Implementing Neural Networks onto Standard Low-Cost Microcontrollers for Sensor Signal Processing
Nicolás J. Medrano-Marqués, B. Martín-del-Brío, Antonio Bono-Nuez, Carlos Bernal-Ruíz · 2006
In this paper we show some sensor linearizing methods based on feed-forward neural networks (multilayer perceptron). These procedures can be easily programmed into a computer-based data acquisition system. Nevertheless, we show that introducing some simplifications in the neural network architecture, these linearization procedures can also be programmed onto low-cost microcontrollers for embedded (portable) applications. In this work, we make use of an NTC sensor as a case study, but the procedure is so general and flexible that it can easily be applied to other non linear sensors. System performance measurements for both simulations and real set up are presented