The Virtues of Redundancy

The Virtues of Redundancy

An Introduction
to
Error-Correcting Codes

Paul H. Siegel
Director, CMRR
University of California, San Diego

2/28/03 1
Outline

• Information Storage and Transmission

• Repetition Codes
• Hamming Code (with a live demo!)
• Shannon’s Coding Theorem
• Compact Audio Disk Code (another live demo!)
• Summary

2/28/03 ECE Staff Luncheon 2

Information

• We all generate and use information:

Text Images

Scientific
Music Measurements

2/28/03 ECE Staff Luncheon 3

Information Storage

• We often store the information for later use:

• Disk drive

• Tape drive

• Zip/diskette drive

• CD-R/W

2/28/03 ECE Staff Luncheon 4

Information Storage

• And we want our storage devices to be:

• Reliable (no errors!)

• Compact (lots of storage capacity in a small space!)

• Inexpensive (almost free!)

2/28/03 ECE Staff Luncheon 5

Information Transmission

• We also send our information to others:

• Email

• Fax

• Wireless telephony

• Satellite broadcast

2/28/03 ECE Staff Luncheon 6

Information Transmission

• And we want the transmission system to be:

• Reliable (no errors!)

• Fast (instant downloads!)

• Inexpensive (almost free!)

2/28/03 ECE Staff Luncheon 7

Data Storage and Transmission

• Data storage and data transmission are two sides of

the same coin – communication systems.

A data storage system communicates information

through time, i.e. , “from now to then.”

A data transmission system communicates information

through space, i.e., “from here to there.”

2/28/03 ECE Staff Luncheon 8

Communication System

INFORMATION
SOURCE TRANSMITTER RECEIVER DESTINATION

SIGNAL RECEIVED
SIGNAL
MESSAGE MESSAGE

CHANNEL

2/28/03 ECE Staff Luncheon 9

Digitized Information

• We often represent the information digitally, as a

sequence of numbers.

• Each number is converted to a unique string of bits

(0’s and 1’s).

• The 0’s and 1’s are converted to an electrical signal

that is used to store or transmit the information.

2/28/03 ECE Staff Luncheon 10

“Noisy” Channels
• When we retrieve or receive the signal, it gets converted back
to a sequence of 0’s and 1’s.

• But an evil force is at work in the channel…NOISE!

• Electrical and magnetic devices can distort the electrical

signal, and the recovered sequence of bits may have errors:

1111111111  1111101111
sent received

2/28/03 ECE Staff Luncheon 11

A Noisy Communication System

INFORMATION
SOURCE TRANSMITTER RECEIVER DESTINATION
CHANNEL

SIGNAL RECEIVED
SIGNAL

MESSAGE
MESSAGE

NOISE
SOURCE

2/28/03 ECE Staff Luncheon 12

Coding to the Rescue
• To combat the effects of noise, we use an

Error Correction Code (ECC).

• The ECC adds redundancy in a special way that allows us

to correct the bit errors caused by the noisy channel.

Data Error Noisy

Compression Correction Channel
Code Code

Remove redundancy Add redundancy

2/28/03 ECE Staff Luncheon 13

Repetition

The simplest form of redundancy is: Repetition!

Sender Receiver
“0” Did she say “1” ?
I said “0” Sounded like “0”
One more time: “0” Sounded like “0” again

She was sending “0”

2/28/03 ECE Staff Luncheon 14
3-Repetition Code

• Encoding rule: Repeat each bit 3 times

Example:
1 . 0 . 1 .  111 . 000 . 111 .
• Decoding rule: Majority vote!
Examples of received codewords:
110 . 000 . 111 .  1 . 0 . 1 . Error-free!
111 . 000 . 010 .  1 . 0 . 0 . Error!

2/28/03 ECE Staff Luncheon 15

How good is this 3-repetition code?

• The code can correct 1 bit error per 3-bit codeword.

• The price we pay in redundancy is measured by the

efficiency or rate of the code, denoted by R:

R= #information bits / # bits in codeword

• For the 3-repetition code: R=33%

2/28/03 ECE Staff Luncheon 16

How good is this 3-repetition code?

Suppose that, on average, the noisy channel flips

1 code bit in 100

Then, on average, the 3-repetition code makes

only 1 information bit error in 3333 bits!

2/28/03 ECE Staff Luncheon 17

Can we do better?
How about repeat 5 times?

On average, only 1 bit error in 100,000 bits.

How about repeat 7 times?

On average, only 1 bit error in 2,857,142 bits

If we let the number of repetitions grow and grow, we can

approach perfect reliability !

2/28/03 ECE Staff Luncheon 18

What’s the catch????
The catch is:

As the number of repetitions grows to infinity, the

transmission rate shrinks to zero!!!

This means: slow data transmission / low storage density.

Is there a better (more efficient) error correcting code?

2/28/03 ECE Staff Luncheon 19

Hamming Code
• Invented by Richard W. Hamming in 1948.

• 7-bit codewords: 4 information bits, 3 redundant bits.

