Specifically, when using K-type thermocouples, what sort of accuracy expectations demand cold-junction compensation as opposed to simple calibration?

Clearly there's no substitute for direct experimentation on the circuit in question, but short of that are there any general rules of thumb?

I'm building a thermal immersion circulator, I want to hold the liquid temperature accurate to 1°C, so the thermocouple readings will need to be at least this accurate.


4 Answers 4


... what sort of accuracy expectations demand cold-junction compensation as opposed to simple calibration?

An ideal thermocouple provides \$\mathrm{V_{out} = k \cdot (T_{hot}-T_{cold})}\$.

The end result cannot be more accurate than the accuracy to which you measure or infer the cold junction temperature, and in practice will be somewhat worse and can of course be much worse.

Therefore, ALL thermocouple temperature measurements require cold junction "compensation".
"Compensation" is really shorthand for "determining the temperature of the cold junction with a level of accuracy that is appropriate for the application". Compensation could consist of saying "this will be used in an air conditioned office which is temperature regulated to 20 degrees C - so that will be the cold junction temperature".


Thermal compensation is needed when you want to measure absolute temperature.

When you are reading a voltage produced by thermocouple, it's proportional to temperature difference between environment where the junction is and ambient temperature. If you want to know the exact temperature of environment you are measuring, you have to add ambient temperature to temperature readings from thermocouple.

This can be done in several ways - analog or digital. There are analog cold junction compensation chips like LT1025, which produce a voltage that is equal to voltage the thermocouple would produce if it's other end would be in ice water at 0C and the junction at ambient. Then you connect thermocouple in series with the output of this chip and amplify it's output.

If you are using a microcontroller, you can get ambient temperature from a digital or analog sensor directly to microcontroller and add it to thermocouple readings in software. This way you can use an opamp with low bias current and low offset voltage to amplify thermocouple output.

There are special chips from Linear Technology and Analog Devices for K-type thermocouples, that amplify thermocouple voltage to some usable level AND provide cold junction compensation out of the box:



They both provide free samples.


A Fluke 2635A data bucket claims 0.1°C accuracy with type-K thermocouples.

An Agilent 34901A data acquisition card (used in the 34907A mainframe) also claims 0.1°C with type-K.

Both technologies use cold-junction compensation to achieve this accuracy. (Fluke has their sensor in the instrument; Agilent puts the compensation on the card).

My experience with these data loggers is that there's usually only 0.1 or 0.2°C variation from thermocouple to thermocouple if they're all approximately the same length and welded/soldered in the same manner. (I'm used to connecting anywhere from 20 to 60 TCs to a a unit under test for thermal characterization.)

I suppose the need for compensation would depend on how well the ambient temperature at the thermocouple terminations is controlled. If it can vary by 10-20°C, you probably need to implement the compensation.


You have a cold junction, whether you want it or not. It will be at a certain temperture when you did the calibration, and it will be at a certain (probably different) temparature when you do your measurement.

The voltage you measure will be proportionally to the temperature difference between your cold junction and the thermocouple. When you don't compensate for the cold junction temperature, the uncertainty (=error) in your calculated temperature will be as large as the uncertainty in your cold junction temperature (difference between the measurement situation versus the calibration situation). If that amount is (far) less than your desired accuracy you can do without the compensation. This will generally be the case if

  • you measure relatively high temperatures
  • with relatively low accuracy
  • the cold junction temperature is relatively constant.

I think your application does not fall within this category: you want 1C accuracy. The best you can hope for the cold-junction temperature stability is 'room temperature', which will vary (much) more than 1C.


Your Answer

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

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