Non-Square Blind Source Separation Under Coherent Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing blind source separation techniques face challenges in non-square cases under coherent noise, particularly in scaling algorithmically and computationally, and in effectively separating mixed source signals from output noise.
Innovation Solution
A method and system for non-square blind source separation under coherent noise, which involves estimating mixing parameters, filtering to reduce output noise, and further filtering to separate the mixed source signal from noise, utilizing a maximum likelihood estimator and W-disjoint orthogonality assumptions to optimize partitioning and parameter computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If non-square blind source separation is performed under coherent noise, then the number of sources and sensors can be unequal, but computational complexity and algorithmic scaling become problematic
Solution Approach 1:
The patent segments the source separation problem into independent frequency bins through Fourier transformation, allowing parallel processing of each frequency component. This segmentation enables the algorithm to handle non-square cases (unequal sources and sensors) independently at each frequency, reducing overall computational complexity while maintaining adaptability to varying source-sensor configurations
Solution Approach 2:
The patent changes the parameter representation by transforming the time-domain signal into the frequency domain using Fourier transform. This parameter transformation allows the mixing matrix to be estimated and processed differently at each frequency bin, enabling non-square separation while controlling computational complexity through frequency-domain operations
2Measurement precision
If maximum likelihood estimation is used to estimate mixing parameters, then parameter accuracy is improved, but the technique assumes noise limit zero or non-isotropic noise fields
Solution Approach 1:
The patent applies local quality by estimating the mixing matrix separately at each frequency bin rather than using a single global estimate. This allows the maximum likelihood estimation to be performed with frequency-specific noise characteristics, improving parameter accuracy while accommodating actual noise conditions (including coherent and isotropic noise) without requiring the noise to be zero or non-isotropic
Solution Approach 2:
The patent uses an iterative feedback mechanism where the mixing matrix estimate is refined through multiple passes: initial estimation, separation, noise subspace identification, and re-estimation. This feedback loop improves parameter estimation accuracy while making the method robust to various noise conditions by continuously adapting the estimates based on the observed data characteristics
3Measurement precision
If filtering is applied to reduce output noise, then signal-to-noise ratio is improved, but separation of mixed source signal from noise becomes more difficult
Solution Approach 1:
The patent introduces an additional dimension by performing separation in the frequency domain rather than solely in the time domain. By transforming to frequency bins and processing each independently, the method creates a new dimension where noise and signal can be distinguished through their spectral characteristics, improving signal-to-noise ratio while maintaining separation capability through frequency-selective processing
Solution Approach 2:
The patent uses the frequency domain as an intermediary space between the raw mixed signal and the final separated output. The Fourier transform acts as an intermediary that reveals the structure of the mixing process at different frequencies, allowing noise reduction filtering to be applied in a way that preserves source separation capability through frequency-specific processing
Data Source
AI summary
A system and method for non-square blind source separation (BSS) under coherent noise. The system and method for non-square BSS estimates the mixing parameters of a mixed source signal and filters the estimated mixing parameters so that output noise is reduced and the mixed source signal is separated from the noise. The filtering is accomplished by a linear filter that performs a beamforming for reducing the noise and another linear filter that solves a source separation problem by selecting time-frequency points where, according to a W-disjoint orthogonality assumption, only one source is active.