Hi Lou,
My version of MATLAB is older and does not support lsqcurvfit and nlinfit functions so can't comment on those. Maybe someone out there can advise.
Simplest strategy is brute force - set a while or for loop for each co-eff with an upper and lower limit and step size (or number of steps). Nest the loops to cover all possibilites in the search space. Then calculate the predicted value versus the measured value of Vsensor for each Tdie/Tobj combination. Store the error metric for each combination of co-efficients in an array and at the end, search for the 'best' value of your error criterion and the co-eff values asociated with it.
Common criteria for best fit are either the minimum max error or the average rms error across a data set. Once an initial candidate set of co-effificents is found, a subsequent search with smaller bounds and finer steps can determine if there is opportunity for significant improvement. It is often helpful to use MATLABs graphing capabilties to determine where in the Tdie/Tobj space the errors are occurring and the magnitude of the errors.
The MATLAB function fminsearch (Please visit the site to view this file) may also be useful. It allows you to specify a function and a data set, then adjust the co-efficients of the function to map the data set. The function should be the one used by the TMP007 as specified in the calibration guide.
Finally, in mid-October we should release EasyCal which will have co-eff generation routines built in.
Regards,
Werner