Sparse Code Decoding for Short-Block Low-Complexity Communication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional channel codes struggle to achieve the Shannon limit for short block lengths, particularly in 5G/6G communication systems with limited power and high latency, due to high decoding complexity and power consumption, especially in environments like factory automation and THz wireless communication.
Innovation Solution
The development of sparse code and orthogonal sparse superposition code encoding and decoding methods that select index sets for information bits, allowing for efficient mapping and decoding of codewords, achieving both coding and power gains by minimizing computation and power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional channel codes (LDPC, polar, convolutional) are used to achieve Shannon limit, then coding performance is improved, but decoding complexity increases linearly with block length, making them unsuitable for short block lengths
Solution Approach 1:
The code block is segmented into multiple sections, and the decoding process is divided into sequential stages. Each stage decodes a portion of the codeword using simplified logic, avoiding the need for complex full-block decoding. This segmentation reduces per-stage computational complexity while maintaining overall coding performance.
Solution Approach 2:
Different decoding strategies are applied to different sections of the codeword based on their specific requirements. Early sections use simpler decoding logic with fewer iterations, while later sections may use enhanced logic to correct residual errors. This local optimization reduces average decoding complexity across the entire block.
2Reliability
If code block length is increased to achieve Shannon limit, then coding gain is improved, but computation amount of encoding and decoding increases linearly with block length
Solution Approach 1:
The encoding and decoding processes are segmented into multiple lightweight stages that process smaller portions of data sequentially. Each stage performs simplified computations rather than one complex full-block operation, reducing the total computational burden while achieving comparable coding gain through cumulative error correction.
Solution Approach 2:
The decoder performs a first pass with limited iterations to achieve basic error correction, then optionally performs additional iterations only for sections that require enhanced correction. This partial action approach achieves sufficient coding gain for most cases while avoiding the full computational cost of exhaustive decoding.
3Reliability
If Reed-Muller codes or Biothogonal codes are used for THz communication with very long code blocks, then Shannon limit is achieved, but decoding complexity does not have linear characteristics with respect to code block length
Solution Approach 1:
The long code block is divided into multiple manageable sections that can be decoded independently or semi-independently. Each section uses a standardized decoding template that scales efficiently, avoiding the exponential complexity growth associated with conventional Reed-Muller and Biothogonal code decoders. This segmentation enables linear scalability with block length.
Solution Approach 2:
The decoding complexity adapts dynamically to the actual channel conditions and error patterns observed. The decoder adjusts the number of iterations and processing depth based on received signal quality, performing minimal processing when conditions are good and scaling up only when necessary. This dynamic adaptation prevents unnecessary computational overhead while maintaining Shannon limit performance.
Data Source
AI summary
The inventive concept relates to method for encoding and decoding sparse codes and orthogonal sparse superposition codes. A sparse code encoding method which is to be performed by an encoding apparatus, according to an embodiment of the inventive concept may include selecting an index set that is a part of a code block by using an information bit, and mapping a codeword less than a preset size to the selected index set.