Optimization of Irregular NB QC-LDPC Block Codes over Small Alphabets

Abstract

We propose a novel approach for optimization of nonbinary (NB) quasi-cyclic (QC)-LDPC codes. In this approach, the base parity-check matrices are constructed by the simulated annealing method, and then labeled while maximizing the so-called generalized girth of the NB LDPC code Tanner graph. Random coding bounds on the ML decoding error probability for ensembles of "almost regular" NB LDPC codes of finite lengths over extensions of the binary Galois field are derived. These bounds are based on the average bit weight spectra for the ensembles of NB LDPC codes. The observed FER performance of the sum-product BP decoding of "almost regular" NB QC-LDPC block codes is presented and compared to the finite-length random coding bounds, as well as to the performance of the optimized binary QC-LDPC block code in the 5G standard. In the waterfall region, the gap between the finite-length bounds on the error probability of the ML decoding and the simulation performance of the BP decoding is about 0.1 – 0.2 dB.