0% found this document useful (0 votes)
107 views3 pages

Leeman Coding Theory 1

The document contains homework problems from a coding theory course. It includes solutions to several problems: 1) A proof that a code C with distance d can correct up to d/2-1 errors but not d/2 errors. 2) A proof that the generator matrix G2 from the document has full rank. 3) A proof that if there exists an [n,k,d] code over F2m, then there exists an [nm,km,d≥d] code over F2, and if there exists an [n,k,d] code over Fq, then there exists an [n-d,k-1,d'≥dd/qe]

Uploaded by

bugra
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
107 views3 pages

Leeman Coding Theory 1

The document contains homework problems from a coding theory course. It includes solutions to several problems: 1) A proof that a code C with distance d can correct up to d/2-1 errors but not d/2 errors. 2) A proof that the generator matrix G2 from the document has full rank. 3) A proof that if there exists an [n,k,d] code over F2m, then there exists an [nm,km,d≥d] code over F2, and if there exists an [n,k,d] code over Fq, then there exists an [n-d,k-1,d'≥dd/qe]

Uploaded by

bugra
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

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

You might also like