· · · Repository hosted by TU Delft Library

The ultimate goal of instantaneous blind signal extraction is to find one source out of an instantaneous mixture of many others, without, or with a minimum of, prior information. Extraction can be performed by first identifying the complete mixing system and subsequentlyinverting that system. The goal of this paper is to describe the problems behind blind extraction and to directly find the extracting solution, without first identifying the complete mixing system. The proposed method uses second order statistics to identify the extracting solution and can be applied to mixing problems with different kinds of temporal structure e.g. non-stationarity, coloredness.

In many acoustic applications, the extraction of only one desired source signal from a mixture of signals is required. We propose a novel method to perform this extraction in the overdetermined case wheremore sensors than sources are available. We apply blind signal processing techniques in a structure that is similar to a Generalized Sidelobe Canceller (GSC). In a GSC the upper branch reproduces the desired source signal with a beamformer while the lower branch reducesthe contribution of both noise and interferers. In this work the upper branch is used to extract the desired source from a noisy mixtureof all sources and the lower branch is solely used to perform noisereduction. Therefore, the underlying optimization problem is essentially different from the conventional GSC. We validate our method bymeans of a computer simulation.

In many practical applications we are interested in the extraction of only one desired signal out of a mixture of signals. A disadvantage of most blind extraction approaches proposed in the literature isthat they are inefficient in the sense that they also separate or extract undesired signals. To deal with this inefficiency we exploit an a priori guess of direction of arrival related parameters of the desired signal, which serves as a mold. Based on this mold we createlinear combinations of noise-free correlation matrices that are usedto construct a single matrix with a specific eigenstructure. The eigenvector that corresponds to the smallest eigenvalue of this matrixis the desired extraction filter. Finally it is shown that this approach paves the way to make the algorithm flexible in the utilization of additional a priori information.