Implementation of an Approximator Based on a Multilayer Perceptron and Wavelet-Neural Network on the STM32 Microcontroller

Neural network approximators are used in digital signal processing, in particular when automating the process of their analysis. The paper is devoted to the implementation of an approximator based on a multilayer perceptron and a wavelet-neural network. Implementation is carried out on the STM32F401RE microcontroller. In the article the speed of training neural networks during the approximation of fragments of biomedical signals is estimated and the influence of the selected activation function of the wavelet-neural network on the approximation error is evaluated. The experimental method shows that for the approximation of electrocardiogram and electroencephalogram when implementing a neural network approximator on the selected microcontroller, the most justified use of the wavelet neural network using the Mexican hat as the activation function of the wavelet is used.