What's wrong with this?

Or more so. The radiation curve has to be pretty high to get decent light out. If you aren't heating the temperature high enough all you get is heat. Lower the watts in and the temperature drops considerably and you get more heat relative to the light.

Rick C.

That's what I meant. It's inevitable since it's so inefficient and designed for the least inefficiency with a full wave.

t out. If you aren't heating the temperature high enough all you get is he at. Lower the watts in and the temperature drops considerably and you get more heat relative to the light.

Radiated power rises as the fourth power of temperature, so less power mean s a lower temperature, and even less radiation at wavelengths we can see.

Most of the power we put into a filament lamp comes out as infra-red radiat ion anyway, and running it cooler makes this worse.

Bill, you and Lasse were right, the opamp was oscillating. However, in a very Heisenberg-like fashion, it was hard to catch it red-handed. The frequency was 2.53MHz. Adding 68nF between VREF and GND silenced it completely, reducing the total offset to a magnificent value of 200uV (end to end, including the ampification stage). Now it is 8 times better than what the datasheet says and super-linear and symmetric in the desired range of load currents. The LTC1407 ADC attached after this stage shines, too.

Thank you for your help, case closed, I owe you all a beer!

A side effect is that I know of a cheap proven Chinese distributor of genuine LTC parts, if somebody is interested. The name of the shop is POCCIO on AliExpress.

Best regards, Piotr

What would that be?

Best regards, Piotr

Pimpom suggested it first,and I was more worried about the comparators oscillating which was a bit silly when they weren't there.

Offsets do have a habit of reflecting oscillations somewhere in the circuit, and scope probes often add enough capacitance to kill the oscillation.

The datasheets mostly list worst cases ...

That's worth knowing. I've been a fan of Linear Technology for a very long time, when the customer needs the performance enough to justify the price.

Probably something designed by John Larkin, at vast expense.

John suffers badly from "not invented here".

Not pimpom, same initial letter, but it was I, piglet :)

piglet

You wouldn't have had to go to all this trouble if you incorporated broken loop gain/phase analysis into your SPICE. Anything less is a wish. Your cir cuit may still be marginal and unstable with environmental and component va lue shifts.

I'm mortified. I actually checked before I posted, but clearly not carefully enough.