Oversampled Subband Adaptive Filtering for Faster Echo Cancellation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fullband adaptive filters face performance degradation with colored interfering signals, slow convergence, and increased computational requirements, especially in applications like acoustic echo cancellation, which demand long filters and high processing power, making them unsuitable for portable devices.
Innovation Solution
The method and system process subband signals using an oversampled weighted overlap-added (WOLA) filterbank, which transforms input signals into oversampled subband signals, whitens them through spectral emphasis or decimation, and employs adaptive filters to improve convergence rates and reduce computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fullband adaptive filters are used to model long return path responses, then cancellation performance is improved, but computational complexity and processing power requirements increase significantly
Solution Approach 1:
The fullband signal is divided into multiple frequency subbands using a filterbank. Each subband is processed independently by a separate adaptive filter, allowing the overall system to achieve long effective filter length while keeping individual filter complexities manageable. The segmentation of the frequency domain enables parallel processing of multiple shorter filters instead of one long filter.
2Reliability
If the length of the LMS filter is increased to model longer return paths, then cancellation performance is improved, but convergence rate decreases and computational requirements increase
Solution Approach 1:
By segmenting the fullband adaptive filter into multiple parallel subband adaptive filters, each operating at a lower sampling rate, the convergence rate is improved because each individual filter processes fewer samples per unit time while collectively achieving the same effective filtering length and performance.
Solution Approach 2:
The problem is transformed from the time domain to the frequency domain through subband decomposition. This dimensional change allows the system to achieve long filter lengths in the time domain by combining multiple shorter filters operating in parallel frequency subbands, thereby improving convergence while maintaining cancellation performance.
3Productivity
If subband signals are maximally decimated by critical sampling to reduce computational complexity, then processing speed is improved, but aliasing distortion is introduced
Solution Approach 1:
An oversampling factor is introduced as an intermediary between the critical sampling rate and the original sampling rate. This intermediate sampling rate allows for reduced computational complexity compared to fullband processing while avoiding the severe aliasing distortion that would result from critical sampling. The oversampled subband signals provide a compromise that maintains signal quality while improving processing efficiency.
4Reliability
If oversampled subband signals are used to avoid aliasing, then signal quality is improved, but computational complexity increases compared to critical sampling
Solution Approach 1:
The sampling rate parameter is optimized to an oversampled rate that is higher than the critical sampling rate but lower than the fullband rate. This parameter change achieves a balance where signal quality is preserved by avoiding critical sampling aliasing, while computational complexity is reduced compared to processing fullband signals. The specific oversampling factor is chosen to optimize the trade-off between these two competing requirements.
Data Source
AI summary
A method and system for processing subband signals using adaptive filters is provided. The system is implemented on an oversampled WOLA filterbank. Inputs signals are oversampled. The system includes an adaptive filter for each subband, and the functionality of improving the convergence properties of the adaptive filter. For example, the convergence property is improved by whitening the spectra of the oversampled subband signals and/or affine projection algorithm. The system is applicable to echo and/or noise cancellation. Adaptive step size control, adaptation process control using Double-Talk detector may be implemented. The system may further implement a non-adaptive processing for reducing uncorrelated noise and/or cross-talk resistant adaptive noise cancellation.