Generalized Canonical Correlation Analysis and Its Application to Blind Source Separation Based on a Dual-Linear Predictor Structure
This work addresses blind source separation for noisy signals, which is important for signal processing applications, but it appears incremental as it builds on existing CCA methods.
The authors tackled the problem of blind source separation in noisy mixtures by generalizing canonical correlation analysis to handle added white noise without requiring noise variance information, and they derived an adaptive blind source extraction algorithm with a dual-linear predictor structure.
Blind source separation (BSS) is one of the most important and established research topics in signal processing and many algorithms have been proposed based on different statistical properties of the source signals. For second-order statistics (SOS) based methods, canonical correlation analysis (CCA) has been proved to be an effective solution to the problem. In this work, the CCA approach is generalized to accommodate the case with added white noise and it is then applied to the BSS problem for noisy mixtures. In this approach, the noise component is assumed to be spatially and temporally white, but the variance information of noise is not required. An adaptive blind source extraction algorithm is derived based on this idea and a further extension is proposed by employing a dual-linear predictor structure for blind source extraction (BSE).