Abstract: A two-stage approach is studied to resolve the noisy independent component analysis(ICA)for noisy blind source separation.In the first stage, particle filtering(PF)is used to estimate noise-free mixtures, and turn the noisy ICA to noise-free ICA.Thereafter, Fast ICA is adopted to extract the independent components from the estimated clean mixtures.The time-varying autoregressive model of clean mixtures and the relationships between noisy mixtures and clean mixtures compose the dynamic state space equations.The characteristics of the equations are multivariable,moreover,the process and observed noises are not restricted to be gaussian.Due to above reasons, particle filtering is applied. Thus, PF+FastICA is put forward.The simulation proves that PF+FastICA outperforms Denoising Source Separation.