A Class of LDPC Erasure Distributions with Closed-Form Threshold Expression

TL;DR

This paper introduces a family of LDPC degree distributions called p-positive distributions, which have a simple closed-form decoding threshold on the BEC, can closely match optimal thresholds, and achieve capacity.

Contribution

The paper defines p-positive LDPC distributions with a closed-form threshold expression, demonstrating their near-optimality and capacity-achieving properties on the BEC.

Findings

Threshold expression is simply [λ'(0)ρ'(1)]^{-1}.

p-positive distributions can closely match the best known thresholds.

Binomial degree distributions are part of the p-positive family.

Abstract

In this paper, a family of low-density parity-check (LDPC) degree distributions, whose decoding threshold on the binary erasure channel (BEC) admits a simple closed form, is presented. These degree distributions are a subset of the check regular distributions (i.e. all the check nodes have the same degree), and are referred to as $p$-positive distributions. It is given proof that the threshold for a $p$-positive distribution is simply expressed by $[\lambda'(0)\rho'(1)]^{-1}$. Besides this closed form threshold expression, the $p$-positive distributions exhibit three additional properties. First, for given code rate, check degree and maximum variable degree, they are in some cases characterized by a threshold which is extremely close to that of the best known check regular distributions, under the same set of constraints. Second, the threshold optimization problem within the $p$-positive class can be solved in some cases with analytic methods, without using any numerical optimization tool. Third, these distributions can achieve the BEC capacity. The last property is shown by proving that the well-known binomial degree distributions belong to the $p$-positive family.