An Extraction Algorithm Based on Second-Order Statistics for Noisy Mixtures

There is a trend to develop blind source extraction (BSE) algorithms for noisy measurements based on second-order statistics. Many existing BSE methods are limited to noise-free mixtures, which is not realistic. Based on a rigorous analysis of the existing BSE method, we address the problem of recovering a desired source from a noisy linear mixture. In this paper we develop a novel cost function, from which the effect of noise is removed. Maximizing the cost function, we obtain an algorithm, which caters for the effects of noise. Through the analysis of artificially synthesized data and real-world electrocardiogram (ECG) data, we illustrate the efficiency of this algorithm in the presence of noise.