Resumen

ZAMBRANO ESCOBAR, Alejandro  y  PINTO MINDIOLA, Lácides. ARTIFICIAL NEURAL NETWORKS APLICATION ON DIGITAL SIGNAL PROCCESORS: INFRARED SENSORS CHARACTERIZATION . uct [online]. 2009, vol.13, n.51, pp.129-136. ISSN 2542-3401.

This paper proposes a method for the characterization of the nonlinear response of an infrared sensor used in measuring distances, through an Artificial Neural Network model with supervised training. The neural network model is developed using the neural networks tools (NNT: Neural Networks Toolbox ®) of MATLAB ™, and then implemented via C language, in a Digital Signal Processor (DSP) for further application in embedded systems. We compare three training algorithms to verify the feasibility of future implementation of online training. Levenberg - Marquardt backpropagation algorithm has yielded the best results in modeling the sensor characteristic curve, getting through in this application, the lowest error in learning, in the lowest number of training times recorded compared with the other methods: Resilient backpropagation and quasi-Newton backpropagation Model results and implementation confirm a satisfactory performance of the method used, which can be extended to the characterization of other sensors.

Palabras clave : Infrared Sensor; Nonlinear Response; Artificial Neural Network; Backpropagation; Digital Signal Processor.