A Kernel-Based Nonlinear Blind Source Separation Algorithm with Reference and Its Application in Satellite Micro-vibration System

this paper, a kernel-based nonlinear blind source separation algorithm with reference information is proposed to identify the harmonic source signals in satellite micro-vibration system. The kernel feature space with reduced dimension constructed by the proposed algorithm can transform the nonlinear blind source separation in the input space into linear blind source separation. In the linear blind source separation phase, aiming at the weak non-Gaussian characteristic of micro-vibration harmonic sources, a new linear blind source separation objective function with reference information is proposed to ensure the accuracy of source identification. The effective estimated signals of the sources are selected from the linear separated signals according to the spectrum correlation coefficient index. The effectiveness of the proposed algorithm is verified by the satellite cabin structure experiment.