SPECTRUM ANALYSER BASED ON NEURAL NETWORKS

摘要

The invention relates to a spectrum analyser based on neural networks. More specifically, the invention consists of a method and device for the spectral analysis of periodic signals. The purpose of the invention is to enable the spectral analysis of periodic signals to be performed with accuracy and great flexibility. The novel spectral analysis method is based on sine-fitting techniques and on artificial neural networks. The signal to be analysed is sampled at corresponding moments during several cycles. The samples obtained are applied as inputs to a neural network which supplies the coefficients of the truncated Fourier series corresponding to the signal being analysed and the spectrum of the signal is established using said coefficients. In turn, the neural network also supplies an estimate of the fundamental frequency of the signal being analysed. The inventive device is suitable for use in all cases requiring accurate spectral analysis with great flexibility.