Vojtech@Brtnik.

eu BCH Coding April 19, 2010

Example of simple BCH error-correcting coding

Problem: Design a 2-error correcting BCH code C of length 13 over the field
GF(27). Find a non-zero code word in C, impose two errors and correct
using an efficient algorithm.

Definition: Let β be a primitive root of the polynomial xn − 1. If we use

the generator polynomial g(x) such that β, β 2 , . . . , β d−1 , are all roots of
g(x), such a code is then called a BCH-code with design distance d.

Part 1: factorization into irreducible factors and listing roots of

those factors.

In GF(3) holds

x13 − 1 = (x + 2)(x3 + 2x + 2)(x3 + x2 + 2)(x3 + x2 + x + 2)(x3 + 2x2 + 2x + 2).

Denote these polynomials as follows

f1 (x) := x + 2,
f2 (x) := x3 + 2x + 2,
f3 (x) := x3 + x2 + 2,
f4 (x) := x3 + x2 + x + 2,
f5 (x) := x3 + 2x2 + 2x + 2.

We will now work in the algebraic field extension of GF(3) such that x13 − 1
is factored into linear polynomials (x − βi ) where βi are primitive roots of
1. Such a field is GF(27).

Let β be a zero of f2 , i.e. f2 (β) = β 3 + 2β + 2 = 0. It also follows that β is

a root of x13 − 1 and thus

β 13 = 1. (1)

Notorically known Lemma says that in the field F of characteristics p is for

each a, b ∈ F and k ∈ N
k k k
(a + b)p = ap + bp . (2)

1
Vojtech@Brtnik.eu BCH Coding April 19, 2010

Because in GF(27) is 23 = 2, we have

f2 (β 3 ) = (β 3 )3 + 2β 3 + 2 = (β 3 + 2β 3 + 2)3 = 0
and we conclude that β 3 is a zero of f2 . Using the same argument β 9 is as
well a zero of f2 .

Let us compute f4 (β 2 ). We know that β 3 = −2β − 2 = β + 1.

f4 (β 2 ) = β 6 + β 4 + β 2 + 2
= (β + 1)2 + β(β + 1) + β 2 + 2
= β 2 + 2β + 1 + β 2 + β + β 2 + 2 = 0.
We can see that β 2 is a zero of f4 . Using the identity (2) in the similar
manner as presented, we find that β 6 is also a zero of f4 . Assume that β 5
is a zero of fi , then using (1) and (2) is also (β 5 )3 = β 15 = β 2 a zero of fi .
We know that β 2 is a zero of f4 and thus also β 5 must be a zero of f4 .

Furthermore
f3 (β 4 ) = β 12 + β 8 + 2
= (β + 1)4 + β 2 (β + 1)2 + 2
= β 4 + 4β 3 + 2β 2 + 4β + 1 + 4β 2 + β 4 + 2β 3 + β 2 + 2 = 0
f3 (β 12 ) = 0

Using (1) and (2) we find that β 7 , (β 7 )3 = β 8 , (β 8 )3 = β 11 are zeros of

the same fi and i must be thus 5. Also β 10 and β 4 are zeros of the same
polynomial.

We can conclude this part stating that f2 has zeros β, β 3 , β 9 ; f3 has zeros
β 4 , β 10 , β 12 ; f4 has zeros β 2 , β 5 , β 6 and finally f5 has zeros β 7 , β 8 and β 11 .

Part 2: Designing the code

The powers of β in GF(27) forms the following table:

1 β β 2 β 3 β 4 β 5 β 6 β 7 β 8 β 9 β 10 β 11 β 12 β 13
1 0 0 1 0 1 1 1 2 2 0 1 2 1
0 1 0 1 1 1 2 2 0 1 2 1 0 0
0 0 1 0 1 1 1 2 2 0 1 2 1 0

2
Vojtech@Brtnik.eu BCH Coding April 19, 2010

We want to develop a code C with the design distance d = 5. According to

the definition of BCH codes, we shall take as a generating polynomial g(x)
product of those fi such that β 1 ,. . . ,β 4 are zeros of this product. Put

g(x) = f2 (x) · f3 (x) · f4 (x) = 2 + x3 + x4 + 2x6 + x7 + 2x8 + x9 .

Because C is a cyclic code over GF(3), the generator matrix G can be written
immediately
 
2 0 0 1 1 0 2 1 2 1 0 0 0
 0 2 0 0 1 1 0 2 1 2 1 0 0 
G=  0 0 2 0 0 1 1 0 2 1 2 1 0 .


