Escolar Documentos
Profissional Documentos
Cultura Documentos
Rong-Jaye Chen
Outline
[1] Introduction [2] Basic assumptions [3] Correcting and detecting error patterns [4] Information rate [5] The effects of error correction and detection [6] Finding the most likely codeword transmitted [7] Some basic algebra [8] Weight and distance [9] Maximum likelihood decoding [10] Reliability of MLD [11] Error-detecting codes [12] Error-correcting codes
p2.
[1] Introduction
Coding theory
The study of methods for efficient and accurate transfer of information Detecting and correcting transmission errors
Information Source
k-digit
n-digit
Noise
n-digit
k-digit
p3.
{0,1}
Channel Channel
{0,1}
011011001
2. Identifying the beginning of 1st word 3. The probability of any digit being affected in transmission is the same as the other one.
p5.
0 1
p: reliability
0 1
p6.
1 p 1 p
In many books, p denotes crossover probability. Here crossover probability(error prob.) is 1-p
C 1 ! {00 , 01 ,10 ,11} Cannot detect any errors !!! C 2 ! {000000 ,010101 ,101010 ,111111 }
source Channel
110101
correct
010101
parity-check digit
110 ?
010
correct
000 ? 011 ?
p7.
[4] Information rate Definition: information rate of code C 1 is defined as log 2 C n where n is the length of C
Examples
2
1 log c1 ! 2
4 ! 1
1 ! 3
c3
2 ! 3
p8.
Then Pr(a word is transmitted incorrectly) = 1-p11 11x10-8 11x10-8(wrong words/words)x107/11(words/sec)=0.1 wrong words/sec
1 wrong word / 10 sec 6 wrong words / min 360 wrong words / hr 8640 wrong words / day
p9.
Pr(at least 2 errors in a word)=1-p12-12 x p11(1-p) 66x10-16 So 66x10-16 x 107/12 5.5 x 10-9 wrong words/sec one word error every 2000 days! The cost we pay is to reduce a little information rate + retransmission(after error detection!)
p10.
reasonable information rates low encoding and decoding costs some error-correcting capabilities
p11.
BSC channel
p d n
J p ( R , Z ) ! p n d (1 p ) d
Example
Code length = 5
Jp (R, R ) ! p5
is sent when
p
is received if
( R , Z ) ! max{
( u , ) : u C}
Theorem 1.6.3 Suppose we have a BSC with < p < 1. Let R 1 and R 2 be codewords and a word, each of lengthn . Suppose that R 1 and disagree in d 1 positions and R 2 and disagree in d 2 positions. Then
J p (R 1 ,Z ) e J p (R 2 ,Z )
iff d1 u d 2
p13.
Example
R !?
channel
Z ! 00110
p ! 0.98
R
01101 01001 10100 10101
Z)
p14.
0 0 ! 0, 0 1 ! 1, 1 0 ! 1, 1 1 ! 0 Multiplication 0 0 ! 0, 1 0 ! 0, 0 1 ! 0, 11 ! 1
the set of all binary words of length n
0 [ ! 0n , 1 [ ! [
0 n zero word
p15.
Kn is a vector space
u , v, w a, b
words of length n scalar
Hamming weight
t (v)
R
wt (000000 ) ! 0
wt (110101) ! 4,
Hamming distance
d (v, w)
R and w disagree
d (01011,00111) ! 2,
d (10110,10110) ! 0
p17.
Some facts
u, v, w
a
digit
words of length n
1. 0 3. 0
wt ( v ) d (v, w )
n v ! 0s n v ! w
2 . wt ( v ) ! 0 iff 4 . d ( v , w ) ! 0 iff
8 . wt ( av ) ! a wt ( v ) 9 . d ( av , aw ) ! a d ( v , w )
w=v+u
channel
CMLD IMLD
Source string x
codeword CMLD
u
Error pattern
decode
If only one word v in C closer to w , decode it to v If several words closest to w, select arbitrarily one of them
IMLD
If only one word v in C closer to w, decode it to v If several words closest to w, ask for retransmission
u !vw
wt (v2 w)
The most likely codeword sent is the one with the error pattern of smallest weight Example Construct IMLD. |M|=3 , C={0000, 1010, 0111}
Received w 0000 1000 0100 0010 0001 1100 1010 1001 0110 0101 Error Pattern 0000 + w 0000 1000 0100 0010 0001 1100 1010 1001 0110 0101 1010 + w 1010 0010 1110 1000 1011 0110 0000 0011 1100 1111 0111 + w 0111 1111 0011 0101 0110 1011 1101 1110 0001 0010 Decode v 0000 0000 0000 1010 0111 0111
p20.
The probability that if v is sent over a BSC of probability p then IMLD correctly concludes that v was sent
w L(v )
Can detect
Example
Error Pattern u 000 100 010 001 110 101 011 111 Can detect Cant detect
C ! {000,111}
v = 000 000 100 010 001 110 101 011 111 v = 111 111 011 101 110 001 010 100 000
u !vw
000 000 ! 000 000 111 ! 111 111 111 ! 000
p22.
Theorem 1.11.14
A code C of distance d will at least detect all non-zero error patterns of weight less than or equal to d-1. Moreover, there is at least one error pattern of weight d which C will not detect.
t error-detecting code
It detects all error patterns of weight at most t and does not detect at least one error pattern of weight t+1 A code with distance d is a d-1 error-detecting code.
p23.
+
u
Error pattern
For all v in C , if it is closer to v than any other word in C, a code C can correct u.
Theorem 1.12.9
A code of distance d will correct all error patterns of weight less than or equal to (d 1) / 2 . Moreover, there is at least one error pattern of weight 1+(d 1) / 2 which C will not correct. It corrects all error patterns of weight at most t and does not correct at least one error pattern of weight t+1 A code of distance d is a (d 1) / 2 error-correcting code.
p24.
t error-correcting code
d !3
Error Pattern 000 + w 000* 100* 010* 001* 110 101 011 111 111 + w 111 011 101 110 001* 010* 100* 000* Decode v 000 000 000 000 111 111 111 111
(3 1) / 2 ! 1
C corrects error patterns 000,100,010,001
p25.