Hardware implementation of the wavelet transform coupled with Artificial Neural Network for quantification purposes

This work presents a hardware implementation of the Discrete Wavelet Transform (DWT) coupled with an Artificial Neural Network (ANN), allowing quantification of three chemical species in liquid environments from voltammetric signals. The chemical species included in this work are Tryptophan (Tpr), Cysteine (Cys) and Tyrosine (Tyr). Due to the voltammetric signals complexity obtained from the blend, the DWT is used as analytical multiresolution tool for voltammograms. The compressed signals are processed by an ANN, which delivers the concentration of threes chemical species. The ANN, previously trained, is a multi-layer perceptron with a 16×60×3 neurons structure, and with a non-linear activation function and a linear activation function for the hidden and the output layer, respectively. Both tools (DWT-ANN) were implemented in the microcontroller dsPIC30F6010 from Microchip®. The developed prototype is part of an analytical system known as voltammetric electronic tongue. The results obtained by the system were similar to MATLAB® results with the same signals at the input. The performance of the system proposed in this work showed that it is available replacing robust systems by compacted systems for quantification purposes and signal processing.