• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Class D must read papers and publications

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
3,011
Likes
6,049
Location
Prague
I would like to open the thread dealing with important publications on class D amplifiers principles, topologies, characteristics and limitations. One of the great sources is the Hypex publications library at

The most explaining and resourceful publication is IMHO the "AES Class D master class" presentation
which explains in detail class D principles and also the main different basic topologies, the hysteresis modulator and the phase shift controlled oscillator. Benefits and limitations of both approaches are explained and it would explain to the reader what he could expect from both approaches.

UcD principle and comparison to hysteresis modulator is also explained in

and NCore improvement over UcD in

and also in

I believe that these publications might help the ASR visitors for better understanding of results of class D reviews here at ASR.

I would be very grateful if @amirm , @BDWoody or @AdamG247 could make this thread sticky, as it links to basic and necessary literature to understanding of class D function and parameters.
----------------------------------------
https://www.cornestech.co.jp/images/uploads/file/products/pdf/oa_product7-4_201503.pdf
Audio Precision white paper "Measuring Switch-mode Power Amplifiers". All the basic measuring methods like THD, THD+N/level, THD+N/frequency, IMD, DIM, noise, damping factor, crosstalk are explained quite in detail. A lot of permanently repeated questions is answered.
 
Last edited:

DanielT

Major Contributor
Joined
Oct 10, 2020
Messages
2,678
Likes
2,139
Location
Sweden - Слава Україні
Thanks PMA.:)

Questions from a beginner, if it's ok with you? :

Is it possible to say anything in general about class d and levels of distortion in the higher frequency registers? Does not the distortion in the higher registers increase quite a lot when the effect is increased? Has not there been (is?) a problem with class d amplifiers? Or has development regarding class d progressed so much that this is solved?It probably differs from manufacturer, model and construction and so on, but is it possible to say something in general? What I'm looking for is what you sometimes hear about class d, that is: Great for subwoofers but not for full-range speakers.

In any case, I think Hypex seems to be really good for full-range speakers. Even world class if I interpret different tests that Amir performed on Hypex class d correctly.:)

I also have in mind that with class d, it is not so easy to get good constructions that can take care of the loads the speakers expose the amplifier to. Do you know if that's right?
 
Last edited:
  • Like
Reactions: pma

dougi

Addicted to Fun and Learning
Joined
May 31, 2020
Messages
618
Likes
505
Location
ACT, Australia
I like this (old, 2001) white paper which has the very basics but focuses on measuring them. The 2019 update on measurements of their chips is also a good measurement reference. To understand the principles, you need to measure too!
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
3,011
Likes
6,049
Location
Prague
@DanielT Again it is a question of circuit topology (not of the manufacturer) and feedback loopgain.

Though hysteresis modulators (TPA and similar) have basically more linear LF transfer function, they suffer from frequency response load dependence which can be fixed only partially.

Phase shift controlled oscillators with global loop taken from load have less linear transfer function, however load invariant frequency response and low output impedance over the audio band. Now it is about feedback, more feedback (higher loopgain) improves linearity up to the level seen in Purifi. Both NCore and Purifi nonlinearity even at high output level is below any audible thresholds. Now it is about stability, if you use a lot of feedback, then loopgain amplitude and phase plots must fulfil stability criteria and this is what both NCore and Purifi were able to achieve, so it makes the the top class in class D. Circuits based on hysteresis modulators have no chance than to be worse in parameters.
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
3,011
Likes
6,049
Location
Prague

Soren

Member
Joined
Feb 27, 2021
Messages
16
Likes
14
Very usefull links, thank you very much.
I´m part of a small group of four nerds, doing our own designs, and amongst that class D design as well.
In doing so we found a few points, which may be of interest for the DIY class D enthusiast.
Lately Mr Putzeys nCore and Purifi modules have become the absolute reference standard for any kind of amplifier when talking about measured performance. Often Putzeys amplifier modules beates the competition with a decade or more when talking of THD+N.
That is really amazing, but there might be a logical explanation to it though.

