In this case (NC500) it is constant.
It's easy to test this by introducing a pure delay into one leg of a differential measurement and noting if a wide-band phase response plot can be flattened. (That's what I did in the second plot of that post.)
If there were phase distortion inside the audio band, this technique would not be successful.
People tend to notice what they want to notice, and not notice what they don't want to. ie, Maty Tinman.
Dave.
It's easy to test this by introducing a pure delay into one leg of a differential measurement and noting if a wide-band phase response plot can be flattened. (That's what I did in the second plot of that post.)
If there were phase distortion inside the audio band, this technique would not be successful.
People tend to notice what they want to notice, and not notice what they don't want to. ie, Maty Tinman.
Dave.
@Davey
You might be right
I think it is because class D is pretty anti intuitive.
Class D has group delay because it uses a modulator often of higher order, which introduces latency.
I think the 10µsec in NC500OEM suggests something like 5th or 6th order or so.
This group delay will not occur in linear amps, thus the phase seems to be more linear, if the group delay in class D is not taken into account.
And like that a lot of common measuring methods and considerations has to be modified, when it comes to class D, or the results will be seriously spooky.
That understood, ICE 1200AS2 is still a very well performing amplifier, and the only one using ICE EDGE chipset.
Best regards
Soren
You might be right
I think it is because class D is pretty anti intuitive.
Class D has group delay because it uses a modulator often of higher order, which introduces latency.
I think the 10µsec in NC500OEM suggests something like 5th or 6th order or so.
This group delay will not occur in linear amps, thus the phase seems to be more linear, if the group delay in class D is not taken into account.
And like that a lot of common measuring methods and considerations has to be modified, when it comes to class D, or the results will be seriously spooky.
That understood, ICE 1200AS2 is still a very well performing amplifier, and the only one using ICE EDGE chipset.
Best regards
Soren
This sort of severe phase shift in Class-D's in red is cause by the switching frequency (600khz) output filter.
76489414-30-CE-471-E-BC96-F9-DD04-C23-DC1 — ImgBB
Too greatly reduce it by 3 times, you have to raise the switching frequency and the output filter corner from 600khz to 1.5mhz as Technics did in their unobtainable flagship Class-D SE-R1. Then you've got just a few degrees at 20khz instead of 70 degrees!!!
Cheers George
76489414-30-CE-471-E-BC96-F9-DD04-C23-DC1 — ImgBB
Too greatly reduce it by 3 times, you have to raise the switching frequency and the output filter corner from 600khz to 1.5mhz as Technics did in their unobtainable flagship Class-D SE-R1. Then you've got just a few degrees at 20khz instead of 70 degrees!!!
Cheers George
Last edited:
I will base my reply on the maths of sysnthesising a sine wave or equivalently a collection of sine waves.
The Fourier Series of a power stage typically used in class D amplifiers has the coefficient of the first term very significantly larger than the coefficients of the second, third, fourth terms until the twentieth term when the coefficients start increasing. However, the series converges, which in mathematical analysis means, it has a finite value for every value of the independent variable (time). The Fourier Series evaluates to an instantaneous voltage, which implies, the phase between the output current and output voltage is immaterial, as long as the output stage is capable of handling the current with sufficient speed. The latter should be the case, as the output stage is always powered and low pass filters block the higher terms/harmonics in the series.
The only limitation I see with class D amplification is when it amplifies low frequencies when the twentieth harmonic is within the audio range. This would suggest class D amplification presents a higher distortion at low audio frequencies.
Since, I do NOT use any class D amplifiers apart from those thrown down my throat in ready made equipment like mobile phones and television sets, I cannot be certain whether more distortion at low frequencies is an issue, especially, for audiophiles. I am assuming, a sufficiently high sampling rate (switching frequency) should minimise any issues.
The Fourier Series of a power stage typically used in class D amplifiers has the coefficient of the first term very significantly larger than the coefficients of the second, third, fourth terms until the twentieth term when the coefficients start increasing. However, the series converges, which in mathematical analysis means, it has a finite value for every value of the independent variable (time). The Fourier Series evaluates to an instantaneous voltage, which implies, the phase between the output current and output voltage is immaterial, as long as the output stage is capable of handling the current with sufficient speed. The latter should be the case, as the output stage is always powered and low pass filters block the higher terms/harmonics in the series.
The only limitation I see with class D amplification is when it amplifies low frequencies when the twentieth harmonic is within the audio range. This would suggest class D amplification presents a higher distortion at low audio frequencies.
Since, I do NOT use any class D amplifiers apart from those thrown down my throat in ready made equipment like mobile phones and television sets, I cannot be certain whether more distortion at low frequencies is an issue, especially, for audiophiles. I am assuming, a sufficiently high sampling rate (switching frequency) should minimise any issues.
I said no such thing.
George, I suggest to simulate, or construct a few LC filters on your bench and see what kind of phase distortion they exhibit in the audio band.
