Tiled Image Data Compression for Memory Bandwidth Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding techniques are not directly applicable to memory-data compression, as they are complex and offer high compression ratios, whereas memory-data compression requires simplicity and a constant compression ratio, and cannot exploit inter-frame correlation.
Innovation Solution
The proposed solution involves dividing frames into blocks, performing integer wavelet transformation on luminance and chrominance data, spatial prediction, quantization, and entropy encoding to achieve lossy or lossless compression, with a focus on reducing memory bandwidth requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If video compression algorithms (MPEG-2, MPEG-4, H.264/AVC) are used to achieve high compression ratios, then compression efficiency is improved, but algorithm complexity increases and real-time processing becomes difficult
Solution Approach 1:
The video frame is divided into multiple tiles, and each tile is independently compressed. This segmentation allows parallel processing of different tiles, reducing the computational complexity for real-time processing while maintaining overall compression efficiency. The independent compression of tiles enables simpler algorithms to be used in each tile without compromising the total compression ratio.
2Reliability
If decoded pictures are stored in external memory for prediction of subsequent pictures, then prediction accuracy is improved, but memory bandwidth requirements increase
Solution Approach 1:
The picture memory is divided into multiple picture memory banks, with each bank storing a specific tile of the decoded picture. This segmentation allows different tiles to be accessed from different memory banks simultaneously, enabling parallel memory access and reducing the bottleneck on memory bandwidth while maintaining prediction accuracy through complete picture storage.
Solution Approach 2:
The patent introduces a temporal dimension to memory access by storing multiple tiles in different memory banks that can be accessed at different times. This allows the system to fetch required tiles for prediction without requiring all tiles to be available simultaneously in a single memory location, thereby reducing peak memory bandwidth requirements.
3Manufacturing precision
If high-resolution video (D1 or higher) is processed in real-time, then video quality is improved, but memory bandwidth requirements exceed available bandwidth
Solution Approach 1:
The high-resolution video frame is divided into multiple tiles that can be processed and stored independently in different memory banks. This allows the memory system to handle high-resolution video by distributing the data access load across multiple banks, preventing any single memory interface from becoming a bandwidth bottleneck while maintaining overall video quality.
Solution Approach 2:
The patent implements dynamic tile-based processing where the compression and memory access operations are adapted based on the specific tile being processed. This dynamic approach allows the system to optimize memory bandwidth usage for high-resolution video by processing tiles in an order and manner that matches the available memory bandwidth, enabling real-time processing of D1 or higher resolutions.
Data Source
AI summary
A method of compressing digital image data is provided that includes, for each image data block in a plurality of image data blocks in the digital image data, transforming image data in the image data block to convert the image data to a low-frequency coefficient and a plurality of high-frequency coefficients, computing a predicted low-frequency coefficient for the image data block based on at least one neighboring image data block in the plurality of image data blocks, computing a residual low-frequency coefficient based on the predicted low-frequency coefficient and the low-frequency coefficient, quantizing the plurality of high-frequency coefficients, and entropy coding the residual low-frequency coefficient and the quantized high-frequency coefficients.