Performance improvement methods for layered decoding of ldpc codes

Low-density parity-check (LDPC) codes are a special type of linear block codes. Although they were originally invented by R. Gallager in 1960, they have captured the attention of scientific community since its rediscovery in 1996. Due to their excellent correction capability, they have been adopted in many communications and storage systems for forward error correction. Gallager defined an iterative two-phase decoding algorithm referred to as "Belief Propagation" (BP). Much study has been conducted to improve the performance of LDPC decoding. Later, another scheduling, "Turbo Decoding Message Passing" (TDMP) is introduced as an alternative to his standard message passing algorithm. TDMP, also called layered decoding, has the advantage of converging faster than the standard BP because it uses more reliable information to update next set of values. Using layered decoding, it is possible to reduce the number of iterations by 50% without any performance degradation. In this thesis, we present several methods in order to improve layered decoding performance of LDPC codes. Methods proposed have been applied to several LDPC codes with different length and rate. According to our results, the biggest performance improvements are achieved when they are applied to small-length, high-rate codes. In addition to proposed methods, a simulation acceleration platform using OpenMP is also described where parallel decoding is implemented on a real multicore hardware platform, obtaining more than 6x speedup compared to single-core version. Keywords: Low-density parity-check, error correction codes, belief propagation, layered decoding, performance improvement, DSP, parallelization, OpenMP