Øyvind Ytrehus

Generalized Hamming weights of linear codes

The generalized Hamming weight, d/sub r/(C), of a binary linear code C is the size of the smallest support of any r-dimensional subcode of C. The parameter d/sub r/(C) determines the codes performance on the wire-tap channel of Type II. Bounds on d/sub r/(C), and in some cases exact expressions, are derived. In particular, a generalized Griesmer bound for d/sub r/(C) is presented and examples are given of codes meeting this bound with equality. >

Coding and Cryptography

Small print The syllabus for the course is defined by the Faculty Board Schedules (which are minimal for lecturing and maximal for examining). I should very much appreciate being told of any corrections or possible improvements and might even part with a small reward to the first finder of particular errors. This document is written in LTEX2e and should be available from my home page http://www.dpmms.cam.ac.uk/ ̃twk in latex, dvi, ps and pdf formats. Supervisors can obtain fairly detailed comments on the exercises at the end of these notes from me or from the secretaries in DPMMS. My e-mail address is twk@dpmms. These notes are based on notes taken in the course of the previous lecturer Dr Pinch. Most of their virtues are his, most of their vices mine. Although the course makes use of one or two results from probability theory and a few more from algebra, it is possible to follow the course successfully whilst taking these results on trust. There is a note on further reading at the end but [7] is a useful companion for the first three quarters of the course (up to the end of section 8) and [9] for the remainder. Please note that vectors are row vectors unless otherwise stated.

An Efficient Algorithm to Find All Small-Size Stopping Sets of Low-Density Parity-Check Matrices

In this work, we introduce an efficient algorithm to find all stopping sets, of size less than some threshold, of a fixed low-density parity-check (LDPC) matrix. The solution is inspired by the algorithm proposed by Rosnes and Ytrehus in 2005 to find an exhaustive list of all small-size turbo stopping sets in a turbo code. The efficiency of the proposed algorithm is demonstrated by several numerical examples. For instance, we have applied the algorithm to the well-known (3, 5)-regular (155, 64) Tanner code and found all stopping sets of size at most 18 in about 1 min on a standard desktop computer. Also, we have verified that the minimum stopping set size of the (4896, 2474) Ramanujan-Margulis code is indeed 24, and that the corresponding multiplicity is exactly 204. Furthermore, we have applied the algorithm to the IEEE 802.16e LDPC codes and determined the minimum stopping set size and the corresponding multiplicity exactly for these codes. Finally, as an application, we present a greedy algorithm to find a small number of redundant parity checks to add to the original parity-check matrix in order to remove all stopping sets in the corresponding Tanner graph of size less than the minimum distance. An extensive case study of the (155, 64) Tanner code illustrates the usefulness of the algorithm, and we present a 110 times 155 redundant parity-check matrix for this code with no stopping sets of size less than the minimum distance. Simulation results of iterative decoding on the binary erasure channel show performance improvements for low-to-medium erasure probabilities when this redundant parity-check matrix is used for decoding.

Bounds on the minimum support weights

The minimum support weight, d/sub r/(C), of a linear code C over GF(q) is the minimal size of the support of an r-dimensional subcode of C. A number of bounds on d/sub r/(C) are derived, generalizing the Plotkin bound and the Griesmer bound, as well as giving two new existential bounds. As the main result, it is shown that there exist codes of any given rate R whose ratio d/sub rd/sub 1/ is lower bounded by a number ranging from (q/sup r/-1)/(q/sup r/-q/sup r-1/) to r, depending on R. >

Turbo Decoding on the Binary Erasure Channel: Finite-Length Analysis and Turbo Stopping Sets

This paper is devoted to the finite-length analysis of turbo decoding over the binary erasure channel (BEC). The performance of iterative belief-propagation decoding of low-density parity-check (LDPC) codes over the BEC can be characterized in terms of stopping sets. We describe turbo decoding on the BEC which is simpler than turbo decoding on other channels. We then adapt the concept of stopping sets to turbo decoding and state an exact condition for decoding failure. Apply turbo decoding until the transmitted codeword has been recovered, or the decoder fails to progress further. Then the set of erased positions that will remain when the decoder stops is equal to the unique maximum-size turbo stopping set which is also a subset of the set of erased positions. Furthermore, we present some improvements of the basic turbo decoding algorithm on the BEC. The proposed improved turbo decoding algorithm has substantially better error performance as illustrated by the given simulation results. Finally, we give an expression for the turbo stopping set size enumerating function under the uniform interleaver assumption, and an efficient enumeration algorithm of small-size turbo stopping sets for a particular interleaver. The solution is based on the algorithm proposed by Garello et al. in 2001 to compute an exhaustive list of all low-weight codewords in a turbo code.

