This is a bad idea, I know

Here's what I know about amplifying really tiny DC voltages:

  • it's impractical to build your own for production
  • there's going to be noise
  • the signal is relative to the cold junction temperature
  • it would be easier with an ASIC than using basic components
  • most op amps aren't rail-to-rail (can't go to 0)

These are all acceptable to me. I want to learn how to do it anyway.


I've just started self-teaching and I thought that measuring the actual temperature in my toaster oven with what I have laying around (batteries, resistors, thermocouples, resistors, and op amps) would be a fairly simple and easy first project to design myself. Maybe I'm wrong, but I believe it's possible and I want to learn how.

Where I've been

I was able to measure between 0.2mV and 0.4mV while holding the K Type (41µV/˚C) thermocouple with my finger (37˚C supposedly) using my multimeter. That checks out: ~0.5mV = 41µV * (37 - 23).

Well, I can't read that in on 0-5v analog input...

At first I thought that perhaps I could use series of transistors, but since there's no current (and the voltage is well below the minimum necessary), that was a bust.

I remembered reading and hearing about op amps, so I thought perhaps that might be the right tool for the job. I watched Dave Jone's excellent EEVblog #600 - OpAmps Tutorial - What is an Operational Amplifier? and that I'd give it a try.

The local university has a plentiful supply LF347-N in a vending machine on the 4th floor of their college of EE, so I grabbed a few of those.

I was able to get a 10 µF capacitor to hold < 1V of charge (tapping to the 9v with a resistor and then draining with a resistor while measuring) and then created the suggested non-inverted feedback loop with each of a 5k1 and 2k resistor on a breadboard with a 9v battery and see the expected 3.55v (1 + RF/R1). It seemed that the op amp was actually charging the cap, despite the no-in/no-out "guideline". Or it was just increasing from residual voltage as it settled.

(only using one for now with the basic "non-inverting feedback loop" for now, but considering trying the "instrumentational amplifier" or "two op amp inamp" next time)

Cool, I think I'm on the right path now.

I swapped out the 3.5x feedback (5k1, 2k) with a 101x feedback (100k, 1k) so that I could test with the thermocouple. I got an initial ~1.2V reading (maybe because I was below the rail limit?), but then it didn't change when I held my finger on it.

... and I could smell the Magic Smoke! Hot resistors. Hot op amp. I don't think it works anymore. I still get some readings, but they seem off and I don't trust it.

Tried a second time, similar result (gotta go back to that vending machine). At various points as I was experimenting I even tried touching my hot soldering iron to the tip, but never saw a response different from holding it with my finger through the op amp output.

What I'd like help with

  • what might I have done to burn up the LF347-N?
    • (it's rated +/- 25v input)
  • how do I bias the thermocouple's 0v starting point?
    • what happens if I don't? (it's not like the first 10˚C matter anyway)
  • is battery "ground" compatible with the thermocouple "ground"?
    • (they're both relative to themselves rather than ground ground)

I'm not ready to give up and use more specialized components like the MAX31855, LT1025 and LTC1049. I want to bang my head against this with "basic" components a little longer.

What else does a noob like me need to know to get on the right path? How might you go about this? What might I be missing?

P.S. Some related posts

  • \$\begingroup\$ schematic? include all the grounds \$\endgroup\$ Commented Mar 15, 2019 at 2:44
  • 1
    \$\begingroup\$ You can build precise DC amplifiers, if you chop the DC input, perhaps at 10,000 cycles per second (use FET analog multiplexors), then amplify that chopped input ---- a lot. Then you use a synchronous detector (the Gilbert Cell Mixer, which operates like an analog exor gate) to recover the DC, highly amplified. These are called "chopper stabilized opamps" today. \$\endgroup\$ Commented Mar 15, 2019 at 2:50
  • 1
    \$\begingroup\$ Yup. We need a full schematic -- including what you did with the other three op-amps in the package. If you're going to use that op-amp, you should probably just start with two 9V batteries for \$\pm\$9V supplies. This can be done, BTW, even with that amp -- it just takes some doing. \$\endgroup\$
    – TimWescott
    Commented Mar 15, 2019 at 2:51
  • \$\begingroup\$ You can use more FET analog multiplexers for the synchronous detector, for that matter. I've done it. Works a treat. \$\endgroup\$
    – TimWescott
    Commented Mar 15, 2019 at 2:53
  • \$\begingroup\$ Or you can trim out the op-amp's offset. Or you can use a plain old switch to manually ground the input, read the voltage, then switch in the thermocouple, read that voltage, then do some math. \$\endgroup\$
    – TimWescott
    Commented Mar 15, 2019 at 2:55

1 Answer 1

  1. No idea, maybe you applied reverse voltage to the supply terminals.

  2. You can use 0V if you give the op-amp +/- supplies. The one you are using requires at least -4V on the negative supply to work near 0V on the inputs. See the data sheet where it talks about "common mode voltage range". The circuit you show has no negative supply so you are not going to get useful results. It still should not damage the op-amp.

  3. The op-amp you are using may be useful for some things, but it's about useless as a thermocouple amplifier. Note the large offset voltage (+/-5mV typical, up to +/-13mV). That could be trimmed out, but the offset voltage change with temperature is a typical 18uV/K (no maximum is given), which means you will have big compensation errors. Chopper (zero drift) op-amps are much better, you can get relatively inexpensive op-amps with drifts of 50nV/K which is several hundred times better. The offset voltage would also be a fraction of a °C equivalent.

  4. Thermocouples work well when the junction is grounded, but it's not necessary, if you isolate (galvanically) the junction you can ground the negative terminal of the thermocouple. Of course you don't want to thermally isolate the junction or it won't follow whatever it is you are measuring. Note that heat travels through the wires so there is work being done to maintain a temperature differential- this is the thermodynamic basis of the thermocouple- you are not getting voltage (and potentially current) from nothing.

With a better op-amp and dual supplies you can get useful results, though you'll have to supply cold junction compensation in some manner (for example by measuring the temperature where the thermocouple metals are attached to copper) and compensating digitally, if you want decent accuracy.

  • \$\begingroup\$ Thanks a lot. Are there any op amps that you could suggest that I should look at? Thanks for translating some of the terminology in the datasheet that I (obviously) didn't understand. :) \$\endgroup\$
    – coolaj86
    Commented Mar 15, 2019 at 16:50
  • \$\begingroup\$ Also, I can buy a meter + thermocouple for $4 from China on ebay, so they can't possibly be using these zero-drift TI op amps that cost $8 each (even if they're steeply discounted in qualtity). Have any ideas on how they do it? \$\endgroup\$
    – coolaj86
    Commented Mar 15, 2019 at 16:57
  • \$\begingroup\$ @CoolAJ86 Probably badly. \$\endgroup\$ Commented Mar 15, 2019 at 17:19
  • \$\begingroup\$ Hahaha. Probably. But what I mean is that there's probably some way of using some capacitors and two or three crappy op amps to be able to accomplish the same task that's "precise enough". \$\endgroup\$
    – coolaj86
    Commented Mar 15, 2019 at 21:56
  • \$\begingroup\$ There are cheaper op-amps that are sort-of okay if you're not fussy. The ancient OP-07 is perhaps 50 fen in modest quantity, and even a penny LM358 is several times better than the one you have. You can also insert an analog switch and periodically re-zero it by periodically disconnecting and shorting the input, and make the correction digitally. That could make the inputs more fragile if done badly. \$\endgroup\$ Commented Mar 15, 2019 at 23:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.