Offline amplifier

Hello everyone,
for a powered subwoofer project I was wondering if it is possible to come up with a Class D configuration directly connected on the mains without isolated power supply.
The system is based on a PFC regulated at 380V, 2 rails are derived from it using big electrolytic capacitors, the 2 rails will be semi regulated at +/-190V.

Of course from a safety point of view the audio input needs absolutely to be isolated from the mains with reinforced insulation and here comes the main problems....

I have think about 2 possible options (see attached pdf):

1) The Class D half bridge power stage will be on the primary side and the
pwm modulator on the secondary isolated side.
The main problem is not the pwm drive itself (possible with a GDT,
optocoupler driver or other) but how to pass the feedback (pre-filter or
post-filter) trough the isolation barrier??
Does anyone has a clue about it?
An optocoupler will be surely too slow and moreover too variable...
How to keep to output offset reasonably low?

2) The Class D half bridge power stage will be on the primary side as
well as the PWM modulator. The amplifier will be a standard class-D
topology (still to decide which one) but with rails as high as +/-190V...
The audio signal can be passed trough the barrier trough a wideband
transformer but what about distortion and sound quality? The transformer
must in any case guarantee the proper distances for safety compliance...
It is probably better to convert the audio in digital, passing the barrier
and then reconvert to analog before reaching the class D modulator.
Do you know any A/D and D/A ICs for audio use which does not need
a bunch of software to control them?

Other questions:

- The application will be subwoofer only, so 20..200Hz at maximum. Which
switching frequency for class D?
I was thinking something around 30..40kHz. Of course the size of the
output filter increases but it is not an issue. Running at low frequency
allows probably to use standard cores like cool-mu of MPP without huge
core losses to manage.

- Power swithes for class D: Since and half bridge class D is basically a
synchronous buck converter with a speaker attached to it; what about
body diodes recovery time....
I will need big 600V FETs possibly with the fastest possible body diode;
Infineon CFD2 series seems promising.
Another way can be by using fast 600V IGBTs (STGW30V60DF for example)
where the co-packaged diode is far superior but the IGBT itself is slower...
Does anyone has experience with class D at such high voltage rails?

- Soft switching: the PFC will be CCM ZVS for sure (design already done for
other purpose), what about making also the class D power stage resonant?
Is it possible? any ideas? It will make a big difference at such high voltage.

- Safety hazards: The speaker will be directly connected to the mains but, at
least in my case, it is inside a 6th order bandpass enclosure not user
accessible. I don't see any safety related issues unless the inputs are
correctly isolated.

Any comment will be highly appreciated

Thank you


  • block_diagram.pdf
    53 KB · Views: 175
Everything connected to mains is potentially lethal if you don't know properly what you are doing....
On this blog there are things much more dangerous than this... just look at some user pictures on how someone wired the AC mains....

So if we can not post stuff like that just restrict the blog to battery powered stuff and the like.

I think that a project like this is not for everyone, if you don't have the right skills to work with potentially lethal voltages please do yourself a favor and stay away from DIYing with these stuff

- Audio input signal can be converted to PWM, transmitted by high speed digital isolators (opto, capacitive, inductive), then converted back to audio.

- Toslink digital audio input.

IGBT works well, I used HGTG20N60A4D.

Resonant ClassD amp? I can't figure out a way to make it reasonable (without producing more trouble then the original). But switching loss can be reduced many ways if you really want (especially if it's only for low freq).
This is a 13 years old design for stereo audio-PWM-isolation-demodulation chain. ICs are 74HC04 and 2*OPA. I don't have a shematic, that time I kept values in head, and wiring is talkful.


  • Decouple.GIF
    64.1 KB · Views: 621
Hello Pafi,
thank you very much for your suggestions.

1) IL300: I have already used HCNR200 (practically the same as IL300) but I
had a very bad experience in switching environment... It was picking up
noise very easily. On top of that what about using it to transmit audio?
Is the linearity good enough to keep THD acceptable

2) PWM: it seems the easier and best way to do it.
I suppose your PCB is about that, I will take a look and let you know, many

3) Toslink: nice idea but I need balanced analog input

IGBT: HGTG20N60A4D is a fast 600V IGBT similar to the ST "V" series I have
proposed, did you used it at which rail voltage?

