Class D Amplifiers 101

[Vhond]

Also don't forget about Mueta : they also played an important role in developing better Class D hardware

 

[Kokoriantz]

I want to add on this reference thread the dead time issue in class D amplifiers.
The dead time is necessary to switch safely on transitions, both upper and lower switches go off the time the drivers delay difference and the switching transistors toggle. Nowadays the fastest can be obtained by using LMG1210 driver 3.5ns and GS61004 GaN transistor 1.5ns to have 5ns dead time.
IRF document AN1071 tutorial gives detailed explanation about it on page 9. As one can see the amplifier because of dead time is horribly nonlinear that any audiophile wouldent rime class d and audio amplifier together. I simulated the exaggerated aspect by 50ns dead time for 2Mhz switching and 1.1A ripple peak current as shown in the picture bellow. The max level the signal should end at 37v. The large flatt plateau is the time that the input signal must reach to modulate the switching pulse by 50ns.
The solutions for this problem have been applied either by increasing the gain or extending the pulse as is the case for TI TPA32XX ultra path series.
I made some measurements on simulated model for 200khz and 2Mhz With 5ns dead time and 1.1A ripple peak current. With 3.7v and 37v output I get the distortions
200Khz: 0.03% and 0.08%. 2Mhz 0.9% and 3.9% for 3.7v (in ZVS) and 37v (hard switching)
This is what one can get nowadays in open loop mode.
If audiophile version is required because class D amps have inherent low output impedance in open loop mode than lowest frequency need to be used along with 4th order linear phase output filter. If compact low cost is needed, than the 2Mhz is best suited for high speed closed loop >60db feedback. With feedback the flat plateau becomes slant but as slew rate saturation it looses high frequency signals contained.
Washboard musical instrument is the most difficult to reproduce with class D amps.
My proposition for dead time correction is the following:
The passage from ZVS to HS can be detected from the negative pulses on switched outputs on a full bridge see picture from youtube

. If one of outputs looses the negative pulse than the amp has switched from ZVS to HS and signals the triangle generator to add an offset, positive if negative side or inverse. As the offset depend only on dead time and the switching frequency than it is constant. Of course the stitching is never a perfection.

 

[boXem]

Or use a self oscillating amplifier with loads of feedback.

 

[Kokoriantz]

How self oscillating nature kills the dead time distortion?
This is the linearity of 2.1Mhz double pole feedback amp MS thesis presented in this thread

This anomaly can be measured by IM 50hz and 20khz where the ZVS/HD crossing provokes slew rate saturation, and the 20khz gets blanked.
Purifi reviewed as best on the web is a self oscillating with 75db NFB. You can listen on youtube indeed it has very detailed sound when the music is simple, but with Wagner's Valkyries, it is a horror.

 

[SIY]

Purifi reviewed as best on the web is a self oscillating with 75db NFB. You can listen on youtube indeed it has very detailed sound when the music is simple, but with Wagner's Valkyries, it is a horror.

Utter nonsense.

 

[boXem]

Loads of feedback kill the dead time distortion. Other sins too. Self oscillation is the *easy* way getting loads of feedback.
Judging an amplifier based on a YouTube video while playing ride of the walkiries on microscopic 2 ways? Interesting...

 

[Kokoriantz]

I just started learning about class D amp design attracted by GaN transistors. My only experience with such amps is few years ago I tried out the TPA3116D2 in PBTL board. I recorded the horrifying sound while I took refuge on the balcony and posted here

I modified the amp by passing the Fs from 400k to 1.2M, modified the output filter and increased the NFB by 20db by feeding the input with a current preamplifier. You can hear the sound of the sublimed TPA3116D2 now

and compare it by SET300b modified 2A3C 40W

.
This why I am convinced that along high NFB high Fs is also necessary.

