Matemáticas Aplicadas a la Informática

Error Correcting Codes

Fco. Javier Lobillo

References

[HP10] W. Cary Huffman and Vera Pless. Fundamentals of Error-Correcting Codes.

Cambridge University Press, 2010.
2 Tema 1

Block codes
Let Fq be a field of q elements.
TEMA 1 An [n, k]q –linear code is a vector subspace C ≤ Fn q of dimension k.
Vectors (a0 , a1 , . . . , an−1 ) ∈ Fn
q are written in the form a0 a1 . . . an−1 .
The vectors in C are called codewords.
Linear codes
Let C be an [n, k]–linear code.
Any G ∈ Fq k×n such that C = {uG | u ∈ Fkq } is called a generator matrix or an
encoder. Any set of k linearly independent columns of G is called an information
set, while the remaining n − k columns are called a redundancy set. An encoder
Simplified communication channel G = [Ik |A] is called a systematic encoder.
Any H ∈ Fqw (n−k)×n such that C = {v ∈ Fn q | vH = 0} is called a parity check
T

matrix.
Noise
Theorem 1. If G = [Ik |A] is a systematic encoder for a [n, k] code C, then H =
[−AT |In−k ] is a parity check matrix for C.
Message Received message
Binary repetition code
This is a binary [n, 1]–code with encoder
Sender Channel Receiver  
G = 1 .n. . 1

and consequently parity check matrix

" #
1
H= .. In−1
.
1
Communication system scheme
[7, 4] Hamming code
Encoder  
Noise 1 0 0 0 0 1 1
0 1 0 0 1 0 1
G=
0 0 1 0 1

1 0
Message Sent signal Received signal Message 0 0 0 1 1 1 1
Parity check matrix
Source Transmitter Channel Receover Target  
0 1 1 1 1 0 0
H = 1 0 1 1 0 1 0
1 1 0 1 0 0 1

F. J. Lobillo
MAI: Error Correcting Codes 3

Dual code Puncturing codes

Let C be an [n, k]q –code over Fq . The dual subspace Let C be an [n, k, d]–code over Fq . A punctured code C{i} is the coded obtained
from C deleting the ith coordinate in each codeword.
C⊥ = {x ∈ Fn
q | hx, ci = 0, ∀c ∈ C}
Theorem 7. Let C be an [n, k, d]–code over Fq , and let C{i} be the code punctured
is called the dual code. at the ith coordinate.
Theorem 2. If G, H are encoder and parity check matrices of a code C, then H, G 1. C{i} is a linear code.
are encoder and parity check matrices of C⊥ .
2. If d > 1, C{i} is an [n − 1, k, d∗ ]–code where d∗ = d − 1 if C has a minimum
A code C is self-orthogonal provided C ⊆ C⊥ , and self-dual provided C = C⊥ . weight codeword with a nonzero ith coordinate and d∗ = d otherwise.
3. When d = 1, C{i} is an [n−1, k, 1]–code if C has no codeword of weight 1 whose
Hamming weight and distance nonzero entry is in coordinate i; otherwise, if k > 1, C{i} is an [n − 1, k − 1, d∗ ]–
Let x, y ∈ Fn
q . The Hamming weight w(x) of x is defined as the number of non zero
code with d∗ ≥ 1.
coordinates of x. The Hamming distance of x, y is defined as d(x, y) = w(x − y).
The puncturing process can be extended to any set of coordinates T . The punctured
Theorem 3. The Hamming distance satisfies for all x, y, z ∈ Fn
q code at T is then denoted by CT .
non-negativity d(x, y) ≥ 0
Extending codes
non-degeneracy d(x, y) = 0 ⇐⇒ x = y
Let C be an [n, k, d]–code over Fq . The extended code b
C is defined as
symmetry d(x, y) = d(y, x)
b
C = {x0 . . . xn−1 xn ∈ Fn+1 | x0 . . . xn−1 ∈ C, x0 + · · · + xn−1 + xn = 0}
q
triangle inequality d(x, y) ≤ d(x, z) + d(z, y)
b b
The (minimum) distance of a code C is defined as the minimum weight of its non– Theorem 8. Let C be an [n, k, d]–code over Fq . Then C is a linear [n+1, k, d]–code,
b b
zero elements. If an [n, k]–code C has distance d, then C is called an [n, k, d]–code. where d = d or d = d + 1.

Weight and parity checking Shortening codes

Theorem 4. Let C be a linear code with parity check matrix H. If c ∈ C, the Let C be an [n, k, d]–code over Fq and let T be a set of t coordinates. Let C(T ) be
columns of H corresponding to the non-zero coordinates of c are linearly depen- the subcode of C consisting on those codewords which are 0 on T . Puncturing C(T )
dent. Conversely, if a linear dependence relation with non-zero coefficients exists on T gives a code of length n − t called the code shortened on T and denoted CT .
among w columns of H, then there is a codeword in C of weight w whose non-zero Theorem 9. Let C be an [n, k, d]–code over Fq . Let T be a set of t coordinates.
coordinates correspond to these columns. Then:
Corollary 5. A linear code has minimum weight d if and only if its parity check 1. (C⊥ )T = (CT )⊥ and (C⊥ )T = (CT )⊥ ,
matrix has a set of d linearly dependent columns any set of d − 1 columns is
linearly independent. 2. if t < d, then CT and (C⊥ )T have dimensions k and n − t − k respectively,
Theorem 6. If C is an [n, k, d]–code, then every n−d+1 coordinate position contains 3. if t = d and T is the set of coordinates where a minimum weight codeword
an information set. Furthermore, d is the largest number with this property. In is non–zero, then CT and (C⊥ )T have dimensions k − 1 and n − d − k + 1
particular d ≤ n − k + 1. respectively.

