> Ah, but that's only part of the problem. The positive side > of the hysterisis > also changes with input voltage. Split R6 into two. Take a diode from it's midpoint (anode) to the reference diode (D11/D12). This clamps the midpoint at 1 diode drop above reference and "stabilises" the on drive through R6_left. Two diodes or lower R6_left = more drive. BUT somewhere about here you say enough is enough and Replace Q2 with two transistors in a long tailed pair. The reference feeds Q2_left_base and the sense signal drives Q2_right_base with any hysteresis added to (probably) Q2_right. This was done by ?Natsemi? long ago. Different than either 'my' GSR or your/Roman's approach. Proper comparator should be much more regulated. *BUT* unfortunately, the thing that makes this circuit almost certainly unsuited to me need is that, as didn't occur to me in my initial enthusiasm, the "regulation" at present occurs on the input and not the output side. ie the ocillation occurs when Iin reaches a limit set by the voltage across R2. At that stage the inductor energy is dumped into the LEDs (and a series diode with a capacitor on the KLED side of it would make the LEDs' life more gentle) and the cycle then repeats. You COULD add LED current sensing to make the process more defined but for now it's a bit open loop. I need something with reasonably well defined LED drive current over the battery operating range and suspect that this would be hard to tame. Maybe not impossible though. My CD40106 circuit uses a $US0.08 (25000 volume IC) and 3 transistors (2 are FET gate drivers, one is for LED current sensing) so is certainly dearer. Efficiency is excellent. > Coincidentally, I'm looking at trying to design a > "constant power" circuit > to discharge a capacitor (33mF, 75V) quickly . Somewhat > similar requirement. Why "constant power"? Would constant current do? Russell -- http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist