A post non-linear source separation algorithm for bounded magnitude sources and its application to ISFETs
Rights accessRestricted access - publisher's policy
The response of ion-sensitive field-effect transistors (ISFETs) can be seriously affected in mixed-ion solutions by different interfering ions. As has been demonstrated, this problem can be addressed using nonlinear semi-blind source separation (BSS) algorithms based on post non-linear mixtures in which nonlinear transforms must be computed using supervised samples, i.e. calibration points for known concentrations of the main ion. In order to eliminate the cost of collecting such samples, this paper introduces a novel non-linear BSS algorithm that employs linearizing transforms computed only with unsupervised information. The scale indeterminacy of this transform is removed using a prior on the sources based on magnitude bounding and, besides, gaussianization is generalized by using a kernel estimator. Experiments with real ISFET measurements demonstrate that this BSS algorithm achieves a level of accuracy similar to that of the semi-blind counterpart based on independent component analysis and outperforms a post-nonlinear BSS algorithm which minimizes the mutual information.
CitationBermejo, S. A post non-linear source separation algorithm for bounded magnitude sources and its application to ISFETs. "Neurocomputing", 19 Gener 2015, vol. 148, p. 477-486.
|A post-non-linear source separation algorithm.pdf||Article definitiu||1,227Mb||Restricted access|