ICA in signals with multiplicative noise

Independent component analysis (ICA) has been shown in the last few years to be a very useful tool in blind separation of sources and feature extraction. However, at least in its simpler form, its utility is reduced to the case that the outputs are linear mixtures of independent sources. This excludes signals with multiplicative noise. In this paper, ICA is extended to this situation. In order to do this, the special structure that appears in this new model is first studied, and then, the multiplicative ICA method is designed such that it uses this structure to find the mixture of the sources in the noisy environment. The local and global convergence properties of the method are also studied and its performance compared with standard ICA methods.