Professional Documents
Culture Documents
Essays On Coding Theory 1St Edition Blake full chapter pdf docx
Essays On Coding Theory 1St Edition Blake full chapter pdf docx
Essays On Coding Theory 1St Edition Blake full chapter pdf docx
Blake
Visit to download the full and correct content document:
https://ebookmass.com/product/essays-on-coding-theory-1st-edition-blake/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://ebookmass.com/product/kantian-commitments-essays-on-
moral-theory-and-practice-barbara-herman/
https://ebookmass.com/product/diplomatic-investigations-essays-
on-the-theory-of-international-politics-herbert-butterfield/
https://ebookmass.com/product/information-theory-coding-and-
cryptography-3rd-edition-ranjan-bose/
https://ebookmass.com/product/blake-wilder-fbi-11-the-house-on-
the-hill-gray/
Private Law and Practical Reason: Essays on John
Gardner's Private Law Theory (Oxford Private Law
Theory) Haris Psarras (Editor)
https://ebookmass.com/product/private-law-and-practical-reason-
essays-on-john-gardners-private-law-theory-oxford-private-law-
theory-haris-psarras-editor/
https://ebookmass.com/product/geometric-patterns-with-creative-
coding-coding-for-the-arts-1st-edition-selcuk-artut/
https://ebookmass.com/product/chipless-rfid-based-on-rf-encoding-
particle-realization-coding-and-reading-system-1st-edition-
arnaud-vena/
https://ebookmass.com/product/eduard-bernstein-on-socialism-past-
and-present-essays-and-lectures-on-ideology-1st-edition-marius-s-
ostrowski/
https://ebookmass.com/product/essays-on-ethics-and-culture-
sabina-lovibond/
Essays on Coding Theory
Critical coding techniques have developed over the past few decades for data storage,
retrieval and transmission systems, significantly mitigating costs for governments and
corporations that maintain server systems containing large amounts of data. This book
surveys the basic ideas of these coding techniques, which tend not to be covered in the
graduate curricula, including pointers to further reading. Written in an informal style,
it avoids detailed coverage of proofs, making it an ideal refresher or brief introduction
for students and researchers in academia and industry who may not have the time to
commit to understanding them deeply. Topics covered include fountain codes designed
for large file downloads; LDPC and polar codes for error correction; network,
rank-metric and subspace codes for the transmission of data through networks;
post-quantum computing; and quantum error correction. Readers are assumed to have
taken basic courses on algebraic coding and information theory.
I a n F. B l a k e is Honorary Professor in the Department of Electrical and
Computer Engineering at the University of British Columbia, Vancouver. He is a
fellow of the Royal Society of Canada, the Institute for Combinatorics and its
Applications, the Canadian Academy of Engineers and a Life Fellow of the IEEE. In
2000, he was awarded an IEEE Millennium Medal. He received his undergraduate
degree at Queen’s University, Canada and doctorate degree at Princeton University in
1967. He also worked in industry, spending sabbatical leaves with IBM and M/A-Com
Linkabit, and working with the Hewlett-Packard Labs from 1996 to 1999. His research
interests include cryptograph and algebraic coding theory, and he has written several
books in these areas.
I a n F. B l a k e
University of British Columbia
www.cambridge.org
Information on this title: www.cambridge.org/9781009283373
DOI: 10.1017/9781009283403
© Ian F. Blake 2024
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press & Assessment.
First published 2024
A catalogue record for this publication is available from the British Library
A Cataloging-in-Publication data record for this book is available from the Library of Congress
ISBN 978-1-009-28337-3 Hardback
Cambridge University Press & Assessment has no responsibility for the persistence or
accuracy of URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Preface page xi
1 Introduction 1
1.1 Notes on Finite Fields and Coding Theory 1
1.2 Notes on Information Theory 16
1.3 An Overview of the Chapters 21
2 Coding for Erasures and Fountain Codes 26
2.1 Preliminaries 27
2.2 Tornado Codes and Capacity-Achieving Sequences 30
2.3 LT and Raptor Codes 44
3 Low-Density Parity-Check Codes 66
3.1 Gallager Decoding Algorithms A and B for the BSC 71
3.2 Performance of LDPC Codes on the BIAWGN Channel 77
3.3 Thresholds, Concentration, Gaussian Approximation, EXIT
Charts 87
4 Polar Codes 97
4.1 Preliminaries and Notation 98
4.2 Polar Code Construction 102
4.3 Subchannel Polarization and Successive Cancellation Decoding 111
5 Network Codes 128
5.1 Network Flows 129
5.2 Network Coding 135
5.3 Construction and Performance of Network Codes 142
6 Coding for Distributed Storage 157
6.1 Performance Limits on Coding for Distributed Storage 158
vii
Index 454
xi
pressure departments face, many of the other topics considered here may not
fare so well.
The idea of this book was to create a series of presentations of modest
length and depth in these topics to facilitate access to them by graduate
students and researchers who may have an interest in them but defer from
making the commitment of time and effort for a deeper understanding. Each
chapter is designed to acquaint the reader with an introduction to the main
results and possibilities without many of the proofs and details. They can
be read independently and a prerequisite is a basic course on algebraic
coding and information theory, although some of the topics present technical
challenges.
There are as many reasons not to write such a book as to write it. A few of
the areas have either excellent monographs or tutorials available on the web.
