Joint non-orthogonal joint diagonalization based on LU decomposition and Jacobi scheme

摘要

We present an algorithm for jointly solving multiple nonorthogonal joint diagonalization (JNJD) problems with linked loading matrices, by using LU decomposition and Jacobi scheme in the updating procedure. The proposed method could be used in applications that involve a set of correlated linear instantaneous blind source separation (BSS) problems (e.g. frequency domain convolutive BSS), and differs from existing joint diagonalization based algorithms in that it incorporates the correlations among the multiple JD sets within the JNJD scenario. Simulations are provided to examine the performance of the proposed algorithm.