The Laws of Cryptography With Java Code

The Laws of
Cryptography
with Java Code
by Neal R. Wagner
Copyright c 2003 by Neal R. Wagner. All rights reserved.
Permission is granted to retrieve a single electronic copy of this book for personal use, but the
permission does not extend to printing a copy of the book or to making a copy, electronic or in
any other form, for any other than personal use.
Table of Contents
Parts and Chapters
Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
I. Preliminaries . . . . . . . . . . . . . . . . . . . . . 2
1. Cryptographers’ Favorites . . . . . . . . . . . . . . . . . . 3
2. Cryptographers’ Favorite Algorithms . . . . . . . . . . . . 14
II. Coding and Information Theory . . . . . . . . . . . . 22

3. Coding and Information Theory . . . . . . . . . . . . . . 23
4. Visualizing Channel Capacity . . . . . . . . . . . . . . . 31
5. The Huffman Code for Compression . . . . . . . . . . . . 35
6. The Hamming Code for Error Correction . . . . . . . . . . 44
7. Coping with Decimal Numbers . . . . . . . . . . . . . . . 48
8. Verhoeff’s Decimal Error Detection . . . . . . . . . . . . . 54
III. Introduction to Cryptography . . . . . . . . . . . . . 60

9. Cryptograms and Terminology . . . . . . . . . . . . . . . 61
10. The One-Time Pad . . . . . . . . . . . . . . . . . . . . 66
11. Conventional Block Cipher Cryptosystems . . . . . . . . . . 70
??. Conventional Stream Cipher Cryptosystems
IV. Public Key Cryptography . . . . . . . . . . . . . . . 78

12. Public Key Distribution Systems . . . . . . . . . . . . . . 79
13. Public Key Cryptography: Knapsacks . . . . . . . . . . . . 82
14. The RSA Public Key Cryptosystem . . . . . . . . . . . . . 86
15. Rabin’s Version of RSA . . . . . . . . . . . . . . . . . . 93
??. Elliptic Curve Cryptosystems
??. Other Public Key Cryptosystems
V. Random Number Generation . . . . . . . . . . . . . 98

16. Traditional Random Number Generators . . . . . . . . . . 99
17. Random Numbers From Chaos Theory . . . . . . . . . . . 106
18. Statistical Tests and Perfect Generators . . . . . . . . . . . 112
VI. The Advanced Encryption Standard (AES) . . . . . . . . 114

iv The Laws of Cryptography
19. Introduction to the AES . . . . . . . . . . . . . . . . . . 115

20. The Finite Field GF(256) . . . . . . . . . . . . . . . . . 119
21. The S-Boxes . . . . . . . . . . . . . . . . . . . . . . . 127
22. Key Expansion . . . . . . . . . . . . . . . . . . . . . . 130
23. Encryption . . . . . . . . . . . . . . . . . . . . . . . 133
24. Decryption . . . . . . . . . . . . . . . . . . . . . . . 136
VII. Hash Functions and Digital Signatures

??. One-Way Functions and Hash Functions
??. Digital Signatures
VIII. Randomization Techniques
??. Simple Randomization
??. More Complex Tricks
??. The Rip van Winkle Cipher and Rabin’s Scheme
IX. Identification and Key Distribution . . . . . . . . . . . 140
25. Passwords . . . . . . . . . . . . . . . . . . . . . . . . 141
26. Zero-Knowledge Protocols . . . . . . . . . . . . . . . . . 144
27. Identification Schemes . . . . . . . . . . . . . . . . . . 150
28. Threshold Schemes . . . . . . . . . . . . . . . . . . . . 153
??. Case Study: the Secure Shell (ssh)
Java Programs . . . . . . . . . . . . . . . . . . . . . . 159

Appendices . . . . . . . . . . . . . . . . . . . . . . . 307
A. Using Printed Log Tables . . . . . . . . . . . . . . . . . . . 309
B. Unsigned bytes in Java . . . . . . . . . . . . . . . . . . . . 310
C. Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
Table of Contents
Laws
I. Preliminaries
1. Cryptographers’ Favorites
Law XOR-1 . . . . . . . . . . . . . . . . . . . . . . . 3
Law XOR-2 . . . . . . . . . . . . . . . . . . . . . . . 5
Law LOG-1 . . . . . . . . . . . . . . . . . . . . . . . 5
Law LOG-2 . . . . . . . . . . . . . . . . . . . . . . . 7
Law GROUP-1 . . . . . . . . . . . . . . . . . . . . . 8
Law FIELD-1 . . . . . . . . . . . . . . . . . . . . . . 9
Law FIELD-2 . . . . . . . . . . . . . . . . . . . . . . 10
Law FERMAT-1 . . . . . . . . . . . . . . . . . . . . . 10
2. Cryptographers’ Favorite Algorithms
Law GCD-1 . . . . . . . . . . . . . . . . . . . . . . . 14
Law EXP-1 . . . . . . . . . . . . . . . . . . . . . . . 17
Law PRIME-1 . . . . . . . . . . . . . . . . . . . . . 19
Law PRIME-2 . . . . . . . . . . . . . . . . . . . . . 20
II. Coding and Information Theory
3. Coding and Information Theory
Law ENTROPY-1 . . . . . . . . . . . . . . . . . . . . 23
Law ENTROPY-2 . . . . . . . . . . . . . . . . . . . . 24
Law INFORMATION-1 . . . . . . . . . . . . . . . . . 25
Law SHANNON-1 . . . . . . . . . . . . . . . . . . . . 29
4. Visualizing Channel Capacity
5. The Huffman Code for Compression
Law SHANNON-2 . . . . . . . . . . . . . . . . . . . . 35
Law COMPRESSION-1 . . . . . . . . . . . . . . . . . 36
6. The Hamming Code for Error Correction
Law HAMMING-1 . . . . . . . . . . . . . . . . . . . 46
7. Coping with Decimal Numbers
Law DECIMAL-1 . . . . . . . . . . . . . . . . . . . . 48
Law DECIMAL-2 . . . . . . . . . . . . . . . . . . . . 53
8. Verhoeff’s Decimal Error Detection
Law DECIMAL-3 . . . . . . . . . . . . . . . . . . . . 57
III. Introduction to Cryptography
9. Cryptograms and Terminology
vi The Laws of Cryptography
Law CRYPTOGRAPHY-1a . . . . . . . . . . . . . . . . 62
Law CRYPTOGRAPHY-1b . . . . . . . . . . . . . . . . 62
Law CRYPTOGRAPHY-2 . . . . . . . . . . . . . . . . 63
Law CRYPTANALYSIS-1 . . . . . . . . . . . . . . . . . 64
10. The One-Time Pad
Law PAD-1 . . . . . . . . . . . . . . . . . . . . . . . 69
11. Conventional Block Cipher Cryptosystems
Law BLOCKCIPHER-1 . . . . . . . . . . . . . . . . . 71
Law BLOCKCIPHER-2 . . . . . . . . . . . . . . . . . 73
IV. Public Key Cryptography
12. Public Key Distribution Systems

13. Public Key Cryptography: Knapsacks
14. The RSA Public Key Cryptosystem
Law RSA-1 . . . . . . . . . . . . . . . . . . . . . . . 86
Law RSA-2 . . . . . . . . . . . . . . . . . . . . . . . 91
15. Rabin’s Version of RSA
Law RABIN-1 . . . . . . . . . . . . . . . . . . . . . . 93
V. Random Number Generation
16. Traditional Random Number Generators

