Jointly Approximate Diagonalization for a set of eigen-matrices
Hai Zhang · Jounal of Xidian University · 2001
In recent years, the jointly approximate diagonalization problem is proposed in blind source separation. This paper gives a new algorithm for jointly approximate diagonalization of a set of eigen matrices on the basis of the biorthogonality between left and right eigen vectors. We propose that the new cost functions are used to find the single eigen component. An efficient iterative algorithm is introduced and used to find the global optimal solution of the cost functions. By the systematic multistage decomposition and multistage reconstruction, we have obtained all the eigen components.