Adaptive Filter Coefficient Update for Stereo Audio Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for updating filter coefficients in adaptive filters for multi-channel signal encoding require a large number of operations, leading to increased computational load, especially when the filter order is high, such as 3200, which is necessary for efficient encoding of stereo audio signals.
Innovation Solution
The proposed solution involves determining a specific update order range for filter coefficient updates based on the cross-correlation function between channel signals, allowing only the filter coefficients in important sections to be updated, thereby reducing the computational load while maintaining prediction performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the filter order is increased to represent acoustic characteristics for sufficient time length (e.g., N=3200 for 100ms at 32kHz sampling), then the encoding performance and accuracy of acoustic characteristic representation is improved, but the computational load for updating filter coefficients increases dramatically
Solution Approach 1:
The patent divides the filter coefficient update process into two distinct phases: an initial learning phase where all N filter coefficients are updated to capture complete acoustic characteristics, and a subsequent maintenance phase where only a limited number of coefficients (e.g., 32 out of 3200) are updated periodically. This segmentation allows the system to achieve high representation accuracy through initial comprehensive learning while reducing computational load during normal operation by updating only essential coefficients at reduced intervals.
2Reliability
If the filter coefficient update frequency is increased to maintain accurate acoustic characteristic tracking, then the prediction performance is improved, but the amount of operations required increases significantly
Solution Approach 1:
The patent implements periodic updates of filter coefficients at predetermined intervals rather than continuous updates at every encoding step. During the learning phase, updates occur frequently to establish accurate acoustic characteristics. Once learning is complete, updates are performed periodically with reduced frequency (e.g., every M steps where M>1), maintaining reliable prediction performance while significantly improving encoding efficiency by reducing the total number of operations.
3Measurement precision
If all filter coefficients are updated at every encoding step to maintain optimal prediction, then the prediction accuracy is improved, but the amount of update operations becomes prohibitively large
Solution Approach 1:
The patent applies partial action by updating only a necessary subset of filter coefficients rather than all coefficients at every step. After the initial learning phase where all coefficients are established, the system updates only the most critical coefficients (those with largest magnitudes or greatest impact on prediction accuracy) at reduced frequency. This partial update approach maintains sufficient prediction accuracy while dramatically reducing processing time and computational resources required.
Data Source
AI summary
An encoding device, a decoding device, and encoding and decoding methods are provided, wherein when a multi-channel signal is encoded with high efficiency, using an adaptive filter, the number of arithmetic operations to update a filter coefficient of the adaptive filter can be reduced. An update range determination unit determines the range of a filter coefficient order (update order range) of a filter coefficient to be updated, among filter coefficients gk(n) of the adaptive filter, on the basis of a mutual correlation function between an input (L) signal and an input (R) signal. The adaptive filter updates the filter coefficient gk(n) of the filter coefficient order (n) to be updated, using a decoding (L) signal and a decoding error (R) signal.