Title: Faster IVA: Update Rules for Independent Vector Analysis based on Negentropy and the Majorize-Minimize Principle

Abstract: Algorithms for Blind Source Separation (BSS) of acoustic signals require efficient and fast converging optimization strategies to adapt to nonstationary signal statistics and time-varying acoustic scenarios. In this paper, we derive fast converging update rules from a negentropy perspective, which are based on the Majorize-Minimize (MM) principle and eigenvalue decomposition. The presented update rules are shown to outperform competing state-of-the-art methods in terms of convergence speed at a comparable runtime due to the restriction to unitary demixing matrices. This is demonstrated by experiments with recorded real-world data.