0 0 0 2 0 0 1 1 0 2 1 2 1

This matrix can be transformed into a canonical form Gc = (I A):

 
1 0 0 0 2 0 0 1 1 0 2 1 2
 0 1 0 0 2 2 0 1 2 1 2 0 0 
Gc =  0 0 1 0
.
0 2 2 0 1 2 1 2 0 
0 0 0 1 0 0 2 2 0 1 2 1 2

From this we get a canonical parity-check matrix Hc = (−AT I)

1 1 0 0 1 0 0 0 0 0 0 0 0
 

 0 1 1 0 0 1 0 0 0 0 0 0 0 


 0 0 1 1 0 0 1 0 0 0 0 0 0 


 2 2 0 1 0 0 0 1 0 0 0 0 0 

Hc = 
 2 1 2 0 0 0 0 0 1 0 0 0 0 .


 0 2 1 2 0 0 0 0 0 1 0 0 0 


 1 1 2 1 0 0 0 0 0 0 1 0 0 

 2 0 1 2 0 0 0 0 0 0 0 1 0 
1 0 0 1 0 0 0 0 0 0 0 0 1

Back to the BCH codes. According to the definition, the parity-check matrix
H of the code C has a form
1 β β 2 β 3 β 4 β 5 β 6 β 7 β 8 β 9 β 10 β 11 β 12
 
 1 β 2 β 4 β 6 β 8 β 10 β 12 β β 3 β 5 β 7 β 9 β 11 
H=  1 β 3 β 6 β 9 β 12 β 2 β 5 β 8 β 11 β
.
β 4 β 7 β 10 
1 β 4 β 8 β 12 β 3 β 7 β 11 β 2 β 6 β 10 β β5 β9

3
Vojtech@Brtnik.eu BCH Coding April 19, 2010

If we substitute vectors from GF(27), we get

 
1 0 0 1 0 1 1 1 2 2 0 1 2
 0 1 0 1 1 1 2 2 0 1 2 1 0 
 
 0 0 1 0 1 1 1 2 2 0 1 2 1 
 
 1 0 0 1 2 0 2 0 1 1 1 2 1 
 
 0 0 1 2 0 2 0 1 1 1 2 1 1 
 
 0 1 1 1 2 1 1 0 0 1 2 0 2
 
H= .

 1 1 1 2 2 0 1 2 1 0 0 1 0 
 0 1 2 1 0 0 1 0 1 1 1 2 2
 

 0 0 1 0 1 1 1 2 2 0 1 2 1
 

 1 0 2 2 1 1 1 0 1 0 0 1 2
 

 
 0 1 0 0 1 2 1 0 2 2 1 1 1 
0 1 2 1 0 2 2 1 1 1 0 1 0

It can be checked that a vector space generated by H has a dimension 9 and

is obviously the same as the vector space generated by Hc .

We conclude that we have designed a cyclic linear code C of the length 13

and dimension 3, over GF(3) with the generator matrix Gc and the parity-
check matrix Hc . It is also a BCH code over GF(27) of the design distance
5 with the parity-check matrix H.

Part 3: characteristics of the code.

Length: 13 Dimension: 4
Information symbols: 9 Check symbols: 4
Rate: 9/13 Designed distance: 5
#code words: |F3 |4 = 81 Capable of correcting: 2 errors

Hilbert bound says that

|F3 |13 313 313

M3 (13, d) ≤ = 13 13 13 = < 17329.
|Be (w)| 1 + 13 + 78




0 + 1 + 2

Because M3 (13, k = 4, d) < 34 = 81, M3 (13, k = 5, d) < 35 = 243, we see,

that the dimension of the code could be greatly improved.

4
Vojtech@Brtnik.eu BCH Coding April 19, 2010

Part 4: theory behind decoding.

This section is written rather briefly, introducing the theory is not the pur-
pose of this paper.

Let us assume that the design distance is d, and that e errors (e ≤ 1/2d)
have occurred when transmitting certain codeword. Let σ(z), ω(z) be poly-
nomials
Y X Y
σ(z) = (1 − β i z) and ω(z) = Ei β i z (1 − β j z),
i i j6=i

where i and j run through the positions of the errors and Ei is the value of
error i. The formal derivative of σ(z) is
X Y
σ 0 (z) = −β i (1 − β j )z.
i

If we let z = β −i and divide ω by σ 0 , we get

ω(β −i )
Ei = β i ,
σ 0 (β −i )

where we still consider only those i where errors have occurred, or at least
where σ(β −i ) = 0 so in both σ 0 and ω all terms but one disappear.