Law RNG-1 . . . . . . . . . . . . . . . . . . . . . . . 99
Law RNG-2 . . . . . . . . . . . . . . . . . . . . . . . 99
Law RNG-3 . . . . . . . . . . . . . . . . . . . . . . 100
17. Random Numbers From Chaos Theory
18. Statistical Tests and Perfect Generators
VI. The Advanced Encryption Standard (AES)
19. Introduction to the AES

Law AES-1 . . . . . . . . . . . . . . . . . . . . . . 115
20. The Finite Field GF(256)
21. The S-Boxes
22. Key Expansion
23. Encryption
Table of Contents: Laws vii
24. Decryption

IX. Identification and Key Distribution
25. Passwords
26. Zero-Knowledge Protocols
27. Identification Schemes
28. Threshold Schemes
Law THRESHOLD-1 . . . . . . . . . . . . . . . . . 157
Java Programs
Appendices
A. Using Printed Log Tables
B. Unsigned bytes in Java
Law JAVA-BYTES-1 . . . . . . . . . . . . . . . . 311
Law JAVA-BYTES-2 . . . . . . . . . . . . . . . . 312
Table of Contents
Java Programs
I. Preliminaries
1. Cryptographers’ Favorites
a. Demonstration of Xor . . . . . . . . . . . . . . . 4 161
b. Formulas for logs . . . . . . . . . . . . . . . . 5 162
c. Fermat’s Theorem Illustrated . . . . . . . . . . . 12 163
2. Cryptographers’ Favorite Algorithms
a. Basic GCD Algorithm . . . . . . . . . . . . . . . 15 165
b. Extended GCD Algorithm . . . . . . . . . . . . . 16 166
c. Extended GCD Algorithm (debug version) . . . . . . 17 167
d. Testing Two Exponential Algorithms . . . . . . . . 19 169
II. Coding and Information Theory
3. Coding and Information Theory

a. Formula for Channal Capacity . . . . . . . . . . . 26 172
b. Table of Channal Capacities . . . . . . . . . . . . 26 173
c. Inverse of the Channal Capacity formula . . . . . . . 26 175
d. Table of Repetition Codes . . . . . . . . . . . . . 28 177
4. Visualizing Channel Capacity
a. The Simulation Program . . . . . . . . . . . . . 33 179
5. The Huffman Code for Compression
a. The Huffman Algorithm . . . . . . . . . . . . . . 41 183
b. Two Distinct Huffman Codes . . . . . . . . . . . . 43 191
6. The Hamming Code for Error Correction
a. The Hamming Algorithm . . . . . . . . . . . . . 47 193
7. Coping with Decimal Numbers
a. U.S. Banking Scheme . . . . . . . . . . . . . . . 53 203
b. IBM Scheme . . . . . . . . . . . . . . . . . . 53 206
c. ISBN mod 11 Scheme . . . . . . . . . . . . . . . 53 209
d. Mod 97 Scheme . . . . . . . . . . . . . . . . . 53 212
e. Hamming mod 11 Scheme, Error Correction . . . . . 53 215
f. Hamming mod 11 Scheme, Double Errors . . . . . . 53 219
8. Verhoeff’s Decimal Error Detection
a. Use of the Dihedral Group . . . . . . . . . . . . . 57 223
b. Verhoeff’s Scheme . . . . . . . . . . . . . . . . 57 226
Table of Contents: Java Programs ix
III. Introduction to Cryptography
9. Cryptograms and Terminology

a. Cryptogram Program . . . . . . . . . . . . . . . 62 229
10. The One-Time Pad
a. Caesar Cipher . . . . . . . . . . . . . . . . . . 67 232
b. Beale Cipher . . . . . . . . . . . . . . . . . . 67 235
c. Generate a One-time Pad . . . . . . . . . . . . . 69 239
d. Wheels to Encrypt/Decrypt With a Pad . . . . . . . 69 242
11. Conventional Block Cipher Cryptosystems
IV. Public Key Cryptography
12. Public Key Distribution Systems

13. Public Key Cryptography: Knapsacks
14. The RSA Public Key Cryptosystem
a. RSA Implementation . . . . . . . . . . . . . . . 90 246
b. Faster RSA, Using Chinese Remainder Theorem . . . 92 251
15. Rabin’s Version of RSA
a. Square Roots mod n = p*q . . . . . . . . . . . . . 94 256
V. Random Number Generation
16. Traditional Random Number Generators

a. Linear Congruence Random Number Generators . . 103 259
b. Exponential and Normal Distributions . . . . . . . 103 262
17. Random Numbers From Chaos Theory
a. The logistic Lattice as a RNG . . . . . . . . . . 110 266
18. Statistical Tests and Perfect Generators
a. Maurer’s Universal Test . . . . . . . . . . . . . 112 270
b. The Blum-Blum-Shub Perfect Generator . . . . . . 112 272
VI. The Advanced Encryption Standard (AES)
19. Introduction to the AES

20. The Finite Field GF(256)
a. Generate Multiplication Tables . . . . . . . . . . 125 273
b. Compare Multiplication Results . . . . . . . . . 126 275
21. The S-Boxes
x The Laws of Cryptography
a. Generate AES Tables . . . . . . . . . . . . . . 127 277

22. Key Expansion
23. Encryption
a. AES Encryption . . . . . . . . . . . . . . . . 135 282
24. Decryption
a. AES Decryption . . . . . . . . . . . . . . . . 138 290
b. Test Runs of the AES Algorithm . . . . . . . . . 138 293

IX. Identification and Key Distribution
25. Passwords and Key Distribution
26. Zero-Knowledge Proofs
27. Identification Schemes
28. Threshold Schemes
a. Shamir’s Threshold Schemes . . . . . . . . . . . 157 299
Foreword
There are excellent technical treatises on cryptography, along with a number of popular books.
In this book I am trying to find a middle ground, a “gentle” introduction to selected topics in
cryptography without avoiding the mathematics. The material is aimed at undergraduate com-
puter science students, but I hope it will be accessible and of interest to many others. The idea
is to cover a limited number of topics carefully, with clear explanations, sample calculations,
and illustrative Java implementations.
The emphasis is on the underlying systems and their theory, rather than on details of the
use of systems already implemented. For example, the notes present material on the RSA
cryptosystem, its theory and a Java implementation, but there is no discussion of a commercial
implementation such as PGP (“Pretty Good Privacy”).
The Java class libraries now give considerable support for commercial cryptography, and
there are whole books just on this subject, but this book doesn’t cover this topic.
The reader should not actively dislike mathematics, although the amount and difficulty of
the mathematics requirements vary. One of my goals is to cover the necessary mathematics
without hiding details, but also without requiring material from an undergraduate mathematics
degree. Also a number of subjects and results do not include full mathematical proofs.
The notes contain “maxims” or “laws” designed to emphasize important points, sometimes
in an amusing way — hence the title of the overall work.
I refer interested readers to the Handbook of Applied Cryptography, by Menezes, van
Oorschot, and Vanstone (CRC Press, 1997). That work gives a comprehensive survey of the
whole field, leaving many details to technical articles that the handbook refers to, and present-
ing “techniques and algorithms of greatest interest to the current practitioner”. In contrast, my
work is more idiosyncratic, occasionally presenting odd or obscure material, and not trying to
be comprehensive.
The Java programs that accompany this book are demonstration implementations to help
readers and students understand the concepts. I have kept the code simple to further this goal,
rather than strive for code that could be included into commercial or open source projects,
which would require far longer and more complex code (and be much harder for me to write).
The complexities then would get in the way of understanding. Readers need some familiarity
with programming and with Java to understand these programs, but most of the exposition is
independent of Java.
The book also contains various tables of values along with sample or “toy” calculations.
In every case I’ve found it easier and quicker to write Java programs to generate this material
rather than to do the calculations by hand. In many cases the Java programs directly output
HTML source to display a table. Tables in this book use Latex source, but I do not include
Java code that outputs Latex, since HTML is far more accessible. Thus when I say: “The Java
program on page xxx creates Table X.Y,” this means that the Java program creates a nearly
identical HTML table.
The Java programs in the book are available online in machine-readable form on the author’s
web page:
xii The Laws of Cryptography
http://www.cs.utsa.edu/˜wagner/lawsbook/
This book was partly inspired by two undergraduate courses in cryptography taught at the
University of Texas at San Antonio during the Spring 2002 and Spring 2003 semesters. The
web page for the course has many links and other information:
http://www.cs.utsa.edu/˜wagner/CS4953/index.html
A one-semester undergraduate course in cryptography might cover the following material:

