On Thu, 18 Nov 2004 17:37:26 GMT, "Kevin Aylward"
wrote:

John Larkin wrote:
On Thu, 18 Nov 2004 10:09:19 GMT, "Kevin Aylward"
wrote:


It's actually not a bad idea to have an opamp per output transistor,
if you do it right, which this guy clearly hasn't. I make a power
amp that uses 32 300-watt fets in the output (16 p-ch, 16 n-ch,
+-200 volt rails) and do just that.

Bloody hell. That's some amp, probably about 20KW out. This will
certainly kick the **** out of you at 50 Hz.

How did you get the opamp voltage rating, or was it a discrete one?


It's a transconductance amp, so the fet sources are at the rails
(through current-sense resistors) and the drains are the output. So
the gate-drive fets work on the +-200 power supply rails (with small
floating local supplies) and only need to swing enough to drive the
gates.

I have recently been playing with both source out and drain out
configurations for that super duper amp I mentioned in one post. One
issue with the drain out is x-over spikes. I was getting 100A thick
x-over spikes in the drain output, verses essentially, none in the
source output. The gates tied together prevents the spike when slewing.
So, worth noting if you want the utmost in speed without frying the
devices in a linear amp.



I'm driving a mostly-inductive load, with overall current feedback, so
the current-out stage makes sense for me; it keeps the loop dynamics
from depending on the load, which is valuable. An audio or such amp is
better of it has an inherently low z-out, sor the same sorts of
reasons.

As far as crossover goes, that is interesting. My output stage drive
signal is grounded, and I have a splitter circuit to convert it into
sinking and sourcing currents to get up to the floating +side and
-side power stages, which are essentially current mirrors. The
splitter must sustain a small idle current on both outputs and also
perfectly split the input signal, so that positive inputs increase the
sink current and negative sigs increase the source current. I've been
tuning this circuit for years, gradually converging on something that
works pretty well. It's an interesting, philosophically disturbing,
problem.

For a voltage-out amp, it would be cool to float a, say, +-12 volt
supply on the output node and hang an opamp to close the loop around
each fet, making an ideal follower. Then you'd be able to run the fets
all the way to the power rails, and you could use very small source
sense resistors, with just millivolts of idle voltage.

John