A new method of eliminating indeterminacy in ICA for conducted EMI separation

According to the software approaches for conducted EMI separation based on independent component analysis (ICA), this paper presents a simple method using signal statistics to eliminate the inherent indeterminacy of permutation, phase and amplitude of the separated sources. The premise of the study is that a line impedance stabilization network (LISN) is used as the conducted interference testing circuit to isolate the EMI noise from the power source. Permutation, phase and amplitude adjustment matrices are obtained by calculating the variance, covariance and correlation of the separated signals and the observed signals. The proposed method is able to identify and recover the common-mode (CM) and differential-mode (DM) noises perfectly from the total noise measured by the LISN. The simulation with synthetic signals and the experiment using a power converter as an equipment under test (EUT) demonstrate the feasibility and effectiveness of eliminating indeterminacy in the ICA for conducted EMI separation.