A simulation model for improving a non-linear sensor response using linear source separation

This paper presents an empirical simulation model for the frequency response of an eddy-current sensor to a distance measurement disturbed by target temperature variations. In previous papers [3, 4], we have shown that source separation [6, 2] could be a method to obtain a good distance measurement independently from temperature variations. The linearisation of the nonlinear frequency response of the sensor using a model based on a calibration curve seems not to be sufficient to always obtain good and precise enough results. The aim of the simulation model is then to represent the mean behavior of the sensor, in order to test the applicability of linear separation techniques for this problem. Hence, exploitation of experimental results permits to obtain mathematically a linear mixing of distance and temperature. We show that despite a mixing matrix close to singularity, good results can be obtained using linear BSS with simulated signals, and comparison is given between separation of real signals and simulated ones.