Dave.
And the latency you rightly point out, "could" be also be the other the reason why myself and so many say they love Class-D for bass and upper bass/ lower mids. But say there just something not right with the upper mids/highs.
To me they can either sound hard or thin even disjointed in the upper half of the range from the rich yet tight sounding bass/upperbass/lower mids, almost like there's two different amps are doing the amplifying.
Cheers George
To me they can either sound hard or thin even disjointed in the upper half of the range from the rich yet tight sounding bass/upperbass/lower mids, almost like there's two different amps are doing the amplifying.
Cheers George
First of all I will say, there is no such thing as class D sound.
You´ll find both very well sounding class D amps, and others that are less impressive, excactly as you´ll experience with any other class of amplifiers.
Class D has some pretty nice upsides though, as Class D has no cross over distortion, just as class A amps don´t, wich raises its potential to zero % distortion in theory.
Class D amplifiers uses a modulator, and the good ones a higher order modulator, which inevetably leads to group delay, and thus you see phase shift.
Compensating for the delay, will show the actual phase.
What is important is, that the delay is constant throughout the audioband.
The choice of Fswitch, is a compromise, where lower Fswitch would lead to steeper slope in the outputfilter, which we do not like, as it complicates the amplifier and filter.
Higher Fswitch will eventually run into problems with induction, and would then make it difficult to turn the switching devices on and off, especially off.
The sweet spot lies arround 550 KHz give or take.
In selfoscillating class D, the Fswitch is not fixed as it is in open loop designs, thus the noise in selfoscillators has a much lower amplitude, and is distributed over a wide frequency band.
In addition the design of especially the feedback loop(s) will have a huge impact on performance and sound of the amp.
And this is simply where the whole game is played.
Feedback technology often can be a gamechanger i.e. it can almost completely remove the need for very short deadtime, and very high fswitch.
And that might be the most antiintuitive thing about class D.
It works at very high frequencies not at all related to audio, thus it will not distort the audiosignal, and not add noise and coloration, if designed well.
It can be thought of as a noiseshaper.
Only a few companies can do his properly, the rest is noisy stuff.
You´ll find both very well sounding class D amps, and others that are less impressive, excactly as you´ll experience with any other class of amplifiers.
Class D has some pretty nice upsides though, as Class D has no cross over distortion, just as class A amps don´t, wich raises its potential to zero % distortion in theory.
Class D amplifiers uses a modulator, and the good ones a higher order modulator, which inevetably leads to group delay, and thus you see phase shift.
Compensating for the delay, will show the actual phase.
What is important is, that the delay is constant throughout the audioband.
The choice of Fswitch, is a compromise, where lower Fswitch would lead to steeper slope in the outputfilter, which we do not like, as it complicates the amplifier and filter.
Higher Fswitch will eventually run into problems with induction, and would then make it difficult to turn the switching devices on and off, especially off.
The sweet spot lies arround 550 KHz give or take.
In selfoscillating class D, the Fswitch is not fixed as it is in open loop designs, thus the noise in selfoscillators has a much lower amplitude, and is distributed over a wide frequency band.
In addition the design of especially the feedback loop(s) will have a huge impact on performance and sound of the amp.
And this is simply where the whole game is played.
Feedback technology often can be a gamechanger i.e. it can almost completely remove the need for very short deadtime, and very high fswitch.
And that might be the most antiintuitive thing about class D.
It works at very high frequencies not at all related to audio, thus it will not distort the audiosignal, and not add noise and coloration, if designed well.
It can be thought of as a noiseshaper.
Only a few companies can do his properly, the rest is noisy stuff.
Last edited:
The choice of Fswitch, is a compromise, where lower Fswitch would lead to steeper slope in the output filter, which we do not like, as it complicates the amplifier and filter.
Higher order steeper output filter, it also can't handle the power if steeper either, unless it's "huge" correct?, as Mark Levison with their Class-D monoblocks found, they it said series up 4 x lower order output filters see the chokes in the pic, to get a steeper slope and far less phase shift down into the audio band, but it was a failure, sounded crap, and didn't take off. They probably interacted and rang.
https://www.stereophile.com/images/1212levin.side.jpg
The only way I see is what Technics did with the SE-R1, and that is to up the frequency to 1.5mhz and use a higher frequency low order output filter with it, result far less phase shift into the audio band, but they say you have to use the far faster GaN transistors to do it.
Cheers George
Higher order steeper output filter, it also can't handle the power if steeper either, unless it's "huge" correct?, as Mark Levison with their Class-D monoblocks found, they it said series up 4 x lower order output filters see the chokes in the pic, to get a steeper slope and far less phase shift down into the audio band, but it was a failure, sounded crap, and didn't take off. They probably interacted and rang.
https://www.stereophile.com/images/1212levin.side.jpg
The only way I see is what Technics did with the SE-R1, and that is to up the frequency to 1.5mhz and use a higher frequency low order output filter with it, result far less phase shift into the audio band, but they say you have to use the far faster GaN transistors to do it.
