Robust initial LLRs for iterative decoders in presence of non-Gaussian noise

We consider the decoding of LDPC codes in presence of non-Gaussian noise, especially a set of ∈-mixture models. For each of these models, the optimal LLRs are presented. We study the performance degradation due to the use of incorrect LLR in presence of a given noise model. Without modifying the existing LDPC decoder, we propose robust initial LLR which require minimum knowledge about the underlying noise model and are computationally less complex. Since BER simulations are computationally heavy, we use density evolution to compare the thresholds of different LLRs.