When looking at datasheets both Purifi and nCore looks quite amazing, but there is one thing, that should catch your attention, and that is the resulting gain of the amplifier mudules. The gain for both purifi and nCore is around 12-13 dB which doesnt reach the standards for required gain in home HI-FI systems. Gain arround 26 dB is normal for "Non Studio" applications.
The result of this is, that external gain is required, and to accomodate the amplifier with gain of sufficient quality, you´ll allways need one of the latest technology op-amps i.e. OPA1612 or like.
Hypex OEM modules can be ordered with what they call "Converterboards", containing this extra gainstage for full signal level.
And this is one of the sources for the extra performance from both brands. Reducing the gain raises performance significantly.
In addition the OPA1612 has even less distortion than the class D modules themselves, at least in unity gain designs.
So the overall distortion of the required external gainstage and the module combined still has superior performance, which eventually couldt be even further enhanced, if someone could create an additional feedback loop around both the external gainstage and the class D amplifier together.
The noise level though doesn´t really show you the reality, as the op-amps noise level should be added to the noise of the module, this raises the noise of i.e. nCore500OEM to more normal levels, also obtained in our own amplifier design with 28dB of gain.
So the point is, that gain is a of trade off. You´ll get the extra gain these days without sacrificing performance, but you´ll have to complicate the design with an external gainstage. And to this day I´ve never experienced that adding such components will give you any further performance enhancement, it´s quite reverse til my opinion.
Operating a class D module directly into the modulator without any unity gain bufferstage, will require a pretty strong pre-amplifier though, because the input impedance for a typpical class D amplifier is around 1K Ohm only for unballanced input and 2 X 1K Ohm for ballanced input. This is due to noise level optimization in the amplifier, higher impedance results in higher noise levels.
The best class D amps have noiselevels @ about 25µV at the output, which means that the entire noise level after a gain @ 30dB more or less equals the noise level in a single resistor. Hypex/Purifi modules need the bufferstage noise added to compare.

Anyways class D really has disrupted the market for high quality amplification, beginning with ICEPowers design back in the mid ninetees.
Low noise levels, very low THD, and the absence of cross over distortion just like class A amplifiers, have raised the bar significantly.
Now we just have to understand, why they sound differently in spite of spaceship data, which for the best class D should be sufficient to claim transparancy.
 

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
2,318
Likes
3,869
Location
Bjerringbro, Denmark.
Now we just have to understand, why they sound differently in spite of spaceship data, which for the best class D should be sufficient to claim transparancy.

Have it been established that they actually do sound different?

Don't get me wrong, I'm all for an investigation if there's something to look for, but wild goose chases should be avoided, IMO.
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
3,011
Likes
6,049
Location
Prague
When looking at datasheets both Purifi and nCore looks quite amazing, but there is one thing, that should catch your attention, and that is the resulting gain of the amplifier mudules. The gain for both purifi and nCore is around 12-13 dB which doesnt reach the standards for required gain in home HI-FI systems. Gain arround 26 dB is normal for "Non Studio" applications.
Hi, thank you. I am not sure if I agree, I have NC252MP OEM module here and as specified it has gain of 26dB (unless you make changes on PCB). This parameter is confirmed by my measurements. Please do not input any misinformation.

I measure noise voltage 36uV (20Hz - 22kHz unw.) at NC252MP output. This makes 1.8uV equivalent input noise voltage and 12nV/rt(Hz) input voltage density. This is a very good value for a power amplifier. Please come here with data if you do not agree, but they have to be well defined data, not an opinion only.

NC252_param.png


and I also have UcD180HG here with the same gain.


 
Last edited:

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
2,318
Likes
3,869
Location
Bjerringbro, Denmark.
@pma Now look at NC500 OEM and 1ET400A.

But yeah, some of the modules has buffers, and some of them don't.
 

Soren

Member
Joined
Feb 27, 2021
Messages
16
Likes
14
Have it been established that they actually do sound different?

Don't get me wrong, I'm all for an investigation if there's something to look for, but wild goose chases should be avoided, IMO.
Well we have used several different models as reference amplifiers alongside with my old school and very bulky class A amplifier.
I.e we used nCORE 500OEM, ICEPower 1200AS2 to compare the audio experience to our own design.
Compared to ICEPower The nCore has the best measured performance, but the ICEPower is a full bridge design, where the nCore is halfbridge.
This leads to different behavior in the real world. Full bridge will not suffer from power supply pumping, but it will be more noisy, as the noise stems from 2 amplifiers in bridged mode.
They also sound different, all though data should be sufficient for transparency (se datasheets)


Our own design would be a bit more like the nCore, since it is also a halfbridge design, but with much more gain (28dB), and can thus be driven by i.e. a unity gain buffer, or a strong pre-amp directly if wanted.

