디지털통신시스템(W14L1)

Coding - Part 2
오류 정정…

전북대학교 전자공학부 송상섭

Error Correcting Codes란?
--- 오류 정정 부호란? ---

 채널은 불완전하다…틀리고…오류가 발생하고…

 건축가  지진에 견디는 건물을 설계…

 ECC 전문가  mathematical insurance against an imperfect digital
world..

 잡음 : lightning, human error, equipment malfunction, voltage surge,…

 Murphy’s Law
“If anything can go wrong, it will go!”

2
Redundancy를 이용

CODNG THEORY IS AN INTRSTNG SUBJCT

 CODING THEORY IS AN INTERISTING SUBJECT

전지공학과  전기공학과, 전자공학과

3
Good ECC(채널 코드)라면…
 오류를 정정할 수 있어야 하고,
 Encoding/ decoding을 고속/ 고효율로 할 수 있고,
 Redundancy를 최소로 사용한다.

Information Channel Modulator

source encoder

C
H
A
N Noise
N
E
L

Use of Channel Demodulator

Information decoder

4
Error Correction Codes

Linear Block Code

Cyclic Code
<Meggitt> Convolutional
Code
BCH Code
Turbo code

CDMA
RS
CD-ROM
Code
QR-code

5
Repetition Code
송신측 수신측 majority vote

m  c0c1c2 r0r1r2  m r0r1r2  m 1 bit error can be

0  000 000  0 110  1 corrected
1  111 001  0 101  1
010  0 011  1
100  0 111  1

6
3 Cards
#1 #2 #4

1357 2367 4567

421
0 000
1 001 2
2 010 3 6
3 011 7
4 100 1 5 4
5 101
6 110
7 111

7
 Encoder b1 b 2 b 3 b 4 b 5 b 6 b7 Bit position
1 0 1 1 4-bit message
b1 : 1 3 5 7 Even parity checks
(7, 4) b2 : 2 3 6 7
b4 : 4 5 6 7
Hamming
0 1 1 0 0 1 1 Tx codeword
Code
X Channel error
0 1 0 0 0 1 1 Received word

Decoder b1 b3 b5 b7 = 0001 fail  1

Syndrome = 011=3
b2 b3 b6 b7 = 1011 fail  1
 b3 is in error
b4 b5 b6 b7 = 0011 pass  0
1 Channel error
0 1 1 0 0 1 1 Recovered word
1 0 1 1 Original message

(7, 4) Hamming code를 사용하면, 7비트 중 1비트의 오류는 정정 가능하다.

8
(7, 4) Hamming Code의 성능
Uncoded message bits (channel) Coded bits

BER(coded) before correction

pc  0.001
Message block 오율
7
7
p M     pcj (1  pc ) 7  j
j
j2  

 1  (1  pc ) 7  7 pc (1  pc ) 6
 1  (0.999) 7  7  0.001(0.999) 6
 1  0.993  0.00696
 4  10 5

BER(coded) after correction

pb  p M / 4  10 5
9
(15, 11) Hamming Code 8421
0 0000

#1 1, 3, 5, 7, 9, 11, 13, 15 1 0001

2 0010
3 0011
#2 2, 3, 6, 7, 10, 11, 14, 15
4 0100
5 0101
#4 4, 5, 6, 7, 12, 13, 14, 15
6 0110
7 0111
#8 8, 9, 10, 11, 12, 13, 14, 15
8 1000
9 1001

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
10 1010
11 1011
p1 p2 1 p4 0 1 1 p8 0 0 1 0 1 0 1
12 1100
13 1101
질문 1. Determine parity checks p1, p2, p3, p4.
14 1110
2. Decode r = 1011 0100 1001 0010 15 1111

Hamming code는 DDR4/DDR5 RAM 등에도 사용된다.

