Channel-Parallel Bitstream Packing for Sparse Data Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multichannel data compression and decompression algorithms are serial processes that handle only one data bit stream at a time, inefficiently handling parallel data storage and retrieval in neural inference accelerator hardware, which leads to high power consumption and storage inefficiencies due to many zeros in the data.
Innovation Solution
A multichannel data packer and unpacker system using a butterfly shuffler with 2-to-1 multiplexers arranged in a specific configuration to permute and pack/unpack bit streams in parallel, allowing for equal bit-stream lengths and random access to packed data, effectively removing zeros and optimizing storage and retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If serial compression algorithms are used to handle one data bit stream at a time, then implementation simplicity is maintained, but processing speed and power efficiency deteriorate due to inability to handle parallel data streams
Solution Approach 1:
The patent divides the parallel compression task into multiple independent serial compression units, each handling one channel. These units are organized in a parallel architecture where each unit independently compresses one bit stream using traditional serial algorithms, allowing simultaneous processing of multiple channels while maintaining algorithm simplicity.
Solution Approach 2:
The patent combines multiple serial compression outputs into a single packed bit stream by merging the compressed data from all channels. The merging process aligns and concatenates the individual channel outputs, creating a unified compressed representation that maintains the benefits of parallel processing while producing a compact output suitable for storage or transmission.
2Productivity
If parallel compression is implemented to process multiple channels simultaneously, then processing efficiency improves, but hardware complexity increases due to need for multiple compression units
Solution Approach 1:
The parallel compression system is segmented into identical modular compression units, each handling one channel independently. This segmentation allows the system to achieve parallel processing throughput while keeping each individual unit simple and standardized, making the overall hardware implementation more manageable despite the increased number of units.
Solution Approach 2:
The patent transitions from serial processing (one-dimensional time sequence) to parallel processing (multi-dimensional channel space). By organizing compression units across multiple channels simultaneously, the system achieves higher throughput by utilizing the channel dimension, effectively converting a time-sequential operation into a spatially-parallel operation.
3Quantity of substance
If data is compressed to remove zeros, then storage requirements and power consumption reduce, but access time increases due to loss of original data structure
Solution Approach 1:
The patent performs preliminary organization of compressed data by channel and position before final packing. Each channel's compressed data is pre-aligned and positioned in the packed stream according to its original spatial relationships, enabling faster random access by preserving the structural information needed for efficient data retrieval without requiring full decompression.
Solution Approach 2:
The packed bit stream structure acts as an intermediary representation that maintains selective accessibility. By organizing compressed data in a structured packed format that preserves channel identities and relative positions, the system enables partial or selective access to specific channels or data regions without requiring complete decompression of the entire stream.
4Quantity of substance
If variable-length bit streams are used to represent different channels, then data compression efficiency improves, but alignment and packing difficulty increases
Solution Approach 1:
The patent changes the parameter of bit stream length from variable to fixed by padding shorter channels with zero bits to match the length of the longest channel. This parameter transformation simplifies the packing process by ensuring all channels have uniform length, eliminating the need for complex variable-length handling logic while maintaining compression efficiency through selective zero-padding only where necessary.
Data Source
AI summary
A multichannel data packer includes a plurality of two-input multiplexers and a controller. The plurality of two-input multiplexers is arranged in 2N rows and N columns in which N is an integer greater than 1. Each input of a multiplexer in a first column receives a respective bit stream of 2N channels of bit streams. Each respective bit stream includes a bit-stream length based on data in the bit stream. The multiplexers in a last column output 2N channels of packed bit streams each having a same bit-stream length. The controller controls the plurality of multiplexers so that the multiplexers in the last column output the 2N channels of bit streams that each has the same bit-stream length.