FIR Filter Coefficient Update via Time-Varying Regularization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Adaptive learning techniques for Finite Impulse Response (FIR) filters in Acoustic Echo Cancellers (AECs) are sensitive to parameter initialization and fail to maintain performance stability, especially in 'double talk' scenarios with near-end non-stationary spoken speech, leading to increased residual echo and performance deterioration.

Innovation Solution

A method for sustainably updating the coefficient vector of FIR filters using a time-varying regularization factor, obtained from the power of microphone signals and effective estimation values of coupling factors, to adaptively update the filter coefficients iteratively, ensuring stability and reliability in signal processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If adaptive variable step-size learning techniques are used to update FIR filter coefficients, then the filter can adapt to changing conditions, but the system becomes sensitive to parameter initialization and fails in double talk situations

Engineering Contradiction:

Improveadaptability to changing conditionsVSAvoidperformance stability

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent introduces a time-varying regularization factor that dynamically changes based on signal conditions. This parameter change allows the filter to adapt to different scenarios (single talk, double talk, non-stationary speech) while maintaining stability. The regularization factor is adjusted according to the estimated near-end speech power and echo path coupling factor, enabling the system to switch between adaptive and stable modes as needed.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If Double Talk Detector (DTD) is used to detect near-end speech signals, then adaptive learning can be stopped to avoid divergence, but processing delay and misjudgment affect AEC performance

Engineering Contradiction:

Improveprevention of coefficient divergenceVSAvoidprocessing delay

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary estimation of near-end speech power and echo path coupling factor continuously, even before double talk detection is needed. This preliminary action prepares the system with necessary information so that when double talk occurs, the regularization factor can be immediately adjusted without waiting for DTD processing, thereby reducing detection delay and avoiding misjudgment effects.

Inventive Principle:

Principle #10Preliminary action

3Productivity

If continuous adaptive learning of filter coefficients is performed, then the filter can track echo path changes, but near-end speech causes coefficient deviation and divergence

Engineering Contradiction:

Improvetracking capability of echo path changesVSAvoidcoefficient accuracy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent makes the learning process dynamic by introducing a time-varying regularization factor that automatically adjusts the learning behavior. During single talk periods, the factor allows aggressive adaptation to track echo path changes. During double talk or non-stationary speech periods, the factor increases to constrain learning and prevent divergence. This dynamic adjustment resolves the contradiction between tracking capability and coefficient accuracy.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS11295752B2Method and device of sustainably updating coefficient vector of finite impulse response filter

Publication Date: 2022.04.05 CHINA ACAD OF TELECOMM TECH

AI summary

A method and a device of sustainably updating a coefficient vector of a finite impulse response FIRfilter. The method includes obtaining (21) a time-varying regularization factor used for iteratively updating the coefficient vector of the FIR filter in a case that the coefficient vector of the FIR filter is used for processing a preset signal; updating (22) the coefficient vector of the FIR filter according to the time-varying regularization factor.