Blind Source Separation for Non-stationary Signal Based on Time-Frequency Analysis
Shi He-Ping, Cao Ji-hua, Liu Xiao · 2011
To separate independent sources from the linear mixed observed data, the traditional blind source separation (BSS) is usually neglected the non-stationary of the signals. With the non-stationary of the signals focused on, a new blind source separation algorithm is proposed by combining the characteristic of time-frequency analysis (TFA) and blind source separation (BSS). In the algorithm, the full time-frequency domain Wigner searching is utilized to find the local maximum of the Smoothed Pseudo Wigner-Ville time-frequency distribution. The simulations show this algorithm not only suppresses cross term interference but also remain time-frequency resolutions. The proposed algorithm provides an effective technology for fault diagnosis of mechanical equipment.