Quick Algorithms for Independent Vector Extraction and Analysis Based on Exact Newton-Raphson Optimization
Zbyněk Koldovský, Václav Kautský · arXiv (Cornell University) · 2019
We propose new algorithms for the joint blind extraction and analysis of independent vector/components in linear complex-valued instantaneous mixtures. An efficient parameterization of the mixing and de-mixing matrix is used, which enables us to derive algorithms for extraction based on the exact Newton-Raphson update rule. For the complete independent vector/component analysis, orthogonally constrained algorithms are run in parallel. In experiments, the proposed methods show fast and stable convergence that is competitive to the state-of-the-art algorithms. Moreover, our approach is applicable also to piecewise determined mixtures with constant separating vectors.