Low range, high accuracy thermometer for medical-like application

[t4ir1]

Hello everyone!

I am looking for a temperature sensor that can offer me a accuracy of ±0.05°C (preferentially ±0.01°C) over temperatures that you'd measure from a human body, I'd say from 30°C to 45°C. I made some fast math and thought that if I have 0.01°C resolution over 15 degrees than I'd have 15,000 measurement points to which I would need 14bits ADC but I could dial it down to 10bits if I measure only 33°C to 43°C (1000 points of measurement).

I started by looking at the LMT70 revision A from TI which proved a viable candidate but if someone can point me in a better direction I would for sure appreciate it very much.

[arcnet]

Hello everyone!

I am looking for a temperature sensor that can offer me a accuracy of ±0.05°C (preferentially ±0.01°C) over temperatures that you'd measure from a human body, I'd say from 30°C to 45°C. I made some fast math and thought that if I have 0.01°C resolution over 15 degrees than I'd have 15,000 measurement points to which I would need 14bits ADC but I could dial it down to 10bits if I measure only 33°C to 43°C (1000 points of measurement).

I started by looking at the LMT70 revision A from TI which proved a viable candidate but if someone can point me in a better direction I would for sure appreciate it very much.

Surface temperature? Implanted?
Melexis has a medical grade infrared sensor MLX90632 (+-0.2 °C accuracy from 15 °C to 40 °C)
QTI offers (disposable) polyimide tube thermistors (+- 0.1 °C accuracy)
There are (for use in animals) implantable RF transponders that can measure temperature (e.g. LifeChip)
If 10 mK accuracy is really needed you should use RTDs and a good amount of calibration (equipment). Even a class AA PT100 has only an accuracy of +-0.1765 °C (calculated at 45 °C). The older 1/10 DIN B class was +-0.0525 (at 45 °C). This is without any errors of the remaining measurement system (temperature drift/self-heating, aging, non-linearity etc.).
The needed voltage resolution (without internal or external signal conditioning) for a PT100 @ 45 °C  is around 3,85 uV/10 mK @ 1 mA and 0.96 uV/10 mK @ 0.25 mA noise free so the ADCs RMS noise should be below or equal to 0,584 uV @ 1 mA and 0,146 uV @ 0.25 mA excitation current.
Depending on the ADC and the signal conditioning used you can use off-the-shelf 24-Bit ADCs without having to select/correct for integral non-linearity. 0.05 °C accuracy should be doable with most of these ADCs without any INL correction/calibration (usually these have INL errors upto 15 ppm).
In short: Definitely doable but depending on what you actually want to measure not achievable.
 

[SvanGool]

Below or equal to ±0.05°C accuracy is already a big challenge in temperature-land for a reasonable price.

I would take a look at a TE/MS 55036 10K precision glass NTC thermistor with a tolerance of ±0.1°C and maybe calibrate it down to ±0.05°C (0°C to +50°C) with a Littelfuse PR103J2 10K epoxy NTC thermistor (if you still can find one).
Why?: as soon as you take the PR103J2 out of its Moisture Barrier Bag, humidity will start influencing its tolerance because of its epoxy housing vs the glass housing of the 55036.
You would also need a very accurate meter for this with an accurate voltage reference

Another reasonable and much easier alternative is the Smartec SMT172 ±0.1°C and can be easily measured by measuring the PWM signal by an MCU.

[Anders Petersson]

Since you mention ADC, I guess you're planning to build the electronics yourself. Then you need to deal with any noise sources and the temperature dependencies of all analog components.
Hence, the resolution of the ADC isn't the biggest problem and won't be a big design issue unless you need to optimize the price for high-volume manufacturing. Just go with a 16-bit or 24-bit ADC and enjoy the noise rejection offered by the extra bits.

Your analog frontend will be the ADC, a stable thermistor and a few passive components. A thermistor gives better resolution for your limited temperature range, and their stability over time can be similar to PRTs, I hear. Achieving accuracy boils down to avoiding different types of drift in this frontend, combined with calibration points spread over the measurement range.

As I've been going through this process myself for a while, I feel confident to recommend that if you only need a few devices, a calibration from a professional lab is worth the cost to save you the headache of doing reliable calibration without prior experience or reference equipment. The calibration will be a list of corresponding temperatures and ADC values.

The design choices depend on factors you didn't mention such as max cost per device, response time, size/weight/power limitations etc. Feel free to post more details or contact me offline.

[MK]

An often forgotten source of error with Pt resitance probes is the dislocation slip when dropped or shocked that causes a permament small resistance change with the wire types, and thin films are prone to something similar too due to slip at grain touching points