Mixed Domain Blind Source Separation for Sensor Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Fourier-domain blind source separation methods for convolutive signal mixtures face output permutation ambiguity, leading to distorted signals when transformed back to the signal domain, and require separate BSS operations for each frequency channel, resulting in high computational complexity and error propagation.
Innovation Solution
A mixed domain method that performs a single blind source separation operation on interleaved time-frequency distributions, ensuring correct output ordering and reducing computational requirements by vectorizing time-frequency data into mixed frequency and time matrices, allowing for simultaneous processing of all frequency channels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate BSS operations are performed on each frequency channel, then the output can be obtained for each channel, but the output permutation ambiguity causes signal distortion and errors propagate to subsequent channels
Solution Approach 1:
The patent merges all frequency channels into a single mixed-domain matrix by vectorizing time-frequency distributions, then performs a single unified BSS operation on the entire matrix. This eliminates the permutation ambiguity problem that occurs when separate BSS operations are performed on each frequency channel independently, as the unified operation ensures consistent output ordering across all channels.
2Adaptability or versatility
If separate BSS operations are performed on each frequency channel, then channel-specific processing is achieved, but computational complexity increases and processing time is extended
Solution Approach 1:
The patent combines all frequency channel data into a single mixed-domain matrix and performs one unified BSS operation, reducing computational complexity from multiple separate operations to a single operation. This merging approach maintains the ability to process each frequency channel while significantly improving processing efficiency and reducing computational burden.
3Extent of automation
If greedy channel matching methods are used to order output channels, then the ordering process is automated, but the method requires signal similarity in adjacent channels which is not always met, resulting in errors
Solution Approach 1:
The patent eliminates the need for greedy channel matching by performing a single unified BSS operation on the mixed-domain matrix that encompasses all frequency channels simultaneously. This approach inherently maintains correct output ordering across all channels without requiring post-processing matching algorithms, thereby eliminating the accuracy limitations of greedy matching methods.
4Productivity
If a single BSS operation is performed on the mixed frequency and time matrix, then computational requirements are reduced and output ordering is guaranteed, but the algorithm complexity increases due to mixed-domain processing
Solution Approach 1:
The patent transforms the traditional frequency-domain-only processing into mixed-domain processing by vectorizing time-frequency distributions and combining them into a unified matrix. This dimensional transformation enables a single BSS operation to process all frequency channels simultaneously, reducing computational requirements while the structured vectorization approach manages the algorithmic complexity.
Data Source
AI summary
A method for increasing accuracy and reducing computational requirements for blind source separation of mixtures of signals in multi-path environments includes receiving a plurality of channel inputs, each channel input comprising a mixture of signals from a plurality of sources, performing a short time Fourier transform on each channel input of the plurality of channels, wherein a respective output of a respective short time Fourier transform on a respective channel is a respective time-frequency distribution for the respective channel, vectorizing each respective time-frequency distribution into a respective mixed frequency and time vector, combining each respective mixed frequency and time vector into a mixed frequency and time matrix, and performing blind source separation on the mixed frequency and time matrix to separate the mixture of signals from the plurality of sources into a plurality of signal source channels, each respective signal source channel comprising signals from a respective source.