Improved belief propagation based on the estimation backtracking

Abstract

We propose a general method to improve the decoding performance of hard-decision and soft-decision belief propagation decoders. The performance of the proposed algorithm is tested on MacKay's widely known LDPC (504,252) code, CCSDS LDPC (256,128) and our irregular LDPC (128,64) code. The bit error rate indicated by our novel solution is up to five times lower for a particular signal-to-noise ratio of tested codes measured through AWGN simulations. Furthermore, a decoding time required for the same bit error rate as the bit error rate of a traditional BP decoder is up to three times lower when our approach is used in a non-pipelined implementation.