Audio Encoding Threshold Control for Transform Coefficient Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting transform coefficients in sound coding require a large amount of calculation, especially when dealing with complex high-band signals, leading to inefficiencies due to the need for numerous branching processes and the difficulty in setting statistically reliable thresholds for extracting a desired number of coefficients.
Innovation Solution
A coding apparatus that dynamically calculates and updates a threshold based on statistical parameters of extension-band transform coefficients, gradually lowering the threshold to extract a predetermined number of coefficients with reduced loops and calculation, ensuring accurate extraction without significant loss of coding quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all transform coefficients in the extension-band subband are used to calculate the correlation value, then the accuracy of correlation calculation is improved, but the amount of calculation increases significantly
Solution Approach 1:
The patent extracts only the N transform coefficients with the largest absolute values from the extension-band subband to calculate the correlation value. This extraction approach maintains adequate correlation calculation accuracy while significantly reducing the number of coefficients processed, thereby reducing the computational power required.
Solution Approach 2:
Instead of using all M transform coefficients in the extension-band subband, the patent uses only a partial subset of N coefficients (where N < M) that have the largest absolute values. This partial action approach provides a good balance between calculation accuracy and computational efficiency.
2Power
If the number of transform coefficients used for correlation calculation is reduced, then the calculation amount is decreased, but the accuracy of correlation value may be compromised
Solution Approach 1:
The patent applies local quality by selecting transform coefficients based on their individual characteristics (largest absolute values) rather than uniformly processing all coefficients. This localized selection ensures that the most significant coefficients are used for correlation calculation, maintaining accuracy while reducing computation.
3Ease of manufacture
If a fixed threshold is used to extract transform coefficients, then the extraction process is simplified, but it becomes difficult to extract a desired number of coefficients accurately
Solution Approach 1:
The patent employs a dynamic sorting and selection process where transform coefficients are ordered by their absolute values and the top N coefficients are selected. This dynamic approach adapts to the actual distribution of coefficient values, ensuring that exactly N coefficients are extracted with high accuracy while maintaining relative process simplicity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Provided is an encoding apparatus. A threshold value calculating unit (32) calculates a threshold value from a statistical amount of conversion factors of an extended band. A representative conversion factor extracting unit (33) uses the calculated threshold value to extract conversion factors having large amplitudes. If the number of extracted conversion factors does not reach a specified number, the threshold value calculating unit (32) determines, in accordance with a lacking number of conversion factors, an amount by which the threshold value should be lowered, and modifies the threshold value accordingly. The representative conversion factor extracting unit (33) uses the threshold value, which has been modified, to extract conversion factors. Such threshold value modification by the threshold value difference calculating unit (32) and such conversion factor extraction by the representative conversion factor extracting unit (33) are repeated until the number of extracted conversion factors reaches the specified number.