Also, it might be argued that an edited book on these topics with chapters
written by acknowledged experts would be of more value. Indeed, such a
volume is A Concise Encyclopedia of Coding Theory, W.C. Huffman, J.-L.
Kim and P. Solé, eds., 2021, CRC Press. However, the entries in such a volume,
as excellent as they usually are, are often of an advanced nature, designed for
researchers in the area to bring them abreast of current research directions. It
was felt that a series of chapters, written from a fairly consistent point of view
and designed to introduce readers to the areas covered, rather than provide
a deep coverage, might be of interest. I hope some readers of the volume will
agree. For many of the areas covered, the influence of the seminal papers on the
subjects is impressive. The attempt here is to explain and put into context these
important works, but for a serious researcher in an area, it does not replace the
need to read the original papers.
Choosing the level of the presentation was an interesting challenge. On
the one hand it was desired to achieve as good an appreciation of the results
and implications of an area as possible. The emphasis is on describing and
explaining contributions rather than proving and deriving, as well as providing
a few examples drawn mainly from the literature. On the other hand the
inclusion of too much detail and depth might discourage reading altogether.
It is hoped the compromise reached is satisfactory. While efforts were made to
render readable accounts for the topics, many readers might still find some of
the topics difficult.
Another problem was to choose a consistent notation when describing
results from different authors. Since one of the goals of the work was to provide
an entrée to the main papers of an area, an effort was made to use the notation
of the seminal works. Across the chapters, compromises in notation had to be
made and the hope is that these were reasonable. Generally there was a bias
toward describing code construction techniques for the areas which tended to
make some sections technically challenging.
I would like to thank the many colleagues around the world who provided
helpful and useful comments on many parts of the manuscript. First among
these are Shu Lin and Frank Kschischang, who read virtually all of the work
and consistently supported the effort. I cannot thank them enough for their
comments and suggestions. I would also like to thank Raymond Yeung, Vijay
Kumar, Eitan Yaakobi, Rob Calderbank, Amir Tasbihi and Lele Wang, who
read several of the chapters and provided expert guidance on many issues.
I would also like to thank the Department of Electrical and Computer
Engineering at the University of British Columbia and my colleagues there,
Vijay Bhargava, Lutz Lampe and Lele Wang, for providing such a hospitable
environment in which to pursue this work.
Finally, I would like to thank my wife Betty without whose love, patience
and understanding this book would not have been written.
Since the early 2000s we have seen the basic notions of coding theory expand
beyond the role of error correction and algebraic coding theory. The purpose
of this volume is to provide a brief introduction to a few of the directions that
have been taken as a platform for further reading. Although the approach is to
be descriptive with few proofs, there are parts which are unavoidably technical
and more challenging.
It was mentioned in the Preface that the prerequisite for this work is a basic
course on algebraic coding theory and information theory. In fact only a few
aspects of finite fields, particularly certain properties of polynomials over finite
fields, Reed–Solomon codes and Reed–Muller codes and their generalizations
are considered to provide a common basis and establish the notation to be used.
The trace function on finite fields makes a few appearances in the chapters
and its basic properties are noted. Most of the information will be familiar
and stated informally without proof. A few of the chapters use notions of
information theory and discrete memoryless channels and the background
required for these topics is also briefly reviewed in Section 1.2. The final
Section 1.3 gives a brief description of the chapters that follow.
which has no zero divisors (i.e., two nonzero elements a,b such that ab = 0).
A Euclidean domain is an integral domain which is furnished with a norm
function, in which the division of an element by another with a remainder of
lower degree can be formulated. Equivalently the Euclidean algorithm (EA)
can be formulated in a Euclidean domain.
Recall that the gcd of two integers a,b ∈ Z is the largest integer d
that divides both a and b. Let F be a field and denote by F[x] the ring of
polynomials over F with coefficients from F. The gcd of two polynomials
a(x),b(x) ∈ F[x] is the monic polynomial (coefficient of the highest power
of x is unity) of the greatest degree, d(x), that divides both polynomials. The
EA for polynomials is an algorithm that produces the gcd of polynomials a(x)
and b(x) (the one for integers is similar) by finding polynomials u(x) and v(x)
such that
d(x) = u(x)a(x) + v(x)b(x). (1.1)
That d(x), the last nonzero remainder, is the required gcd is established by
tracing back divisibility conditions. Furthermore, tracing back shows how two
polynomials u(x) and v(x) are found so that Equation 1.1 holds.
A similar argument holds for integers. The gcd is denoted (a,b) or
(a(x),b(x)) for integers and polynomials, respectively. If the gcd of two
integers or polynomials is unity, they are referred to as being relatively prime
and denoted (a,b) = 1 or (a(x),b(x)) = 1.
If the prime factorization of n is
e
n = p1e1 p2e2 · · · pkk , p1,p2, . . . ,pk distinct primes,
then the number of integers less than n that are relatively prime to n is given
by the Euler Totient function φ(n) where
k
φ(n) = piei −1 (pi − 1). (1.2)
i=1
irreducible polynomial f (x), i.e., the remainder after division by f (x). The
inverse of a nonzero polynomial a(x) ∈ Fp [x]/f (x) is found via the EA as
before. That is since by definition (a(x),f (x)) = 1 there exist polynomials
u(x),v(x) such that
u(x)a(x) + v(x)f (x) = 1
and the inverse of a(x) ∈ Fp [x]/f (x) is u(x). Algebraically this structure
might be described as the factor field of the ring Fp [x] modulo the maximal
ideal f (x).
