Blind separation of non-negative source signals using multiplicative updates and subspace projection

Authors:
Alexander Bertrand;Marc Moonen
Affiliations:
Department of Electrical Engineering (ESAT-SCD), Katholieke Universiteit Leuven, Kasteelpark Arenberg 10, B-3001 Leuven, Belgium;Department of Electrical Engineering (ESAT-SCD), Katholieke Universiteit Leuven, Kasteelpark Arenberg 10, B-3001 Leuven, Belgium
Venue:
Signal Processing
Year:
2010

Citing 7
Cited 2

Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications

Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
Multiplicative updates for non-negative projections

Neurocomputing
Sliding window adaptive SVD using the unsymmetric householder partial compressor

Signal Processing
Handbook of Blind Source Separation: Independent Component Analysis and Applications

Handbook of Blind Source Separation: Independent Component Analysis and Applications
Blind separation of multichannel biomedical image patterns by non-negative least-correlated component analysis

PRIB'06 Proceedings of the 2006 international conference on Pattern Recognition in Bioinformatics
Sliding window adaptive SVD algorithms

IEEE Transactions on Signal Processing
Global convergence analysis of a discrete time nonnegative ICA algorithm

IEEE Transactions on Neural Networks

On-line learning parts-based representation via incremental orthogonal projective non-negative matrix factorization

Signal Processing
A convergent algorithm for orthogonal nonnegative matrix factorization

Journal of Computational and Applied Mathematics

Quantified Score

Hi-index	0.08

Visualization

Abstract

In this paper, we consider a noise-free blind source separation problem with independent non-negative source signals, also known as non-negative independent component analysis (NICA). We assume that the source signals are well-grounded, which means that they have a non-vanishing pdf in a positive neighborhood of zero. We propose a novel algorithm, referred to as multiplicative NICA (M-NICA), which uses multiplicative updates together with a subspace projection based correction step to reconstruct the original source signals from the observed linear mixtures, and which is based only on second order statistics. A multiplicative update has the facilitating property that it preserves non-negativity, and does not depend on a user-defined learning rate, as opposed to gradient based updates such as in the non-negative PCA (NPCA) algorithm. We provide batch mode simulations of M-NICA and compare its performance to NPCA, for different types of signals. It is observed that M-NICA generally yields a better unmixing accuracy, but converges slower than NPCA. Especially when the amount of data samples is small, M-NICA significantly outperforms NPCA. Furthermore, a sliding window implementation of both algorithms is described and simulated, where M-NICA is observed to provide the best results.