A. Gutiérrez and S. Marco (Eds.): Biologically Inspired Signal Processing, SCI 188, pp. 137–167. springerlink.com © Springer-Verlag Berlin Heidelberg 2009 9 Multivariate Calibration Model for a Voltammetric Electronic Tongue Based on a Multiple Output Wavelet Neural Network R. Cartas 1 , L. Moreno-Barón 1 , A. Merkoçi 1 , S. Alegret 1 , M. del Valle 1 , J.M. Gutiérrez 2 , L. Leija 2 , P.R. Hernandez 2 , and R. Muñoz 2 1 Sensors & Biosensors Group, Department of Chemistry, Autonomous University of Barcelona, Bellaterra, Catalonia, Spain Raul.Cartas@campus.uab.es 2 Bioelectronics Section, Department of Electrical Engineering, CINVESTAV, Mexico City, Mexico Abstract. Electronic tongues are bioinspired sensing schemes that employ an array of sensors for analysis, recognition or identification in liquid media. An especially complex case happens when the sensors used are of the voltammetric type, as each sensor in the array yields a 1- dimensional data vector. This work presents the use of a Wavelet Neural Network (WNN) with multiple outputs to model multianalyte quantification from an overlapped voltammetric signal. WNN is implemented with a feedforward multilayer perceptron architecture, whose activation functions in its hidden layer neurons are wavelet functions, in our case, the first derivative of a Gaussian function. The neural network is trained using a backpropagation algorithm, adjusting the connection weights along with the network parameters. The principle is applied to the si- multaneous quantification of the oxidizable aminoacids tryptophan, cysteine and tyrosine, from its differential-pulse voltammetric signal. WNN generalization ability was validated with train- ing processes of k-fold cross validation with random selection of the testing set. 9.1 Introduction An electronic tongue is a chemical analysis system that employs sensors in a novel way, in order to accomplish quantification, classification or identification in liquid media. Conceptually, it relies on the use of a chemical sensor array, with some cross- sensitivity features plus a chemometric processing tool, needed to decode the gener- ated multivariate information. This scheme corresponds to how olfaction and taste senses are organized in animals, allowing for the identification of thousands of differ- ent compounds with a reduced number of differentiated receptors, so it is clearly bio- inspired. Main types of sensors used in electronic tongues are potentiometric and voltammetric, which yield very different responses. When the nature of the sensors used is voltammetric, a 1-dimensional data vector is generated for each electrode, making extremely complex the chemometric processing of the generated signals. A powerful bioinspired processing tool used with electronic tongues is Artificial Neural Networks (ANNs), although more suited to simpler input information. The use of an ANNs with these signals might then imply some kind of preprocessing stage for data