Coherence-Based Noise Estimation for Audio Signal Separation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing audio systems struggle to efficiently separate noise from microphone signals, particularly in environments with multiple acoustic sources, leading to suboptimal noise reduction and comfort noise generation.

Innovation Solution

The technology estimates the power spectral density of noise signals by using coherence calculations and matrix operations in the frequency domain, allowing for the separation of noise from system audio contributions using a single microphone.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If adaptive filtration techniques are used to separate noise from audio signals, then noise reduction capability is improved, but the system malfunctions in the presence of correlated sources and requires significant computational resources

Engineering Contradiction:
Improvenoise estimation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces complex time-domain adaptive filtration operations with frequency-domain coherence-based calculations. By transforming the signal processing approach from iterative time-domain filtering to direct frequency-domain spectral analysis using coherence functions, the system achieves comparable noise estimation accuracy with significantly reduced computational complexity and without the malfunction issues associated with correlated sources

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the domain of signal processing from time-domain to frequency-domain operations. By computing power spectral densities and coherence functions in the frequency domain rather than performing time-domain adaptive filtering, the system fundamentally alters the processing parameters to achieve both accuracy and computational efficiency

Inventive Principle:
Principle #35Parameter changes

2Loss of information

If time waveform determination is performed for signal analysis, then temporal information is obtained, but computing resources are wasted without direct frequency-specific information

Engineering Contradiction:
Improvefrequency-specific information availabilityVSAvoidcomputational resource consumption
Core Design Contradiction:
Loss of informationVSUse of energy by moving object

Solution Approach 1:

The patent extracts frequency-specific information directly from the signal using spectral analysis and coherence calculations in the frequency domain. By taking out and directly computing the power spectral density and coherence functions, the system obtains the needed frequency-specific information without the intermediate step of determining time waveforms, thereby avoiding wasted computational resources

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent performs preliminary frequency-domain transformation and coherence calculation to obtain frequency-specific information before any further processing. By pre-computing the power spectral densities and coherence functions, the system makes frequency-specific information immediately available without requiring subsequent time-domain analysis

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3963578B1Signal component estimation using coherence
Publication Date: 2025.06.04 BOSE CORP
  • EP3963578B1 patent drawingFigure 1
  • EP3963578B1 patent drawingFigure 2
  • EP3963578B1 patent drawingFigure 3

AI summary

Systems, methods, and machine-readable storage devices that receive an input signal representing audio captured using a microphone. The input signal includes portions that represent acoustic output from one or more audio sources, and a portion that represents other acoustic energy in the environment. A frequency domain representation of the input signal is iteratively modified to substantially reduce effects due to all but a selected one of the portions, from which an estimate of the power spectral density, PSD, of the selected portion is determined. Based upon the estimated PSD a noise or echo component is reduced, or a replacement noise is provided.The iterative modification involves a diagonalization of the cross-spectral density matrix to remove content coherent with a first audio input from the auto and cross-spectra of other signals.