John Larkin wrote:
On Fri, 19 Nov 2004 13:16:44 GMT, "Kevin Aylward"
wrote:

Again, potential issue of stability. I do have an interesting,
somwhat original output circuit that does this with a first order
loop enclosing the outputs. It gets me wonderfully low distortion at
20Khz. In one sense topologically its essentially the same as other
compound pairs, but this does behave a little different. The
feedback diodes are zeners at 10V. The drive circuit on its own
0.0001% THD 20Khz, according to spice that is. Note the feedback
transisters can be low voltage ones, hence fast ones. DC loop gain
is 135db, open loop more!

http://www.anasoft.co.uk/Mospoweramp.jpg



One problem here is current sharing. Linear-mode paralleled mosfets
don't do it very well.

Yes, yes yes I do know all about this:-)

....but they're not that bad either, when compared to bipolar.

One of my earlier gradient amps was done with
paralleled fets (with source resistors, at least) and we wound up
making a fixture so we could match sets of fets for production... a
real pain in the sternparts.

On might well be surprised just how well they match in same batches. Its
unlikely to be more than a 2/1 mismatch at say 8 volts drive. Dropping a
volt or so at max current via source resisters is always an option. It
depends on the application. In audio power amps, the reality is, if its
running max output and not clipping, then the average power is way way
less for music signals. Indeed, in my MOSFET1000 (500W per chan) the
transformer was only rated at 800W. I can tell you, transformers never
blow under these conditions, despite worst case saying the will. Its
knowing when to engineering cheat. My own MOSFET1000 has been going
strong for nearly 20 years now, and it has no source resisters in the
design.

For real applicatons where a amp is actually going to be outputing its
ratings contineously, one has to be a bit more carefull.

The advantage of an opamp per fet is that
the gate drive becomes very simple and sharing is forced to be
perfect.

Yes, but as I noted, there is the pole of the opamp, and the pole of
rout with the fets, plus all the other op-amp parasitics. Its the old
frequency verses accuracy trade off. Anyway, its cheating to use an
op-amp:-)

My none output feedback version is
here.http://www.anasoft.co.uk/Mospoweramp2.jpg. This is quite a bit
faster as it needs less compensation. It still does 0.005% at 20k,
8ohms, 500 Watts, in spice of course.

Why do you think I tied the caps at the gates, instead of he push pull
emitters?

Kevin Aylward

http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.