An Improved Algorithm for Dynamic Independent Component Analysis

The inner time-structure information is considered for the dynamic independent component analysis (DICA) model, and the new concept of improved and corresponding algorithm are presented. When the improved is applied in DICA, a new linear mixing model of improved DICA (IDICA) can be obtained in which the assumptions of independence and non-Gaussianity for the basic ICA model are also satisfied, and the non-Gaussianity of latent components increase. What’s more, the mixing matrix (or demixing matrix) estimated from IDICA model is also available for original DICA model. Then the new algorithm of (IDICA) is presented, and the effects of IDICA are discussed. It is validated that it can accelerate the convergence rate with the increasing of the latent component’s non-Gaussianity, but has little influence on the convergence accuracy. In performing improved, the adaptive filter based on Minimum Square Error (MSE) is proposed. The experimental results show the new method’s efficient and available.