That is at the gate of the mosfet. The wave shown in post 10 is what the output looks like without any modulation-thats what I meant by "the opamps wave". As input voltage/current draw increases, the whole waveform compresses and flattens out. When the square wave (which looks great) is added at low level, the waveform goes crazy.
I battled with the small switcher I'm using to run the whole thing, as well as the brushless fan's emi. I added a pi filter between the fan and added filter caps at the opamp board and it helped a lot. However I'm considering using a seperate supply for the opamp as I'm not convinced psrr is not still having an affect.
I'm using star grounding for the whole thing, but a ground problem is certainly possible.

Oh, well that's a problem then. Funny, the first version of this actually used a precision 2.5v reference with the pots for the reference signal.
Since the supply is 12v on the money and wouldn't have any headroom for a 12v reg, could I use a 9v battery or 7809 to power the opamp and provide reference voltage to the pots? The only problem I can see with a battery is if I go back to a big power module, they need a good bit of drive current. More than a 9v could supply for too long.
Everything runs to the star ground except the fan and the power led. The 510ohm resistor literally ties directly to star gnd and pin3.

Ok, linear wall wart it is followed by regulation.

When I get another triple darlington module (only about $20 surplus) for the largest of the three units I'm building, I'll put the common collector stage back in to get the needed current. I'll report back tonight. Thanks!

i'm a little late to the party, but do you have a list of specifications that this load generator needs to meet? are there any load slew rate requirements?

Also, I would suggest - unless the testing requires longer on-times - to cap the duty cycle at 10%. There's still a lot of power to dissipate, but low duty cycles should help to keep that to a minimum. The period will have to be determined by the specs of the supply - expected settling time, etc.

Hi OutToLunch,
I'm actually building three different sizes of these. One to about 500W, one to about 1.8kW, and the third Probably to 5kW. The reason for three is location and portability. These will be used on existing supplies to periodically check general health.

Duty cycle will mostly be short, say 15mins, but could be all day if we're torture testing. The units (except for the largest) are surprisingly compact already.

Ok, did a lot to this last night and today. Built a new psu with good regulation and better isolation from the cooling fan. Fixed reference voltage section, and replaced the 100k pot with a 1meg.
The first pic is 6Hz across the sense resistor without compensation.
The second is with .03uF compensation.
Seems like that value is a good compromise, but how far should I go with capacitance here? I found that with enough, it did exactly what a large value should do- squash and flatten the square wave.

Well thats encouraging. I did have the scope on ac coupling. Should I try viewing it on dc?
I can't load it up to where the other designs have failed until I get parts, but I'll report back when I do. Thanks!

Ok, I'll try raising the frequency.
I've ordered a few of the large power modules for the units because I got them for less than a couple tubes of the mosfets and I don't have to worry about matching. Probably just need to add a big emitter resistor when I parallel a couple of them in the largest unit.
Now that I have a clean supply and better stability at low power (which I didn't have before) I'll be interested to see what it does as well. All three designs failed when I put an old 24v 750W switcher on them. It was ok until I hit an imaginary line and it became unstable VERY fast.

Raising the frequency will likely cause more heat dissipation in the pass element due to increased switching losses. I would recommend reducing the duty cycle to no more than 10%. You had mentioned your duty cycle is 15mins - i mean the duty cycle of the individual load pulses. For example - if the loading period is 1msec, the time that the load should be applied would be 100usec. You can obviously increase the period but keep the duty cycle low to reduce power loss across the pass device. This should still allow you to see how the power supply responds to the load.