10
에러 패턴과 채널 부호…
 에러 패턴
 산발적 에러 (random error) : 에러가 떨어져 있으면 채널 부호가
고칠 수 있으나…
 연속적 에러 (burst error) : 일단 에러가 몰려 있으면…

 채널 부호
 해밍 부호, RS 부호, 터보 부호, LDPC 부호, …
 Galois field, …

 Coded system의 성능은?

11
BPSK…
BPSK Eb BPSK BER
변조 N0 복조 PB  Q( 2 Eb / N 0 )

BER
PB  1.02  10 5 Coherent
BPSK

9.6 dB
12
sn=0.1 BER=0.762*10-23 sn=0.2 BER=0.287*10-6

sn=0.3 BER=0.434*10-3 sn=0.5 BER=0.0228

13
Coded BPSK…

BPSK Eb BPSK BER(uncoded)

변조 N0 복조 PB ,u  Q( 2 Eb / N 0 )

(n, k) BPSK BPSK t 개까지

부호화 변조 E c  k  Eb 복조 오류수정
 
N0  n  N0
BER(coded) before correction
p  Q( 2 Ec / N 0 )

BER(coded) after correction

1 n n j
PB ,c   j   p (1  p ) n  j
n j t 1  j 

14
 1 p correct
PB,coded with (n, k, t) Code… p error

n
1
j 
n j
PB ,c 
n

j  t 1 

j
p (1  p ) n  j

1 n 2
When t=1, PB ,c  2  p (1  p ) n  2
n 2
 (n  1) p 2 (1  p ) n  2
 (n  1) p 2 , when p is small

B. Sklar, Digital communications: fundamentals and applications, 2nd ed., PH, 2001

15
 BER(uncoded)
Coded BPSK… PB ,u  Q( 2 Eb / N 0 )  Q( 18.24 )
(n, k, t) = (15, 11, 1) 부호  1.02  10  5
Eb
 10 0.96  9.12(9.6 dB)
N0
E c  k  Eb Ec 11 Eb
    6.69(8.3 dB)
N0  n  N0 N 0 15 N 0
BER(coded) before correction
p  Q ( 2 Ec / N 0 )
p  Q( 13.38 )  1.36  10  4

PB ,c  ( n  1) p 2 BER(coded) after correction

when p is small PB ,c  (n  1) p 2
 14  (1.36  10  4 ) 2
 2.59  10  7

16
Matlab code
Eb_No = 10^-1:0.1:1000;
Ec_No = Eb_No*11/15; PB ,c  (n  1) p 2
ENR = 10*log10(Eb_No);
PBu = Q(sqrt(2*Eb_No));
semilogy(ENR,PBu,'k');
hold on;
p = Q(sqrt(2*Ec_No));
PBc = 14 * p^2;
semilogy(ENR,PBc,'r');
grid on;
axis([0,15,10^-9,1])
xlabel('Eb/No');
ylabel('Coded BPSK and uncoded BPSK');
% function Qfunct = Q(x)
% Qfunct = 0.5*erfc(x/sqrt(2));

17
Better Performance…

BER(uncoded)
PB ,u  1.02  10 5

BER(coded) after correction

PB ,c  2.59  10  7 9.6 dB

18
Coding Gain…
Or lower Eb/N0 for the
same performance !!

BER(uncoded)
PB ,u  1.02  10 5

BER(coded)
PB ,c  1.02  10 5 8.1 dB 9.6 dB
Coding gain of 1.5 dB
19
And More …
Decoded bit-error probability, pb

Hamming (7, 4) t = 1
Hamming (15, 11) t = 1
Hamming (31, 26) t = 1
Extended Golay (24, 12) t = 3
BCH (127, 36) t = 15
BCH (127, 64) t = 10

Uncoded

Greater CG
Eb/N0 dB

20
요약
 3 codings
 Source codes(압축) – mp3, jpeg, mpeg
 Secrecy codes(보안) – DES, AES, RSA
 Channel codes(보호) – block(Hamming, BCH, RS, LDPC),
convolutional, turbo, …

21
요약…

22