LDPC Decoding with Partial Residual Clustering for Faster Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The conventional dynamic scheduling decoding algorithm for LDPC codes performs global searches, leading to inefficient resource allocation and error floors due to its greedy nature and ring structures, which limits decoding performance and increases computing resources.
Innovation Solution
The proposed decoding method employs partial average residual belief propagation (PARBP), which calculates cluster sizes and pre-computes edge residuals to prioritize updates within clusters, reducing spatial search and minimizing error floors by focusing on the largest edge residuals within each cluster.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If global search is performed to find the largest edge residual, then decoding performance is improved, but computing resources are excessively consumed
Solution Approach 1:
The patent divides the global search space into local search spaces by partitioning edges into clusters. Instead of searching all edges globally, the algorithm performs localized searches within each cluster to find the largest edge residual, significantly reducing computational complexity while maintaining decoding performance.
2Speed
If greedy update strategy is used to prioritize largest edge residuals, then convergence speed is improved, but error floors occur due to ring structures
Solution Approach 1:
The patent segments edges into clusters to break ring structures that cause error floors in greedy algorithms. By performing localized searches within clusters rather than global searches, the algorithm avoids getting trapped in local optima caused by ring structures while maintaining fast convergence through prioritized updates.
3Ease of operation
If uniform update order is used across all edges, then implementation is simplified, but decoding efficiency is reduced
Solution Approach 1:
The patent implements a dynamic update order within each cluster based on edge residual values. The algorithm dynamically selects which edge to update next by comparing residuals, allowing flexible adaptation to the current decoding state while maintaining systematic organization through cluster-based structure.
Data Source
AI summary
A decoding method of low-density parity-check (LDPC) codes based on partial average residual belief propagation includes the following steps: S1: calculating a size of a cluster π in a protograph based on a code length m and a code rate of a target codeword; S2: pre-computing an edge residual rc<sub2>i</sub2>→v<sub2>j </sub2>corresponding to each edge from a variable node to a check node in a check matrix H; S3: calculating, based on π, a partial average residual (PAR) value corresponding to each cluster in the check matrix H; S4: sorting m/π clusters in descending order of corresponding PAR values, and updating an edge with a largest edge residual in each cluster; S5: updating edge information mc<sub2>i</sub2>→v<sub2>i </sub2>from a check node ci to a variable node vj, and then updating a log-likelihood ratio (LLR) value L(vj) of the variable node vj; and S6: after the updating, making a decoding decision.