Convergence Analysis of Non-Negative Matrix Factorization for BSS Algorithm

Shangming Yang, Zhang Yi

Abstract: In this paper the convergence of a recently proposed BSS algorithm is analyzed. This algorithm utilized Kullback–Leibler divergence to generate non-negative matrix factorizations of the observation vectors, which is considered an important aspect of the BSS algorithm. In the analysis some invariant sets are constructed so that the convergence of the algorithm can be guaranteed in the given conditions. In the simulation we successfully applied the algorithm and its analysis results to the blind source separation of mixed images and signals.