It follows the set Fp [x]/f (x) forms a finite field with p n elements. It is
conventional to denote q = p n and the field of pn elements as either Fpn or
Fq . Every finite field can be shown to have a number of elements of the form
q = p n for some prime p and positive integer n and that any two finite fields of
the same order are isomorphic. It will be noted that an irreducible polynomial
of degree n will always exist (see Equation 1.4) and so all finite fields can be
constructed in this manner.
In general, suppose q = pm and let f (x) be a monic irreducible polynomial
over Fq of degree m (which will be shown to always exist). The set of q m
polynomials over Fq of degree less than m with multiplication modulo f (x)
will then be a finite field with q m elements and designated Fq m . For future
reference denote the set of polynomials of degree less than m by Fq <m [x]
and those less than or equal by Fq ≤m [x]. Since it involves no more effort,
this general finite field Fq m will be examined for basic properties. The subset
Fq ⊆ Fq m is a field, i.e., a subset that has all the properties of a field, a subfield
of Fq m .
The remainder of the subsection contains a brief discussion of the structure
of finite fields and polynomials usually found in a first course of coding theory.
It is straightforward to show that over any field F (x m − 1) divides (x n − 1)
iff m divides n, written as
(x m − 1)(x n − 1) iff m | n.
By definition the elements are distinct. Each of these elements could be taken
modulo f (x) (which is zero in the field) which would result in the field
elements being all polynomials over Fq of degree less than k. Multiplication
in this field would be polynomials taken modulo f (x). The field element x
is a primitive element. While this is a valid presentation, it is also common to
identify the element x by an element α with the statement “let α be a zero of the
primitive polynomial f (x) of degree k over Fq .” The two views are equivalent.
There are φ(q k − 1) primitive elements in Fq k and since the degree of
an irreducible polynomial with one of these primitive elements as a zero is
necessarily k, there are exactly φ(q k − 1)/k primitive polynomials of degree
k over Fq .
Suppose f (x) is an irreducible nonprimitive polynomial of degree k over
Fq . Suppose it is of order n < q k − 1, i.e., f (x) | (x n − 1). One can define the
field Fq k as the set of polynomials of degree less than k
Fq k = ak−1 x k−1 + ak−2 x k−2 + · · · + a1 x + a0, ai Fq
with multiplication modulo f (x). The element x is not primitive if n < (q k −1)
but is an element of order n,n | q k − 1 (although there are still φ(q k − 1)
primitive elements in the field).
Let α ∈ Fq m be an element of maximum order (q m − 1) (i.e., primitive) and
denote the multiplicative group of nonzero elements as
F∗q m = α = 1,α,α 2, . . . ,α q −2 .
m
= f (x q )
and, as above, g(x) has coefficients in Fq , i.e., g(x) ∈ Fq [x]. It follows that
g(x) must divide f (x) and since f (x) was assumed monic and irreducible it
must be that g(x) = f (x). Thus if one zero of the irreducible f (x) is in Fq m ,
all are. Each conjugacy class of the finite field corresponds to an irreducible
polynomial over Fq .
By similar reasoning it can be shown that if f (x) is irreducible of degree k
m m
over Fq , then f (x) | (x q − x) iff k | m. It follows that the polynomial x q − x
is the product of all monic irreducible polynomials whose degrees divide m.
Thus
m
xq − x = f (x).
f (x)irreducible
over Fq
degreef (x)=k|m
f3 (x) x6 + x5 + x2 + x + 1 63
f4 (x) x6 + x3 + 1 9
f5 (x) x3 + x2 + 1 7
f6 (x) x +x +x +x +1
6 5 3 2 63
f7 (x) x6 + x4 + x3 + x + 1 63
f8 (x) x6 + x5 + x4 + x2 + 1 21
f9 (x) x2 + x + 1 3
f10 (x) x +x +x +x+1
6 5 4 63
f11 (x) x3 + x + 1 7
f12 (x) x6 + x5 + 1 63
the monic irreducible polynomial of least degree that has β as a zero. From
the above discussion, every element in a conjugacy class has the same minimal
polynomial.
Further notions of finite fields that will be required include that of a
polynomial basis of Fq n over Fq which is one of a form {1,α,α 2, . . . ,α n−1 }
for some α ∈ Fq n for which the elements are linearly independent over Fq .
2 n−1
A basis of Fq n over Fq of the form {α,α q ,α q , . . . ,α q } is called a normal
basis and such bases always exist. In the case that α ∈ Fq n is primitive (of
order q n − 1) it is called a primitive normal basis.
since each element of Fq n is a zero of the LHS and exactly one term of
the RHS.
Also, suppose [8] L(·) is a linear function from Fq n to Fq in the sense that
for all a1,a2 ∈ Fq and all α1,α2 ∈ Fq n
L(a1 α1 + a2 α2 ) = a1 L(α1 ) + a2 L(α2 ).
Then L(·) must be of the form
L(α) = Trq n |q (βα) = Lβ (α)
for some β. Thus the set of such linear functions is precisely the set
Lβ (·), β ∈ Fq n
and these are distinct functions for distinct β.