Now, we get
∞ ∞
ω(z) X Ei β i z X X
ij j
X X
(z) = = = E i β z = zj Ei β ij .
σ(z) 1 − βiz
i i j=1 j=1 i

Because i Ei β ij equals to the inner product of the received word with the
P
jth row of the check matrix, exactly d − 1 coefficients of (z) can be found.
The inner product must be equal to zero for a correct word. Because ω and
σ are at most of degree e and have constant terms 0 and 1, respectively, we
only have 2e ≤ d − 1 unknowns. We find those by solving the equation

ω(z) = σ(z)(z).

This gives a system of d − 1 linear equations in the unknown coefficients of

σ and ω.

5
Vojtech@Brtnik.eu BCH Coding April 19, 2010

Part 5: find a non-zero code word in C, impose two errors and

correct.

In this part, we index elements from zero. The first row of a matrix is thus
in our notation 0th row.

For simplicity assume that a word represented by the generating polynomial

g(x) = g(x0 x1 . . . x12 ) was sent. We shall call this vector s,

s = (2 0 0 1 1 0 2 1 2 1 0 0 0).

Furthermore assume that during the transition two errors occurred, on po-
sitions 3 and 4, i.e. bits x3 and x4 were changed. We have thus received for
instance the word

s0 = (2 0 0 2 0 0 2 1 2 1 0 0 0).

Based on the theory, we want to compute the coefficients of the polynomial

∞
!
X X
(z) = zj · Ei β ij ,
j=1 i

where Ei is the value of error on position i. The j-th coefficient is equal

to the inner product of s0 and j-th row hj of the parity-check matrix H.
With a little abusive notation, we denote this coefficient (β j+1 ). When
calculating the coefficient we can cheat and use the fact that s · hj = 0
for each j = 0, . . . , 3. We have to consider only the third column where 1
was during transition changed to 2 and the fourth column, where one was
changed to 0. It means that s0 · hj = hj,3 − hj,4 = hj,3 + 2hj,4 . A computer
cannot cheat since it does not know where the position of errors are, but
the results that it gets are naturally the same.

(β) = β 3 − β 4 = β 3 + 2β 4 = 2β 12
(β 2 ) = β 6 − β 8 = β 6 + 2β 8 = 2β 5
(β 3 ) = β 9 − β 12 = β 9 + 2β 12 = 2β 10
(β 4 ) = β 12 − β 3 = β 12 + 2β 3 = β 6

Using the key equation ω(z) = σ(z) · (z), we will find the error positions
and will be able to correct the transition. Expanding the equation we get

ω1 z + ω2 z 2 = (1 + σ1 z + σ2 z 2 )(2β 12 z + 2β 5 z 2 + 2β 10 z 3 + β 6 z 4 ).

6
Vojtech@Brtnik.eu BCH Coding April 19, 2010

We shall decompose this into the following equations

ω1 = 2β 12
ω2 = 2β 5 + 2σ1 β 12
0 = 2β 10 + 2σ1 β 5 + 2σ2 β 12
0 = β 6 + 2σ1 β 10 + 2σ2 β 5 .

Solving the linear system in four variables, we get that ω1 = 2β 12 , ω2 = 0,

σ1 = 2β 6 , σ2 = β 7 .

Based on the theory, we write:

σ(z) = 1 + 2β 6 z + β 7 z 2 = (1 − β 3 z)(1 − β 4 z)
σ 0 (z) = 2β 6 + 2β 7 z
ω(z) = 2β 12 z 2

The roots of σ are β −3 and β −4 and we conclude that errors occurred on

positions 3 and 4, i.e. s03 and s04 have to be corrected. To calculate the
syndrome Ei , we will use the formula

ω(β −i )
Ei = −β i
σ 0 (β −i )
ω(β −3 ) 2β 6 3 β
6
E3 = −β 3 0 −3 = −β 3 6 = −β = β −3
σ (β ) 2β + 2β 4 2β 12
ω(β −4 ) 2β 4 4 β
4
E4 = −β 4 0 −4 = −β 4 6 = −β = 2β −4
σ (β ) 2β + 2β 3 β 12

The syndrome is thus

s00 = (0 0 0 1 2 0 0 0 0 0 0 0 0),

and if we subtract it from the received word

s0 = (2 0 0 2 0 0 2 1 2 1 0 0 0),

we get
s = (2 0 0 1 1 0 2 1 2 1 0 0 0),
exactly what was sent. This concludes the exercise.