LDPC Min-Sum Decoding With Degree-Specific Neural Weights

Solution Overview

Problem

Existing LDPC decoders require significant memory and training to find optimal weights for edges in bipartite graphs, leading to inefficient performance.

Innovation Solution

Constrain neural networks to use the same weight for edges connecting nodes with the same degree, specifically check-to-variable and variable-to-check messages, reducing training requirements and improving decoding efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If neural networks are used to find optimal weights for edges in bipartite graphs, then decoding accuracy is improved, but memory requirements and training complexity increase significantly

Engineering Contradiction:

Improvedecoding accuracyVSAvoidtraining complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by assigning different weight values to edges based on the degrees of connected nodes. Specifically, edges connecting nodes with different degrees have different weights, while edges connecting nodes with the same degree share the same weight. This localized differentiation improves decoding accuracy by capturing structural properties of the bipartite graph without requiring full neural network complexity.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent applies homogeneity by constraining edges with the same node degree characteristics to share identical weight values. This reduces the number of independent parameters in the system, thereby reducing memory requirements and training complexity while maintaining the ability to capture important graph structural features for accurate decoding.

Inventive Principle:

Principle #33Homogeneity

2Measurement precision

If neural networks are used to find optimal weights for edges in bipartite graphs, then decoding accuracy is improved, but memory requirements increase significantly

Engineering Contradiction:

Improvedecoding accuracyVSAvoidmemory requirements

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies local quality by assigning different weight values to edges based on the degrees of connected nodes. Specifically, edges connecting nodes with different degrees have different weights, while edges connecting nodes with the same degree share the same weight. This localized differentiation improves decoding accuracy by capturing structural properties of the bipartite graph without requiring full neural network complexity.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent applies homogeneity by constraining edges with the same node degree characteristics to share identical weight values. This reduces the number of independent parameters in the system, thereby reducing memory requirements and training complexity while maintaining the ability to capture important graph structural features for accurate decoding.

Inventive Principle:

Principle #33Homogeneity

Data Source

PatentUS12574052B2Neural-network-optimized degree-specific weights for LDPC minsum decoding

Publication Date: 2026.03.10 CALIFORNIA INST OF TECH

AI summary

Neural Normalized MinSum (N-NMS) decoding for improving frame error rate (FER) performance on linear block codes over conventional normalized MinSum (NMS). Dynamic multiplicative weights are assigned to each check-to-variable message in each iteration to efficiently provide training parameters of N-NMS that support N-NMS for longer block lengths. Embodiment are described for neural two-dimensional normalized MinSum (N-2D-NMS) decoders requiring fewer training parameters. The N-2D-NMS approaches for example use the same weight for edges with the same check and/or variable node degree. Simulation results indicate that that this LDPC decoding performs similarly to previous techniques while substantially reducing the amount of training necessary.