Noise in LM35 temperature circuit

I'm afraid that in this case, you need to take Sturgeon's Law into account - 90% of everything is rubbish. This is particularly important on the usernet, where simple majorities often contain a disproportionate number of the simple-minded.

The idea of treating every transducer output as differential is extravagant, expensive, time-consuming and complicated, and if you can get away without it, you save big-time.

Mostly you can't get away without it - and knowing when you can takes much more expertise than knowing how to set up a differential system in the first place.

It's your neck - and you'll have to do the rewiring if it doesn't work out.

That should give you the best chance of getting the scheme to work. Don't forget to twist the ground and signal wires (as well as the positive power supply wire) running between each LM35 and the multiplexer.

Try reading page 7 of the LM35 data sheet. The LM35 can oscillate if you hang e more that 50pF of capacitance between the output pin and ground - that's about a metre of twisted pair. A 2k resistor between the output pin and the cable is one way of preventing this oscillation

- the data sheet also suggests the alternative of loading the output with 75R in series with 1uF to ground.

The best single-point calibration for you is with an ice-bath, with the transducers (LM35s) totally immersed in wet, crushed, ice. See page 13 of

This gets you within 1mK of 0.0C - if you use de-ionised water and ice made from de-ionised water - with a minimal amount of effort. You may want the negative supply during calibration ...

Sounds good.

The other problems with IC sensors is that their thermal time constants are doodoo, and that unless you do something special, their temperature is basically set by the leads, rather than the ambient. I like the interchangeable YSI glass-bead parts too, though I usually use them for jobs like millikelvin temperature stabilization for diode lasers, where I don't care too much about their nonlinearity--for that job, you just have the SW search over a few degrees for a sweet spot and sit there. If you mount the thermistors properly, you can get bandwidths of about 1 Hz or slightly better, versus 0.1 Hz at most for an IC sensor--and often closer to 0.01 Hz. Of course, their advantage is much less if you just drill a hole and epoxy them into it.

An even better method for some things is to forward-bias the monitor photodiode of the diode laser and look at its forward voltage drop. Most lasers nowadays have huge monitor PD currents, which is inconvenient--you can turn the laser off briefly to do this, or compensate for the effect of photocurrent. It's worth it, though--you can get bandwidths of 5 Hz or faster, because everything is brazed together with metal in a very small package. Because thermal conduction distance goes like the square root of time, thermal things tend to speed up quadratically as you decrease the separation of the sensor element from the object.

Insulation and thermal grounding help, but when there's power being dissipated on the cold plate, there's nothing like bandwidth to improve temperature control.

Cheers,

Phil Hobbs

If you put a second thermistor on the heat sink of the Peltier junction, you can - in theory - do feedforward correction to improve the rejection of changes in ambient temperature in my paper in Meas. Sc. and Technol. volume 7 (1996) pages 1653-1664). E-mail me - my address is real - if you want more details or a reprint.

If you would just get the datasheet from the vendor (National Semi) you would see how to solve most of the difficulties.