A useful property of the trace function ([8], lemma 3.51, [11], lemma 9.3)
is that if u1,u2, . . . ,un is a basis of Fq n over Fq and if
Trq n |q (αui ) = 0 for i = 1,2, . . . ,n, α ∈ Fq n ,
then α = 0. Equivalently if for α ∈ Fq n
Trq n |q (αu) = 0 ∀u ∈ Fq n , (1.5)
then α = 0. This follows from the trace map being onto. It will prove a useful
property in the sequel. It also follows from the fact that
⎡ ⎤
q q n−1
u1 u1 · · · u1
⎢ ⎥
⎢u uq · · · uq n−1 ⎥
⎢ 2 2 2 ⎥
⎢. . . .. ⎥
⎢ .. .. .. . ⎥
⎣ ⎦
q q n−1
un un · · · un
is nonsingular iff u1,u2, . . . ,un ∈ Fq n are linearly independent over Fq .
A formula for the determinant of this matrix is given in [8].
If μ = {μ1,μ2, . . . ,μn } is a basis of Fq n over Fq , then a basis ν =
{ν1,ν2, . . . ,νn } is called a trace dual basis if
1 if i = j
Trq n |q (μi νj ) = δi,j =
0 if i = j
and for a given basis a unique dual basis exists. It is noted that if μ =
{μ1, . . . ,μn } is a dual basis for the basis {ν1, . . . ,νn }, then given
n n
y= ai μi then y= Trq n |q (yνi ) μi , ai ∈ Fq . (1.6)
i=1 i=1
It can be shown that for a given normal basis, the dual basis is also normal.
A convenient reference for such material is [8, 12].
That the minimum distance bound of this code, d = 2t + 1, follows since any
2t × 2t submatrix of H is a Vandermonde matrix and is nonsingular since the
elements of the first row are distinct.
A cyclic Reed–Solomon (n,k,d = n − k + 1)q code can be generated by
choosing a generator polynomial of the form
n−k
g(x) = (x − α i ), α ∈ Fq , α primitive of order n.
i=1
That the code has a minimum distance d = n − k + 1 follows easily from the
above discussion.
A standard simple construction of Reed–Solomon codes over a finite field
Fq of length n that will be of use in this volume is as follows. Let u =
{u1,u2, . . . ,un } be a set, referred to as the evaluation set (and viewed as a
set rather than a vector – we use boldface lowercase letters for both sets and
⊥ (u,v,q) = GRS
GRSn,k n,n−k (u,w,q)
= w1 g(u1 ),w2 g(u2 ), . . . ,wn g(un ), g ∈ F<n−k
q [x] .
(1.7)
In other words, for any f (x) ∈ F<k q [x] and g(x) ∈ Fq
<n−k
[x] for a given
evaluation set u = {u1, . . . ,un } and multiplier set v = {v1, . . . ,vn } there is
a multiplier set w = {w1, . . . , wn } such that the associated codewords cf ∈
GRSn,k (u,v,q) and cg ∈ GRSn,n−k (u,w,q) are such that
−1
wi = vi (ui − uj ) . (1.8)
j =i
n
e(x) = (x − ui ) and ei (x) = e(x)/(x − ui ) = (x − uk ),
i=1 k=i
It follows that for any polynomial h(x) ∈ Fq [x] of degree less than n that takes
on values h(ui ) on the evaluation set u = {u1,u2, . . . ,un } can be expressed as
n
ei (x)
h(x) = h(ui ) .
ei (ui )
i=1
The coefficient of x n−1 on the left side is 0 while on the right side is 1 (as ek (x)
is monic of degree (n − 1)) and hence
vk−1
n n
1
0= f (uk )g(uk ) = (vk f (uk )) g(uk )
ek (uk ) ek (uk )
k=1 k=1
n
= (vk f (uk ))(wk g(uk )) (by Equation 1.8)
k=1
= (cf,cg ) = 0.
Reed–Muller Codes
Reed–Muller (RM) codes are discussed in some depth in most books on coding
(e.g., [3, 4]) with perhaps the most comprehensive being [2] which considers
their relationship to Euclidean geometries and combinatorial designs. The
properties of RM codes are most easily developed for the binary field but the
general case will be considered here – the Generalized Reed–Muller (GRM)
codes (generalized in a different sense than the GRS codes). The codes are
of most interest in this work for the construction of locally decodable codes
(Chapter 8) and their relationship to multiplicity codes introduced there.
Consider m variables x1,x2, . . . ,xm and the ring Fq [x1,x2, . . . ,xm ] = Fq [x]
of multivariate polynomials over Fq (see also Appendix B). The set of all
monomials of the m variables and their degree is of the form
xi = x1i1 x2i2 · xm
im
, i ∼ (i1,i2, . . . ,im ), degree = ij . (1.9)
j
the balls in d+m
m ways in which to place markers corresponding to m + 1
bins. As before subtract a ball from each bin. The contents of the last cell are
regarded as superfluous and discarded to take into account the “at most” part
of the enumeration. The contents of the first m cells correspond to a partition
and the number of monomials on m variables of total degree at most d is
d +m
= {i1 + i2 + · · · + im ≤ d, ij ∈ Z≥0 }. (1.11)
m
Note that it follows that
d
j +m−1 d +m
= ,
m−1 m
j =1
= 2 (bivariate
m polynomials) this subclass of GRM codes has the parameters
d+2 2
q , d ,q − dq . Note that the rate of the code is
2
q
d +2
q 2 ≈ d 2 /2q 2 = (1 − )2 /2.
2
Thus the code can have rate at most 1/2.
For a more complete analysis of the GRM codes the reader should consult
([2], section 5.4). Properties of GRS and GRM codes will be of interest in
several of the chapters.
and unless otherwise specified all logs will be to the base 2. It represents the
amount of uncertainty in the outcome of a realization of the random variable.
