Projection-pursuit-based method for blind separation of nonnegative sources

Area of Science:

• Signal Processing
• Machine Learning
• Data Analysis

Background:

• Blind source separation is crucial for isolating individual signals from mixed observations.
• Existing methods often rely on statistical independence or nonnegativity assumptions, limiting their applicability.

Purpose of the Study:

• To develop a novel and efficient projection pursuit (PP) based method for blind separation of nonnegative sources.
• To improve upon existing methods in terms of speed and applicability to cross-correlated sources.

Main Methods:

• A new mixing model is constructed by mapping the observation matrix.
• The projection pursuit (PP) method is employed to estimate the mixing matrix column by column.
• Optimal projection is utilized for enhanced computational efficiency.

Main Results:

• The proposed PP method demonstrates significantly faster performance compared to Chan's method.
• The method effectively separates cross-correlated sources without needing source statistical information.
• Successful recovery of sources is achieved through iterative estimation of the mixing matrix.

Conclusions:

• The developed PP-based method offers a superior and faster alternative for blind separation of nonnegative sources.
• This approach expands the applicability of blind source separation by not requiring nonnegative mixing matrices.
• The method shows robust performance in simulations, outperforming existing techniques.