Graph model-aided optimal iterative decoding technique for LDPC in optical fiber communication.

Area of Science:

• Digital Communications
• Machine Learning for Signal Processing
• Error Correction Coding

Background:

• Neural networks enhance low-density parity-check (LDPC) decoding performance.
• Increasing code complexity leads to higher neural network computational demands.
• Existing methods require extensive feature extraction and large datasets.

Purpose of the Study:

• To introduce a novel iterative LDPC decoding technique, the graph model neural network-belief propagation (GMNN-BP).
• To leverage graph models to bridge deep learning and belief propagation (BP) for improved decoding.
• To reduce the reliance on direct codeword category learning and large training datasets.

Main Methods:

• Developed the GMNN-BP algorithm, integrating graph models with BP.
• Utilized graph models to link deep learning and BP, combining their strengths.
• Tested the GMNN-BP algorithm using IEEE 802.3ca standard LDPC codewords.

Main Results:

• GMNN-BP demonstrates superior performance compared to traditional BP-based iterative decoding.
• Achieved a maximum performance gain of 1.9dB under identical iteration counts.
• Required only half the number of iterations compared to other algorithms for equivalent performance.

Conclusions:

• GMNN-BP offers significant advantages over conventional neural network decoders.
• The proposed method reduces training data requirements and computational complexity.
• GMNN-BP represents an efficient and effective approach for advanced LDPC decoding.