✤ Part I. Introductory Material on Functions and Algorithms, referring back to it as
needed.
✤ Part II. Coding and Information Theory, without the Huffman or Hamming codes,
and with emphasis on Verhoeff’s detection method.
✤ Part III. Introduction to Cryptography, covered quickly.
✤ Part IV. Public Key Cryptography, the first four chapters.
✤ Part V. Random Number Generation, the first two chapters.
✤ Part VI. The Advanced Encryption Standard (AES), all.
✤ Plus selected remaining topics as desired.
The author would like to thank his mother for giving birth to him, but can’t think of anyone
else to thank at this time.
San Antonio, Texas

June, 2003
Introduction
Mankind has used the science of cryptography or “secret messages” for thousands of years to
transmit and store information needing secrecy. Until recently the military expended most of the
effort and money involved. However, starting in 1976 with the introduction in the open literature
of public key cryptography by Diffie and Hellman, the non-military and academic pursuit of
cryptography has exploded. The computer revolution has given people the means to use far
more complicated cryptographic codes, and the same revolution has made such widespread
and complex codes necessary. At the start of a new millennium, even non-technical people
understand the importance of techniques to secure information transmission and storage.
Cryptography provides four main types of services related to data that is transmitted or
stored:
✤ Confidentiality: keep the data secret.

✤ Integrity: keep the data unaltered.
✤ Authentication: be certain where the data came from.
✤ Non-repudiation: so someone cannot deny sending the data.
Consider first confidentiality. This is just a big word meaning “secrecy” — keeping the
data secret. For this one uses encryption, a process of taking readable and meaningful data, and
scrambling or transforming it so that someone who happens to intercept the data can no longer
understand it. As part of the process, there has to be a way for authorized parties to unscramble
or decrypt the encrypted data.
Integrity means keeping the data in unaltered form, while authentication means to know
where the data came from and who sent it. Neither of these services has anything to do with
secrecy, though one might also want secrecy. Consider, for example, the transfer of funds
involving U.S. Federal Reserve Banks (and other banks). While secrecy might be desirable, it is
of small importance compared with being sure who is asking for the transfer (the authentication)
and being sure that the transfer is not altered (the integrity). One important tool that helps
implement these services is the digital signature. A digital signature has much in common with
an ordinary signature, except that it works better: when properly used it is difficult to forge,
and it behaves as if the signature were scrawled over the entire document, so that any alteration
to the document would alter the signature. In contrast, ordinary signatures are notoriously easy
to forge and are affixed to just one small portion of a document.
The final service, non-repudiation, prevents someone from claiming that they had not sent
a document that was authenticated as coming from them. For example, the person might claim
that their private key had been stolen. This service is important but difficult to implement, and
is discussed in various of the books referred to in the references.
Refinements and extensions of these basic services fall into a category I call cryptographic
trickery: clever capabilities that might initially seem impossible, such as public keys, zero
knowledge proofs, and threshold schemes. I include examples of this material to entice readers
into the fascinating field of cryptography.
xiv The Laws of Cryptography
Taken all together, cryptography and its uses and implementations have become essential
for mankind’s technical civilization. The future promise is for the smooth functioning of these
and other services to allow individuals, businesses, and governments to interact without fear in
the new digital and online world.
Part I
Favorites
2
1
The Laws of Cryptography
Cryptographers’ Favorites
1.1 Exclusive-Or.
The function known as exclusive-or is also represented as xor or a plus sign in a circle, .
The expression
means either or but not both. Ordinary inclusive-or in mathematics
means either one or the other or both. The exclusive-or function is available in C / C++ /
Java for bit strings as a hat character: ˆ. (Be careful: the hat character is often used to mean
exponentiation, but Java, C, and C++ have no exponentiation operator. The hat character also
sometimes designates a control character.) In Java ˆ also works as exclusive-or for boolean
type.
Law XOR-1:
The cryptographer’s favorite function is Exclusive-Or.
Exclusive-or comes up continually in cryptography. This function plays an essential role
in the one-time pad (Chapter 10), stream ciphers (Chapter ??), and the Advanced Encryption
Standard (Part VI), along with many other places.
Recall that the boolean constant true is often written as a 1 and false as a 0. Exclusive-or
is the same as addition mod 2, which means ordinary addition, followed by taking the remainder

on division by .

For single bits and , Table 1.1 gives the definition of their exclusive-or.
The exclusive-or function has many interesting properties, including the following, which

hold for any bit values or bit strings , , and :
Exclusive-Or

Table 1.1 Definition of Exclusive-Or.
4 I. Favorites
ci
m mi
i (insecure line)
r ri
i
RNG RNG
seed
seed
(secure line)
Figure 1.1 Cryptosystem Using XOR.

, where is bit complement.
(commutativity)
(associativity)

if , then and .
Beginning programmers learn how to exchange the values in two variables a and b, using
a third temporary variable temp and the assignment operator = :
temp = a;
a = b;
b = temp;
The same result can be accomplished using xor without an extra temporary location, re-
garding a and b as bit strings. (A Java program that demonstrates interchange using exclusive-
or is on page 161).
a = a xor b;
b = a xor b;
a = a xor b;
For an example of exclusive-or used in cryptography, consider taking the xor of a pseudo-

random bit stream with a message bit stream
to give an encrypted bit stream , where

. To decrypt, xor the same pseudo-random bit stream with to give back:
. Figure 1.1 illustrates this process.
1. Cryptographer’s Favorites 5
Law XOR-2:
Cryptographers love exclusive-or because it immediately
gives a cryptosystem.
1.2 Logarithms.

By definition,

means the same as

. One says: “ is the logarithm of to base

,” or “ is the log base of .” Stated another way,

you raise to in order to get . Thus
(also known as ) is the exponent
. In more mathematical terms, the logarithm
is the inverse function of the exponential.
Law LOG-1:
The cryptographer’s favorite logarithm is log base 2.

One uses logs base in cryptography (as well as in most of computer science) because of

the emphasis on binary numbers in these fields.

