A method for constructing a frequecy analyzer of quasi-stationary signals

A method for frequency analysis of polyharmonic signals based on the time-frequency processing of the spectral pattern is proposed. The amplitude spectrum is determined in two steps in the proposed method. The amplitude spectrum of the analyzed signal is estimated by a comb of digital bandpass filters with highly overlapping frequency characteristics at the first stage.In order to find the amplitude spectrum, the output signals of the bandpass filters are rectified and smoothed by lowpass filters. At the second stage, the obtained spectrum is processed using a single-layer network of neural-like elements with delayed lateral inhibitory connections. The use of such a network allows to improve the characteristics of the frequency analyzer, namely, for a given accuracy and frequency resolution, to significantly reduce both the spectral envelope pulsations and the analysis time, which makes it possible to analyze shorter signals. The performance of the proposed method was tested with examples of harmonic and speech signal analysis for the case of using a comb of second-order digital bandpass filters with equal relative bandwidths and central frequencies evenly spaced on a logarithmic frequency scale. Such a choice of filter parameters of a frequency analyzer ensured equal accuracy of the analysis of the harmonic components of the signal under study.