I do believe in good measured performance, I´ve never heard HI-FI gear with problematic measured performance giving af superior audio experience, but it is also my experience, that you cannot always cook everything down to measurements alone.
Different design principles can also contribute to the final experience, despite they might have almost equal performance in the lab.
The example above with half bridge vs full bridge is just one of theese, they sound different maybe because they are different, but they do not differ that much in the testlab.
Some other class D designs performs very well because of i.e. very low dead-time, this is a very important aspect in class D design.
Excessive dead-time leads directly to distortion, thus dead-time is important, and should be lowered, but this is not easy to do.
First of all insufficient dead-time leads to shoot through in the switching devices and is fatal in milliseconds, very low dead-time rases the risc of excactly that. It also develops excessive heat, if the dead-time is to low, so dead-time has to be sufficient, allowing higher distortion, if you consider reliabillity important.
The solution to this problem is feedback technology, advanced feedback loops of higher order can reduce the distortion stemming from dead-time to very low values, and at the same time reducing stress placed on the switces and adjacent components significantly, thus improving reliabillity.
Also these two ways of reducing distortion sounds different, despite that measured performance indicates that this should not be the case.
In this example one could say, that you have "Clever engineering" vs "Precision engineering".

Low levels of distortion, Zout, noise and any unwanted signals are always preferable, but this will not always result in excactly what you expected.
This is not spooky, but probably just something that doesn´t show significantly in a normal set of measurements, or what do I know, and that is yet to be discored I think.
 

Soren

Member
Joined
Feb 27, 2021
Messages
16
Likes
14
Hi, thank you. I am not sure if I agree, I have NC252MP OEM module here and as specified it has gain of 26dB (unless you make changes on PCB). This parameter is confirmed by my measurements.

View attachment 176152

and I also have UcD180HG here with the same gain.



That´s true, but the nCore showed is not one of the ultimate performance ones.
The nCore 500 has a noise level of only 9µV without input buffer.
With the supplied buffer noise raises to 27µV as far as I remember.
Our own design does 25µV with 28 dB gain and no inputbuffer.
Your module has probably an onboard gainstage which kan be bypassed.
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
3,011
Likes
6,049
Location
Prague
I do believe in good measured performance,
That's your problem and of course of many others. The key is usually a limited set of measured parameters and a lack of ability to interpret the measured data. Sighted impressions then come into the game. The other usual reason are business interests. Enough for me to argue today, thank you.
 

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
2,318
Likes
3,869
Location
Bjerringbro, Denmark.
I.e we used nCORE 500OEM, ICEPower 1200AS2 to compare the audio experience to our own design.

Afraid to ask, but "compare the audio experience" means exactly what?

In this example one could say, that you have "Clever engineering" vs "Precision engineering".

I agree. Good engineering is mostly about making sensible compromises.
 

Soren

Member
Joined
Feb 27, 2021
Messages
16
Likes
14
Afraid to ask, but "compare the audio experience" means exactly what?
That means listening to the amplifiers in various loads and with a lot of different music.
So we are down to subjective impressions of the amplifiers behavior.
 

Soren

Member
Joined
Feb 27, 2021
Messages
16
Likes
14
Sighted listening with no controls?
Sighted mostly, but unsighted as well.
But it really should not surprise anyone, that subjective differences occurs in different designs.
In the example with a full - and a halfbridge amplifier, the measurements will not take into account, that a speaker delivers some amounts of back EMF to the amplifier.
Back EMF in a full bridge amplifier will be converted to heat in the output devices, as where back EMF in a half bridge design instead, will pump up the supply rails. When designing power supplies for class D amplifiers, this has to be adressed, because else the elctrolytic capacitors will die most suddenly due to excessive voltage.
This is just one reallity, that will have no impact in the lab, but surely will, when listened to in a real world environment.
So even though things look alike, they don´t always have to be.
 

Shazb0t

Addicted to Fun and Learning
Joined
May 1, 2018
Messages
602
Likes
1,120
Location
NJ
Sighted mostly, but unsighted as well.
But it really should not surprise anyone, that subjective differences occurs in different designs.
In the example with a full - and a halfbridge amplifier, the measurements will not take into account, that a speaker delivers some amounts of back EMF to the amplifier.
Back EMF in a full bridge amplifier will be converted to heat in the output devices, as where back EMF in a half bridge design instead, will pump up the supply rails. When designing power supplies for class D amplifiers, this has to be adressed, because else the elctrolytic capacitors will die most suddenly due to excessive voltage.
This is just one reallity, that will have no impact in the lab, but surely will, when listened to in a real world environment.
So even though things look alike, they don´t always have to be.
You hand waive and generalize a lot, but these explanations are not legitimate to prove your claims.
 
Top Bottom