9/24/2013

Advanced Coding Theory

ECE581

Overview
Codes?
Digital Communication

jaj
Linear Block Codes Signal Detection
Channel
Hamming Codes
Nikesh Bajaj Shannon Theorem
nikesh.14730@lpu.co.in
Digital Signal Processing
Lovely Professional University 2 By Nikesh Bajaj

Channel Coding Ba

Block Codes
Words, Codeword
Code rate
Hamming distance/Min d*
sh
Non binary
Hamming weight/ Min w*
d(c1,c2)=w(c1-c2)
Block length (n)
Code(n,k) M=qk
3 By Nikesh Bajaj 4 By Nikesh Bajaj
ke

Linear Block Codes Encoding

All Zeros C ={0000 1010 0101 1111}
Ni

Sum of any two codewords is a codeword

d* = w* Proof
Example
C ={00000 10100 11110 11001}
C={0100 1111}
C ={0000 1010 0101 1111}
C={12, 21} over GF(3)

5 By Nikesh Bajaj 6 By Nikesh Bajaj

1
 9/24/2013

Encoding: Matrix form G Linear Codes

1 0 1 Encoding
G

jaj
0 1 0

7 By Nikesh Bajaj 8 By Nikesh Bajaj

Generator Matrix (Properties)
Size kxn,
rank k
Ba
Rows are linearly independent and basis vector
Find linear code

Q: S={1100 0100 0011} over GF(2)
Q: S={12 21} over GF(3)
sh
Any set of basis vector can be used to generate code span
Generator matrix is not unique for given linear code
Saves memory space
Sufficiency

9 By Nikesh Bajaj 10 By Nikesh Bajaj

ke

Systematic Codes Find systematic G

Ni

Code(5,2) over GF(3) . 0 1 2 1

G= 1 0 1 0
1 2 2 1

11 By Nikesh Bajaj 12 By Nikesh Bajaj

2
 9/24/2013

Linear Codes Parity check matrix H

Size==??
Encoding
purpose
Limitation

jaj

Uniqueness
Check matrix or Parity Check Matrix H Sufficiency
G =[I | P] or G = [P | I] d* =min rows in HT or column of H whose sum is
H = [-PT | I ] or H = [ I | -PT ] zero (in linear code) and binary

13 By Nikesh Bajaj 14 By Nikesh Bajaj

Hamming Code
1
0
G
0

0
0 0 0 1 0 1

0 1 0 0 1 0
0 0 1 0 1 0
Ba
1 0 0 1 1 1

Parity check matrix H
Q: find for H

. 0 1 2 1
G= 1 0 1 0
1 2 2 1
sh
15 By Nikesh Bajaj 16 By Nikesh Bajaj
ke

What you can find? Error Correction and Detection

Ni

17 By Nikesh Bajaj 18 By Nikesh Bajaj

3
 9/24/2013

Singleton Bound Error detecting capability

Singleton Bound For code(n,k) Code (n, k, d*)
d* n k + 1 d*-1

jaj

Hamming Sphere C={000 111}

C={001 110 101}
Maximum Distance Separable (MDS) code

d* =n-k+1

19 By Nikesh Bajaj 20 By Nikesh Bajaj

Decoding

Spheres of t radius, t errors
Vector space, q-ary, n-tuples
d* 2t+1
Ba Decoding

Coders
sh
Worst case Incomplete Coder
Nearest neighborhood decoding Complete Coder
C={00000 01010 10101 11111}
Tx=11111--- Rx =11110
Tx=00000 --- Rx =01000

21 By Nikesh Bajaj 22 By Nikesh Bajaj

ke

Coset Coset
Ni

For C (n, k) over GF(q) and a be any vector of Coset Leader: Vector of minimum weight
length n then the set If two, choose any one
Coset a + C ={c1+a, c2+a } c1, c2 C Find Cosets for G=[1 0 1; 0 1 0]
Theorems (Prove) Ref: 106 in RB GF(qn) = C (a1+C) (a2+C) (at+C)
Every vector b of length n is in some Coset t =qn-k -1
Each Coset contain exactly qk vector
Two Cosets are either disjoint or Concide
If a+C is Coset of C and b a+C then b+C =a+C
a and b are said to be in same Coset if (a-b) C

