Video Coding Motion Compensation Memory Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital service providers face increased costs and reduced scalability when transmitting high definition video due to the need for expensive high-bandwidth memory, which limits their ability to serve more subscribers with high definition services at a lower cost.
Innovation Solution
The solution involves reordering video data to utilize less expensive memory by extracting motion vectors from a video stream, sorting them in a different order than raster scan, and distributing memory accesses between high and low bandwidth memory, allowing for efficient spatial translation of reference video data into predicted video data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If high definition video is transmitted using conventional memory architecture, then video quality is maintained, but memory cost and processing complexity increase significantly
Solution Approach 1:
The reference picture buffer is divided into multiple regions (first region and second region) with different memory bandwidth characteristics. Motion estimation and compensation operations are segmented to access different regions based on their requirements, allowing high-bandwidth memory to be used only where necessary while reducing overall system complexity and cost.
2Productivity
If more expensive high-bandwidth memory is used to service more subscribers, then high definition service capability is improved, but operational cost increases
Solution Approach 1:
Different regions of the reference picture buffer are allocated with different memory bandwidth qualities. The first region uses high-bandwidth memory to support computationally intensive motion compensation operations, while the second region uses lower-bandwidth, lower-cost memory for operations requiring less bandwidth. This local differentiation allows the system to service more subscribers with high definition content while minimizing overall memory costs.
3Speed
If memory bandwidth is increased to handle high definition video processing, then processing speed is improved, but memory expense increases
Solution Approach 1:
The reference picture buffer is segmented into multiple regions with different bandwidth characteristics. Motion estimation and compensation operations are divided to access appropriate regions based on their bandwidth requirements. This segmentation allows high processing speed to be achieved only in regions where it is necessary, while reducing overall memory expense by using lower-bandwidth memory in other regions.
4Device complexity
If conventional memory access patterns are used for motion compensation, then implementation simplicity is maintained, but memory cost increases
Solution Approach 1:
The system implements different memory access patterns for different regions of the reference picture buffer. The first region uses access patterns optimized for high-bandwidth memory, while the second region uses patterns suitable for lower-bandwidth memory. This local differentiation maintains implementation simplicity within each region while reducing overall memory cost through strategic allocation.
Data Source
AI summary
A video coding system includes a decoder configured to decode an encoded reference picture, a primary memory, and a secondary memory configured to store the decoded reference picture. The video coding system also includes a logic device configured to extract motion vectors for generating a predicted picture of the reference picture and to store the extracted motion vectors in memory bins. The reference picture includes a plurality of predefined regions and each memory bin stores motion vectors for a particular region of the plurality of predefined regions. For each of the plurality of predefined regions of the reference picture, the logic device is configured to use the memory bin corresponding to a reference region to read the reference region from the secondary memory and write the reference region into the primary memory. The written reference region and the motion vectors of the corresponding memory bin are used to spatially translate the written reference region into the predicted picture.