SystemD_2kW, any interest for an open design?

But really a lot.of people are waiting a monster from you. :)

...you mean something powered from a three phase PFC delivering a 600V rail and then running a full bridge with paralleled SiC devices? :D
Well, I just had a short look into the concept - and then decided it's time
to learn something new.

Microcontrollers. Of course not new, but my skills in this regard are too poor for my taste. And at the same time uC have evolved to become really fantastic devices.
Not just suitable for controlling simple house keeping tasks, but available for attractive costs with fantastic peripherals like fast ADC, DAC, comparators... and cool timer structures which allow to generate highly flexible gate drive patterns including nice cycle by cycle controls for various kinds of power conversion .... :D
...got addicted short after starting to dig...
Since that my brain is permanently close to a synaptic collaps (work + learning uC + general survival).
And when looking to the capabilities of modern DSP or SoC - entering a world.

I am sorry. No monster amp in preparation.
 
...available for attractive costs with fantastic peripherals like fast ADC, DAC, comparators... and cool timer structures which allow to generate highly flexible gate drive patterns including nice cycle by cycle controls for various kinds of power conversion .... :D
...got addicted short after starting to dig...
Since that my brain is permanently close to a synaptic collaps (work + learning uC + general survival).
And when looking to the capabilities of modern DSP or SoC - entering a world...

Oooo, kewl! :cool: So can we expect a new design in the next few years that has an ADC in, FET-driver out, and all software in between? Maybe a digital EQ, X-over, and speaker protection while you're in there? :D

You might have chosen a chip already, and you wouldn't necessarily have to use my suggestion either, but this one looks interesting to me for its built-in power supply and FET drivers:
Hydra-X | Active-Semi
The on-chip ADC is "only" 10-bit natively, but maybe you could do some Delta-Sigma trickery to fill in some more bits...or use a for-real audio ADC as a separate chip, which basically does that anyway.
 
... a new design in the next few years that has an ADC in, FET-driver out, and all software in between? Maybe a digital EQ, X-over, and speaker protection while you're in there? :D
:snail::snail:
...well, my uC playground in the moment is not in audio, but in RC boats.
And the only audio thing which I need there is some engine sound etc. For this it turned out sufficient to use the internal ADC + timer/PWM and as power stage an external gate driver. Surprising how well this thing worked - but I stepped away from that, because I was not able to generate satisfying engine sounds without a DSP. Finally I ended up a sound board which allows to use modified records of real engine sounds.
And this board already has an amp as well.


You might have chosen a chip already...
Currently I am playing with XMC family from Infineon.
They also provide an IDE (called DAVE) based on Eclipse and have available already pretty some powerful API.
Furtheron the real time debugger system uC-Probe from Micrium is already included to DAVE and licensed for the XMC family.

Overall really a powerfull and comfortable IDE which you get automatically, when going for these XMC devices.
Of course IDEs like DAVE do not end up in the smallest code, but
that's no issue for me in the moment.
In any case there are various interesting uC and IDE out there.
The Hydra-X might be another, but I never looked into this.

...or use a for-real audio ADC as a separate chip...
Yup, for real audio applications a separate chip would be my choice.
While the integrated ADCs of the XMC are 12 bit and allow conversion times in the sub microsecond range - for audio IMHO it is more suitfull and
also well established to go for a separate 24bit/192kHz device ..and of course there is more speed and resolution, if you are willing to spend serious money on the ADC.
Also the integrated 16bit timer units for the PWM with 7ns resolution of the XMC4700 might not be a high end audio choice, but it is amazing to see that a todays standard industry micro controller can already provide a bread and butter level of a class D modulator simply as a side task.

Well may be with the dithering options to enhance the effective resolution of the PWM the XMC could do even better than just bread and butter level , I never tried.
Also one could superimpose small 500kHz AC to the inputs of the ADC and average multiple readings to get more pseudo resolution and/or use multiple of the ADC inputs for synchronized but slightly time shifted conversions and calculate higher pseudo resolution from that.
 
:snail::snail:
...well, my uC playground in the moment is not in audio, but in RC boats...

Oh, okay.

...Yup, for real audio applications a separate chip would be my choice.
While the integrated ADCs of the XMC are 12 bit and allow conversion times in the sub microsecond range - for audio IMHO it is more suitfull and
also well established to go for a separate 24bit/192kHz device ..and of course there is more speed and resolution, if you are willing to spend serious money on the ADC.
Also the integrated 16bit timer units for the PWM with 7ns resolution of the XMC4700 might not be a high end audio choice, but it is amazing to see that a todays standard industry micro controller can already provide a bread and butter level of a class D modulator simply as a side task.

Yeah, it's pretty much a no-brainer to use an external ADC for the audio-in, just so you know it works and don't have to re-invent that wheel. Feedback from the amp might be interesting though. *Maybe* you can compensate for the inherent delay in the ADC chip, but that's more of an open-loop design with real-time adjustments than a true error-amp. But it does at least answer the question of what to do with the other channel of a stereo ADC. (one for input, one for feedback)

Well may be with the dithering options to enhance the effective resolution of the PWM the XMC could do even better than just bread and butter level , I never tried.
Also one could superimpose small 500kHz AC to the inputs of the ADC and average multiple readings to get more pseudo resolution and/or use multiple of the ADC inputs for synchronized but slightly time shifted conversions and calculate higher pseudo resolution from that.

