Implementation of Universal Neural Network Approximator on a ULP Microcontroller for Wavelet Synthesis in Electroencephalography

The paper describes the implementation of universal neural network approximator on microcontroller MSP430G2553 with ultralow power consumption. A multilayer perceptron is used as an approximating artificial neural network. Approximation is one of the stages of building wavelet neural network models. The approximation of electroencephalogram fragments allows obtaining wavelets adapted for its continuous wavelet transform. Implementation of neural network approximator on this microcontroller is one of the stages to create a portable system for automated electroencephalogram analysis. Two ways of implementation a neural network approximator are presented in the paper. The first way is based on the experiment with the neural network approximator without its training. The second way involves the experiment based on the interaction with the neural network approximator with training algorithm. The studies of the approximation rate depending on the number of iterations and the core clock frequency of the microcontroller have been conducted.