Chapter 18: Belief Propagation — The Sum-Product Algorithm

Learning Objectives

• Write the sum-product message updates for an arbitrary factor graph
• Decode LDPC codes using log-domain BP with min-sum approximation
• Derive and implement Gaussian belief propagation for jointly Gaussian models
• Distinguish sum-product (marginals) from max-product (MAP) and apply each correctly
• Use density evolution to predict the decoding threshold of an LDPC code ensemble