A special case will be important for later use, that of a binary ensemble
{p,(1 − p)} which has an entropy of
H2 (p) = −p log2 p − (1 − p) log2 (1 − p) (1.12)
referred to as the binary entropy function. It is convenient to introduce the
q-ary entropy function here, defined as
x logq (q − 1) − x logq x − (1 − x) logq (1 − x), 0 < x ≤ θ = (q − 1)/q
Hq (x) =
0, x=0
(1.13)
an extension of the binary entropy function. Notice that Hq (p) is the entropy
associated with the q-ary discrete symmetric channel and also the entropy of
the probability ensemble {1 − p,p/(q − 1), . . . ,p/(q − 1)} (total of q values).
The binary entropy function of Equation 1.12 is obtained with q = 2.
Similarly the entropy of a joint ensemble P (xi ,yi ),i = 1,2, . . . } is
given by
H (X,Y ) = − P (xi ,yi ) log P (xi ,yi )
xi ,yi
and measures the amount of information knowledge that one of the variables
gives on the other. The notation {X;Y } is viewed as a joint ensemble. It will
often be the case that X will represent the input to a discrete memoryless
channel (to be discussed) and Y the output of the channel and this notion
of mutual information has played a pivotal role in the development of
communication systems over the past several decades.
Similarly for three ensembles it follows that
P (xi ,yj ,zk )
I (X;Y,Z) = P (xi ,yj ,zk ) log .
P (xi )P (yj ,zk )
i,j,k
or alternatively
P (xi ,yj ,zk )P (zk )
I (X;Y | Z) = P (xi ,yj ,zk ) log .
P (xi ,zk )P (yj ,zk )
i,j,k
W (y | x) = 1 for each x ∈ X.
y∈Y
1−p 1−δ
0 0 0 0
p δ . ..
E X = {0,1} .. . Y
p δ
1 1 1 1
1−p 1−δ
N ∼ N(0,σ 2 )
+
X ∈ {±1} Y =X+N
Figure 1.2 The binary-input additive white Gaussian noise channel
tractable solutions and are designated additive white Gaussian noise (AWGN)
channels. The term “white” here refers to a flat power spectral density function
of the noise with frequency. The binary-input AWGN (BIAWGN) channel,
where one of two continuous-time signals is chosen for transmission during
a given time interval (0,T ) and experiences AWGN in transmission, can be
represented as in Figure 1.2:
Yi = Xi + Ni , and Xi ∈ {±1}, BIAWGN,
where Ni is a Gaussian random variable with zero mean and variance σ 2 ,
denoted Ni ∼ N(0,σ 2 ). The joint distribution of (X,Y ) is a mixture of discrete
and continuous and with P (X = +1) = P (X = −1) = 1/2 (which achieves
capacity on this channel) and with p(x) ∼ N(0,σ 2 ). The pdf p(y) of the
channel output is
2 2
1 1 − (y+1) 1 1 − (y−1)
p(y) = ·√ e 2σ 2 + · √ e 2σ 2
2 2π σ 2 2 2π σ 2
(1.16)
1 (y+1)2 (y−1)2
=√ exp − 2σ 2
+ exp − 2σ 2
8π σ 2
and maximizing the expression for mutual information of the channel (equally
likely inputs) reduces to
1
CBI AW GN = − p(y) log2 p(y)dy − log2 (2π eσ 2 ). (1.17)
y 2
The general shape of these capacity functions is shown in Figure 1.3 where
SNR denotes signal-to-noise ratio.
Another Gaussian channel of fundamental importance in practice is that of
the band-limited AWGN channel (BLAWGN). In this model a band-limited
signal x(t),t ∈ (0,T ) with signal power ≤ S is transmitted on a channel
band-limited to W Hz, i.e., (−W,W ) and white Gaussian noise with two-sided
power spectral density level No /2 is added to the signal in transmission. This
channel can be discretized via orthogonal basis signals and the celebrated and
much-used formula for the capacity of it is
CBSC
1.0 CBI AW GN
1.0
0.5
0 p 0
0 0.5 1.0 SN R ∼ 1/σ 2
(a) (b)
Figure 1.3 Shape of capacity curves for (a) BSC and (b) BIAWGN
are not erasure-correcting codes. Rather they are codes that can efficiently
recreate a file from several random combinations of subfiles. Such codes led
to the important concept of Raptor codes which have been incorporated into
numerous standards for the download of large files from servers in a multicast
network while not requiring requests for retransmissions of portions of a file
that a receiver may be missing, a crucial feature in practice.
Certain aspects of low-density parity-check (LDPC) codes are then dis-
cussed. These codes, which derive from the work of Gallager from the early
1960s, have more recently assumed great importance for applications as
diverse as coding for flash memories as well as a wide variety of communi-
cation systems. Numerous books have been written on various aspects of the
construction and analysis of these codes. This chapter focuses largely on the
paper of [13] which proved crucial for more recent progress for the analytical
techniques it introduced.
The chapter on polar codes arose out of the seminal paper [1]. In a deep
study of information-theoretic and analytical technique it produced the first
codes that provably achieved rates approaching capacity. From a binary-input
channel with capacity C ≤ 1, through iterative transformations, it derived
a channel with N = 2n inputs and outputs and produced a series of N C
sub-channels that are capable of transmitting data with arbitrarily small error
probability, thus achieving capacity. The chapter discusses the central notions
to assist with a deeper reading of the paper.
