Diversity of non-binary cluster-LDPC codes using the EMS algorithm

We propose a new decoding procedure for non-binary cluster-LDPC codes based on the observation that different non-binary Tanner graphs for the same binary code may result in different convergence behaviours. We apply this concept locally at the check nodes and use multiple instances of a check node in parallel. Their outputs are joined together to get a single extrinsic message for each edge of the node. Each instance uses the famous Extended-Min-Sum (EMS) algorithm with very low complexity. We are able to obtain better decoding performance in the error-floor region with equal or lower complexity. We obtain the results via Mont-Carlo simulations and we make a comparison of the complexities as well.