Stability analysis on independent vector analysis
Takashi Itahashi, Kiyotoshi Matsuoka · 2011
Independent vector analysis (IVA) is a new method for solving the permutation problem in the frequency-domain independent component analysis. IVA assumes inner dependency among the frequency components of each source; then, it treats a set of the frequency components of a source as a source vector. Although the effectiveness of IVA has been demonstrated by its applications, there are very few mathematical analyses of its algorithm. This paper shows a remarkable result that proves the validity of IVA: if the desired demixing process is a local minimum of a measure for independence among the source vectors, any permuted one never becomes a local minimum.