Uniformly reweighted APP decoder for memory efficient decoding of LDPC Codes

Abstract

In this paper we propose a uniformly reweighted a posteriori probability (APP) decoder. The APP decoder is well-known to be suboptimal compared to the BP decoder. Here, we derive the APP decoder as an algorithm of approximate Bayesian inference on the LDPC code graph and introduce a correction parameter to overcome the suboptimaly of the APP decoder. We optimize numerically the correction parameter and show that it improves the BER performance of the APP decoder compared to its non-corrected version. In addition, the original APP decoder requires memory that is linear in the number of edges in the code graph. Here, we propose a memory efficient implementation of the algorithm that requires memory that is linear only in the codeword length.