Subband Acoustic Echo Cancellation With Diagonal Regularization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current acoustic echo cancellation techniques for spatial audio communication and teleconferencing on mobile devices face challenges due to high computational complexity and instability, particularly in handling multi-channel and stereo playback scenarios with long echo impulse responses.

Innovation Solution

Implementing subband adaptive filters using diagonally regularized correlation matrices to iteratively determine gain vectors and update filter coefficients, reducing echoes by processing subband signal sequences, and employing a technique to diagonally regularize the inverse correlation matrix to ensure robustness and computational efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If time-domain filter implementations are used for acoustic echo cancellation, then the filtering operation is straightforward, but the computational complexity becomes very high due to long echo impulse responses requiring thousands of taps

Engineering Contradiction:
Improvefiltering operation simplicityVSAvoidcomputational complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent divides the frequency spectrum into multiple subbands and processes each subband separately using shorter filters. This segmentation approach transforms a single complex time-domain filter with thousands of taps into multiple simpler filters operating in parallel on divided frequency segments, thereby reducing overall computational complexity while maintaining echo cancellation effectiveness

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from time-domain filtering to frequency-domain filtering by applying Fourier transforms. This dimensional change allows the system to exploit the convolution theorem, where time-domain convolution becomes frequency-domain multiplication, significantly reducing the computational burden from O(N²) to O(N log N) complexity

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Reliability

If multiple adaptive filters are applied in parallel to every frequency bin, then the acoustic echo cancellation effectiveness is improved, but the CPU usage and computational load increase significantly

Engineering Contradiction:
Improveacoustic echo cancellation effectivenessVSAvoidCPU usage
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent segments the frequency spectrum into multiple subbands and applies adaptive filtering only within each subband rather than across the entire frequency range. This segmentation reduces the number of adaptive filters needed from one per frequency bin to one per subband, significantly reducing CPU usage while maintaining cancellation effectiveness through localized adaptation

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies adaptive filtering selectively to specific subbands rather than uniformly to all frequency bins. By concentrating computational resources on subbands where echo cancellation is most needed and using shorter filter lengths, the system achieves sufficient cancellation effectiveness with reduced overall computational load

Inventive Principle:
Principle #16Partial or excessive action

3Productivity

If diagonal regularization of the correlation matrix is applied, then the computational efficiency and stability are improved, but the filtering accuracy may be reduced compared to full matrix inversion

Engineering Contradiction:
Improvecomputational efficiencyVSAvoidfiltering accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent modifies the correlation matrix by applying diagonal regularization, which changes the matrix parameters from full covariance to diagonally-loaded form. This parameter change adds a scaled identity matrix to the correlation matrix, ensuring numerical stability and enabling efficient inversion through closed-form solutions, while the scaling factor is optimized to maintain filtering accuracy

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent applies different processing strategies to different parts of the correlation matrix structure. By focusing on the diagonal elements and applying regularization specifically there, the system achieves computational efficiency without significantly degrading the overall filtering accuracy, as the off-diagonal elements (cross-correlations) are handled through the subband decomposition

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4708836A1Acoustic echo cancellation based on one or more diagonally regularized correlation matrices, and related devices, methods and computer programs
Publication Date: 2026.03.11 NOKIA TECHNOLOGIES OY
  • EP4708836A1 patent drawingFigure 1
  • EP4708836A1 patent drawingFigure 2~3
  • EP4708836A1 patent drawingFigure 4

AI summary

Devices, methods and computer programs for acoustic echo cancellation (AEC) based on one or more diagonally regularized correlation matrices are disclosed. At least some example embodiments may allow a smart and efficient solution for multi-channel and/or stereo AEC, providing improved performance and ease of implementation.