Just for fun: UcD on a shoestring

[luka]

any feedback would be hard to add, am I right?
it seems you can only use complementary pairs or you'll get offset, is this corrent?

[Elvee]

Originally Posted by luka

any feedback would be hard to add, am I right?

But there is already feedback!!

R3-R6 and R6-R8 set the theoretical closed loop gain to 11, but since the open loop gain is of the same order, the resulting closed loop gain is approximately the half of that.

You could set a lower closed loop gain, this would result in an even better distortion.

Note that with this simple design, you need good quality supplies: the average gate voltage is a direct proportion of the supply voltage, and because of C6 crossing the gates, any noise on the supply rails will appear in the G-S voltages. This requires a good decoupling.

[luka]

Originally Posted by Elvee

But there is already feedback!!

R3-R6 and R6-R8 set the theoretical closed loop gain to 11, but since the open loop gain is of the same order, the resulting closed loop gain is approximately the half of that.

You could set a lower closed loop gain, this would result in an even better distortion.

Note that with this simple design, you need good quality supplies: the average gate voltage is a direct proportion of the supply voltage, and because of C6 crossing the gates, any noise on the supply rails will appear in the G-S voltages. This requires a good decoupling.

That is when you have monitor too close... then I simulated myself and I've seen it.
any reason why not to use higher switching freq?
I changed 150p to 33p or 47p

[Elvee]

C7 does not actually set the oscillating frequency, its role is to provide some positive feedback, to make the oscillations clean and strong.

As in other UcD schemes, the frequency is set by C4, C5, the output filter, and other marginal delays in the loop, like the lowpass filtering caused by the MOS input capacitance and resistors R1 and R2.
If you attempt to act on a single element, the performances will decrease because this lowpass will tend to dominate, and will also increase the losses.

You could add follower stages before the MOS transistors, but then it would become a six-transistor UcD, not a 4...

[elevator]

Elvee,
clever little circuit you have tossed our way here .
After hours of sim to learn it, here is a version with +-20V rails. Keeping the changes from your original to the minimum to allow fair comparisons:
- limited the gate drive to 8.2V
- tweaked the emitter biases by keeping the voltage steady for proper handling of crossover over a range of supply voltages. Not the final solution (I created a 1.2V Zener...) but works for sim purposes.
- a couple of minor value changes.
With the same input as in your OP I was able to get 0,04% distortion, and even close to 0.01% by fiddling with the Zener but that’s obviously not practical.

Next I plan to analyse the PSR behaviour. On the practical side, I am getting ready to make a PCB if I may; thinking of the following (please comment):
- make it stereo, inverting the phase on one channel to avoid pumping issues
- add an input stage to provide some (needed IMO) gain, and for the phase inversion
- add PS filtering and decoupling, the usual stuff
- depending on how the PSR looks I could add a rectifier bridge for a linear unregulated supply.
The goal is to make it the (cue Mr Evil soundtrack) cheapest class-D amp on the planet (but not the worst), so suggestions in this direction are welcome, eg for dirt cheap output mosfet pairs (say TO-220).
Given the circuit’s simplicity, a one layer board is probably overkill Do they make half-layers PCBs?
Again, great job Elvee.
E

[elevator]

Forgot, I haven't looked at thermal issues at all.
One thing is that the "zeners" should probably track the Vbe downwards, as things heat up. So I'm thinking of another Vbe somewhere just for that. Ok, a transistor here, another transistor there...

[luka]

what did you use R12, 13 for?

[elevator]

Originally Posted by luka

what did you use R12, 13 for?

for simulation, to measure source currents (and cross-conduction)
E

[Elvee]

Originally Posted by elevator

- tweaked the emitter biases by keeping the voltage steady for proper handling of crossover over a range of supply voltages. Not the final solution (I created a 1.2V Zener...) but works for sim purposes.

Forward biased diodes or Vbe multipliers would be OK I think, they will thermally compensate the input transistors.

Next I plan to analyse the PSR behaviour. On the practical side, I am getting ready to make a PCB if I may; thinking of the following (please comment):
- make it stereo, inverting the phase on one channel to avoid pumping issues
- add an input stage to provide some (needed IMO) gain, and for the phase inversion
- add PS filtering and decoupling, the usual stuff
- depending on how the PSR looks I could add a rectifier bridge for a linear unregulated supply.

Yeah, that looks sensible.
At the very least, an input buffer is necessary as the source impedance needs to be low enough.
I am afraid the PSR will be rather poor with such a low loop gain, and it will require either a very good filtering or a regulated supply.

The goal is to make it the (cue Mr Evil soundtrack) cheapest class-D amp on the planet (but not the worst), so suggestions in this direction are welcome, eg for dirt cheap output mosfet pairs (say TO-220).

Good luck!

[savu]

What is the role of R2 and R10?

I am simulating it to work on +\- 45V and i want to understand what do those resistors do.

regards