Blind Source Separation Using PICA Network
Min Wan, Xinli Zhang, Yi Zhang · 2008
The principal independent component analysis (PICA) network is used to the real-valued source signals blind separation with a reference. It's proved in this paper that when a reference signal $r$ is available, the blind source separation can be transformed to the eigenvalue eigenvector decomposition of a real symmetric matrix. When generalized to the multi-reference case, a similar result is obtained. By these results, corresponding algorithms are proposed. Due to existing efficient eigen value decomposition techniques, these algorithms have faster computing speed than other algorithms. Simulations verify the efficiency of the algorithms.