Sequential blind identification of underdetermined mixtures using a novel deflation scheme

Area of Science:

• Signal Processing
• Information Theory
• Statistical Signal Processing

Background:

• Blind identification addresses source separation without prior knowledge of the mixing system.
• Underdetermined instantaneous mixtures (more sources than sensors) pose significant challenges for traditional methods.

Purpose of the Study:

• To propose a new blind identification algorithm for underdetermined instantaneous mixtures.
• To enhance the flexibility and robustness of blind identification techniques.

Main Methods:

• A novel sequential blind identification algorithm is developed.
• The algorithm utilizes a rank-1 detecting device to reformulate identification as a constrained optimization problem.
• A generalized eigenvalue decomposition-based deflation method is employed for identifying subsequent mixing matrix columns.

Main Results:

• The proposed algorithm sequentially estimates the mixing matrix, reducing identification to iterative optimization tasks.
• The deflation method effectively avoids error accumulation, a common issue in sequential approaches.
• Comparative simulations show superior performance over simultaneous blind identification algorithms.

Conclusions:

• The sequential blind identification algorithm offers enhanced flexibility and robustness.
• The method demonstrates superior performance in underdetermined instantaneous mixture scenarios.
• This approach advances blind source separation techniques in challenging conditions.