Hi
This is a good question and requires some engineerng judgement based on the particular data set.
The answer depends on the range over which you want the calibration to be valid. For each measurement, graph the values of Tdie on one axis and Tobj on the other axis. This should give you a good idea over what range the calibration will be valid (remember - calibration is valid for interpolation only!). It is OK if all the Tdie values are close, it just means that your calibration will be valid only for a particular value of Tdie.
Based on how the data points cluster, you can decide how to generate the calibration data set. If the data forms tight clusters, then use the data in a cluster averaged together to estimate a Vsensor value for a particular Tdie/Tobj combination. Depending on the number of data points in a cluster, this will incease the confidence in the correct value of Vsensor being assigned to a given Tdie/Tobj combination.
If there is not much clustering then use each data point as an entry in the calibration data set. The confidence that the correct value of Vsensor will be associated with a particular Tdie/Tobj combination is lower due to noise inherent in the measurement, but the larger number of points may tend to 'smooth' out the fit.
A helpful guide is to hold the Tdie and Tobj as fixed as you can in your system, then take approx 20 samples of Vsensor as shown in section 2.3, fig 4 of the TMP007 Calibration Guide. This will allow you to estimate the noise and also see how the mean trends over time. If the mean is flat, then variation in Tdie probably not an issue. If it is a relatively linear trend in mean then try a first order linear correction as shown in section 6.2 and use the corrected data as the calibration data set.
I have had cases where depending on the particular combination of Tdie/Tobj some regions were highly clusterd and could use averaging, some required linear correction, some had low clustering and used individual measurements and some were chaotic and ecluded from beng a valid region of calibration (often a sign of convection or condensation on the device).
Hope this is helpful,
Regards,
Werner Metz
IR Thermal Systems Engr