 

[cjm2077]

I just started learning about class D amp design attracted by GaN transistors. My only experience with such amps is few years ago I tried out the TPA3116D2 in PBTL board. I recorded the horrifying sound while I took refuge on the balcony and posted here

I modified the amp by passing the Fs from 400k to 1.2M, modified the output filter and increased the NFB by 20db by feeding the input with a current preamplifier. You can hear the sound of the sublimed TPA3116D2 now

and compare it by SET300b modified 2A3C 40W

.
This why I am convinced that along high NFB high Fs is also necessary.

You're missing the most basic element of a class D audio amp. It doesn't switch at the desired output frequency in a sine wave shape. It switches at a much higher frequency square wave shape and that output is then low pass filtered to create a waveform. Turning the fets all the way on or all the way off , and therefore setting the output voltage to one rail or the other, keeps the resistance to a minimum and the losses at a minimum as well. Deadtime is irrelevant to the final output waveform, it is only there to protect the transistors from shoot-through.

 

[DonH56]

As said by others, dead time is not a big concern for most class D amps due to fast switching times and the output filter. Duty cycle distortion from varying dead time can increase distortion, but that's what feedback fixes quite handily. A glance at the specs and test results of any competent modern class D design shows excellent performance, usually comparable to or better than class A or AB designs.

Self-oscillating designs are preferred not to do anything about dead time but to avoid the need for a precision triangle-wave generator and simplify the design. The triangle-wave generator makes it easy to simulate and is what I used in the original post so it was easy to understand the operation, but I doubt any current amp uses that approach. My first class D amp was around 1978~1979 and used a triangle wave generator because I didn't know any better. It was for PA applications so low distortion was not a criteria. A little later I tried again and worked for quite a while to make a good oscillator with some success. But modern designs are self-oscillating, switch much faster, operate at much higher clock rates, and have much better feedback design compared to those early efforts (commercial -- mine were never commercial products).

 

[Kokoriantz]

Hello Don, I truly appreciate this thread you have started it, at on stop and someone has all he needs to know about class D.
In 1979 a colleague in R&D department built a class D amp from the Hitachi MOSFET databook for portable guitar amp, It had so screaming sound that made disgust from this class. Lately, in a nearby commercial center was exhibiting a wonderful air suspension but very dynamic speakers. I could withstand listening only 10 minutes because it was powered by a class D amp the new NAD3020.
On simulator if the amp has zero dead time it has zero distortion at full power 200khz Fs, with 5ns it is 0.08%, 10ns 0.17%. This distortion is worst than crossover in AB as it occurs twice at each zero crossing and as in class AB, the feedback can reduce it but you can convince any one that it is as good as class A.
My first intention, as now we have very fast transistors, is the open loop mode and it looks I have still lot to discover when I see the specs of GS evaluation board
https://gansystems.com/wp-content/u...fier-and-SMPS-EVB-User-Guide-Rev-20200227.pdf
It uses the GaN gs61008 with 1.5ns switching but 10ns skew driver. It can be selected open or closed loop.
The open loop distortion is only 0.008% in ZVS zone where I have 0.03% with half of the frequency and half of the dead time, this shows I have still lot to learn. BTW, the feedback makes things worst at low power.

 

[NTK]

The figures below are taken from a paper by Bruno Putzeys et al. PWM is analog, meaning that they aren't simply 1's and 0's. The 1.1's, 0.9's, 0.1's, etc. totally matter to what will become the filtered output. If you want to get by without feedback and good performance, you'll need perfect control of your supply voltage rail, regardless of the current the amplifier needs to output.

(Source: https://www.hypex.nl/img/upload/doc/an_wp/WP_All_amps_are_analogue.pdf)

 

[boXem]

Self-oscillating designs are preferred not to do anything about dead time but to avoid the need for a precision triangle-wave generator and simplify the design. The triangle-wave generator makes it easy to simulate and is what I used in the original post so it was easy to understand the operation, but I doubt any current amp uses that approach.

One can not do post filter feedback above a few dB if the amplifier is not self oscillating
The phase shift induced by the filter being close to 180 deg. And yes it simplifies the design in term of part count, not in term of engineering, The variable oscillation frequency presents interesting challenges.
I perfectly understand the use of a triangle generator for this presentation.While also easy to simulate, a basic self oscillating amplifier is also more complicated to understand in the context of a class D 101. A lot of actual design still use triangle generators BTW.

