Title :
A time shift parameter setting of temporal decorrelation source separation for periodic signals
Author :
Amishima, Takeshi ; Hirata, Kazufumi
Author_Institution :
Inf. Technol. R&D Center, Mitsubishi Electr. Corp., Kamakura, Japan
fDate :
Oct. 29 2012-Nov. 2 2012
Abstract :
Temporal Decorrelation Source SEParation (TDSEP) [1] is a blind separation scheme that uses the time structure of the sources signals. The merit of this TDSEP is that it can separate Gaussian distributed signals, as long as they have some time structural properties, such as periodicity, whereas other non-Gaussianity based methods [2] can not. The drawback is that the separation performance heavily depend upon the time shift parameters. This motivates us to investigate the optimal time shift parameters for periodic Gaussian signals. This paper proposes a method to automatically select the time shift parameters, that provides high separation performance.
Keywords :
Gaussian distribution; blind source separation; correlation methods; Gaussian distributed signals; TDSEP; blind separation; periodic Gaussian signals; separation performance; source signals; temporal decorrelation source separation; time shift parameter setting; Correlation; Equations; Estimation; Indexes; Mathematical model; Monte Carlo methods; Vectors;
Conference_Titel :
Antennas and Propagation (ISAP), 2012 International Symposium on
Conference_Location :
Nagoys
Print_ISBN :
978-1-4673-1001-7