An Adaptive Short-time Frequency Domain Algorithm for Blind Separation of Non-stationary Convolved Mixtures (CROSBI ID 488769)
Prilog sa skupa u zborniku | izvorni znanstveni rad | međunarodna recenzija
Podaci o odgovornosti
Kopriva, Ivica ; Devcic, Zeljko ; Szu, Harold ;
engleski
An Adaptive Short-time Frequency Domain Algorithm for Blind Separation of Non-stationary Convolved Mixtures
Many of the algorithms for the blind separation of convolved mixtures are iterative and consequently not suitable for real time (closed loop) applications. Furthermore, either the unknown mixture sources have a slow turning maneuver or the turbulent propagation media have non-stationary scintillation in time, we require a wavelet-like window to reduce the relative non-stationary effect such that within a finite filter length to be determined the data become piecewise stationary. Such a divide-and-conquer strategy can be implemented by using the short-time windowed FFT or equivalently the discrete z-transform of adaptive finite filter length. Here, we present a frequency domain algorithm derived for windowed-adaptive blind separation of convolved sources. Signal separation (filtering) is performed in short-time-windowed-frequency domain in terms of a finite filter length L obtaining faster convergence and better performance compared with the strictly time domain algorithms. In order to avoid whitening effect the recurrent neural network, similar to one proposed by Back and Tsoi [1], is employed. Statistical independence test is done in time domain in order to determine the relative time-varying effect and solve the permutation indeterminacy problem for which it is known that can deteriorate performance seriously. Corrections of the learning rules are introduced for which it is shown to improve separation performance significantly. Additionally, the results developed by Amari [2, 3] for the instantaneous mixtures are applied making learning equations computationally more efficient. To resolve the permutation problems at the neural network outputs algorithm developed by Markowitz and Szu [21] is applied.
blind source separation; convolutive mixtures; neural networks
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
Podaci o prilogu
424-429-x.
2001.
objavljeno
Podaci o matičnoj publikaciji
Proceedings of the INNS-IEEE Joint Conference on Neural Networks
Piscataway (NJ): Institute of Electrical and Electronics Engineers (IEEE)
Podaci o skupu
INNS-IEEE 2001 Joint Neural Network Conference
poster
15.07.2001-19.07.2001
Sjedinjene Američke Države