So

means the same as

, and a logarithm base of is the exponent you
!
" #$
!%' & )(
raise to in order to get . In symbols: if , then
) (

. In particular

+*
means the same as
is not defined for . - ,
. Notice that

for all , and inversely
Here are several other formulas involving logarithms:
. 0/ 1 for all 1 *

32 . 54 1 for all 1 *
62 . 87 % 4 1 for all *
9 1 for all * 81
0/ (Oops! No simple formula for this.)
Table 1.2 gives a few examples of logs base .

:
Some calculators, as well as languages like Java, do not directly support logs base . Java

does not even support logs base , but only logs base , the “natural” log. However, a log

base is just a fixed constant times a natural log, so they are easy to calculate if you know the
“magic” constant. The formulas are:
;< 2 ; , (mathematics)

Math.log(x)/Math.log(2.0); (Java).
!; 8=?>@A $(CB $D EFE)@@(!EGA @( HB A , or 62 ;
=I(( >@JE )( DJDD@>AK( BAEL@@ "(> . (Similarly, M %'& 2 %'& , and %'&
The magic constant is:
8=?A @@@E)>>A@D $( .)

6 I. Favorites

# base
Logarithms 2

! 0 undefined
0
Table 1.2 Logarithms base 2.

A Java program that demonstrates these formulas is found on page 162.

Here is a proof of the above formula:
. . take ; of each side)

;
, or ; (then(then
; ; (then solveuse properties of logarithms)
; 2 ; (then substitute
for y)
for y)
;< 2 ; <=
Law LOG-2:
The log base 2 of an integer x tells how many bits it takes
to represent x in binary.
A8= DBB AD , so it takes $( bits to represent in binary. (In fact,
%'& .) Exact powers of are a special
case: )( , but it
Thus
takes bits to represent

)( in binary, as .
Similarly,
gives the number of decimal digits needed to represent .
%'&
1.3 Groups.
A group is a set of group elements with a binary operation for combining any two elements to
get a unique third element. If one denotes the group operation by , then the above says that

for any group elements and , is defined and is also a group element. Groups are also
associative, meaning that
, for any group elements , , and . There
:
has to be an identity element satisfying : :
:
for any group element . Finally,

any element must have an inverse satisfying

.
If

for all group elements and , the group is commutative. Otherwise it is
non-commutative. Notice that even in a non-commutative group, might sometimes

be true — for example if or is the identity.
A group with only finitely many elements is called finite; otherwise it is infinite.
Examples:

4
1. The integers (all whole numbers, including and negative numbers) form a group us-

ing ordinary addition. The identity is and the inverse of is . This is an infinite
commutative group.
62
2. The positive rationals (all positive fractions, including all positive integers) form a group

7%
if ordinary multiplication is the operation. The identity is and the inverse of is
. This is another infinite commutative group.
integers mod n form a group for any integer

*
=$=$= 4
3. The . This group is often denoted
. Here the elements are , , , ,

and the operation is addition followed by
8 I. Favorites
+ 0 1 2 3 4 5 6 7 8 9
0 0 1 2 3 4 5 6 7 8 9
1 1 2 3 4 5 6 7 8 9 0
2 2 3 4 5 6 7 8 9 0 1
3 3 4 5 6 7 8 9 0 1 2
4 4 5 6 7 8 9 0 1 2 3
5 5 6 7 8 9 0 1 2 3 4
6 6 7 8 9 0 1 2 3 4 5
7 7 8 9 0 1 2 3 4 5 6
8 8 9 0 1 2 3 4 5 6 7
9 9 0 1 2 3 4 5 6 7 8
Table 1.3 Addition in the integers mod 10,

%'& .

remainder on division by
. The identity is and the inverse of is
4 (except for
which is its own inverse). This is a finite commutative group.
4. For an example of a non-commutative group, consider 2-by-2 non-singular matrices of
real numbers (or rationals), where the operation is matrix multiplication:

Here , , , and are real numbers (or rationals) and must be non-zero (non- 4
singular matrices). The operation is matrix multiplication. The above matrix has inverse

4
4 4
and the identity is

=
This is an infinite non-commutative group.
5. The chapter on decimal numbers gives an interesting and useful example of a finite non-
commutative group: the dihedral group with ten elements.
Law GROUP-1:
The cryptographer’s favorite group is the integers mod n,
Zn.

In the special case of
/
, the operation of addition in can be defined by
mod , that is, divide by 10 and take the remainder. Table 1.3shows how one can
%'&
also use an addition table to define the integers modulo 10:
1.4 Fields.
/
A field is an object with a lot of structure, which this section will only outline. A field has
/
two operations, call them and (though they will not necessarily be ordinary addition and
4
multiplication). Using , all the elements of the field form a commutative group. Denote the
identity of this group by and denote the inverse of by
. Using , all the elements of the

7%
field except must form another commutative group with identity denoted and inverse of

/ / /
denoted by . (The element has no inverse under .) There is also the distributive identity,
linking and : , for all field elements , , and . Finally,

one has to exclude divisors of zero, that is, non-zero elements whose product is zero. This is
equivalent to the following cancellation property: if is not zero and , then .
Examples:
1. Consider the rational numbers (fractions) Q, or the real numbers R, or the complex num-
bers C, using ordinary addition and multiplication (extended in the last case to the com-
plex numbers). These are all infinite fields.

A E B
A 6B $@ A )@ =$=$= /
2. Consider the integers mod p, denoted , where p is a prime number ( , , , , ,
, , , , , ). Regard this as a group using (ordinary addition followed by

remainder on division by ). The elements with left out form a group under (ordinary
multiplication followed by remainder on division by ). Here the identity is clearly ,

but the inverse of a non-zero element is not obvious. In Java, the inverse must be an

element satisfying . It is always possible to find the unique element
, using an algorithm from number theory known as the extended Euclidean algorithm.
This is the topic in the next chapter, but in brief: because is prime and is non-zero, the

5 /
4
greatest common divisor of and is . Then the extended Euclidean algorithm gives
ordinary integers and satisfying , or
, and this
says that if you divide

by , you get remainder , so this is the inverse of . (As
an

integer, might be negative, and in this case one must add to it to get an element of
.)
Law FIELD-1:
The cryptographer’s favorite field is the integers mod p,
denoted Zp , where p is a prime number.
The above field is the only one with elements. In other words, the field is unique up to
renaming its elements, meaning that one can always use a different set of symbols to represent
the elements of the field, but it will still be essentially the same.
10 I. Favorites
There is also a unique finite field with elements for any integer

* .

, denoted

* E>
Particularly useful in cryptography is the special case with , that is, with elements for

. The case is used, for example, in the
new U.S. Advanced Encryption Standard
(AES). It is more difficult to describe than the field . The chapter about multiplication for the
E> D
AES will describe this field in more detail, but here are some of its properties in brief for now:
It has elements, represented as all possible strings of bits. Addition in the field is just the
same as bitwise exclusive-or (or bitwise addition mod ). The zero element is , and

B
the identity element is . So far, so good, but multiplication is more problematic:
one
has to regard an element as a degree polynomial with coefficients in the field (just a or

a ) and use a special version of multiplication of these polynomials. The details will come in
the later chapter on the AES.
Law FIELD-2:
The cryptographer’s other favorite field is GF(2n).
1.5 Fermat’s Theorem.

