OP you will not get high accuracy due to thermistor self-heating. You'll find the mW dissipated in it warms it much more that 0.1°C and readings will be sensitive to airflow across the thermistor. Drove me nuts trying to figure that out error, but switching power on only when taking a reading, and off otherwise fixed that.

True. But - - -

Whatever is doing the power switching now is in series with the thermistor, and adds three more variables to the error budget. If you switch power to both sides of the measurement bridge through a single switch device, you reduce one of those error effects. But there still is an extra voltage drop between the sensor power source and the sensor, caused by a bipolar transistor Vcesat, a FET's Rdson, the output voltage of a uC GPIO pin, etc.; and that voltage drop has both a non-zero value and a temperature coefficient.

ak

 

If you make a switchable constant-current source to drive the thermistor, then you add the error from the constant current source, but you lose the error in the pull-up resistor, and you also lose some of the complexity in the maths, as the voltage on the input is directly proportional to the thermistor resistance.

 

ironically, I'd think "self heating" is accounted by the thermistor itself as that is what it is, the rise in temperature changes its resistance. Hence, I'd think this is normally managed by using rather high resistances and lower voltages, this should alleviate the 'self-heating' problem somewhat.

one problem is RF noise that is received, I've done something similar using a semiconductor temp sensor, and it turns out ADCs can receive radio waves, even without amplification. those can be as high as more than 10 mV sampled on the adc. they could easily account for a degree change for those few millivolts differences. a way out is to low pass filter it and oversampling and averaging (similar to low pass filter but is digital). Keeping the sensor probe short, I'd think also matters.

 

ironically, I'd think "self heating" is accounted by the thermistor itself as that is what it is, the rise in temperature changes its resistance.

Nope.

Correct that the thermistor's resistance is determined directly by its temperature, but ambient air temperature and internal temperature rise caused by I^2R (current-squared times resistance) internal power dissipation are independent. With a constant system voltage or current source, the i2r power dissipation will decrease with increasing thermistor design point (25C) resistance, but so will system noise.

I like thermistors; they are quirky, but very well-behaved. But when you are down in the tenths of a degree, everything matters.

ak

 

"Self-heating" is taken into account once you perform the one-time calibration.

 

If you make a switchable constant-current source to drive the thermistor, then you add the error from the constant current source, but you lose the error in the pull-up resistor, and you also lose some of the complexity in the maths, as the voltage on the input is directly proportional to the thermistor resistance.

Did that in one system. Connected a 10 K, 0.01% resistor in place of the sensor, and adjusted the current source for exactly the desired output voltage with a 6-1/2 digit Keithley meter. Bourns used to (and maybe still does) make trimpot-and-fixed-resistor hybrids with excellent temperature tracking.

ak

 

"Self-heating" is taken into account once you perform the one-time calibration.

Only at one temperature.

My favorite part is 10 K at +25C. At +41.xC its resistance is 5 K. With a constant current exciting source, the self-heating component of the device's internal temperature decreases by 50%. Depending on the accuracy you are trying to achieve, that may or may not be an issue. But still . . .

ak

 

I would perform calibration across the entire usable range and then perform a polynomial fit.

 

Did that in one system. Connected a 10 K, 0.01% resistor in place of the sensor, and adjusted the current source for exactly the desired output voltage with a 6-1/2 digit Keithley meter. Bourns used to (and maybe still does) make trimpot-and-fixed-resistor hybrids with excellent temperature tracking.

ak

And - I forgot to mention - returning to post #35 - Now you DO need an accurate reference!
(Best if the constant current reference and the A/D reference are the same)