# Very low temperature (-25C) on ds18b20 precision

I'm busy building a multi-probe temperature sensor and log system using a couple of ds18b20 probes on a Raspberry Pi.

They will be monitoring freezer temperatures that should be between -23°C and -18°C.

I'm a little concerned about the precision of the probes though as the datasheets only give the precision between the -10°C to +85°C Range (±0.5°C).

There's an error curve on page 18 here, and another almost identical one on page 20 here. Both of them however start a 0°C and only go up.

As I'll be operating outside of this range, how/where can I find the expected accuracy?

The freezers that the probes will be monitoring are used for storing human tissue, and I'll need this before I can put forward a tender/proposal. Any Ideas?

• The obvious solution would be too use another temp sensor more suitable your temperature range. Dec 22, 2016 at 10:49

The error is guaranteed to be within +/-2 degrees C over the useful range.

That (and only that) is what you can depend on. If that's not accurate enough, use another sensor or adjust the reading with individual per-sensor calibration (not usually a good way to go if you have any option at all because giving up interchangeability means you can't just replace parts in the field).

An advantage of this type of sensor is that the error guarantee of this mixed-signal digital-output part pretty much includes the entire system error. Using analog sensors such as an RTD would require signal conditioning and the evaluation of the total system accuracy will involve a bunch of engineering calculations. The sensor error of a Class A Pt-RTD is about +/-0.2 degrees C at your target temperature.

If +/-2 degrees C is good enough (probably you just need to guarantee the maximum storage temperature) then maybe it's an acceptable sensor, though it makes sense that the energy cost of maintaining a given maximum temperature with error +/-2 degrees C will be ~3% higher than if you can measure the temperature to within +/-0.5 degrees C (assuming Newtonian heat transfer and average Ta of 25 degrees C).

In the first table on page 17 it says: Terr = +/- 2°C for -55°C - +100°C.

So you can expect an error of up to +/- 2 °C.

The error curve on page 18 contains statistical data valid between 0°C and 70°C. It shows the expected spread in accuracy for many samples at a certain temperature. This curve is not very relevant for you.

If that +/- 2°C is not accurate enough for you you can either use a different (more accurate) sensor or you could calibrate your sensors. At a known temperature, read their output and compensate (add or subtract) a value to arrive at the correct temperature. You will need to do this for each sensor on an individual basis for the best accuracy.

• I must admit, I didn't consider calibration before you mentioned it. I will definitely be doing so.
– Jim
Dec 22, 2016 at 14:25