In cryptography, one often wants to raise a number
to a power, modulo another number. For the
integers mod where is a prime (denoted ), there is a result know as Fermat’s Theorem,

discovered by the 17th century French mathematician Pierre de Fermat, 1601-1665.
Theorem (Fermat): If
is a prime and is any non-zero number less than , then

7 % mod
Law FERMAT-1:
The cryptographer’s favorite theorem is Fermat’s Theorem.
Table 1.4 illustrates Fermat’s Theorem for . Notice below that the value is always A

by the time the power gets to , but sometimes the value gets to earlier. The initial run up

A ( >
to the value is shown in bold italic in the table. The lengths of these runs are always numbers

> B
that divide evenly into , that is, , , , , or . A value of for which the whole row is

4
bold italic is called a generator. In this case , , , and are generators.

L4
Because to a power mod always starts repeating after the power reaches , one can

reduce the power mod
might be,
and still get the same answer. Thus no matter how big the power

mod mod 7 % mod =
p a % %'& % % %
13 2 2 4 8 3 6 12 11 9 5 10 7 1
13 3 3 9 1 3 9 1 3 9 1 3 9 1
13 4 4 3 12 9 10 1 4 3 12 9 10 1
13 5 5 12 8 1 5 12 8 1 5 12 8 1
13 6 6 10 8 9 2 12 7 3 5 4 11 1
13 7 7 10 5 9 11 12 6 3 8 4 2 1
13 8 8 12 5 1 8 12 5 1 8 12 5 1
13 9 9 3 1 9 3 1 9 3 1 9 3 1
13 10 10 9 12 3 4 1 10 9 12 3 4 1
13 11 11 4 5 3 7 12 2 9 8 10 6 1
13 12 12 1 12 1 12 1 12 1 12 1 12 1
Table 1.4 Fermat’s Theorem for .
L4
$A
Thus modulo in the expression requires modulo in the exponent. (Naively, one might

expect to reduce the exponent mod , but this is not correct.) So, for example, if as
mod $A % mod A

above, then
mod
mod A<=
The Swiss mathematician Leonhard Euler (1707-1783) discovered a generalization of Fer-
mat’s Theorem which will later be useful in the discussion of the RSA cryptosystem.
Theorem (Euler): If
is any positive integer and is any positive integer less
than
with no divisors in common with
, then

mod
1
where

is the Euler phi function:

4 62 % =$=$= 4 6 2 1

and ,
% ==$=
, are all the prime numbers that divide evenly into
, including
itself in case it is a prime.
If
is a prime, then using the formula,

, so Euler’s 4 2 7 % 4
result is a special case of Fermat’s. Another special case needed for the RSA cryptosystem
comes when the modulus is a product of two primes:
. Then

4 6 2 4
12 I. Favorites
n a % %'& % % % % %
15 2 2 4 8 1 2 4 8 1 2 4 8 1 2 4
15 3 3 9 12 6 3 9 12 6 3 9 12 6 3 9
15 4 4 1 4 1 4 1 4 1 4 1 4 1 4 1
15 5 5 10 5 10 5 10 5 10 5 10 5 10 5 10
15 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6
15 7 7 4 13 1 7 4 13 1 7 4 13 1 7 4
15 8 8 4 2 1 8 4 2 1 8 4 2 1 8 4
15 9 9 6 9 6 9 6 9 6 9 6 9 6 9 6
15 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
15 11 11 1 11 1 11 1 11 1 11 1 11 1 11 1
15 12 12 9 3 6 12 9 3 6 12 9 3 6 12 9
15 13 13 4 7 1 13 4 7 1 13 4 7 1 13 4
15 14 14 1 14 1 14 1 14 1 14 1 14 1 14 1
Table 1.5 Euler’s theorem for and

.
62 - 4 K4 . Table 1.5 illustrates

theorem for
6E A8E , with
6 E 6 E 4 62A
4 62E A 4 E04 D . Notice here that a is reached
Euler’s
when the power gets to D (actually in this simple case when the power gets to or ( ), but only
for numbers with no divisors in common with E . For other base numbers, the value never gets
to .
The tables above were generated by the Java program on page 163.
In a way similar to Fermat’s Theorem, arithmetic in the exponent is taken mod

, so

that, assuming has no divisors in common with
,

mod
mod mod =
6E as above, then

D , and if neither A nor E divides evenly into , then

If
D . Thus for example,

mod 6E mod mod 6E mod 6E<=
The proof in Chapter 14 that the RSA cryptosystem works depends on the above fact.
Exercises

1. For any bit string , what is equal to?
2. Prove in two ways that the three equations using exclusive-or to interchange two values
work. One way should use just the definition of xor in the table, and the other way
should use the properties of xor listed above. (On computer hardware that has an xor
instruction combined with assignment, the above solution may execute just as fast as the
previous one and will avoid the extra variable.)
3. Use the notation to mean “inclusive-or”, to mean “and”, and to mean “not”. With
this notation, show, using either truth tables or algebraically that
, and

=
4. Show how to use exclusive-or to compare the differences between two bit strings.

5. Given a bit string , show how to use another mask bit string
of the same length to

reverse a fixed bit position , that is, change to and to , but just in position .
6. How many bits are needed to represent a number that is 100 decimal digits long? How
many decimal digits are needed to represent a number that is 1000 bits long? How many
decimal digits are needed to represent a number that is 100 decimal digits long?
7. Write a Java function to return the log base
of , where * and * . Test your
function.
8. In the example of 2-by-2 matrices, verify that the product of a non-zero element and its
inverse in the identity.
2
Cryptographers’ Favorite Algorithms
2.1 The Extended Euclidean Algorithm.

The previous section introduced the field known as the integers mod p, denoted or .

Most of the field operations are straightforward, since they are just the ordinary arithmetic

7%
operations followed by remainder on division by . However the multiplicative inverse is not

intuitive and requires some theory to compute. If is a non-zero element of the field, then
can be computed efficiently using what is known as the extended Euclidean algorithm, which
gives the greatest common divisor (gcd) along with other integers that allow one to calculate
the inverse. It is the topic of the remainder of this section.
Law GCD-1:
The cryptographer’s first and oldest favorite algorithm is
the extended Euclidean algorithm, which computes the greatest
common divisor of two positive integers a and b and also sup-
plies integers x and y such that x*a + y*b = gcd(a, b).
D @ (> D @ A A B $ A

The Basic Euclidean Algorithm to give the gcd: Consider the calculation of the greatest

(> "A )B A )B
common divisor (gcd) of and . One could factor the numbers as:
and
, to see immediately that the gcd is . The problem with this
method is that there is no efficient algorithm to factor integers. In fact, the security of the RSA
cryptosystem relies on the difficulty of factoring, and we need an extended gcd algorithm to
implement RSA. It turns out that there is another better algorithm for the gcd — developed 2500
years ago by Euclid (or mathematicians before him), called (surprise) the Euclidean algorithm.
The algorithm is simple: just repeatedly divide the larger one by the smaller, and write an
equation of the form “larger = smaller * quotient + remainder”. Then repeat using the two
numbers “smaller” and “remainder”. When you get a remainder, then you have the gcd of the
original two numbers. Here is the sequence of calculations for the same example as before:
D @ (> / AEB

(> AEB / E
(Step 0)

AEB E A / (
(Step 1)

E ( /
(Step 2)

( /
(Step 3, so GCD = 21)
(Step 4)
2. Cryptographer’s Favorite Algorithms 15
The proof that this works is simple: a common divisor of the first two numbers must also
be a common divisor of all three numbers all the way down. (Any number is a divisor of , sort
of on an honorary basis.) One also has to argue that the algorithm will terminate and not go on
forever, but this is clear since the remainders must be smaller at each stage.
Here is Java code for two versions of the GCD algorithm: one iterative and one recursive.
(There is also a more complicated binary version that is efficient and does not require division.)
Java function: gcd (two versions)
public static long gcd1(long x, long y) {
if (y == 0) return x;
return gcd1(y, x%y);
}

while (y != 0) {
long r = x % y;
x = y; y = r;
}
return x;
}
A complete Java program using the above two functions is on page 165.
D @ ( >
D @3/ (> . D @81 (>
The Extended GCD Algorithm: Given the two positive integers and , the extended
Euclidean algorithm finds unique integers and so that
4 @
$D @ / > $(> .
In this case, .

