Blind Source Separation Using Multi-Technique Combiner
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio signal processing technologies face challenges in efficiently performing broadband blind source separation for convolutive mixtures, which is essential for acoustic scene analysis, due to limitations in exploiting all fundamental statistical signal properties like nonstationarity, nongaussianity, and nonwhiteness simultaneously.
Innovation Solution
The proposed solution involves an audio signal processing apparatus and method that uses a combination of multiple blind source separators and a combiner to iteratively compute and adjust coefficients, allowing for the simultaneous exploitation of nonstationarity, nongaussianity, and nonwhiteness, using techniques such as Independent Component Analysis (ICA) and performance measures like the Karhunen-Loeve transform to optimize output signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single blind source separator based on one technique is used, then the device complexity is reduced, but the ability to simultaneously exploit multiple statistical signal properties (nonstationarity, nongaussianity, nonwhiteness) is limited
Solution Approach 1:
The patent combines multiple blind source separators using different separation techniques (e.g., ICA for nongaussianity, subspace methods for nonwhiteness, time-frequency methods for nonstationarity) into a unified system. Each separator processes the mixture signals independently and their outputs are merged to achieve simultaneous exploitation of multiple statistical properties, resolving the contradiction between versatility and complexity.
Solution Approach 2:
The system implements a universal blind source separation framework where multiple specialized separators (each optimized for specific statistical properties) work together through a common architecture. This multi-functional approach allows the system to handle diverse signal characteristics without requiring a completely different system for each property, balancing versatility with manageable complexity.
2Measurement precision
If multiple blind source separators are used to exploit all statistical properties, then the separation accuracy and convergence speed are improved, but the computational complexity increases
Solution Approach 1:
The patent segments the blind source separation task into multiple parallel processing paths, each handling specific statistical properties. By dividing the overall computation into specialized sub-tasks (nongaussianity handling, nonwhiteness handling, nonstationarity handling), the system achieves high separation accuracy while managing computational complexity through modular, independent processing units that can be selectively activated.
Solution Approach 2:
The system implements partial exploitation of statistical properties by allowing selective activation of different blind source separators based on the specific application requirements. Not all statistical properties need to be exploited simultaneously in every scenario, enabling the system to achieve sufficient separation accuracy with reduced computational burden when full exploitation is not necessary.
3Productivity
If multiple blind source separators are used to improve convergence speed, then the productivity of acoustic scene analysis is enhanced, but the device complexity increases
Solution Approach 1:
The patent implements continuous iterative adaptation of the blind source separators where each separator continuously processes the input signals and updates its parameters based on feedback from the separation performance. This continuous action across multiple parallel separators accelerates convergence by maintaining persistent processing activity, achieving high productivity while the modular structure keeps complexity manageable through standardized processing units.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a signal processing apparatus (200) for separating a plurality of mixture signals (201) from a MIMO system to iteratively obtain a plurality of output signals (203), the plurality of mixture signals (201) being the response of the MIMO system to a plurality of source signals. The signal processing apparatus (200) comprises: a plurality of blind source separators (205A-N) including a first blind source separator (205A) based on a first blind source separation technique and a second blind source separator (205N) based on a second blind source separation technique, wherein the first blind source separator (205A) is configured to compute a first plurality of preliminary output signals (202A) on the basis of a first set of coefficients describing the MIMO system and wherein the second blind source separator (205N) is configured to compute a second plurality of preliminary output signals (202N) on the basis of a second set of coefficients describing the MIMO system; a combiner (207) configured to combine the first plurality of preliminary output signals (202A) with the second plurality of preliminary output signals (202N) on the basis of a set of combiner coefficients to obtain the plurality of output signals (203), wherein each combiner coefficient is associated with a blind source separator of the plurality of blind source separators (205A-N); and an adjuster (209) configured to adjust the first set of coefficients of the first blind source separator (205A) and the second set of coefficients of the second blind source separator (205N) on the basis of the plurality of output signals (203).