Auxiliary-function-based independent vector analysis with power of vector-norm type weighting functions

In this paper, we present an auxiliary-function-based independent vector analysis (AuxIVA) based on the Generalized super Gaussian source model or Gaussian source model with time-varying variance. AuxIVA is a convergence-guaranteed iterative algorithm for independent vector analysis (IVA) with a spherical and super Gaussian source model, and the source model can be characterized by a weighting function. We show that both of the generalized Gaussian source models with the shape parameter 0 <; β ≤ 2 and the Gaussian source model with time-varying variance unifiedly yield a power of vector-norm type weighting functions. A scaling and a clipping technique for numerical stability are discussed. The dependency of the separation performance on the source model is also investigated.