That's pretty much the Delta-Sigma technique that the pre-fab audio ADC's use anyway:
1. Grossly oversample (in the MHz range) with relatively low resolution and a cheap/simple analog lowpass/anti-alias filter
2. Digital "brickwall" lowpass to finish anti-aliasing for the final kHz rate and to fill in the lower bits
- (from a DSP perspective, averaging is simply an FIR lowpass with all coefficients equal; there are lots of other ways to do it too: FIR, IIR, tweaking coefficients, etc.)
3. Throw away most of the samples to arrive at the final kHz rate

It probably doesn't help at all (and may hurt by a few instructions) to use multiple ADC channels if they're actually MUX'ed into a single converter...unless it's easier to offset multiple input pins by a different analog constant than to add a noise generator to a single one. (the noise generator doesn't have to be sync'ed with the ADC; in fact, the theory itself uses "uncorrelated" aka "white" noise that is simply band-limited away from the "interesting" spectrum)
 
Last edited:
Disabled Account
Joined 2005
...you mean something powered from a three phase PFC delivering a 600V rail and then running a full bridge with paralleled SiC devices? :D

Well, not that such beast actually :D
Even I am a dumb, but stil I prefer a realistic amp size. My choice is a 4 ohm loaded amp eough to drive two big 18" or 21" speaker per channel. In this case a fullbridge class D with 1 set only mosfet i.e. IRFP4668 like yours is enough.

Fullbridge vs BTL?
When we are in kilowatt amp sometimes we get uncounted problem. Small problem sometimes become big problem. In this case BTL is not as stable as fullbridge because of switching speed slip between two channel may cause big problem. Unless both channels are in synched and locked. I have experience that fullbridge much more reliable in this application.
 
It probably doesn't help at all (and may hurt by a few instructions) to use multiple ADC channels if they're actually MUX'ed into a single converter...
The XMC4700 has four independant ADCs of which each can be muxed to 8 inputs (well, somehow limtited to 26 in total - the XMC family is suffering from unspleasant restrictions in pin routing and not all ports have all capabilities...)
Anyhow - there are really four independant ADCs available. However I did not dig into their timing restrictions.
 
I think I'm getting a rough architecture in mind, with a DSP and two Hydra-X's (each chip has three half-bridge FET drivers and a DC-DC buck controller):

1x Hydra-X for the main power supply and tweeter amp
- Two half-bridges for an active rectifier - can accept AC or DC input power with minimal loss
- Remaining half-bridge for the tweeter
- DC-DC for digital supply

1x Hydra-X for midrange and woofer amps
- Two half-bridges for a full-bridge woofer
- Remaining half-bridge for midrange
- DC-DC for analog supply (balanced input, anti-alias, etc.)

1x DSP for audio processing
- ADC in
- System EQ
- 3-way X-over
- Driver EQ's / Gains
- Driver delays? (for a horn tweeter and direct-radiated woofer, for example - the woofer would need to be delayed in that case by the speed of sound over the length of the horn)
- Out to Hydra-driven amps

The amps themselves would have no audible processing as long as everything works normally (no component failures, user doesn't crank it and walk away, etc.), but their software should lean towards the pessimistic side in terms of detecting problems and preventing damage. And maybe their corrective actions can be communicated back to the DSP so that the end result can still sound somewhat reasonable. For example, if the woofer ends up, say, 6dB down for some reason (maybe it's been clipping for a while and getting hot), reduce the mids and highs by that much so that it still sounds "full".

Now...if the full-bridge woofer is more-or-less two of Choco's 2kW design adapted for digital control, and it's now dedicated to the woofer alone... :D:D:D \m/

I know it's a lot of work, but does anyone see a technical problem with that?
 
..does anyone see a technical problem with that?
Whenever you do something which is new for you - problems lurk around at the most unexpected corners. :D

One thing you definitely should add to your concept is the safety isolation.
Our audio signal sources are not suited to be connected directly to offline circuitries. So you will need something like safety isolation amplifiers at the input.
Or do you skip the AC option and run from floating batteries only?
Still then you have to ensure that your signal source is not lifted much vs earth and you have to ensure that circuit parts with lethal voltage levels and conductive speakers chassis cannot be touched.
 
Whenever you do something which is new for you - problems lurk around at the most unexpected corners. :D

Yeah, I'm sure that'll happen! Several times even.

One thing you definitely should add to your concept is the safety isolation.

YES!!! Definitely!

Our audio signal sources are not suited to be connected directly to offline circuitries. So you will need something like safety isolation amplifiers at the input.
Or do you skip the AC option and run from floating batteries only?
Still then you have to ensure that your signal source is not lifted much vs earth and you have to ensure that circuit parts with lethal voltage levels and conductive speakers chassis cannot be touched.