23 By Nikesh Bajaj 24 By Nikesh Bajaj

4
 9/24/2013

Standard Array Standard array: Decoding

Standard Array: for Code C (n,k), qn-k x qk array of all STEPS
.
vector In GF(qn) C={0000 1011 0101 1111}

jaj
If Rx v = 1101
c= v - e

25 By Nikesh Bajaj 26 By Nikesh Bajaj

Standard array: Decoding Ba Example

sh
27 By Nikesh Bajaj 28 By Nikesh Bajaj
ke

Syndrome: Decoding Syndrome: Decoding

Ni

Why syndrome?

x and y are in same coset iff both have same

syndrome

29 By Nikesh Bajaj 30 By Nikesh Bajaj

5
 9/24/2013

Example: Syndrome Decoding Bounds and Codes

Singleton Bound
Hamming Bound

jaj

Perfect Code
Hamming Code
Optimal Code
MDS Code

31 By Nikesh Bajaj 32 By Nikesh Bajaj

Sphere and # vector
A sphere of radius t (0<=t<=n) contains vectors
Ba
Perfect Codes
Perfect Code is one which achieves the Hamming
Bound
MV = qn
sh
Example: # of vector of distance 2 from a vector of length
4 in GF(2)
Hamming Bound
MV <= qn

33 By Nikesh Bajaj 34 By Nikesh Bajaj

ke

Equivalent Codes Equivalent Linear Codes

Ni

Two q-ary codes are called equivalent codes if one Two Linear q-ary codes are called equivalent codes
can be obtained from other by one or both if one can be obtained from other by one or both
Permutation of components Multiplication of components by any non-zero scalar
Permutation of position of codewords Permutation of position of codewords

2 0 1 2 2 2

C 1 2 0 C 1 1 1
0 1 2 0 0 0

35 By Nikesh Bajaj 36 By Nikesh Bajaj

6
 9/24/2013

Hamming Codes (binary*) Optimal Linear Codes

n = 2m 1 For optimal code (n, k, d*)
k = 2m m 1 No (n-1, k, d*),

jaj
(n, k) =(2m 1, 2m m 1) (n+1, k+1, d*) or
(n+1, k, d*+1) exist

1 0 0 0 1 0 1
(24, 12, 8)
0 1 0 0 1 1 1
G 23, 12, 8 X 23 12 7
0 0 1 0 0 1 0

0 0 0 1 0 1 0 25, 13, 8 X
25, `12, 9 X
37 By Nikesh Bajaj 38 By Nikesh Bajaj

MDS Codes

Ba
(n, n-r, d*) code satisfy d* =r+1

(n, n-r, r+1) is MDS Codes

Linear Codes
.

Set S,
Extended field GF(qn ) linear span <S>
linear span <S> is sub space of GF(qn )
sh

Linear code C= <S>

39 By Nikesh Bajaj 40 By Nikesh Bajaj

ke

Channel Coding
Ni

41 By Nikesh Bajaj 42 By Nikesh Bajaj

7
 9/24/2013

Repetition Code Coding Gain

0000
1111

jaj
Code rate =1/n
Example 2.4 from Ranjan boss

43 By Nikesh Bajaj 44 By Nikesh Bajaj

Channel Coding

Words and Codeword
Ba
Code(n,k) or code(n,k,d)
Code rate k/n
Code C = {000 111 }
sh

Hamming distance d
Hamming weight w
Error detection capability ed
Error Correction Capability ec

45 By Nikesh Bajaj 46 By Nikesh Bajaj

ke
Ni

47 By Nikesh Bajaj