Digital signal processing for biaxial position measurement with a pyroelectric sensor array

This paper proposes a new full digital approach to estimate biaxial position with a pyroelectric sensor array. The previously developed analog interface shows its limits in the calibration procedure, requiring several trimming adjustments. A DSP-based hardware has been developed to experimentally evaluate three digital methods: radial basis function (RBF) neural network, best fitted plane (BFP), and look-up table (LUT) in the least mean square (LMS) error sense. Experimental results show that no dramatic improvements are obtained by the RBF, despite the long training required and the external PC support for weight calculation. The BFP reaches performances comparable to the analog processing system using only nine calibration points, but the best tradeoff has been found with the LUT technique. Actually, with a 64-point calibration set, LUT gives a root mean square error (RMSE) of 0.5% with respect to full scale (FS), offering a valid in-circuit compensation of array structural defects.