For this example, to calculate the magic and , just work backwards through the original
equations, from step 3 back to step 0 (see above). Below are equations, where each shows two

/ 1
numbers and from a step of the original algorithm, and corresponding integers and (in
bold), such that . Between each pair of equations is an equation in
italics that leads to the next equation.
1*105 + (-2)* 42 = 21 (from Step 3 above)
(-2)*357 + (-2)(-3)*105 = (-2)*42 = (-1)*105 + 21 (Step 2 times -2)
(-2)*357 + 7*105 = 21 (Combine and simplify previous equation)
7 *462 + (7)(-1)* 357 = 7*105 = 2*357 + 21 (Step 1 times 7)
7*462 + (-9)*357 = 21 (Combine and simplify previous equation)
(-9)*819 + (-9)(-1)*462 = (-9)*357 = (-7)*462 + 21 (Step 0 * (-9))
(-9)*819 + 16*462 = 21 (Simplify -- the final answer)
It’s possible to code the extended gcd algorithm following the model above, first using a
loop to calculate the gcd, while saving the quotients at each stage, and then using a second
loop as above to work back through the equations, solving for the gcd in terms of the original
two numbers. However, there is a much shorter, tricky version of the extended gcd algorithm,
adapted from D. Knuth.
Java function: GCD (extended gcd)
public static long[] GCD(long x, long y) {
long[] u = {1, 0, x}, v = {0, 1, y}, t = new long[3];
while (v[2] != 0) {
long q = u[2]/v[2];
16 I. Favorites
for (int i = 0; i < 3; i++) {

t[i] = u[i] - v[i]*q; u[i] = v[i]; v[i] = t[i];
}
}
return u;
}
A complete Java program using the above function is on page 166.

The above code rather hides what is going on, so with additional comments and checks, the
code is rewritten below. Note that at every stage of the algorithm below, if w stands for any of
the three vectors u, v or t, then one has x*w[0] + y*w[1] = w[2]. The function check
verifies that this condition is met, checking in each case the vector that has just been changed.
Since at the end, u[2] is the gcd, u[0] and u[1] must be the desired integers.
Java function: GCD (debug version)
long[] u = new long[3];
long[] v = new long[3];
long[] t = new long[3];
// at all stages, if w is any of the 3 vectors u, v or t, then
// x*w[0] + y*w[1] = w[2] (this is verified by "check" below)
// vector initializations: u = {1, 0, u}; v = {0, 1, v};
u[0] = 1; u[1] = 0; u[2] = x; v[0] = 0; v[1] = 1; v[2] = y;
while (v[2] != 0) {
long q = u[2]/v[2];
// vector equation: t = u - v*q
t[0] = u[0] - v[0]*q; t[1] = u[1] - v[1]*q; t[2] = u[2] - v[2]*q;
check(x, y, t);
// vector equation: u = v;
u[0] = v[0]; u[1] = v[1]; u[2] = v[2]; check(x, y, u);
// vector equation: v = t;
v[0] = t[0]; v[1] = t[1]; v[2] = t[2]; check(x, y, v);
}
return u;
}
public static void check(long x, long y, long[] w) {

if (x*w[0] + y*w[1] != w[2])
System.exit(1);
}
Here is the result of a run with the data shown above:

q u[0] u[1] u[2] v[0] v[1] v[2]}
1 0 1 462 1 -1 357
1 1 -1 357 -1 2 105
3 -1 2 105 4 -7 42
2 4 -7 42 -9 16 21
2 -9 16 21 22 -39 0
gcd(819, 462) = 21
( @ "( $( :

(-9)*819 + (16)*462 = 21
Here is a run starting with and
q u[0] u[1] u[2] v[0] v[1] v[2]}
1 0 1 24140 1 -1 16762
1 1 -1 16762 -1 2 7378
2 -1 2 7378 3 -5 2006
3 3 -5 2006 -10 17 1360
1 -10 17 1360 13 -22 646
2 13 -22 646 -36 61 68
9 -36 61 68 337 -571 34
2 337 571 34 -710 1203 0
gcd(40902, 24140) = 34
(337)*40902 + (-571)*24140 = 34
A complete Java program with the above functions, along with other example runs appears on
page 167.
2.2 Fast Integer Exponentiation (Raise to a Power).
)(
A number of cryptosystems require arithmetic on large integers. For example, the RSA public
key cryptosystem uses integers that are at least bits long. An essential part of many of the
algorithms involved is to raise an integer to another integer power, modulo an integer (taking
the remainder on division).
Law EXP-1:
Many cryptosystems in modern cryptography depend on
a fast algorithm to perform integer exponentiation.
"(
It comes as a surprise to some people that in a reasonable amount of time one can raise a
%'&
bit integer to a similar-sized power modulo an integer of the same size. (This calculation
"(
can be done on a modern workstation in a fraction of a second.) In fact, if one wants to calculate

(a -bit exponent), there is no need to multiply by itself
%'&
times, but one only needs

(a
) (
to square and keep squaring the result times. Similarly, squarings yields
-bit exponent), and an exponent with bits requires only that many squarings if it is an

exact power of . Intermediate powers come from saving intermediate results and multiplying
them in. RSA would be useless if there were no efficient exponentiation algorithm.

There are two distinct fast algorithms for raising a number to an integer power. Here is
pseudo-code for raising an integer to power an integer :
Java function: exp (first version)
int exp(int x, int Y[], int k) {
// Y = Y[k] Y[k-1] ... Y[1] Y[0] (in binary)
int y = 0, z = 1;
18 I. Favorites
for (int i = k; i >= 0; i--) {

y = 2*y;
z = z*z;
if (Y[i] == 1) {
y++;
z = z*x;
}
}
return z;
}

The variable is only present to give a loop invariant, since at the beginning and end of
terminates is also true, so at the end,

each loop, as well as just before the if statement, the invariant
. To find
holds, and after the loop

mod
one should add a
remainder on division by
to the two lines that calculate .
Here is the other exponentiation algorithm. It is very similar to the previous algorithm, but
differs in processing the binary bits of the exponent in the opposite order. It also creates those
bits as it goes, while the other assumes they are given.
Java function: exp (second version)
int exp(int X, int Y) {
int x = X, y = Y, z = 1;
while (y > 0) {
while (y%2 == 0) {
x = x*x;
y = y/2;
}
z = z*x;
y = y - 1;
}
return z;
}
The loop invariant at each stage and after the each iteration of the inner while statement is:

=

Here is a mathematical proof that the second algorithm actually calculates . Just before
the while loop starts, since
, , and , it is obvious that the loop invariant is
true. (In these equations, the is a mathematical equals, not an assignment.)

