Improving belief propagation on graphs with cycles

• Published in: IEEE communications letters Vol. 8; no. 1; pp. 57 - 59
• Main Authors: Yazdani, M.R., Hemati, S., Banihashemi, A.H.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.01.2004; Institute of Electrical and Electronics Engineers; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Additives; Algorithms; Applied sciences; Associate members; Back propagation; Belief propagation; Bipartite graph; Block codes; Coding, codes; Computer simulation; Decoding; Digital communication; Errors; Exact sciences and technology; Graphs; Information, signal and communications theory; Iterative algorithms; Iterative decoding; Maximum likelihood decoding; Messages; Parity check codes; Processor scheduling; Signal and communications theory; Telecommunications and information theory; Performance evaluation; iterative decoding; low-density parity-check (LDPC) codes; offset belief propagation; Signal processing; Iterative method; message-passing decoding algorithms; Decoding; Bipartite graph; Algorithm; normalized belief propagation; Belief propagation
• ISSN: 1089-7798; 1558-2558
• DOI: 10.1109/LCOMM.2003.822499

In this letter, we propose two modifications to belief propagation (BP) decoding algorithm. The modifications are based on reducing the reliability of messages throughout the iteration process, and are particularly effective for short low-density parity-check codes, where the existence of cycles makes the original BP algorithm perform suboptimal. The proposed algorithms, referred to as "normalized BP" and "offset BP," reduce the absolute value of the outgoing log-likelihood ratio messages at variable nodes by using a multiplicative factor and an additive factor, respectively. Simulation results show that both algorithms perform more or less the same, and both outperform BP in error performance.