An underdetermined blind source separation algorithm based on clustering analysis and time-frequency representation

To overcome the disadvantages of traditional blind source separation algorithm in solving the problem of undetermined blind source separation, a new algorithm based on clustering analysis and time-frequency representation is proposed. Firstly, Ensemble Empirical Mode Decomposition algorithm was utilized to decompose the observed signal into a series of eigenfunctions. K-means as well as singular value decomposition clustering was used for estimating source number in observed signals. Then, the estimation of the mixed matrix was realized by Short Time Fourier Transform(STFT) and fuzzy c-mean clustering. Simulation experiments on linear and nonlinear mixture and the separation experiment for vibration signal of rolling bearing datasets demonstrate the efficiency and feasibility of the proposed algorithm.