Finite alphabet iterative decoders approaching maximum likelihood performance on the Binary Symmetric Channel

We introduce a generic approach for improving the guaranteed error correction capability of regular low-density parity check codes. The method relies on operating (in serial or in parallel) a set of finite alphabet iterative decoders. The message passing update rules are judiciously chosen to ensure that decoders have different dynamics on a specific finite-length code. The idea is that for the Binary Symmetric Channel, if some error pattern cannot be corrected by one particular decoder, there exists in the set of decoders, another decoder which can correct this pattern. We show how to select a plurality of message update rules so that the set of decoders can collectively correct error patterns on the dominant trapping sets. We also show that a set of decoders with dynamic re-initializations can approach the performance of maximum likelihood decoding for finite-length regular column-weight three codes.