F. J. Lobillo
4 Tema 1

The (u, u + v) construction Reed–Muller codes

Let Ci an [n, ki , di ]–code over Fq for i = 1, 2. The code Let r ≤ m be a nonnegative and a positive integer. The rth order Reed–Muller
(RM) code is the binary code R(r, m) of length 2m defined recursively by
C = {(u, u + v) | u ∈ C1 , v ∈ C2 }
R(0, m) is the binary repetition code of length 2m ,
is called the (u, u + v) construction.
m
Theorem 10. Let Ci an [n, ki , di ]–code over Fq for i = 1, 2 and let C be the (u, u+v) R(m, m) is the entire subspace of F22 ,
construction. Then C is a [2n, k1 + k2 , min(2d1 , d2 ))]–code with encoder and parity
check matrices for 1 ≤ r < m, R(r, m) = {(u, u + v) | u ∈ R(r, m − 1), v ∈ R(r − 1, m − 1)}
   
G1 G1 H1 0
,
0 G2 −H2 H1
Theorem 13. The following hold:
where Gi , Hi are an encoder and a parity check matrix for Ci , i = 1, 2.
1. R(i, m) ⊆ R(j, m), if 0 ≤ i ≤ j ≤ m.
Permutation equivalence m

m

m


2. The dimension of R(r, m) equals 0 + 1 + ··· + r .
Two linear codes C1 and C2 are permutation equivalent provided there is a per-
mutation of coordinates which sends C1 to C2 . 3. The minimum weight of R(r, m) equals 2m−r .
Theorem 11. Let C be a linear code. 4. R(r, m)⊥ = R(m − r − 1, m), if 0 ≤ r < m.
1. C is permutation equivalent to a code which has a systematic encoder.
Decoding strategies
2. If I and R are the information and redundancy positions, respectively, for C,
then R and I are the information and redundancy positions, respectively, for c ∈ Fn is sent and y = c + e ∈ Fn is received.
q q
the dual code C⊥ .
Maximum a posteriori decoder: Choose b c = c such that prob(c|y) is maximum.
Monomial equivalence
Maximum likelihood decoder: Choose b c = c such that prob(y|c) is maximum.
Two codes C1 and C2 are monomially equivalent if there exists a monomial matrix1
M such that for each encoder G1 of C1 , G1 M is an encoder for C2 . Nearest neighbor decoder: Choose b c = c such that d(c, y) is minimum.

Hamming codes
Let n = 2r − 1, with r ≥ 2. Let Hr be the r × n matrix whose columns, in order, The sphere of radius r centered in u ∈ Fn
q is
are the numbers 1, 2, . . . , 2r − 1 written as binary numerals. Any code permutation
equivalent to a code with parity check matrix Hr is the [n = 2r − 1, k = n − r]– Sr (U) = {v ∈ Fn
q | d(u, v) ≤ r}.
Hamming code, and denoted H2,r = Hr . Pr n


Its cardinality is i=0 i (q−1)i . They are disjoint as far as their radius is chosen
Theorem 12. 1. The Hamming code H3 has distance 3. small enough,
2. Any binary [2r − 1, 2r − 1 − r, 3]–code is permutation equivalent to Hr . Theorem 14. If d is the minimum distance of a code C and t = b(d − 1)/2c, then
1A monomial matrix has only one non-zero entry in each row and column the spheres of radius t about distinct codewords are disjoint.

F. J. Lobillo
MAI: Error Correcting Codes 5

Packing radius
The packing radius of a code is the largest radius of spheres centered at codewords
so that the spheres are pairwise disjoint.
Theorem 15. Let C be an [n, k, d]–code over Fq . The following hold:
1. The packing radius of C equals t = b(d − 1)/2c.

2. The packing radius t of C is characterized by the property that nearest neigh-

bor decoding always decodes correctly a received vector in which t or fewer
errors have occurred but will not always decode correctly a received vector
in which t + 1 errors have occurred.

Syndrome
Let C be an [n, k, d]–code over Fq , with parity check matrix H. The syndrome of
y ∈ Fn T n
q is syn(y) = Hy . Let’s consider the equivalent relation in Fq associated
to C, i.e. x, y are related if and only if x − y ∈ C. It is clear that x, y are related if
and only if syn(x) = syn(y).

Suppose a codeword sent over a communication channel is received as a vector y.

Since in nearest neighbor decoding we seek a vector e of smallest weight such that
y − e ∈ C, nearest neighbor decoding is equivalent to finding a vector e of smallest
weight in the coset containing y. Such an element is called a coset leader.

The Syndrome Decoding Algorithm

Let C be an [n, k, d]–code over Fq with parity check matrix H.
1. For each syndrome s ∈ Fn−kq , choose a coset leader es of the coset es + C.
Create a table pairing the syndrome with the coset leader.

2. After receiving a vector y, compute its syndrome s = syn(y) = yHT .

3. y is then decoded as the codeword y − es .

F. J. Lobillo