A Non-Disjoint Group Shuffled Decoding for LDPC Codes

TL;DR

This paper introduces a novel grouping technique for group shuffled BP decoding of LDPC codes, significantly accelerating convergence without increasing computational complexity, as validated by analysis and experiments.

Contribution

A new grouping method for group shuffled BP decoding that improves convergence speed of LDPC decoders.

Findings

Faster convergence rate than conventional and existing group shuffled BP decoders.

Verification through Gaussian approximation analysis and numerical experiments.

Maintains the same computational complexity while enhancing decoding speed.

Abstract

To reduce the implementation complexity of a belief propagation (BP) based low-density parity-check (LDPC) decoder, shuffled BP decoding schedules, which serialize the decoding process by dividing a complete parallel message-passing iteration into a sequence of sub-iterations, have been proposed. The so-called group horizontal shuffled BP algorithm partitions the check nodes of the code graph into groups to perform group-by-group message-passing decoding. This paper proposes a new grouping technique to accelerate the message-passing rate. Performance of the proposed algorithm is analyzed by a Gaussian approximation approach. Both analysis and numerical experiments verify that the new algorithm does yield a convergence rate faster than that of existing conventional or group shuffled BP decoder with the same computing complexity constraint.