engleski

Nonstationary signal blind source separation using clustering algorithms

In this paper an advanced method for blind source separation of nonstationary signals applied to time frequency distributions is presented. The signal spectrogram has been generated in order to identify different spectral components, where a component is defined as a continuous energy concentration. Background noise has been filtered using K-means clustering algorithm. For component extraction and classification Hoshen-Kopelman clustering and labeling algorithm has been used. The obtained results show the suitability of the method for noisy nonstationary signals analysis, offering interesting prospective for real life applications.

Time-frequency analysis; signal component; blind source separation; K-means clustering algorithm; Hoshen-Kopelman; cluster classification/labeling

nije evidentirano