The chapter on network coding is devoted to the somewhat surprising idea
that allowing nodes (servers) in a packet network to process and combine
packets as they traverse the network can substantially improve throughput of
the network. This raises the question of the capacity of such a network and
how to code the packets in order to achieve the capacity. This chapter looks at
a few of the techniques that have been developed for multicast channels.
With the wide availability of the high-speed internet, access to information
stored on widely distributed databases became more commonplace. Huge
amounts of information stored in distributed databases made up of standard
computing and storage elements became ubiquitous. Such elements fail with
some regularity and methods to efficiently restore the contents of a failed
server are required. Many of these early distributed storage systems simply
replicated data on several servers and this turned out to be an inefficient
method of achieving restoration of a failed server, both in terms of storage
and transmission costs. Coding the stored information greatly improved
the efficiency with which a failed server could be restored and Chapter 6
reviews the coding techniques involved. The concepts involved are closely
related to locally repairable codes considered in Chapter 7 where an erased
References
[1] Arıkan, E. 2009. Channel polarization: a method for constructing capacity-
achieving codes for symmetric binary-input memoryless channels. IEEE Trans.
Inform. Theory, 55(7), 3051–3073.
[2] Assmus, Jr., E.F., and Key, J.D. 1992. Designs and their codes. Cambridge Tracts
in Mathematics, vol. 103. Cambridge University Press, Cambridge.
[3] Blahut, R.E. 1983. Theory and practice of error control codes. Advanced Book
Program. Addison-Wesley, Reading, MA.
[4] Blake, I.F., and Mullin, R.C. 1975. The mathematical theory of coding. Academic
Press, New York/London.
[5] Forney, G.D., and Ungerboeck, G. 1998. Modulation and coding for linear
Gaussian channels. IEEE Trans. Inform. Theory, 44(6), 2384–2415.
[6] Gallager, R.G. 1968. Information theory and reliable communication. John Wiley
& Sons, New York.
[7] Huffman, W.C., and Pless, V. 2003. Fundamentals of error-correcting codes.
Cambridge University Press, Cambridge.
[8] Lidl, R., and Niederreiter, H. 1997. Finite fields, 2nd ed. Encyclopedia of Mathe-
matics and Its Applications, vol. 20. Cambridge University Press, Cambridge.
[9] Ling, S., and Xing, C. 2004. Coding theory. Cambridge University Press,
Cambridge.
[10] MacWilliams, F.J., and Sloane, N.J.A. 1977. The theory of error-correcting
codes: I and II. North-Holland Mathematical Library, vol. 16. North-Holland,
Amsterdam/New York/Oxford.
[11] McEliece, R.J. 1987. Finite fields for computer scientists and engineers. The
Kluwer International Series in Engineering and Computer Science, vol. 23.
Kluwer Academic, Boston, MA.
[12] Menezes, A.J., Blake, I.F., Gao, X.H., Mullin, R.C., Vanstone, S.A., and
Yaghoobian, T. 1993. Applications of finite fields. The Kluwer International Series
in Engineering and Computer Science, vol. 199. Kluwer Academic, Boston, MA.
[13] Richardson, T.J., and Urbanke, R.L. 2001. The capacity of low-density parity-
check codes under message-passing decoding. IEEE Trans. Inform. Theory, 47(2),
599–618.
[14] Sudan, M. 1997. Decoding of Reed Solomon codes beyond the error-correction
bound. J. Complexity, 13(1), 180–193.
[15] Ungerboeck, G. 1982. Channel coding with multilevel/phase signals. IEEE Trans.
Inform. Theory, 28(1), 55–67.
26
2.1 Preliminaries
It is convenient to note a few basic results on coding and DMCs for future
reference. Suppose C = (n,k,d)q is a linear block code over the finite field
of q elements Fq , designating a linear code that has k information symbols
(dimension k) and (n − k) parity-check symbols and minimum distance d.
Much of this volume is concerned with binary-input channels and q = 2.
Suppose a codeword c = (c1,c2, . . . ,cn ) is transmitted on a BEC and the
received word is r = (r1,r2, . . . ,rn ) which has e erasures in positions E ⊂
{1,2, . . . ,n}, | E |= e. Then ri = E for i ∈ E for E the erasure symbol. The
unerased symbols received are assumed correct.
A parity-check matrix of the code is an (n − k) × n matrix H over Fq
such that
H · ct = 0tn−k
for any codeword c where 0n−k is the all-zero (n − k)-tuple, a row vector over
Fq . If the columns of H are denoted by h(i), i = 1,2, . . . ,n, then H · ct is
the sum of columns of H multiplied by the corresponding coordinates of the
codeword, adding to the all-zero column (n − k)-vector 0t . Similarly let He be
the (n−k)×e submatrix of columns of H corresponding to the erased positions
and ce be the e-tuple of the transmitted codeword on the erased positions. Then
He · cte = −yte
x1
x2
0 0 0 1 1 1 1 0 x3 c1
0 1 1 0 0 1 1 0 x4 c2
H = 1 0 1 0 1 0 1 0 x5 c3
1 1 1 1 1 1 1 1 x6 c4
x7
x8
(a) (b)
Figure 2.1 (a) The Hamming (8,4,4)2 code and (b) its Tanner graph
requests could overwhelm the source trying to satisfy requests from a large
number of receivers, a condition referred to as feedback implosion. This is the
multicast situation where a transmitter has a fixed number of packets (binary
n-tuples) which it wishes to transmit to a set of receivers through a network.
