LM34 to A/D

Why not use an LM71, digital serial SPI temp sensor? By the time you buy an LM34, and an opamp, and a couple of precision resistors, the LM71 starts to look like a bargain, and a number of analog gotchas disappear.

John

I have only 1 SPI bus, non-selectable. (I'm out of GPIO; I'm already muxing and doing double duty....) I need at least 2 sensors, preferrably 3. I have 5 A/D channels.... So it's more of a do with what I have rather than do the best thing...

Pure luxury, at least in my field. We have to make do with diodes and thermistors. If we were extra good, we get to enjoy a reference chip alongside. As long as it doesn't cost more than 10c, that is.

OK, do this:

Gain is 1 + (15K/2K) = 8.5, and the resistors need to be 1% or better to keep up the accuracy.

Don't run the LM34 from more than +5... they misbehave if you do.

Any decent "single supply" or "rail/rail input" opamp should work.

The output RC keeps ADC mux spikes from freaking out the opamp, and reduces adc noise as well.

It's prudent to oversample and average to reduce residual noise.

John

Hell, I buy gain-of-four amplifiers for $190 a pop, and single flip-flops for $32.

John

$190? That's 30 six-packs of Porter. Toirrty of 'em!

(when they are on sale)

Sometimes it's amazing how little is inside "lab grade" amps. A few BJTs, some 0603 resistors, a milled case. Then it's all marked up by a few thousand percent. Just like us their mfgs only cook with water, they just use bigger pots.

Or by 30 if it contains tubes, says "Super High Fidelity" on there and comes with a guitar input.

Don't forget to multiply by 10:1 or 20:1 for the rad-hard versions..

Best regards, Spehro Pefhany

I wonder what's the most expensive IC you can buy. I've heard that some high-end FPGA's, not even rad-hard, run close to $10K.

John

So, does that op-amp have a charge pump built in? ;-)

It's going to be a trick getting 10V out of that circuit, and you have no offset, wasting 40% of the range.

Oh, yeah, run the opamp from whatever higher supply you have, if you need to swing higher.

John

Yup, the opamp needs a higher supply if the adc needs more swing. But since the LM34 is only good to +-1 deg F, there's not a lot of advantage to a gain above a few, and, actually, not any real advantage of gain at all. The ADC lsb is 2 mV, and the 1 deg F error corresponds to 5 LSBs.

Actually, unless you want to spring for 0.1% resistors, the gain stage will probably make accuracy worse.

If the LM34 is connected directly to the ADC, inserting an R-C lowpass is still a good idea. ADC mux's commonly kick out spikes, and the LM34 series has a weak pulldown and lots of quirks.

John

The $190 amp is a tiny IC, not a box.

We use four of them in one of our products.

John

Well, there you go then. LM34 is only good to 1 degree, does 10mV per degree, your A/D is 10/2^12 = 2.4mV.... plug it in and rock your shiftyware..

Seriously, accuracy and resolution and other stuff are relative things. If you only want the answer within +/-1.24 degrees (barp) or less then don't be bovverred.

If you want more then you will be calibrating. Shiver.

DNA

The most expensive FPGA Digikey lists is $4,177.00 (6 million gates and 684 I/O). OTOH, the Fairchild CCD595 is reputed to cost around

I guess the most expensive IC is one that has a lot of masks and doesn't work properly.

Best regards, Spehro Pefhany

The gain makes the difference between induced noise on the board amounting to several degrees or a fraction of a degree. The input range of interest is 50-120 oF or 0.7V so a gain to full range of the ADC is approximately x14. This means the ADC will see 140mV/oF versus 10mV/oF, making reasonable amounts of onboard noise equivalent to a fraction of a degree versus several whole degrees.

John, just for everyone's benefit, what have you seen the LM34 do at higher voltages?

(BTW, all that care that must be taken regarding this device's sensitivity to capacitive loading and the fact that you need to included a negative supply or other roundabout methods for reading temperatures below zero has made me try the LM50 if I need an analog type sensor.)

Gary Peek, Industrologic, Inc.

But you could oversample and average. Temperature changes slowly, and code is cheap.

John