Linearly-constrained multichannel interference suppression algorithms derived from a minimum mutual information criterion

In this contribution, a generic framework for linearly-constrained multichannel noise and interference suppression algorithms is presented. It is derived from a linearly-constrained minimum mutual information (LCMMI) criterion between mutually statistically independent desired and undesired components, which also accounts for three fundamental signal properties characteristic, e.g., for speech and audio signals: Nonwhiteness, nonstationarity, and nongaus-sianity. We demonstrate links to prominent second order statistics-based algorithms such as the linearly-constrained minimum variance (LCMV) filter and its realization as a generalized sidelobe canceller (GSC). Additionally, we will show how specific supervised constrained and unconstrained multichannel algorithms result as special cases. The presented LCMMI concept leads to new insights for the development of improved adaptation algorithms for noise and interference suppression.