Now suppose that at the start of one of the iterations of the while loop, the invariant holds.
Use , , and for the new values of , , and after executing the statements inside one
iteration of the inner while statement. Notice that this assumes that is even. Then the following
are true:

82 (exact integer because y is even)

=

This means that the loop invariant holds at the end of each iteration of the inner while
statement for the new values of , , and . Similarly, use the same prime notation for the
variables at the end of the while loop.
.
4

7 %
=
So once again the loop invariant holds. After the loop terminates, the variable
must be ,
so that the loop invariant equation says:

& =
For a complete proof, one must also carefully argue that the loop will always terminate.
A test of the two exponentiation functions implemented in Java appears on page 169.
2.3 Checking For Probable Primes.

For 2500 years mathematicians studied prime numbers just because they were interesting, with-
out any idea they would have practical applications. Where do prime numbers come up in the
real world? Well, there’s always the 7-Up soft drink, and there are sometimes a prime number
of ball bearings arranged in a circle, to cut down on periodic wear. Now finally, in cryptography,
prime numbers have come into their own.
Law PRIME-1:
A source of large random prime integers is an essential
part of many current cryptosystems.
Usually large random primes are created (or found) by starting with a random integer
,
and checking each successive integer after that point to see if it is prime. The present situation
is interesting: there are reasonable algorithms to check that a large integer is prime, but these
algorithms are not very efficient (although a recently discovered algorithm is guaranteed to
produce an answer in running time no worse that the number of bits to the twelth power). On
the other hand, it is very quick to check that an integer is “probably” prime. To a mathematician,
it is not satisfactory to know that an integer is only probably prime, but if the chances of making
a mistake about the number being a prime are reduced to a quantity close enough to zero, the
users can just discount the chances of such a mistake.
Tests to check if a number is probably prime are called pseudo-prime tests. Many such
tests are available, but most use mathematical overkill. Anyway, one starts with a property of

7%
a prime number, such as Fermat’s Theorem, mentioned in the previous chapter: if is a prime

and is any non-zero number less than , then mod . If one can find a number
20 I. Favorites

for which Fermat’s Theorem does not hold, then the number in the theorem is definitely not

a prime. If the theorem holds, then is called a pseudo-prime with respect to , and it might
A
actually be a prime.
So the simplest possible pseudo-prime test would just take a small value of , say or ,
and check if Fermat’s Theorem is true.

A 7%
Simple Pseudo-prime Test: If a very large random integer (100 decimal digits
or more) is not divisible by a small prime, and if mod
, then the number
is prime except for a vanishingly small probability, which one can ignore.
A
One could just repeat the test for other integers besides as the base, but unfortunately
E)> A 6B
there are non-primes (called Carmichael numbers) that satisfy Fermat’s theorem for all values

of even though they are not prime. The smallest such number is , but
7 %
these numbers become very rare in the larger range, such as 1024-bit numbers. Corman et
al. estimate that the chances of a mistake with just the above simple test are less than ,
although in practice commercial cryptosystems use better tests for which there is a proof of the
low probability. Such better tests are not really needed, since even if the almost inconceivable
happened and a mistake were made, the cryptosystem wouldn’t work, and one could just choose
another pseudo-prime.
Law PRIME-2:
Just one simple pseudo-prime test is enough to test that a
very large random integer is probably prime.
Exercises
1. Prove that the long (debug) version of the Extended GCD Algorithm works.
(a) First show that u[2] is the gcd by throwing out all references to array indexes 0
and 1, leaving just u[2], v[2], and t[2]. Show that this still terminates and just
calculates the simple gcd, without reference to the other array indexes. (This shows
that at the end of the complicated algorithm, u[2] actually is the gcd.)
(b) Next show mathematically that the three special equations are true at the start of the
algorithm, and that each stage of the algorithm leaves them true. (One says that they
are left invariant.)
(c) Finally deduce that algorithm is correct.
Part II
Coding and
Information Theory
22
3
Coding and Information Theory
3.1 Shannon’s Information Theory and Entropy.

The term information theory refers to a remarkable field of study developed by Claude Shannon
in 1948. Shannon’s work was like Einstein’s gravitation theory, in that he created the whole field
all at once, answering the most important questions at the beginning. Shannon was concerned
with “messages” and their transmission, even in the presence of “noise”. For Shannon, the
word “information” did not mean anything close to the ordinary English usage. Intuitively, the
information in a message is the amount of “surprise” in the message. No surprise means zero
information. (Hey, that’s somewhat intuitive. However, it is not intuitive that a random message
has the most information in this sense.)
Shannon defined a mathematical quantity called entropy which measures the amount of
information in a message, if the message is one of a collection of possible messages, each
with its own probability of occurrence. This entropy is the average number of bits needed to
% $1 =$==1
represent each possible

1$=$=$=1
message, using the best possible encoding. If there are
messages
%
then the entropy

, with probabilities of occurrence:
of this set of messages is:
(with sum equal ),
) 62 /.= = ="/ ) 62 =

% %

Intuitively, the entropy is just the weighted average of the number of bits required to repre-
sent each message, where the weights are the probabilities that each message might occur.
Law ENTROPY-1:
The entropy of a message is just the number of bits of
information in the message, that is, the number of bits needed
for the shortest possible encoding of the message.
It is possible to list reasonable properties of any entropy function and to prove that only the
above formula gives a function with those properties.
For example, if we have two messages 1
male female, each having probability , 62
then the entropy is

62 6 2 62 / 62 62 62
24 II. Coding and Information Theory
62 / 2

62 / 2 =
Thus in this case, as we would intuitively expect, there is one bit of information in such a
%
message.
Suppose

and the remaining probabilities are zero. In this case the entropy works
%
out to be as one would expect, since one is only going to receive the message
is no information and no surprise in receiving this message. The actual message
, so there
might be %
complex, with many bits representing it, but its probability is , so only this message can occur,
entropy, the term 62
with no information or “surprise” on its receipt, even if it is complex.
comes up, which looks like

!
(In the calculation of
. This term would be
indeterminate, but the first part tends to much faster than tends to , so that in

A % 2 6 2)( 62)(
practice such terms are regarded as .)
As another example, suppose
= E
and , , and .
%
,
Then the entropy works out to be
, and
. It is possible to encode these messages as follows:

. In this case the average code length is the same as the entropy.
One doesn’t normally expect to be able to represent a collection of messages with a code whose
average length is exactly equal to the entropy; it is never possible to get the average length less
than the entropy.
Finally, suppose there are
equally probably messages. Then the entropy is:
62 " 2 62 / = = =6/ 62 ) 2 62

62 /.= = =6/ 6 2

62 6
@8=?@>EBD)( DE<=
Thus the entropy value of these messages means that there are nearly bits of information
is
in each message. Similarly, if there are
equally likely messages, then the entropy of a message

. The equal probable case gives the largest possible value for the entropy of a collection
of messages.
Law ENTROPY-2:
A random message has the most information (the greatest
entropy). [Claude Shannon]
3.2 The Three Kinds of Codes.