Cycle-logical treatment for "Cyclopathic" networks

This correspondence addresses the problem of finding the network encoding equations for error-free networks with multiple sources and sinks. Previous algorithms could not cope with cyclic networks. Networks that are cyclic in three different senses are considered in this correspondence, and two extensions of the polynomial time Linear Information Flow (LIF) algorithm are presented. The first algorithm will produce the network encoding equations for a network which can be cyclic, unless the actual flow paths form cycles. The second algorithm will work also when the flow paths form simple cycles. Finally an example of a third kind of cyclic network, where the previous algorithms will fail, is given. However, a binary encoding is provided also in this case.

Addendum to “An Efficient Algorithm to Find All Small-Size Stopping Sets of Low-Density Parity-Check Matrices”

In an earlier transactions paper, Rosnes and Ytrehus presented an efficient algorithm for determining all stopping sets of low-density parity-check (LDPC) codes, up to a specified weight, and also gave results for a number of well-known codes including the family of IEEE 802.16e LDPC codes, commonly referred to as the WiMax codes. It is the purpose of this short paper to review the algorithm for determining the initial part of the stopping set weight spectrum (which includes the codeword weight spectrum), and to provide some improvements to the algorithm. As a consequence, complete stopping set weight spectra up to weight 32 (for selected IEEE 802.16e LDPC codes) can be provided, while in previous work only stopping set weights up to 28 are reported. In the published standard for the IEEE 802.16e codes there are two methods of construction presented, depending upon the code rate and the code length. We compare the stopping sets of the resulting codes and provide complete stopping set weight spectra (up to five terms) for all IEEE 802.16e LDPC codes using both construction methods.

Runlength-limited codes for mixed-error channels

The mixed-error channel (MC) combines the binary symmetric channel and the peak shift channel. The construction of (d, k) constrained t-MC-error-correcting block codes is described. It is demonstrated that these codes can achieve a code rate close to the (d, k) capacity. The encoding and decoding procedures are described. The performance of the construction depends on a particular partitioning of (d, k) constrained block codes. This partitioning is discussed and various tables of codes are included. Examples on encoding/decoding and on code performance are given. >

On maximum length convolutional codes under a trellis complexity constraint

We look at convolutional codes with maximum possible code length for prescribed redundancy, conditioned on constraints on the free distance and on the bit-oriented trellis state complexity. Rate (n - 1)/n codes have been studied to some extent in the literature, but more general rates have not been studied much. In this work we consider convolutional codes of rate (n - r)/n, r ≥ 1. Explicit construction techniques for free distance dfree = 3 and 4 codes are described. For codes with r = 2, an efficient exhaustive search algorithm is outlined. For the more general case with r ≥ 2, a heuristic approach is suggested. Several new codes were found for r = 1 and in particular for r = 2 and 3. Compared to previously known codes of similar free distance and complexity constraints, the new codes have either strictly higher rate, or the same rate but smaller low distance multiplicities.

Improved coding techniques for preceded partial-response channels

A coset of a convolutional code may be used to generate a zero-run length limited trellis code for a 1-D partial-response channel. The free squared Euclidean distance, d/sub free//sup 2/, at the channel output is lower bounded by the free Hamming distance of the convolutional code. The lower bound suggests the use of a convolutional code with maximal free Hamming distance, d/sub max/(R,N), for given rate R and number of decoder states N. In this paper we present cosets of convolutional codes that generate trellis codes with d/sub free//sup 2/>d/sub max/(R,N) for rates 1/5/spl les/R/spl les/7/9 and (d/sub free//sup 2/=d/sub max/(R,N) for R=13/16,29/32,61/64, The tabulated convolutional codes with R/spl les/7/9 were not optimized for Hamming distance. Instead, a computer search was used to determine cosets of convolutional codes that exploit the memory of the 1-D channel to increase d/sub free//sup 2/ at the channel output. The search was limited by only considering cosets with certain structural properties. The R/spl ges/13/16 codes were obtained using a new construction technique for convolutional codes with free Hamming distance 4. Newly developed bounds on the maximum zero-run lengths of cosets were used to ensure a short maximum run length at the 1-D channel output. >