Blind source separation

In signal processing the received data can be modeled as a superposition of a finite number of elementary source signals and an additive noise. Generally, in a multi-sensor environment application, such as underwater acoustics, the principal objective is estimating the number and directions of radiating sources. In the last decade, eigenstructure-based methods yielding resolution have been proposed. However, these methods need the model of sources. In many cases, this model is not available, indeed, there is the appearance of the blind methods. Blind technique refers to lack of knowledge about the mixing system. It assumes very weak assumptions either for signals or mixtures. Blind source separation consists in identifying the sensors/sources transfer matrix and retrieving the source signals without resorting a priori information about the above matrix. It exploits only the information carried by the received signals themselves. The potential benefit of such a blind approach is that source separation is essentially unaffected by errors in the propagation model or in array calibration whereas the classical array processing methods assume that the array geometry is known and the sources are in the far field of the array. This contribution deals with blind source separation using second-order statistics. [This work was partially funded by FEDER and PRAXIS XXI, under Contract No. PRAXIS/3/3.1/TPR/23/94.]