On-line Algorithm for Blind Signal Extraction of Arbitrarily Distributed, but Temporally Correlated Sources Using Second Order Statistics ANDRZEJ CICHOCKI 1 and RUCK THAWONMAS 2 1 Laboratory for Open Information Systems, Brain Science Institute, RIKEN, 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan; Warsaw University of Technology, 00-661 Warsaw, Poland; E-mail: cia@open.brain.riken.go.jp 2 Department of Information Systems Engineering, Kochi University of Technology, 185 Miyanokuchi, Tosayamada-cho, Kami-gun, Kochi 782-8502, Japan; E-mail: ruck@info.kochi-tech.ac.jp Abstract. Most of the algorithms for blind separation/extraction and independent component analysis (ICA) can not separate mixtures of sources with extremely low kurtosis or colored Gaussian sources. Moreover, to separate mixtures of super- and sub-Gaussian signals, it is necessary to use adaptive (time-variable) or switching nonlinearities which are controlled via computationally intensive measures, such as estimation of the sign of kurtosis ofextracted signals. In this paper, we develop a very simple neural network model and an ef¢cient on-line adaptive algorithm that sequentially extract temporally correlated sources with arbitrary distributions, including colored Gaussian sources and sources with extremely low values (or even zero) of kurtosis. The validity and performance of the algorithm have been con¢rmed by extensive com- puter simulation experiments. Key words: adaptive learning algorithms, blind signal processing, neural networks 1. Introduction Recently, Blind Source Separation (BSS) has attracted a great deal of attention in many areas of science and engineering [1^18]. The task in BSS is to recover original signals from linear mixtures of them, received at the sensors, when the mixing coef¢cients are not known. This problem can be formulated as follows: let the sensor signals at discrete time t (t = 0, 1, 2, ...) be described by xt Ast; 1 where xt is an n  1 sensor vector, st is an m  1 unknown source vector having statistically independent and zero-mean elements, and A is an n  m (with n X m) unknown mixing matrix. Neural Processing Letters 12: 91^98, 2000. 91 # 2000 Kluwer Academic Publishers. Printed in the Netherlands.