From the comments you gave me it seems you are better oriented towards
OPTION2; am I right? OPTION1 for me will create a lot of troubles trying to
isolate the feedback introducing poles, zeros, delays and whatever else....

Resonant class-D: it would be nice to have it but currently no idea on what to do exactly.... I was not intending resonant like an LLC converter (sinusoidal current shape) but more ZVS (or ZCS) transitions using an active snubber. Of course keeping Fsw low will reduce the switching losses and probably there is no need to complicate our life by making it soft-switching...
By the way do you think that a fsw around 30..40kHz will be ok for subwoofer applications?

Thank you very much

You're welcome!

Yes, IL300 is not perfectly immune to common mode disturbance, maybe 2 in differential mode help to avoid noise pick-up. I don't know how linear they are. Maybe you can increase linearity by high freq biasing and differential transmittion.

I used HGTG20N60A4D at about +-130V rail voltage and 4 ohm load, with additional hyperfast diodes, and +/-160 V at 8 ohms. Fsw was about 140 kHz at idle.

Another way can be by using fast 600V IGBTs (STGW30V60DF for example)
where the co-packaged diode is far superior but the IGBT itself is slower...

This IGBT is very fast and powerful, the weakness is still the diode unfortunately.

For bass 30..40 kHz can be enough from the aspect of control, but I prefer higher switching freq, because I don't like too big output chokes, nor high idle current. With BD modulation it would be better, but the extra complexity not neccessarily worth the effort. (Parallel operation with BD modulation needs very precise pulse width matching, or additional control loop for current going from one leg to the other.)
Hello Pafi,
I have analyzed you circuit and it is a simple self oscillating PWM modulator based on a opamp (integrator) and inverter gate (comparator). On
the secondary side you are reconstructing the signal with a sallen-key active filter. Very nice solution, I like it.
Just to know with your component values at which frequency is it oscillating at idle?

I think I will use a circuit similar to this one but I will use ADuM isolator because I will have also other signals to cross the isolation barrier.

IGBTs: What about a STGW30V60F (no diode) with a LXA20T600 in parallel?
I have started the schematics and in attached you can find a part of the amplifier section.

1) Power stage: I have chosen IGBTs + Qspeed diode, I think that it will be
enough at 30kHz switching frequency. The driver is a bootstrapped type that allows to set the dead time by changing R25

2) Output filter: 4th order Butterworth, cutoff at around 4kHz at 8R load.
The carrier attenuation is 60dB and the peak-peak ripple current in the first
inductance is 8App. Not yet done the magnetic design but I think to choose
Hi-flux or MPP for the first choke; iron powder -26 for the second.
Capacitors are Panasonic ECWF polypropilene

3) Audio input isolation: I have chosen Pafi's solution (thank you) with some modifications. Don't look too much at the components values here since I have yet to simulate for the right switching frequency.

4) Class-D modulator: I have tried to replicate IRS2092; no experience is class-D except this IC... Probably the circuit can be converted to UCD but how to deal with a 4th order filter???
The values around the integrator are not calculated at all... I have still to simulate that part to reach around 30kHz FSW.

5) Protections: It is protected against overcurrent (using an Hall sensor) and against output offset. Overcurrent event (>35A) will stop the switching, if this condition is continuous for more that 2sec, everything is stopped (Overload).

6) PFC: still to be done, but at 90% complete from another design

7) Auxiliary supply: still to be done. It will be a simple flyback with lots of
outputs on the primary and on the secondary side

8) Audio processing: still to be done. I was thinking about a variable low
pass filter and possibly also a soft clipping stage (limiter). I would like to avoid to clip the class-D power stage at these voltage levels... I prefer to clip the signal at low level before the output reach the supply rails of the class-D

@Chocoholic: very nice stuff indeed these FETs.... I have also experimented
with similar stuff with impressive results. This project is private stuff just for fun and the price of the components is not a problem at all. Maybe I will experiment with these beasts to see what they can do.


