I've been away from audio for a while, and it seems a typical 'digital' setup these days might consists of a music library, something like Roon, sending PCM (perhaps via wifi or ethernet) to a DAC, which outputs an analogue signal. This then goes to an amp, and in the case of 'digital or class-D' amps, the analogue signal is effectively converted back to digital, modulated to produce a high-frequency digital PWM signal to drive high power switches, then filtered into an analogue signal and sent to the speakers.
'
It seems to me that the entire DAC, and the ADC/digital sampling stage of the amp could be omitted. The source PCM could be dithered to SDS/PDM or something similar, digitally, and used to directly drive the high frequency amplifier output switches. Particularly the new GaN switches would allow very high frequency PDM.
Does such a thing exist? For DIY?
'
It seems to me that the entire DAC, and the ADC/digital sampling stage of the amp could be omitted. The source PCM could be dithered to SDS/PDM or something similar, digitally, and used to directly drive the high frequency amplifier output switches. Particularly the new GaN switches would allow very high frequency PDM.
Does such a thing exist? For DIY?
Last edited:
Something like this: https://www.maximintegrated.com/en/design/technical-documents/app-notes/5/5461.html
When you look at the block diagram of the MAX98355, you see that they simply put the DAC on the same chip as the class-D amplifier. The amplifier itself is apparently analogue.
See post #32 of https://www.diyaudio.com/community/threads/pcm-to-pwm-conversion-101.205496/page-2#post-6898514 for an example that apparently uses the direct route: PCM to noise shaped PWM that drives a class-D stage. As a result, the amplifier is extremely sensitive to any disturbances on the supply, among other things.
See post #56 of the same thread, https://www.diyaudio.com/community/threads/pcm-to-pwm-conversion-101.205496/page-3#post-6902928 , for a far better way to make a digital-input class-D amplifier with as much as possible done in the digital domain.
See post #56 of the same thread, https://www.diyaudio.com/community/threads/pcm-to-pwm-conversion-101.205496/page-3#post-6902928 , for a far better way to make a digital-input class-D amplifier with as much as possible done in the digital domain.
Last edited:
Any thoughts on PDM vs PWM?
From 10,000ft, PDM seems like it should be better.
And closed-loop feedback might be easier with analogue-PDM.
From 10,000ft, PDM seems like it should be better.
And closed-loop feedback might be easier with analogue-PDM.
The Axign chip seems to be available in a DIY module, but not exactly cheap: https://truebluebox.com/portfolio-item/power-amplifier-da-input/
The TACT Millenium amplifier released back in 1999 already achieved this (and I bought it). I have no idea why this isn't more popular 20 years later. I assume it is expensive to build something like this really well.
I'm not sure why it would be much different than existing class-D, and I think PDM might be slightly more forgiving than PWM?
What you describe in post #1 has no negative feedback around the output stage to reduce errors caused by the output stage and its supply. That's the main difference with a normal class-D amplifier and with the Axign thing.
Yes I see, I described an open loop setup, but my point was really that the class-D power stage, the switches, should be controlled by PDM, rendered/dithered directly from PCM, rather than sampling analogue from a DAC.
With or without a feedback loop. Probably with feedback would be better, as you say, everything needs to be almost perfect otherwise.
With or without a feedback loop. Probably with feedback would be better, as you say, everything needs to be almost perfect otherwise.
If you want to generate the switching signals digitally, yet have feedback from the class-D amplifier output, you need a low-latency ADC in the feedback path. That's what the Axiom/Axign folks made.
Regarding pulse density modulation: when it's generated digitally, it has to be quantized in time, so it will presumably be a single-bit sigma-delta modulate rather than pure, non-time-quantized PDM. There are several problems with single-bit sigma-delta modulators. The ones with the most effective noise shaping are high-order sigma-delta modulators, but sigma-delta modulators with an order greater than two (and chaotic ones of any order) become unstable when you try to drive them to too high or too low pulse densities. A typical rule of thumb is to keep the percentage of ones coming out of them between 25 % and 75 %. You therefore can't use a plain old high-order sigma-delta when you want to reach very high and low duty cycles for your class-D output stage. The other issue is that you can't dither them properly, which leads to issues with idle tones that get frequency modulated by the signal. Both issues are much reduced when you make a sigma-delta modulator with a multibit quantizer and a pulse width modulator embedded in its noise shaping loop.
I wrote a bit more about these issues in section 3 of https://linearaudio.net/sites/linearaudio.net/files/03 Didden LA V13 mvdg.pdf , although the article is about DACs rather than class-D amplifiers.
