Bounds and New Constructions for Girth-Constrained Regular Bipartite Graphs

TL;DR

This paper investigates bounds and novel constructions for regular bipartite graphs with girth constraints, aiming to improve LDPC code design by balancing girth, node set sizes, and sparsity.

Contribution

It derives bounds on vertex set sizes related to girth and introduces two new constructions for girth 8 bipartite graphs, including an asymptotically optimal semi-regular graph.

Findings

Bounds on check node growth with girth increase

Two girth-8 bipartite graph constructions

Sparse parity-check matrices for high-rate codes

Abstract

In this paper, we explore the design and analysis of regular bipartite graphs motivated by their application in low-density parity-check (LDPC) codes specifically with constrained girth and in the high-rate regime. We focus on the relation between the girth of the graph, and the size of the sets of variable and check nodes. We derive bounds on the size of the vertices in regular bipartite graphs, showing how the required number of check nodes grows with respect to the number of variable nodes as girth grows large. Furthermore, we present two constructions for bipartite graphs with girth $\mathcal{G} = 8$; one based on a greedy construction of $(w_c, w_r)$-regular graphs, and another based on semi-regular graphs which have uniform column weight distribution with a sublinear number of check nodes. The second construction leverages sequences of integers without any length-$3$ arithmetic progression and is asymptotically optimal while maintaining a girth of $8$. Also, both constructions can offer sparse parity-check matrices for high-rate codes with medium-to-large block lengths. Our results solely focus on the graph-theoretic problem but can potentially contribute to the ongoing effort to design LDPC codes with high girth and minimum distance, specifically in high code rates.