Acoustic Echo Suppression via Spatial Mixing Matrix and Frequency Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Full-duplex systems with multi-channel audio reproduction face high computational complexity and numerical instability in adaptive filtering due to correlated input signals, particularly in acoustic echo cancellation, where existing methods struggle to effectively suppress echoes below the spatial aliasing frequency without affecting human perception.
Innovation Solution
The method involves defining a mixing matrix to suppress spectral components below the spatial aliasing frequency, using a combination of spatial and temporal preprocessing to create quiet zones, and employing Multichannel Frequency-Domain Adaptive Filtering (MC-FDAF) for efficient echo reduction, with phase modulation above the aliasing frequency to decorrelate loudspeaker signals and identify echo paths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If adaptive filtering is used for acoustic echo cancellation in multi-channel systems, then echo suppression is achieved, but computational complexity increases due to P×Q signal paths requiring several thousand filter coefficients
Solution Approach 1:
The patent divides the frequency spectrum into multiple bands using filter banks, processing each frequency band separately rather than handling all P×Q channels simultaneously across the entire spectrum. This segmentation reduces the computational burden by breaking down the complex multi-channel problem into simpler parallel sub-problems that can be processed independently.
Solution Approach 2:
The patent replaces traditional time-domain adaptive filtering with frequency-domain processing using filter banks and spectral analysis. This substitution transforms the computational approach from direct time-domain convolution to更高效 frequency-domain operations, reducing the number of required filter coefficients and computational operations.
2Object-affected harmful factors
If adaptive filtering is used for acoustic echo cancellation, then echo suppression is achieved, but numerical instability occurs due to correlated input signals from few independent sources
Solution Approach 1:
By segmenting the signal processing into independent frequency bands using filter banks, the patent ensures that correlation issues in one band do not propagate to other bands. Each frequency band is processed independently with its own adaptive filter, isolating numerical instability problems to localized frequency regions rather than affecting the entire spectrum.
Solution Approach 2:
The patent transforms the signal representation from time domain to frequency domain parameters using spectral analysis. This parameter transformation changes the mathematical properties of the input signals, reducing correlation effects that cause numerical instability in time-domain adaptive filtering while maintaining the ability to suppress echoes effectively.
3Productivity
If psychoacoustically motivated phase modulation is used for signal decorrelation, then adaptation efficiency is improved, but perceptual quality is affected in the lower frequency range below two kHz
Solution Approach 1:
The patent applies different processing strategies to different frequency regions: frequency-domain filtering and adaptive processing are applied primarily to higher frequency bands where phase modulation has less perceptual impact, while lower frequency bands below 2 kHz are handled with care to preserve perceptual quality. This localized approach allows adaptation efficiency improvements in frequencies where it matters most while protecting perceptually critical low frequencies.
Solution Approach 2:
The patent uses frequency-selective parameter transformation, applying frequency-domain processing parameters selectively to different frequency bands based on their perceptual importance and correlation characteristics. This allows optimization of adaptation efficiency in less critical frequency regions while maintaining natural perception in the most important low frequency range.
Data Source
Figure 1~2
AI summary
The method involves defining a mixing matrix, and mapping the source signals from a remote chamber (1) to specific loudspeaker signals, such that a wave field in a close space (2) is generated. The wave field in specific zone and/or in direction of a certain near-space is suppressed, and a receiving system is placed in the specific zone and/or in specific direction. An independent claim is included for device for avoiding acoustic echo in full duplex systems.