I'm thinking to keep all of the electronics completely inaccessible from the outside (heatsinking could be interesting) and use an audio transformer at the input. Then (and only then) I can rectify the AC mains directly, without a power transformer because the DC rails are supposed to be that anyway, or use an equivalent high-voltage battery. I already have a 1:1 isolation transformer that I can use for testing; at least that'll let me touch one thing at a time or put a 'scope on it.

The more I think about it, the more I like the "dual-nature" idea from an active rectifier.
 
Last edited:
Are there any potential pitfalls that you would like to share about building a class D power stage? Maybe some problems that you ran into, together with your solutions?

I know that dead-time/shoot-through is important, as is the correct switching frequency (higher than audible, but slow enough to stay out of the linear region for most of the time), but is there anything else that you had a problem with?
 
With your last question you are opening 50% of class D knowledge base.

Layout is one half - control theory the other.
Regarding layout of the power stage you can have a look to posting #144 and #150 where I modeled the power stage with parasitics and different snubberings. Starting from this I iterated my layout and the simulation with the parasitics according to the layout until it looked reasonably from theory - then I stepped into the real build.
In the layout I enabled various kinds of snubberings and in reality I also used them finally.
In case you intend to make your own layout and such switching stages might not be your key expertise, then be prepared to run through various iterations until you get it reliable. You will also need a reasonably fast scope with a bandwidth of at least 100MHz.
 
Hmm...It may take me a while to wrap my head around all that, especially with the other projects that I need to finish just to avoid having too much to do all at once. (custom DMX lighting controller, toaster-oven controller to reflow SMT parts, 48V phantom-powered guitar pickup, etc.)

I'm also thinking about having an array of FETs for each switch instead of just one. That'll improve Rds_on, heatsinking, and probably switching speed (by using smaller devices) at the expense of PCB real-estate.

If someone else gets to this before I do, please post the results like Choco did!
 
Go for fullbridge topology
I ensure you will not regret :)

Well, according to Choco (and I believe it), there's a fair amount of extra complexity involved with bridging class D, mostly from coordinating the switching times. (so if they're controlled by software, I'd say they *must* come from the same processor) But the advantage is 2x the output voltage swing and thus 4x the power from the same supply voltage. Each "channel" in that bridge sees half the impedance, and the supply sees two of those per bridge, so the extra power does come from somewhere (e.g. not a free lunch), but if you're going for power without a ridiculous HV supply, and you're willing to work with the added complexity, then that's the way to go.

I am designing simple fullbridge class D UcD with IR2110.
With discrete folded LTP it is easy to combine with IR2110.
Still concept.

Okay, a google search for that part number looks to me like a generic half-bridge FET driver. Two of them (and 4 N-channel FETs) could give you full-bridge. Anything special that I'm missing?

As for the file, not all of us can read specialized file formats, even if the software is free that uses it. (and no, we're not going to install all the free software on the planet just to read everyone's files) Can you please convert it to something more generic, like an image? Keep the original in case someone does want to play with it, but a read-only image is sufficient for most of us. Thanks!
 
Disabled Account
Joined 2005
The file is ltspice.
You.can download free.

Fullbridge vs high volt halfbridge definitely fullbridge more simple in term of switching behaviour because of working in low level voltage. The inverting is so simple as well. No need.so fanncy schematic. Both halfbridge and fullbridge also.have equal risk of cross conduction of mosfet. It is wrong that fullbridge has higher risk.
 
The file is ltspice.
You.can download free.

Like I said, we're not going to download all the free software on the planet just to read everyone's files. You might have found the most wonderful tool in the world (for your specific needs), but that doesn't matter. Screenshots please. (Alt+PrntScrn key, then Ctrl+V in MS Paint, save as PNG, upload that...or print to PDF like Choco did)

And yes, I've used LTspice before, so I know what it is and what it does. I just happen to like something different.

Fullbridge vs high volt halfbridge definitely fullbridge more simple in term of switching behaviour because of working in low level voltage. The inverting is so simple as well. No need.so fanncy schematic. Both halfbridge and fullbridge also.have equal risk of cross conduction of mosfet. It is wrong that fullbridge has higher risk.

That makes no sense to me at all. High-voltage is not more complicated; just more dangerous. The physics still work the same way. Maybe you're thinking about complexity in terms of keeping people out of it? Or maybe you're confusing/swapping the definitions of complexity and risk?

Yes, inverting is simple, but not necessary at all in HB. So FB *is* more complicated because you need two HB's per speaker instead of just one, and one of those HB's needs to be inverted...at least in a naive sense. I'm toying with the idea of actively switching only one HB in a FB pair while the other HB stays grounded (low-side solid-on). When the signal changes polarity, the HB's swap roles. That could greatly reduce or perhaps even eliminate the switch-coordination problem. It could be done with an analog circuit that's designed for FB exclusively (no dual-HB option), but I think the marginal cost is even less if you're already driving it from software.

Yes, HB and FB both carry the same risk of cross-conduction. That's a HB thing, and it carries over into FB because you physically have two HB's in a FB. (and if the HB's are linear (class-AB), then they always cross-conduct a little bit anyway)