifi 1ET400A blue, Right red) and no phase response in this plot. Am I missing something? No phase axis, no phase plots.

 

[March Audio]

I can only see 2 plots of amplitude response (Purifi 1ET400A blue, Right red) and no phase response in this plot. Am I missing something? No phase axis, no phase plots.

No its not there but it is linear phase in audio band. If the phase was off you would also expect the FR to be off. Bruno has confirmed the phase response later in that review thread IIRC.