Recovery of single source signal from noisy and reverberant environments using second-order statistics

摘要

In this paper, a new approach for blind system identification is presented. Our model involves a single unknown source and two signal receivers under a reverberant and noisy environment, and the system is described by an FIR equation plus an additive noise. We mainly focus on the recovery of the source signal and the estimation of the FIR coefficients in the presence of a larger amount of noise by making use of the received data. Under some reasonable assumptions, we have found that the second-order statistics is sufficient for the identification. In our experimental work, we have investigated the effects of some factors on the accuracy of estimating the FIR coefficients. Additionally, we have compared our method with the Widrow's approach and found that our algorithm has a better performance and there is no limitation on the type and the strength of noise even when the SNR is very small (e.g. SNR = -11 dB). Our proposed recursive algorithm also inherits better convergence properties.