[Paper Review] Free Distance Bounds for Protograph-Based Regular LDPC Convolutional Codes

David G. M. Mitchell, Alí Emre Pusane+2 more|ArXiv.org|Apr 28, 2008

Error Correcting Code Techniques4 references4 citations

TL;DR

This paper establishes asymptotic lower bounds on the free distance to constraint length ratio for protograph-based regular LDPC convolutional codes using protograph weight enumerators and nonuniform cuts. It demonstrates that the free distance growth rate of these convolutional codes exceeds that of their block code counterparts, with new bounds showing significant improvement for previously unexplored (J,K) pairs, particularly via nonuniform cutting methods.

ABSTRACT

In this paper asymptotic methods are used to form lower bounds on the free distance to constraint length ratio of several ensembles of regular, asymptotically good, protograph-based LDPC convolutional codes. In particular, we show that the free distance to constraint length ratio of the regular LDPC convolutional codes exceeds that of the minimum distance to block length ratio of the corresponding LDPC block codes.

Motivation & Objective

• To derive lower bounds on the free distance to constraint length ratio for regular, asymptotically good, protograph-based LDPC convolutional code ensembles.
• To show that the free distance growth rate of these convolutional codes exceeds the minimum distance growth rate of corresponding LDPC block codes.
• To develop efficient methods for computing tight lower bounds on free distance growth rates, especially for (J,K) pairs not previously analyzed.
• To compare the effectiveness of mini-ensemble sampling and nonuniform cutting techniques in deriving improved free distance bounds.
• To validate that periodically time-varying, unterminated LDPC convolutional codes achieve higher free distance growth than their tail-biting and block code counterparts.

Proposed method

• Utilizes asymptotic analysis based on Divsalar's method for protograph-based LDPC block codes to derive free distance bounds for convolutional ensembles.
• Applies the copy-and-permute operation to generate protograph-based LDPC block code ensembles from a base protograph with n_v variable nodes and n_c check nodes.
• Employs nonuniform cuts of the protograph parity-check matrix to construct unterminated, periodically time-varying LDPC convolutional codes, with cutting vectors ξ chosen to maximize the lower bound on δ_free.
• Uses the relation δ_free ≥ (R/(1−R))λδ_min to compute lower bounds on the free distance growth rate, where δ_min is the minimum distance growth rate of the corresponding block code.
• For gcd(n_c, n_v) = 1, compares two unwrapping methods: random sampling of a 2^{n_v n_c}-member mini-ensemble and nonuniform cuts with specific cutting vectors.
• Validates that the free distance of a periodically time-varying convolutional code is lower bounded by the minimum distance of its tail-biting termination.

Experimental results

Research questions

• RQ1 Does the free distance growth rate of protograph-based regular LDPC convolutional codes exceed that of their corresponding LDPC block codes?
• RQ2 Can nonuniform cuts of the protograph parity-check matrix yield tighter lower bounds on the free distance growth rate than mini-ensemble sampling?
• RQ3 What are the achievable free distance growth rates for (J,K)-regular LDPC convolutional code ensembles with previously unexplored (J,K) pairs?
• RQ4 How do the bounds derived via asymptotic methods compare to known bounds such as the Gilbert-Varshamov and Costello bounds?
• RQ5 What is the relationship between the free distance of unterminated convolutional codes and the minimum distance of their tail-biting terminated versions?

Key findings

• For rate 1/2, (4,8)-regular LDPC convolutional codes, the lower bound on the free distance growth rate is δ_free ≥ 0.191, significantly exceeding the block code’s δ_min = 0.063.
• For the (3,5)-regular ensemble with rate 2/5, the nonuniform cut method yields δ_free ≥ 0.119, outperforming the mini-ensemble median of δ_free = 0.097.
• The (5,6)-regular ensemble achieves a lower bound of δ_free ≥ 0.317, which is substantially higher than its block code’s δ_min = 0.254.
• All investigated (J,K)-regular ensembles show that δ_free bounds exceed δ_min, confirming superior asymptotic distance growth in convolutional codes.
• Nonuniform cuts consistently produce tighter lower bounds than mini-ensemble sampling, especially for gcd(n_c, n_v) = 1 cases.
• The results confirm that the free distance to constraint length ratio of regular LDPC convolutional codes exceeds that of the corresponding LDPC block codes across all tested (J,K) pairs.