  • AMP009_V1.0_p1.pdf
    300.3 KB · Views: 170
Hello, i ve done some full bridge sub amp on 620VDC rail 15 years ago. 20khz with 1200V IGBTs. amp is still working.
At 400V rail today, just for subs, you can go as low as 25khz and use fast 600v igbts, try out the wolfspeed series what is WAY overkill, or use Standard coolMos plus antiparallell carbide diode plus blocking diode to avoid the cool's body diode ever gets conducting (what would be less cool if it must be recovered).
I have quite finished the schematic of the amplifier.
Some comments:

-page1: There is all class-D related circuits. With the values shown the idle
frequency should be 37kHz (from simulation). I have included also a "starter" like in the IRS2092 chip, based around JFET Q7. The purpose is to charge the integrator capacitors C16 and C17 before enabling the power stage switching. This should in theory eliminate "click" noise at power cycling.... we will see if it works. I have also included a limiter circuit based on Bob Cordell
"Klever Klipper". There is also a output DC presence protection (at +/- 5V) and an ouput overcurrent protection (at +/-35A).
Some toughts about the output filter.... there are 8App of ripple in a 400uH inductor.... not yet done the design but seems a big piece of iron...

-page2: Audio processing stage on a separate PCB: input balanced stage and a 4th order variable Linkwitz-Riley low pass filter tunable from 42Hz to 165Hz (all the 4 pots are ganged on the same shaft).
-Various LEDs for status indication and PAFI's PWM modulator. I have changed with 74HC14 plus the possibility to add positive hysteresis (R99) if 74HC04 is used.
-Protection circuit recycled from other design, should monitor the status of the amp and switch it off in case of problems.
-Fan and thermal protection management still to be done

-page3: PFC: sorry I cannot share the full schematic but it should be enough clear what I mean.... It is a 40kHz CCM PFC based on IGBTs. The IGBT always works in ZVS to avoid recovery losses from the diode.
I have made this design for 1kW continuous power and the efficiency at 230V input is about 98%.

-page4: System supply: simple flyback with a lot of outputs for all the needed supplies.... It will be nice to wind that transformer....
All power supplies used for signal are post filtered with a 7815/7915 LDOs
The 15V for driver is regulated on the primary side with DZ5 and Q34

@ViennaTom: Full bridge at 620VDC.... what a monster did you built? Is it intended for 4/8ohm speakers or else?

@Chocoholic: I like these SiC Mosfets but here I think that these IGBTs + parallel hyperfast will be enough at 33kHz... Can be anyway nice to see I we can have some performance gain....

Any comments and/or suggestions are welcome.




  • AMP009_V1.0.pdf
    713.4 KB · Views: 153
I am now designing the output filter and from the calculations my first 400uh choke becomes enormous...
2 stacked mpp cores with 61mm OD to keep the core losses and partial saturation mangeable.
I have some thought about the inductance change at high output current which will create thd since my feedback is before the filter.
Now some questions:
-I have chosen a 4th order filter with a cutoff at 4kHz, switching frequency is 33kHz. Is it a good choice?
-It is 4th order to keep the residual output ripple low; with this I should have about 60dB attenuation. With such high supply rails I think that a 2nd order filter is not enough, what do you think?
-I have tried to simulate with post filter NFB; if I take the feedback before the second coil it seems to work fine
Tried also to include the whole filter with no succees.... I need to boost the phase of more than 180deg to make it oscillate at the right frequency....
Does anyone has already done post filter NFB with 4th order filter?
What about closing the feedback after the first coil and keep the second LC out of the loop?
Thank you
"-It is 4th order to keep the residual output ripple low; with this I should have about 60dB attenuation. With such high supply rails I think that a 2nd order filter is not enough, what do you think?"

There is much freedom in choosing the filter if you do not intend feedback after the filter AND if your signal bandwidth is only 200Hz.

In my case I did some 650Hz damped 2nd order filter using air coils. Drawback is size of course, as you can see from the pics...
Generally you should not worry about residual output ripple too much:
a) it is inaudible / plus as a bonus the subwoofer is totally incapable of reproducing anyway.
b) the subwoofer will present a high impedance, so there are maybe some single extra milliwatts heating up the voicecoil - forget about that as long as it is less than a few watts.