Motion Compensation Apparatus Memory Transfer Overhead Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion compensation methods face inefficiencies in memory transfer due to the inability to batch transfer reference pixel data for small blocks without prior knowledge of their source positions, limiting the reduction of memory transfer overhead.
Innovation Solution
A method that performs motion compensation by specifying and reading out reference pixel data for each block and sub-block using their respective motion vectors, allowing for efficient transfer and storage of prediction pixel data, and temporarily storing sub-block data for later output.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion compensation is performed on small blocks, then prediction accuracy is improved, but memory transfer overhead increases
Solution Approach 1:
The patent combines multiple small blocks into a larger block unit for motion compensation processing. By treating multiple small blocks as a single unit with shared motion vectors, the system achieves the prediction accuracy benefits of small block processing while reducing memory transfer overhead through batched operations on larger units.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating and storing motion vectors for larger blocks that encompass multiple small blocks. This allows the motion compensation unit to access pre-computed motion information without repeatedly transferring reference pixel data for each small block individually, thereby reducing memory transfer overhead while maintaining prediction accuracy.
2Measurement precision
If reference pixel data is read out for each small block separately, then prediction accuracy is maintained, but memory transfer efficiency decreases
Solution Approach 1:
The patent merges the reading operations for reference pixel data by processing multiple small blocks as a single unit. Instead of reading reference data separately for each small block, the system reads reference data once for the larger block unit and then distributes it to the constituent small blocks, significantly improving memory transfer efficiency while maintaining prediction accuracy through shared motion vector information.
Solution Approach 2:
The patent creates a universal approach where a single motion vector and reference pixel data set serve multiple small blocks within a larger block unit. This multi-functional use of reference data allows the same read operation to support multiple prediction operations, improving overall memory transfer efficiency without sacrificing the prediction accuracy that would require separate readings for each small block.
Data Source
AI summary
A motion compensation method includes: specifying, using a motion vector of a current block, a reference image area for use in motion compensation for each of the current block and a current sub-block spatially adjacent to the current block, to read out reference pixel data of the specified area from a frame memory; performing motion compensation using the reference pixel data read out in the readout, to generate prediction pixel data for each of the current block and the current sub-block; outputting the prediction pixel data for the current block, generated in the performing of motion compensation; and temporarily storing the prediction pixel data for the current sub-block, generated in the performing of motion compensation.