Some problems of Graph Based Codes for Belief Propagation decoding

TL;DR

This paper surveys graph-based codes like LDPC, focusing on properties affecting belief propagation decoding, and introduces spectral methods for analyzing trapping sets to improve decoding reliability.

Contribution

It presents fast spectral methods to analyze trapping sets and pseudocodewords, aiding in understanding and improving belief propagation decoding of graph-based codes.

Findings

Spectral classification simplifies trapping set analysis.

Methods relate minimal trapping set weight to code distance.

Challenges for improving decoding of sparse and dense codes are discussed.

Abstract

Short survey about code on the graph by example of hardware friendly quasi-cycle LDPC code. We consider two main properties of code: weight enumerator (well known from classic code theory) and Trapping sets pseudocodewords weight spectrum (a subgraph in code graph's which become reasone of decoding failure under Belief propagation). In paper we show fast methods to measure first components of TS enumerator using EMD=ACE constrains for high girth codes on the graph using graph spectral classification method. This approach simplify solving trouble of agreed between minimal TS pseudocode weight and code distance (which can be measure using knowledge of Authomorphism for algebraic code design methods or measure using lattice reduction(LLL,BKZ) for random graph design methods). In the end of article author raise several challenge problems to improve Sparse and Dense parity-check codes decoding under Belief propagation.