capacitive loading

Do you have room to connect a .1 uF ceramic in series with a 10 ohm resistor? I think it would be more predictable than the tantalum.

Yes, if that's the best fix. We tried the network on the National datasheet, 1 uF + 75 ohms, and a couple of units still oscillated.

I've got to stop using these things. The SPI version, LM71, seems to work OK, but it's a pain to bit-bang, and I had an ADC mux input available...

John

I have had good results with such devices by loading them with a series RC that has a time constant near the period of peak impedance of the device with no load over 2*pi. So if the peak impedance frequency is about 200kHz, the RC time constant would come out to about 0.8 us.

And just exactly what output waveform corresponds to "go nuts," and what error defines "settling time?" And does the "oscillation" with the 1K occur without the mux stimulus or on its own? Do ya' thunk there may be a conditional stability type of deal going on here, do you even have a bypass on the package? Something is wrong if the 75+1u with VCC bypass causes oscillation or msec duration output stage transient response because multiple LM45's driving mux inputs has to be the most common application for the part.

A big positive jump with a slow, ringy exponential recovery.

How about a 5 degree C tempearture error?

Don't know; testing folks reported it on some units.

absolutely

Not far away. This ain't RF.

What's wrong is that this is a crappy part. It has lots of other problems, all related to the emitter-follower thing. Bob Pease promised me he'd fix it, about a decade ago.

A 47 uF tantalum to ground looks promising. Zout looks just like 20 ohms || 47uF, and forced step response is nice and RC-type soft, no peaking or ringing. Noise goes way down, too.

John

If you know it's a crappy part then how many times do you have to get burned before you realize that you need to thoroughly test it for each and every application circuit with which you do not have prior experience? Answer me this.

I *told* you I screwed up on this one. We make a temperature controller (2U rackmount) and a separate gradient driver (same size) for NMR apps, and the customer demanded that we pack both into a 2U box, with a new universal power supply, a new digital interface, an absurd budget, and six weeks from start to the final working unit and start of CE testing. I had to redesign most everything (no Keven-ing) and got the custom magnetics, the firmware, the thermocouple stuff, the heatsinking, the PLDs all right. I messed up the LM45 thing and missed one pullup on the CPU. Wanna see a pic?

I guess I should cut over to the LM71, the SPI version, but it's more expensive and a pain to bit-bang when timing is tight.

John

That sounds like a positive charge injection sufficiently large to turn the output emitter follower Off, a slow recovery as the miserably large emitter resistors try to do a pull-down, mixed in with oscillations as the emitter follower comes back On again.

Try a heavier pull down, (perhaps as large as 1mA and maybe even to a -ve rail), and/or an RC low pass filter, 1k in series and 0.1uF to 1uF across the muxer input.

Yup. Probably the previous mux channel was acquiring a biggish positive voltage. I *did* resequence the mux scan from an earlier version of the code.

Pulling an LM45 negative at any substantial current *will* make them go into SCR latchup, plus the self-heating error. And some oscillate with a 1K to ground. The real fix would be a series R-C, but pcb cuts would be inconvenient now... we'd have to disassemble all the units to get at the bottom of the boards.

There is a general issue of hanging big caps on device outputs, but that hasn't evoked any interest, I guess.

John