Blind signal separation and independent component analysis
暂无分享,去创建一个
Independent component analysis (ICA) is emerging as a new paradigm in signal processing and data analysis. ICA has been proposed as a solution to the blind source separation problem (BSS) in which sensory signals are given as mixtures of unknown source signals. Since neither the sources nor the mixing environment are known a priori there is no general analytical solution to this problem. However, since the last decade researchers have found good approximate solutions based on assumptions about the source statistics. They were able to formulate solutions or algorithms that were derived from well-known cost functions including maximum likelihood estimation, information maximization and higher-order cumulants. The ICA framework touches many theoretical research areas, and therefore receives attention in several research communities including machine learning, neural networks, statistical signal processing and Bayesian modeling. More recently, there have been numerous applications of ICA in adaptive 0ltering, speech signal processing, biomedical signal processing, computational neuroscience, image coding, text data modeling, and 0nancial data analysis. The standard ICA formulation has, in general, strong assumptions or simpli0cations such as: the number of sensors has to be equal or greater than the number of sources, there is no additive noise signal in the sensors, and the sources are modeled as random variables. However, when applying ICA to real-world problems it becomes evident that those restrictions are cumbersome and although the standard ICA algorithms give good approximate solutions, there is great need in going beyond the standard assumptions and develop new algorithms that can relax some of the assumptions. This special issue on ICA and BSS is a timely publication covering important recent 0ndings in ICA theory as well as new applications of ICA in data analysis. Leading researchers present their work to advance this 0eld beyond traditional borders and show how this 0eld can bridge disciplines and yield new methods that lead to mutual bene0ts and advances in science and engineering. This issue is composed of a review article and two parts: the 0rst part presents new theoretical 0ndings while the second part presents novel applications. Often, there is no such sharp division into these two parts since illustrative simulations support theoretical 0ndings and practical methods include algorithmic modi0cations to design and evaluate the application-speci0c methods. In “Frontiers of research in BSS=ICA”, V. David S: anchez A. formulates some of the basic concepts and problems in the areas of blind signal separation and independent component analysis. He then moves to a concise description of key methods of data preprocessing and algorithms that provide solutions to the instantaneous mixture and the