Variable Bit-Depth Increase for Video Coding Rounding Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fixed-point transforms in video coding introduce rounding errors due to cumulative bit loss during cascaded butterfly operations, which degrade the precision of transform output signals, especially as the size of the transform increases.
Innovation Solution
Applying a variable localized bit-depth increase by selecting a constant value based on the size of the fixed-point transform and adjusting the bit-depth of the input and output signals through left-shifting and right-shifting operations, respectively, to mitigate rounding errors and maintain higher precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If fixed-point transforms are used to reduce computational complexity, then device complexity is reduced, but manufacturing precision deteriorates due to cumulative rounding errors
Solution Approach 1:
The patent applies preliminary action by increasing the bit-depth of the input signal before the transform operation. This pre-processing step provides additional precision headroom that compensates for the rounding errors that will occur during the fixed-point transform computation, thereby maintaining output precision while keeping the transform computationally efficient.
Solution Approach 2:
The patent changes the parameter of bit-depth dynamically based on the transform size. Larger transforms that are more prone to cumulative rounding errors receive a greater bit-depth increase. This parameter adjustment optimizes the balance between computational complexity and output precision for different transform configurations.
2Productivity
If transform size is increased to improve compression efficiency, then productivity is improved, but manufacturing precision deteriorates due to increased cumulative rounding errors
Solution Approach 1:
The patent dynamically adjusts the bit-depth parameter based on transform size. As transform size increases to improve compression efficiency, the patent compensates by increasing the input bit-depth proportionally, thereby maintaining output precision despite the larger transform size and associated cumulative rounding errors.
Solution Approach 2:
Before performing large-scale transforms that improve compression efficiency, the patent preliminarily increases the bit-depth of input signals. This preparatory step ensures that sufficient precision is available throughout the computation chain, preventing the degradation of output precision that would otherwise occur with larger transforms.
3Manufacturing precision
If bit-depth is increased throughout the system to maintain precision, then manufacturing precision is improved, but loss of energy increases due to higher computational requirements
Solution Approach 1:
Instead of uniformly increasing bit-depth throughout the entire system, the patent applies bit-depth increase locally and selectively. The bit-depth is increased only at specific points in the processing chain (at the transform input) and only for specific transform sizes, thereby achieving the necessary precision improvement without the excessive energy cost of system-wide high-precision processing.
Solution Approach 2:
The patent changes the bit-depth parameter dynamically based on transform size requirements. Rather than using a fixed high bit-depth throughout the system, the patent adjusts the bit-depth parameter to match the actual precision needs of each transform operation, optimizing the trade-off between precision and computational energy consumption.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This disclosure describes techniques for mitigating rounding errors in a fixed- point transform associated with video coding by applying a variable localized bit-depth increase at the transform. More specifically, the techniques include selecting a constant value based on a size of a fixed-point transform in a video coding device and applying a variable localized bit-depth increase at the transform with a value equal to the constant value. Applying the variable localized bit-depth increase includes left-shifting a transform input signal by a number of bits equal to the constant value before the fixed- point transform, and right-shifting a transform output signal by a number of bits equal to the constant value after the fixed-point transform. The constant value is selected from a plurality of constant values stored on the video coding device. Each of the constant values is pre-calculated for one of a plurality of different transform sizes supported by the video coding.