Coding Theory: Homework 1

Due on Sept 16, 2014

Felipe Voloch

Ethan Leeman

1
Ethan Leeman Coding Theory (Felipe Voloch ): Homework 1 Chapter 1: Exercise 6

Chapter 1: Exercise 6
Let C be a code with distance d for even d. Then argue that C can correct up to d/2 1 many
errors but cannot correct d/2 errors. Using this or otherwise, argue that if a code C is t-error
correctable then it either has a distance of 2t + 1 or 2t + 2.
C being distance d means that every code words differ in more than d 1 entries, and there are two
codewords which differ in exactly d entries. If we consider closed balls of distance d/2 1 around the code
words, I claim these balls do not intersect: If Bd/21 (c1 ) and Bd/21 (c2 ) intersected at a point c0 , then
d(c0 , c1 ) d/2 1 and d(c0 , c2 ) d/2 1 so by the triangle inequality d(c1 , c2 ) d(c0 , c1 ) + d(c0 , c2 ) = d 2,
which is a contradiction of C being distance d. This implies that C corrects d/2 1 errors.
We have two codewords, named 1 , 2 , which are exactly distance d apart. Note that C cannot correct
d/2 errors, since we may exchange the entries of 1 to entries of 2 for d/2 entries. This word is exactly
distance d/2 from 1 and from 2 , so it could be a word with d/2 errors for either 1 or 2 . Regardless if
our decoding maps this word to 1 or 2 , it will not be able to correct d/2 errors. To summarize, C corrects
d/2 1 errors, but not d/2.
Suppose C is t error correctable. If the distance is odd, by proposition 1.4.1, then we have (d 1)/2 = t
or 2t + 1 = d. If d is even, then by the above we have t = d/2 1, so d = 2t + 2.

Chapter 1: Exercise 13
Argue that in any binary linear code, either all codewords begin with a 0 or exactly half the
codewords begin with a 0.
Suppose we have a basis for v1 , . . . , vk for this code. If all the vi begin with 0, then all the codewords
begin with 0 since the codewords are linear combinations of the vi . Suppose one of the vi begins with a 1.
Without loss of generality, let this be v1 . Every code word is a uniquely written as a linear combination (over
F2 ) of the vi s. Say there are n vectors which are linear combinations of v2 , . . . , vk , and m of them begin
with 1. Then there are 2n vectors in C, because we can either add v1 or not add it. Of those, the ones that
begin with 1 are the m vectors that start with 1 and when we do not add v1 , and the ones that do not begin
with 1, of which there are n m, when we do add v1 . So there are n vectors that begin with 1, out of the 2n
vectors. In summary, either all the codewords begin with 0, or exactly half the codewords begin with 0.

Chapter 2: Exercise 4
Prove that G2 from (2.3) has full rank.
Not sure what to make of this problem,

1 0 0 0 0 1 1

0 1 0 0 1 0 1
G=
0
.
0 1 0 1 1 0
0 0 0 1 1 1 1

It clearly has full column rank because the first 4 columns are linearly independent, and there are 4 rows,
so the rank is 4.

Chapter 2: Exercise 16
part (2): Prove if there exists an [n, k, d]2m code, then there also exists an [nm, km, d0 d]2 code.
part (3): If there exists an [n, k, d]q code, then there also exists an [n d, k 1, d0 dd/qe]q .
Part (2): F2m ' F2 [x]/q(x), where q is an irreducible of order m. Therefore, there is an isomorphism
between F2m and Fm m
2 as additive groups, and we can represent F2m in Z2 by defining the multiplication by
the isomorphism with F2 [x]/q(x).

Page 2 of 3
Ethan Leeman Coding Theory (Felipe Voloch ): Homework 1 Chapter 2: Exercise 16

Suppose we have an [n, k, d]2m with full rank G generating matrix of size k n, which exists by Theorem
2.2.1. Create a new matrix G0 of size km nm by the following procedure: For each a(i, j) in G will be
replaced by an m m block matrix. The first row will be the representation of a(i, j) under the isomorphism
above, the second row will be the representation of x a(i, j), and the mth row will be the representation of
xm1 a(i, j).
This matrix G0 has full rank. This is because if we had a linear dependence of the rows of G0 , this would
imply a linear dependence in the rows of G. If the sum includes the i1 , . . . , ir rows that is generated by a
single vector v, then that sum, by the homomorphism, is the same as the scalar multiple (xi+1 +. . .+xi+r )v.
Lastly, we know the distance of the G code is d, so every element which is a linear combination of the
rows of G has at least d non-zero entries. Therefore, every linear combination of rows in G0 has at least d
sets of m entries which are non-zero. This implies that d0 d.
Part (3): C has a non-zero vector with weight d. Let us rearrange the entries of C so that v has v1 , . . . , vd
non-zero and the rest zero. Extend {v} to a basis to create a generating matrix G. We construct G0 by
considering the submatrix that is deleting the first row and the first d columns. I claim G0 generates a
[n d, k 1, d0 dd/qe]q code.
Firstly, G0 is full rank: Suppose there is a linear combination of row vectors of G0 that equals 0. This
implies that there is a linear combination in G with non-zero entries only in the first d entries, that is linearly
independent from v (otherwise, G would not have been a generating matrix). However, this implies that
there is a vector in G with fewer than d non-zero entries, which is a contradiction, so G0 has full rank, and
generates a [n d, k 1, d0 ]q code based on the dimensions.
We now show that d0 dd/qe. Let w0 be the smallest vector in G0 of weight d0 . This is a linear combination
of the rows of G0 , so by construction, we have a vector w in G which is in C, but has d0 non-zero entries
outside of the first d entries. For every one of the first d entries, we calculate what constant we would
multiply v by to cancel it out. By the pigeonhole principle, one group has size at least dd/qe. But since this
linear combination is in G, it must have weight d itself. This implies that d0 dd/qe.

Page 3 of 3