Design of experiments and data-fitting techniques applied to calibration of high-frequency electromagnetic field probes

Characterization of electromagnetic field sensors requires that the behaviour of each sensor is investigated over the whole functional space, i.e., the range in which the device is expected to work. In most cases, for the sake of time reduction, experiments are run only over a subset of the functional space. The reduced extension of the experimental space limits the knowledge of the actual behaviour of the probe represented by the calibration factor K. Linear regression is a common technique to represent measurement data, but it is generally not applied to sensor calibration. Furthermore, no comparative study of experimental plans differing in the number and location of experimental points has ever been carried out nor has a methodology to determine the optimal degree of the regression polynomial been outlined. This paper compares the error variance between experimental points and fitted curve obtained using different experimental plans and regression polynomials.