• Corrects 1 bit error per codeword

(like 3-repetition code)

• Code efficiency: R = 4/7  57%

(compared to rate R  33% for 3-repetition code)

2/28/03 ECE Staff Luncheon 20

Hamming Code

2 3
1
7 6
4

2/28/03 ECE Staff Luncheon 21

Hamming Code: Encoder

• Simple encoding rule:

5

1. Insert data bits in 1,2,3,4. 2 3

1
2. Insert “parity” bits in 5,6,7 7 6
to ensure an even number 4
of 1’s in each circle.

2/28/03 ECE Staff Luncheon 22

Hamming Code

5
Information: 0 1 0 0 1

2 3
Parity bit 5: 1 1 0
0 1
Parity bit 6: 0 0
1
Parity bit 7: 1
0
7 4 6

Codeword: 0 1 0 0 1 0 1

2/28/03 ECE Staff Luncheon 23

Hamming Code: Decoder

• Simple decoding rule:

1. Check the parity of each circle – 5

is the number of 1’s even or odd ?
2 3
The pattern of parities is called 1
the syndrome . 7 6
4
2. Look up the syndrome in the
decoding table.

2/28/03 ECE Staff Luncheon 24

Hamming Code: Decoding Table

Syndrome Flipped bit

even even even None!
odd odd odd 1 5
odd even odd 2
odd odd even 3 2 3
1
even odd odd 4 7 6
odd even even 5 4
even odd even 6
even even odd 7
2/28/03 ECE Staff Luncheon 25
How good is the Hamming code?
Suppose that, on average, the noisy channel flips

1 bit in 100

Then, on average, the Hamming code makes

On average, only 1 error in 1111 !!

(Not as good as 3-repetition, but efficiency is 57% compared to 33%)

2/28/03 ECE Staff Luncheon 26

Matrix description of Hamming Code

Bit # 1 2 3 4 5 6 7 5
Red circle 1 1 1 0 1 0 0
2 3
Green circle 1 0 1 1 0 1 0 1
Blue circle 1 1 0 1 0 0 1 7 6
4

This is the “parity-check matrix” for the code.

Encoding and decoding can be described algebraically .

2/28/03 ECE Staff Luncheon 27

Practical ECC

• Lots of “algebraic” codes can be designed by

choosing different parity-check matrices.

• Many useful and powerful codes have been found

and are used in many data transmission and data
storage applications.

2/28/03 ECE Staff Luncheon 28

Compact Disc ECC
• The compact disk uses a powerful code with 75%
efficiency

• It can correct a burst of about 4000 bits, which

corresponds to a scratch of length about 1-tenth of
an inch (2.5 mm).

• Using the redundancy in musical signals, the CD

player can “fix” a burst of 12,300 bits - a scratch
of length about 3-tenths of an inch (7.5 mm)!

2/28/03 ECE Staff Luncheon 29

CD Demonstration
(Warning: do not attempt this at home!)

8 mm thick line
2.5 mm thick 4 mm thick line

2/28/03 ECE Staff Luncheon 30

How “good” can a code be?

• What is the best tradeoff between efficiency

(transmission speed / storage density) and error-
correcting capability (reliability)?

• Does the efficiency have to go to zero in order for

the average number of decoder errors to approach
zero (as in the repetition codes)?

Amazingly, the answer is NO !

2/28/03 ECE Staff Luncheon 31

Shannon’s Coding Theorem

• A mathematical result proved by Claude E. Shannon (Bell

Labs) - in 1948 (coincidentally!)

• For any noisy channel, there is a maximum achievable

efficiency (larger than 0), called the channel capacity, and
denoted by C.

• For any rate R smaller than C, there exist codes that can
approach perfect reliability!

• For any rate R greater than C, perfect reliability is

IMPOSSIBLE !

2/28/03 ECE Staff Luncheon 32

Shannon’s Coding Theorem

For the channel that flips 1 bit in 100 bits,

the capacity is C = 91.92% !!!

Achievable
Region
More
Errors H
C=0.9192
3R Impossible
5R
Region
7R
Perfect
Reliability
Higher rate R

2/28/03 ECE Staff Luncheon 33

Claude E. Shannon

(CMRR Lobby)

2/28/03 ECE Staff Luncheon 34

The Inscription

2/28/03 ECE Staff Luncheon 35

The Formula on the “Paper”

The paper shows the formula

for the capacity of a discrete
noisy channel [Shannon, 1948]

C  Max (H(x) – Hy (x))

2/28/03 ECE Staff Luncheon 36

Capacity-approaching codes

• Shannon proved that lots of such “good” codes exist…

but he did not show how to construct the encoders and
decoders!

• Finding such practical good codes has been the “holy

grail” of coding theorists since 1948.

• Turbo codes (1993) and Low-Density Parity-Check codes

(2001) finally have come close to achieving Shannon’s
theoretical limits !!! Hooray!!

• Are coding theorists out of work? No way….

2/28/03 ECE Staff Luncheon 37

Summary

• Coding theory is fun and useful!

• You’ve been a nice audience – Thanks!

2/28/03 ECE Staff Luncheon 38