# Thermistor in Wheatstone configuration and ADC offset

Based on my rough analysis, I realized that by measuring a thermistor in a Wheatstone configuration, I could technically reduce the ADC offset error LSBs because we are substracting two divider voltages.

This subtraction should eliminate the ADC offset error term and should allow me to use a much lower-cost ADC. However, I don't see anyone discussing it nor any Google search result talking about it.

I am wondering if my analysis is correct or if I have misunderstood it.

--Update--

Let me add my analysis here. Take note that I have simplified a lot in order to fit them into a single page. For my case, the supply of the bridge are connected to VREF of the ADC and, R1=10k,R3 and R4 are 20k.

I also made assumption that the measured ADC value are summation of error free ADC result plus offset error (Eo in the picture). Maybe this is the part I was wrong?

• if the bridge is balanced, the ADC will still read its own offset voltage. Could you clarify? Jan 9, 2023 at 4:34
• What kind of thermistor? Jan 9, 2023 at 5:03
• @tobalt I have added more details of my analysis. Could you take a look? Jan 9, 2023 at 7:00
• @TimWilliams its a NTC thermistor Jan 9, 2023 at 7:01
• see edit in my answer Jan 9, 2023 at 7:06

You are not eliminating ADC offset error by using differential inputs. In fact you may be doubling it if the offsets of the two inputs happen to have different signs. Noise (including quantization noise) too.

You may be able to use a lower ADC reference (or higher PGA gain) to get more resolution and perhaps more accuracy.

Also keep in mind that many ADCs do not have differential input. Also, if the error in your resistors is worse than the linearity error of the ADC you may be increasing the uncalibrated error. Unless your ADC is particularly crude, resistors of that accuracy and stability (eg. for a 16-bit ADC) will not be inexpensive.

Thermistors in most cases are very high output devices relative to their accuracy, so tricks are not necessary or desirable unless you're looking for very small changes. Nonlinearity is often more of a concern- when the thermistor resistance is much higher or lower than R1 (eg. over a wide temperature measurement range) the LSBs per degree C is greatly reduced, which also accentuates ADC offset errors near the top or bottom of the range.

As to few discussing it, Charle Wheatstone invented (or popularized) the bridge in the mid-1800s and most of the chatter has since died down. It's just another tool in the toolbox, to be used when appropriate. You're more likely to see this arrangement in conjunction with a (relatively) low-output sensor such as a platinum RTD.

A wheatstone bridge makes the differential output voltage signal much smaller (ideally 0) than reading directly from a thermistor. Therefore (using suitable voltage gain after the bridge), you can use the full ADC range for your sensor range of interest and use an ADC with fewer bits for the same target sensor resolution.

But the ADC offset is not affected by that approach. It will be still added to the result. As you will use higher analog gain before the ADC, the offset will result in a smaller temperature error, though.

The error in your analysis is, that you assume the same offset voltage Eo for both ADC inputs. This indeed will cancel in substraction, but is an unrealistic scenario.