 

[Kokoriantz]

You're missing the most basic element of a class D audio amp. It doesn't switch at the desired output frequency in a sine wave shape. It switches at a much higher frequency square wave shape and that output is then low pass filtered to create a waveform. Turning the fets all the way on or all the way off , and therefore setting the output voltage to one rail or the other, keeps the resistance to a minimum and the losses at a minimum as well. Deadtime is irrelevant to the final output waveform, it is only there to protect the transistors from shoot-through.

Thank you for posting a user's introduction for class D . If you read the IRF tutorial you can see there is much more to talk about. The output inductance has a ripple current of triangular shape with peak current alternating positive and negative at 50% DT. When both switches go off for dead time the induction continues its current through the diodes creating an opposite voltage at the output to the precedent state. In other words the inductance completes the dead time switching to next state beforehand, every thing goes well. The problem occurs when the output provides the load more current in either direction than the ripple, here what happens.

In this case the output current in the inductor remains always positive, the upper switch is pouring positive current in the first stage, next both switches go off, the inductor pulls its current from the lower diode provoking negative voltage. Next the lower switch is on, the output already negative remains negative but it will not be able to reverse the polarity of the current. Now when both switches are off as the current didn't go negative, no positive voltage can be created and the inductor will continue by the lower diode creating negative voltage. By this the positive part of the duty cycle gets truncated by the dead time pulling the output less positively. The inverse occurs for the negative currents bellow the ripple current.

 

[cjm2077]

Thank you for posting a user's introduction for class D . If you read the IRF tutorial you can see there is much more to talk about. The output inductance has a ripple current of triangular shape with peak current alternating positive and negative at 50% DT. When both switches go off for dead time the induction continues its current through the diodes creating an opposite voltage at the output to the precedent state. In other words the inductance completes the dead time switching to next state beforehand, every thing goes well. The problem occurs when the output provides the load more current in either direction than the ripple, here what happens.
View attachment 153046In this case the output current in the inductor remains always positive, the upper switch is pouring positive current in the first stage, next both switches go off, the inductor pulls its current from the lower diode provoking negative voltage. Next the lower switch is on, the output already negative remains negative but it will not be able to reverse the polarity of the current. Now when both switches are off as the current didn't go negative, no positive voltage can be created and the inductor will continue by the lower diode creating negative voltage. By this the positive part of the duty cycle gets truncated by the dead time pulling the output less positively. The inverse occurs for the negative currents bellow the ripple current.
View attachment 153047

I'm just trying to help you understand this. I've designed class D (and E) amps for a living. You seem to have some basic gaps here. But if you don't want my help I won't give it then.

 

[Kokoriantz]

I did not invent the class d amplifiers.
https://www.infineon.com/dgdl/an-1071.pdf?fileId=5546d462533600a40153559538eb0ff1
page 9-11

 

[cjm2077]

I did not invent the class d amplifiers.
https://www.infineon.com/dgdl/an-1071.pdf?fileId=5546d462533600a40153559538eb0ff1
page 9-11

That's where you're going wrong. You're looking at an example of a waveform where the switching frequency is equal to the output frequency, and while this is how some application of Class D amps that don't care much about THD work, it is not how audio range class D amps work.

 

[Kokoriantz]

This is the TI patent used in TPA32XX series.
https://patents.google.com/patent/US7795970B2/en
@cmj2017 The dead time provokes nonlinear transfer function independent from the input frequency. You get a pair of flat on every zero crossing even with 0.001Hz. Google dead time class d.

 

[Kokoriantz]

This is 1khz modulating 2Mhz with 50ns dead time with the current of the inductor. Remark that the flat occurs when the peak current crosses the 0A.

 

[Offler]

So after reading the whole topic, i assume that the point about "introduced distortion" is not valid in current and next-gen D amplifiers?

 

[NTK]

So after reading the whole topic, i assume that the point about "introduced distortion" is not valid in current and next-gen D amplifiers?

You mean this distortion?