[Paper Review] Pseudocodewords of Tanner graphs

Christine A. Kelley, Deepak Sridhara|ArXiv.org|Apr 5, 2005

Error Correcting Code Techniques13 references4 citations

TL;DR

This paper provides a comprehensive analysis of pseudocodewords in Tanner graphs, distinguishing between those arising on computation trees and finite lifts. It establishes lower bounds on minimum pseudocodeword weight for the BEC, BSC, and AWGN channels, identifies graph structures that yield problematic pseudocodewords, and shows that redundant Tanner graph representations can significantly improve iterative decoding performance by increasing pseudocodeword minimum weight beyond the code's minimum distance.

ABSTRACT

This papers presents a detailed analysis of pseudocodewords of Tanner graphs. Pseudocodewords arising on the iterative decoder's computation tree are distinguished from pseudocodewords arising on finite degree lifts. Lower bounds on the minimum pseudocodeword weight are presented for the BEC, BSC, and AWGN channel. Some structural properties of pseudocodewords are examined, and pseudocodewords and graph properties that are potentially problematic with min-sum iterative decoding are identified. An upper bound on the minimum degree lift needed to realize a particular irreducible lift-realizable pseudocodeword is given in terms of its maximal component, and it is shown that all irreducible lift-realizable pseudocodewords have components upper bounded by a finite value $t$ that is dependent on the graph structure. Examples and different Tanner graph representations of individual codes are examined and the resulting pseudocodeword distributions and iterative decoding performances are analyzed. The results obtained provide some insights in relating the structure of the Tanner graph to the pseudocodeword distribution and suggest ways of designing Tanner graphs with good minimum pseudocodeword weight.

Motivation & Objective

• To understand the role of pseudocodewords in iterative decoding failure on finite-length LDPC codes.
• To distinguish between pseudocodewords arising on the decoder’s computation tree versus finite-degree lifts.
• To derive lower bounds on minimum pseudocodeword weight for key channels (BEC, BSC, AWGN).
• To identify structural graph properties that lead to poor iterative decoding performance.
• To show that redundant Tanner graph representations can improve pseudocodeword distribution and decoding performance.

Proposed method

• Analyzes pseudocodewords via the graph-covers-polytope definition, comparing it to actual computation tree behavior.
• Distinguishes between lift-realizable pseudocodewords and those arising on the computation tree, showing the polytope is incomplete for min-sum decoding.
• Uses irreducible pseudocodewords as fundamental units, since their weight bounds the weight of all composite pseudocodewords.
• Derives an upper bound on the minimum lift degree needed to realize a given irreducible pseudocodeword in terms of its maximal component.
• Establishes that all irreducible lift-realizable pseudocodewords have components bounded by a finite value t dependent on graph structure.
• Empirically evaluates different Tanner graph representations of [7,4,3] and [15,11,3] Hamming codes to compare pseudocodeword distributions and iterative decoding performance.

Experimental results

Research questions

• RQ1 How do pseudocodewords arising on the computation tree differ from those realizable via finite lifts in their impact on min-sum decoding?
• RQ2 What are the lower bounds on minimum pseudocodeword weight for the BEC, BSC, and AWGN channels?
• RQ3 Which structural features in a Tanner graph lead to the existence of low-weight, harmful pseudocodewords?
• RQ4 Can redundant Tanner graph representations improve pseudocodeword distribution and iterative decoding performance?
• RQ5 What is the maximum component size of any irreducible lift-realizable pseudocodeword, and how does it relate to graph structure?

Key findings

• The graph-covers-polytope definition of pseudocodewords is incomplete for min-sum iterative decoding, as it excludes some harmful pseudocodewords that appear on the computation tree.
• For the [7,4,3] Hamming code, representation B—featuring redundant check nodes—achieved performance identical to maximum-likelihood decoding on the BSC, correcting all one-bit error patterns.
• On the BIAWGNC, representation B of the [15,11,3] Hamming code showed no detected errors, while representation A had a high error rate, demonstrating the impact of pseudocodeword distribution on performance.
• Adding up to order-two redundant parity check equations (as in representation C) significantly reduced the number of low-weight pseudocodewords compared to the standard form (representation A).
• All irreducible lift-realizable pseudocodewords have components bounded above by a finite value t that depends on the Tanner graph structure.
• The minimum lift degree required to realize a given pseudocodeword is bounded above by a function of its maximal component.