MSP430FR5969 Internal Temperature Sensor Accuracy

Mark Ackerson said:

Data sheet indicates 30°C ± 3°C and 85°C ± 3°C using TLV calibration values.

If the datasheet indicates that, than that is all you can get. There is no more informationa vailable than what is in the datasheets. i don't think that many people have the resources to check a large enough number of devices for more precise information.

However, I don't think a single-point calibration will do, as you have offset and gain, both unknown, and with single-point calibration you cannot tell one from the other. You only know their sum at the calibration point.

Mark,

The +/-3C accuracy is a result of a two point calibration using the TLV calibration values. For a single point calibration we do not have any data to indicate the accuracy.

Does the customer intend to use only a single point from the TLV table or make their own measurement?

Ideally if they are making an absolute measurement, I recommend using a line-fit type algorithm to do the two point calibration.

If making a relative measurement (where accuracy to a few degrees is not critical) then a single point may suffice.

Regards,

Priya

Priya Thanigai said:

For a single point calibration we do not have any data to indicate the accuracy.

Well, a single-point calibration is 100% precise for this single point and at the supply voltage the calibration was done with. For any other temperature, the outcome is unknown, as you don't know offset and gain part. For a few degrees, the error might be relatively small, but can go beyond 100% soon.

Priya Thanigai said:

The +/-3C accuracy is a result of a two point calibration using the TLV calibration values.

The +-3C from datasheet is IIRC for the full operating and supply voltage range. Between the two calibration points and at the same supply voltage as the calibration was taken with, the error should be much smaller.
I think the two values stored in the TLV struction are pretty much exact. At the same supply voltage. But then, the datasheet lists the worst case error.

If the supply voltage differs too much, precision is lost. Another 2-point calibration should be done then.

Also, if the temperature range of interest is not between or around the two calibration points, maybe two other points can be used for a new calibration and better results. However, putting the two points too close will make the offset/gain calculation less precise. Especially since the sensor is assumed to be more or less linear.

Finally, it depends on the application, how much precision is required and how much can be provided. If you need the temperature 1000 times a second, doing a complex double-precision floating point formula each time will most likely eat up all your CPU time.