Robust Techniques for BSS and ICA with Noisy Data
Andrzej S Cichocki, Шун-ичи Амари · 2002
In this chapter we focus mainly on approaches to blind separation of sources when the measured signals are contaminated by large additive noise. We extend existing adaptive algorithms with equivariant properties in order to considerably reduce the bias caused by measurement noise for the estimation of mixing and separating matrices. Moreover, we propose dynamical recurrent neural networks for simultaneous estimation of the unknown mixing matrix, source signals and reduction of noise in the extracted output signals. The optimal choice of nonlinear activation functions for various noise distributions assuming a generalized-Gaussian-distributed noise model is also discussed. Computer simulations of selected techniques are provided that confirms their usefulness and good performance. The main objective of this chapter is to present several approaches and derive learning algorithms that are more robust with respect to noise than the techniques described in the previous chapters or that can reduce the noise in the estimated output vector of independent components