It is assumed receivers are not able to contact the transmitter to make requests
for retransmissions of missing packets. The receiver is able to construct the
complete set of transmitted information packets from the reception of any
sufficiently large set of coded packets, not a specific set. Typically the size
of the set of received packets is just slightly larger than the set of information
packets itself, to ensure successful decoding with high probability, leading to
a very efficient transmission and decoding process.
Most of the algorithms in the chapter will have linear complexity (in
codeword length or the number of information symbols), making them very
attractive for implementation.
As the sun died down, the moon arose above the hills and lighted
the travellers along a path winding by the shores of an irregular
mountain lake, and overhung by a multitude of cherry trees in full
blossom.
“Look!” cried Mrs. Brande, joyfully, “there in front you see the lights
of the Dâk Bungalow at last. You will be glad of your dinner, and I’m
sure I shall.”
Two men, who sat in the verandah of the same rest-house, would
also have been most thankful for theirs. The straggling building
appeared full of soldiers and their wives, and there seemed no
immediate prospect of a meal. The kitchen had been taken
possession of by the majestic cook of a burra mem sahib, who was
shortly expected, and the appetites of a couple of insignificant
strangers must therefore be restrained.
These travellers were, of course, Captain Waring and Mark Jervis,
whom the former invariably alluded to as “his cousin.” It was a
convenient title, and accounted for their close companionship. At first
Mark had been disposed to correct this statement, and murmur, “Not
cousins, but connections,” but had been silenced by Clarence
petulantly exclaiming—
“Cousins and connections are the same thing. Who cares a straw
what we are? And what’s the good of bothering?”
“I’m nearly mad with hunger,” groaned Captain Waring. “I’ve eaten
nothing for ten hours but one hard-boiled egg.”
“Smoke, as the Indians do,” suggested his comrade unfeelingly,
“or draw in your belt a couple of holes. Anyway, a little starvation will
do you no harm—you are getting fat.”
“I wonder, if I went and sat upon the steps with a placard round my
neck, on which was written, ‘I am starving,’ if this good lady would
give us a dinner? Hunger is bad enough, but the exquisite smell of
her roast mutton aggravates my pangs.”
“You have only to show yourself, and she will invite you.”
“How do you know, and why do you cruelly raise my hopes?”
“Because I hear that she is the soul of hospitality, and that she has
the best cook on the hills.”
“May I ask how you discovered this really valuable piece of
information?”
“From the harum-scarum youth who passed this afternoon. He
forgot to mention her name.”
“Here she comes along by the weir,” interrupted Waring. “Mark the
excitement among the servants—her meal will be ready to the
minute. She must be truly a great woman, and has already earned
my respect. If she asks me to dinner, I shall love her. What do you
say, Mark?”
“Oh, I think, since you put it in that way, that I should find it easier
to love the young lady!”
“I thought you fought shy of young ladies; and you must have cat’s
eyes if you can see one at this distance.”
“I have the use of my ears, and I have had nothing to do, but
concentrate my attention on what is evidently to be the only meal of
the evening. I heard the cook telling the khitmatghar to lay a place
for the ‘Miss Sahib.’”
“What a thing it is to be observant!” cried Captain Waring. “And
here they are. By George! she is a heavy weight!” alluding to Mrs.
Brande, who was now let down with a dump, that spoke a whole
volume of relief.
The lady ascended the verandah with slow and solid steps, cast a
swift glance at the famishing pair, and went into her own well-
warmed room, where a table neatly laid, and adorned with cherry-
blossoms, awaited her.
“Lay two more places,” were her first commands to the salaaming
Khitmatghar; then to her niece, “I am going to ask those two men to
dinner.”
“But you don’t know them, Aunt Sara!” she expostulated rather
timidly.
“I know of them, and that is quite enough at a dâk bungalow. We
are not so stiff as you are in England; we are all, as it were, in the
same set out here; and I am sure Captain Waring will be thankful to
join us, unless he happens to be a born idiot. In this bungalow there
is nothing to be had but candles and jam. I know it of old. People
who pass up, are like a swarm of locusts, and leave nothing behind
them, but empty tins and bottles. Now I can give him club mutton
and champagne.”
Having carefully arranged her dress, put on her two best diamond
rings, and a blue cap (N.B.—Blue had always been her colour), Mrs.
Brande sailed out into the verandah, and thus accosted the
strangers—
“I shall be very happy if you two gentlemen will dine with me in my
rooms.”
“You are really too good,” returned Captain Waring, springing to
his feet and making a somewhat exaggerated bow. “We shall be
delighted, for there seems no prospect of our getting anything to eat
before to-morrow.”
“You shall have something to eat in less than five minutes,” was
Mrs. Brande’s reassuring answer, as she led the way to her own
apartment.
“This,” waving her hand towards Honor, “is my niece, Miss
Gordon, just out from England. I am Mrs. Brande—my husband is in
the Council.”
“We have had the pleasure of meeting Miss Gordon before,” said
Captain Waring; “this will not be the first time we have sat at the
same table,” and he glanced at her, with sly significance.
“Yes,” faltered Honor, with a heightened colour, as she bowed and
shook hands with Mark. “This is the gentleman of whom I told you,
Aunt Sara, who rescued me when I was left alone in the train.”
