Experiments on Decoding LDPC Codes Using Trees and Random Forests

This paper explores two alternatives to the standard message passing approach to decoding LDPC codes: regression trees and random forests. Regression trees are capable of approximating complicated functions. This paper reports on the performance of a tree-based low-density parity-check code (LDPC) decoder in which a regression tree is trained to approximate the message passing update function. This approach offers potential advantages over the message passing decoder. Single trees as well as groups of trees (i.e. forests) produce their outputs using only threshold comparisons. Therefore, the proposed decoders are arithmetic free.