Regarding pulse density modulation: when it's generated digitally, it has to be quantized in time, so it will presumably be a single-bit sigma-delta modulate rather than pure, non-time-quantized PDM. There are several problems with single-bit sigma-delta modulators. The ones with the most effective noise shaping are high-order sigma-delta modulators, but sigma-delta modulators with an order greater than two (and chaotic ones of any order) become unstable when you try to drive them to too high or too low pulse densities. A typical rule of thumb is to keep the percentage of ones coming out of them between 25 % and 75 %. You therefore can't use a plain old high-order sigma-delta when you want to reach very high and low duty cycles for your class-D output stage. The other issue is that you can't dither them properly, which leads to issues with idle tones that get frequency modulated by the signal. Both issues are much reduced when you make a sigma-delta modulator with a multibit quantizer and a pulse width modulator embedded in its noise shaping loop.
I wrote a bit more about these issues in section 3 of https://linearaudio.net/sites/linearaudio.net/files/03 Didden LA V13 mvdg.pdf , although the article is about DACs rather than class-D amplifiers.
Last edited:
A PWM signal is analog by itself.
From a PCB design perspective, there are some nice benefits for going switched mode all the way.
Also being said by many other very great developers (also here on the forum), GaN gives very little advantages compared to regular MOSFETs.
Another great alternative would be using a FPGA.
These days maybe even some microcontrollers would be fast enough.
From a PCB design perspective, there are some nice benefits for going switched mode all the way.
Also being said by many other very great developers (also here on the forum), GaN gives very little advantages compared to regular MOSFETs.
Another great alternative would be using a FPGA.
These days maybe even some microcontrollers would be fast enough.
What are those benefits?
Time-quantized PWM can be interpreted and generated as a digital signal (generated with a few counters and some combinatorial logic), continuous-time PWM can not.
Last edited:
Oh just the fact that you don't really have to care about a separate (low voltage) analog signal on your board anymore, next to a high current high freq switching signal. So less additional passives (capacitors), easier to voltage shift etc.
Also being able to use any jelly bean logic gate for certain additional functionality etc.
It just makes the whole design a lot more neat.
It would be absolute killer to have something like Axign is doing inside a complete driver (like an IRS20957 or something) for high power applications, or even a step further with something like a HTSSOP package (like a TPA32xx ) with integrated fet's or GaN (I mean it also doesn't really matter at this point).
At this point it's a bit cumbersome, since you need some kind of additional power stage anyway.
Might as well just make an analog amp at that point? (probably cheaper as well, with giving the same amount of performance)
I am working on it at the moment, DSD/PDM is converted to a form of PWM to be amplified, it is part of my new smart amplifier design project. My question to you is there any interest out there for such an amplifier? As I am not up to speed with what has been going on in the world high-end audio. From where I am coming from do not see a need for a DSD/PDM DAC, when it is possible to conversion to Class D, Class I or Class O, as they all use some form of PWM. As for switching speeds my RF amplifiers work very well up to 10 MHz using LDMOS, therefore converting this design to audio will not be a problem. Like wise I am now looking into GaN switching device to replace the LDMOS for improved performance.
I have a couple of questions about that:
1. Why start with DSD rather than PCM?
2. What are classes I and O?
3. If it is an open-loop system, how do you keep the reference/supply clean?
1. Why start with DSD rather than PCM?
2. What are classes I and O?
3. If it is an open-loop system, how do you keep the reference/supply clean?
Marcel, Class I is same as interleaved PWM (Crown, Harman), never heard of Class O though !!!!
https://sound-au.com/articles/137234.pdf
https://sound-au.com/articles/137234.pdf
"Why start with DSD rather than PCM?"
I can do both, I am use a universal pulse modulator, as for DSD I see PS audio and other are pushing this format with their audio equipment.
"What are classes I and O?"
Class O used zero crossing pulse width modulation ZPWM, which provides a positive and negative identities that can be used for analog signal processing,
* note I have a number articles on LinkedIn that go into how this all works, I am not sure if I can upload them here or not as a PDF files?
"If it is an open-loop system, how do you keep the reference/supply clean"
This design used a feedback loops to control various aspects of the amplification process.
In this type of amplifier the supply is part of the amplification process, that can be looked at as a form of an envelope drive like Class H, as the power power supplies track the amplitude of the audio. I do it this way as it is possible to provide higher efficiency at radio frequencies, as I testing this design as a AM radio QAM transmitter, I am now looking operating at the same switching speeds (1 to 10 MHz) for audio frequencies (20 to 20kHz).
I can do both, I am use a universal pulse modulator, as for DSD I see PS audio and other are pushing this format with their audio equipment.
"What are classes I and O?"
Class O used zero crossing pulse width modulation ZPWM, which provides a positive and negative identities that can be used for analog signal processing,
* note I have a number articles on LinkedIn that go into how this all works, I am not sure if I can upload them here or not as a PDF files?
"If it is an open-loop system, how do you keep the reference/supply clean"
This design used a feedback loops to control various aspects of the amplification process.
In this type of amplifier the supply is part of the amplification process, that can be looked at as a form of an envelope drive like Class H, as the power power supplies track the amplitude of the audio. I do it this way as it is possible to provide higher efficiency at radio frequencies, as I testing this design as a AM radio QAM transmitter, I am now looking operating at the same switching speeds (1 to 10 MHz) for audio frequencies (20 to 20kHz).