“Ah! indeed,” said Mrs. Brande, sitting down as she spoke, and
deliberately unfolding her serviette, “I’m sure I’m greatly obliged to
him,” but she secretly wished that on that occasion Honor had been
befriended by his rich associate.
“Let me introduce him to you, Mrs. Brande—his name is Jervis,”
said Captain Waring, with his most jovial air. “He is young, idle, and
unmarried. My name is Waring. I was in the Rutlands, but I chucked
the service some time ago.”
“Well, now we know all about each other” (oh, deluded lady!) “let
us begin our dinner,” said Mrs. Brande. “I am sure we are all
starving.”
Dinner proved to be excellent, and included mahseer from the
lake, wild duck from the marshes, and club mutton. No! Mrs.
Brande’s “chef” had not been over praised. At first every one
(especially the hostess and Clarence Waring) was too frankly hungry
to talk, but after a time they began to discuss the weather, the local
insects, and their journey—not in the formal manner common to
Britons on their mournful travels—but in a friendly, homely fashion,
suitable to a whitewashed apartment, with the hostess’s bed in one
corner.
Whilst the two men conversed with her niece, Mrs. Brande
critically surveyed them, “took stock” as she said to herself. Captain
Waring was a man of five or six and thirty, well set up, and soldierly
looking; he had dark cropped hair, bold merry eyes, and was
handsome, though sunburnt to a deep tan, and his face was deeply
lined—those in his forehead looking as if they had been ruled and
cut into the very bone—nevertheless, his habitual expression was as
gay and animated as that of Toby Joy himself. He had an extremely
well-to-do air (undoubtedly had never known a money care in his
life), he wore his clothes with ease, they fitted him admirably, his
watch, studs, and linen were of the finest quality; moreover, he
appreciated a good dinner, seemed to accept the best of everything
as a matter of course, and looked about intelligently for peppers and
sauces, which were fortunately forthcoming.
“The companion,” as Mrs. Brande mentally called him, was a
younger man, in fact a mere youth of about two and twenty, well set
up, squarely built, with good shoulders and a determined mouth and
chin. He wore a suit of flannels, a silver watch, with a leather chain,
and looked exactly what he was—an idle, poor hanger-on!
Mrs. Brande left him to talk to Honor, and indeed entirely
neglected him for his more important kinsman. Her niece was
secretly aware of (and resented) her aunt’s preference, and
redoubled her efforts to entertain her slighted fellow-traveller. She
had a fellow-feeling for him also. Were they not both dependents—
both poor relations?
“Well, Captain Waring, so you are coming up to see Shirani?” said
Mrs. Brande, with her most gracious air.
“Yes, and I rather want to recall old times out here, and have a
nice lazy summer in the hills.”
“Then you have been in India all the winter?” (The inspection of his
kit the crafty lady kept to herself.)
“Yes. We came out in October. Had a bit of a shoot in Travancore,
and had a couple of months in Calcutta.”
“Then perhaps you came across a Miss Paske, there? Though I
don’t suppose she was in the Government House set. Her uncle is a
nobody.”
“To be sure. We know Miss Paske, don’t we, Mark? She was very
much in the Government House set. All the A.D.C’s adored her. A
little bit of a thing, with tow-coloured, fluffy hair, and a nez retroussé.”
“I know nothing about her nose or hair, but she is at Shirani now.”
“You don’t say so! I am delighted to hear it. She is capital fun!”
Mrs. Brande’s face fell. She sat crumbling her bread for some
seconds, and then said absently, “Did you notice those monkeys on
the way up?”
She had a peculiar habit of suddenly jumping from one topic to
another, figuratively, at the opposite pole. She declared that her
ideas travelled at times faster than her speech. Possibly she had her
own consecutive, if rapid, train of thought, and may thus have
connected Miss Paske with apes.
“Yes, swarms of those old grey fellows with black faces. I suppose
they have a fair club at Shirani, and keep up the whist-room? Are
there many men who play?”
“Only too many. I don’t approve of cards—at least gambling. I do
love a game of whist—I play a half-anna stamp on the rubber, just to
give it a little interest.”
“Do they play high at Shirani?” he asked with a touch of
impatience.
“Yes, I believe they do; and that horrid old Colonel Sladen is the
worst of all.”
“What! is he still up here? he used to play a first-class rubber.”
“He will play anything—high or low stakes—at either night or day
—he pays—his wife pays,” concluded Mrs. Brande, looking quite
ferocious.
“Oh, is she out again? Nice little woman.”
“Out again! She has never been home yet,” and she proceeded to
detail that lady’s grievances, whilst her companion’s roving eyes
settled on his cousin and Miss Gordon.
She was a remarkable-looking, even fascinating girl, quite different
to his impression of her at first sight. She had a radiant smile,
wonderfully expressive eyes (those eyes alone made her beautiful,
and lifted her completely out of the commonplace), and a high-bred
air. Strange that she should be related to this vulgar old woman, and
little did the vulgar old woman guess how she had been championed
by her English niece. The moon shining full on the lake tempted the
whole party out of doors. Captain Waring made a basely ungrateful
(but wholly vain) attempt to exchange ladies with his friend. Mrs.
Brande, however, loudly called upon him to attend her, as she paced
slowly down to the road; and as he lit his cigar at his cousin’s, he
muttered angrily under his moustache—
“I call this beastly unfair. I had the old girl all dinner time. You’ve
got six to four the best of it!”
CHAPTER XV.
A PROUD MOMENT.