Cheers George
Last edited:
Mark Levinson did, as far as I´m informed, use interleaved design, which is in fact not switching faster than most designs, but instead have two stages in parallel, switching in turn, therefor it is called interleaved output stages.
It is very difficult to make that work, and it is highly impractical.
I don´t think it will be difficult to design a outputfilter capable of the power even from high Fswitch.
The technics approach is not a good solution, because there actually is no need for higher Fswitch. The only reason they do so, is because they don´t have the technology to do something more sensible.
1,5MHz Fswitch will prevent distribution because of EMC trouble.
High frequencies and high power calls for problems, and evenmore there is no real intrinsic problem with phase in class D amplifiers when well done.
It all points towards the group delay, which is both harmless and very misunderstood.
Feedback in class D amplifiers cannot be compared to feedback in linear amplifiers. It serves different purposes.
In a modern selfoscillating class D amp, the feedback loop is causing the oscillation needed for it to function at all.
The known feedback technologies until recently was 4 types:
Pre filter feedback (New Class D by end user choise)
Post filter feedback (Hypex, Purifi)
Pre - and post filter feedback (ICE Power) Twin Loop
Pre - and post filter feedback combined with a tap in the middle of the filter (TC Electronics) Tripple Loop
There might be even more vendors in some of the groups.
Open loop class D is very rare, as they are clocked at a fixed frequency, resulting in high amplitude noise, in addition the PSRR is 0 dB, and this we do not like.
which is in fact not switching faster than most designs
Didn't say they switched faster they used the same 600khz switching frequency, I was told it was a steeper slope (higher order) series up output filters, hence the 4 x red chokes they used. But it sounded crap anyway.
And as far as feedback goes, each to his own on that one, a little local to me is preferred.
Puting the output filter into it, may as well put the speaker wire and speaker into the feedback loop as well, like Kenwood tried many moons ago with their monoblocks, and basically turned them into an oscillator just about all blew up.
Cheers George
Didn't say they switched faster they used the same 600khz switching frequency, I was told it was a steeper slope (higher order) series up output filters, hence the 4 x red chokes they used. But it sounded crap anyway.
And as far as feedback goes, each to his own on that one, a little local to me is preferred.
Puting the output filter into it, may as well put the speaker wire and speaker into the feedback loop as well, like Kenwood tried many moons ago with their monoblocks, and basically turned them into an oscillator just about all blew up.
Cheers George
Last edited:
One can have a lot of loop gain with multi-order feedback loops, in a Class D amplifier & have a pretty good sounding amplifier, like Purifi.
The F-word or, why there is no such thing as too much feedback, by Bruno Putzeys: https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
The F-word or, why there is no such thing as too much feedback, by Bruno Putzeys: https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
Attachments
Last edited:
You cannot use a little local feedback in class D amplifiers.
You have to understand the way a modern selfoscillating amplifier works, to suggest different feedback schemes.
There has to be a global feedback loop, to make at oscillate, or it will be very silent and very boaring in the long run
A little local feedback suits linear amplifiers, but is not at all feasible in class D amps, which are only processing the signal at frequencies way beyond the audioband.
The TI TPA3255 amplifier ICs use a driven PWM control (not self oscillating), but there is a local oscillator that drives the PWM modulator. It uses some local feedback inside the IC. This is a limitation of these designs since it hard to get very much global feed back to include the inductor. This leads to higher distortion levels, than is possible with self oscillating Class D amplifiers with a lot of loop gain and large amount of global feedback (as used on Hypex & Purifi).
This App note adds a very limited global feedback, which improves distortion, but not by much in some regions of operation....TPA324x and TPA325x Post-Filter Feedback: https://www.ti.com/lit/pdf/slaa788
The large amount of global feedback in the Hypex & Purifi designs helps reduce the magnetic hystresis distortion of the output inductor, which is hard to do with the TPA3255 amplifiers.
Blog - PURIFI
This App note adds a very limited global feedback, which improves distortion, but not by much in some regions of operation....TPA324x and TPA325x Post-Filter Feedback: https://www.ti.com/lit/pdf/slaa788
The large amount of global feedback in the Hypex & Purifi designs helps reduce the magnetic hystresis distortion of the output inductor, which is hard to do with the TPA3255 amplifiers.
Blog - PURIFI
Last edited:
You cannot use a little local feedback in class D amplifiers.
You have to understand
I didn't say use little feeback in Class-D. I was referring to putting everything but the kitchen sink under a feed back loop to fix problem, saying it not good to use it as a band-aid fix for problems that have other way of fixing them.
Cheers George
You have to understand
I didn't say use little feeback in Class-D. I was referring to putting everything but the kitchen sink under a feed back loop to fix problem, saying it not good to use it as a band-aid fix for problems that have other way of fixing them.
Cheers George