Decoder-Side Motion Refinement With Reference Block Prefetch
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Video coding systems face high memory-access latency due to the time-consuming process of fetching reference blocks from off-chip memory for decoder-side motion estimation and compensation, which hinders high-throughput decoder implementations.
Innovation Solution
Implement a two-pass motion-data reconstruction process, decoupling decoder-side motion estimation from direct memory access by using a coarse motion-data reconstruction block to fetch all required reference blocks before decoder-side motion estimation, allowing for pre-fetching and reducing memory latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reference blocks are fetched from off-chip memory for decoder-side motion estimation and compensation, then motion compensation can be performed, but memory-access latency increases
Solution Approach 1:
The patent applies preliminary action by performing a first pass of motion-data reconstruction to obtain coarse motion vectors before the actual decoder-side motion estimation. This allows the system to pre-identify and prefetch reference blocks to an on-chip buffer before they are needed for the second pass motion estimation, thereby eliminating memory-access latency during the critical computation phase.
2Measurement precision
If decoder-side motion estimation is performed, then motion vector accuracy is improved, but processing time increases due to sequential operations
Solution Approach 1:
The patent segments the motion-data reconstruction process into two distinct passes: a first pass that generates coarse motion vectors and identifies reference block locations, and a second pass that performs precise decoder-side motion estimation. This segmentation allows the first pass to prepare data in advance, enabling the second pass to operate with higher accuracy without sequential bottlenecks, thereby improving overall decoder throughput.
Solution Approach 2:
The first pass of motion-data reconstruction serves as a preliminary action that prepares coarse motion vectors and reference block identifiers before the second pass. This preliminary computation enables the system to prefetch necessary data and organize it in on-chip buffers, allowing the second pass to execute faster with improved accuracy without creating sequential dependencies that would reduce throughput.
3Loss of time
If all reference blocks are pre-fetched before decoder-side motion estimation, then memory latency is reduced, but memory bandwidth consumption increases
Solution Approach 1:
The patent uses preliminary action by performing motion-data reconstruction in a first pass to obtain coarse motion vectors and identify specific reference block locations. This identification step enables selective prefetching of only the necessary reference blocks to on-chip buffers before the second pass motion estimation, rather than fetching all possible reference blocks. This approach reduces memory latency while controlling memory bandwidth consumption through targeted data retrieval.
Data Source
AI summary
A system includes memory and at least one processor coupled to the memory. The processor processes a received bitstream to generate quantized data and control data. The process also generates decoded motion data based on a portion of the control data, fetches one or more reference blocks associated with a current prediction unit of a decoder pipeline region based on the decoded motion data and generates refined motion data based on the decoded motion data and the one or more reference blocks. The processor further generates one or more inter-prediction blocks based on the refined motion data and the one or more reference blocks by performing a motion compensation operation.