The terms code and coding refer to ways of representing information. We will usually be using
binary codes, that is, codes that use only the binary bits and . There are three kinds of
coding:
3. Coding and Information Theory 25
1. Source Coding. This usually involves data compression: representing the data with as
few bits as possible. Notice that one always needs at least as many bits to encode a
message as the entropy of the message. The next chapter talks more about source coding,
and presents one special example: the Huffman code.
2. Channel Coding. Here one uses error detection and error correction to improve the
reliability of the channel. This is accomplished by adding extra redundant bits. The rest
of this chapter presents material on channel capacity and error correction codes. A later
chapter talks about one particular error correcting code: the Hamming code.
3. Secrecy Coding. For secrecy, one uses cryptography to scramble the message so that
it may not be intelligible to an eavesdropper. Much of the rest of the material in these
notes is concerned with cryptographic coding. Often the scrambled message has the same
number of bits as the original message.
Law INFORMATION-1:
In all coding theory, information transmission is essen-
tially the same as information storage, since the latter is just
transmission from now to then.
It is possible to have a single code that combines two or even all three of these functions,
but the codes are usually kept separate. Normally one would compress a message (making the
message smaller, to save storage or channel bandwidth), then transform it cryptographically for
secrecy (without changing the message length), and finally add bits to the message to allow for
error detection or correction.
3.3 Channel Capacity.

Shannon also introduced the concept of channel capacity, which is the maximum rate at which
bits can be sent over an unreliable (noisy) information channel with arbitrarily good reliability.
The channel capacity is represented as a fraction or percentage of the total rate at which bits
can be sent physically over the channel. Shannon proved that there always exist codes that
will signal arbitrarily close to the channel capacity with arbitrarily good reliability. Thus by
choosing a larger and more complicated code, one can reduce the number of errors to as small
a percentage as one would like, while continuing to signal as close as one wants to 100% of
the channel capacity. In practice the theory does not provide these good codes, though they are
known to exist. It is not possible to signal with arbitrarily good reliability at a rate greater than
the channel capacity.
, ,
The simplest example of such a channel is the binary symmetric channel. Here every time

4
a bit is transmitted, there is a fixed probability , with such that a transmitted is

received as a with probability and received as a with probability
. The errors occur
at random.

Probability

Channel Capacity

or

or

or

or

or

or

or

or

or

or
or

Table 3.1 Channel Capacity.
For example, if
there are no errors at all on the channel, and the channel capacity is

8=E
(meaning 100%). If , the capacity is still as long as you realize that all bits are reversed.
If
, then on receipt of a bit, both and are equally likely as the bit that was sent, so
one can never say anything about the original message. In this case the channel capacity is
and no information can be sent over the channel.
For binary symmetric channels there is a simple formula for the capacity
(a Java program
that calculates channel capacity is on page 172):

/ / 4 4 =
Alternatively, one can write this formula as:
,

4
where consists of two messages with probabilities and 04

. One can argue intuitively
that this formula makes use of the amount of information lost during transmission on this noisy
channel, or one can show this formula mathematically using concepts not introduced here.
Table 3.1 gives values of the channel capacity and was produced by the Java program on
page 173:
Exercise: Use a numerical approximation algorithm to find the inverse of the channel capacity
formula. Then write a Java program implementing the algorithm and printing a table of channel
capacities and corresponding probabilities (just the reverse of the above table). The results
might look like Table 3.2 .
[Ans: A Java program that uses the simple bisection method to print the table is on page
175.]
Channel Probability

Probability

Capacity

Table 3.2 Inverse channel capacity.
3.4 A Simple Error Correcting Code.

8=BE
E
Let us work out a simple example of the binary symmetric channel with . Keep in
mind that this means the error rate is : an extremely high figure. In such a case every
8 =BE
fourth bit on the average will be reversed, with the reversals occurring completely at random.
The capacity of the binary symmetric channel with is:
/ 6 2)( 62)( / A2)( A2)(

/ 62)( 4 / A2)( A04

<= DDB =$=$=

A channel capacity of 8= DDB means that one can at best signal at a rate of slightly less
than 19% of the true channel bandwidth, because there are so many errors.
Here is a simple code to improve the error rate: send each bit 3 times and let the majority

decide at the other end. In this case a single error can be corrected, since if is transmitted and
something like is received, the majority rule would interpret the received 3 bits as a .

With a success rate of , there will be errors in the 3 bits with probability , and there

will be error also with probability . Then errors will occur of the time and errors A
%
%'&
will occur of the time. Only the last two cases represent an actual error with this triple
redundant system, so the new error rate is or about 16%. In summary, by reducing the
transmission rate from 100% to 33%, one can reduce the error rate from 25% to 16%. (This is
a very simple-minded and inefficient code.)
One could continue in this way, using more and more duplicates and letting the majority
rule. (An even number of duplicates is not a good choice because the result is indeterminate in
Repetition Codes, Error Probability is 8= E

Number of Transmission Error Success
Duplicates Rate Rate Rate
1 100% 25% 75%
3 33% 16% 84%
5 20% 10% 90%
7 14% 7% 93%
9 11% 5% 95%
11 9.1% 3.4% 96.6%
25 4.0% 0.337% 99.663%
49 2.04% 0.008% 99.992%
99 1.01% 0.0000044% 99.9999956%
199 0.5025% 1.805E-12% 99.9999999999982%
Table 3.3 Repetition codes.

case equal numbers of s and s are received.) Table 3.3 gives the results (a Java program that
prints the table is given as the answer to the next exercise).
It should be clear that you can get as low an error rate as you like by using more and
more duplicates, reducing the transmission rate to near zero. At a little less than the channel
capacity (7 duplicates and a transmission rate of 14%), you can get the error rate down to 7%.
Shannon’s theory says that there are other more complicated codes that will also take the error
rate arbitrarily close to zero, while maintaining a transmission rate close to 18%. (These other
codes can get very large and complicated indeed. No general and practical codes have ever
been discovered.)
Exercise: Find the channel capacity for

[Ans: 0.08170.] Do up a table for

similar to the one for
. [Ans: see the program on page 177.]
3.5 A Slightly Better Error Correcting Code.

Instead of having code words just for the two bits and , it is possible to find codes for longer
blocks of bits: for example, to take two bits at a time and use some distinct code word for each

of the four possibilities: , , , and . Table 3.4 shows a scheme that encodes blocks of
three bits at a time:
% %

, , , where

%
, , and %
If the information bits are designated , , and , then the code word bits are: , , ,
.
These code words have the property that any two of them differ from one another in at least
three positions. (One says that the Hamming distance between any two of them is at least 3.)
If there is a single error in the transmission of a code word, there is a unique code word that
Information Code
Word Word
000 000 000
100 100 011
010 010 101
001 001 110
011 011 011
101 101 101
110 110 110
111 111 000
Table 3.4 Error correcting code.
differs from the transmitted word in only one position, whereas all others differ in at least two
positions. For this reason, this code is single error correcting, just like the previous triple code,
where each bit is transmitted three times.
Notice a difference between this code and the triple transmission code: this code has a
transmission rate of 50%, while the triple code has a rate of only 33.3%, even though both do
single-error correction.
In the previous code, after transmitting each bit 3 times, one got a better error rate by
transmitting each bit 5 times, or 7 times, or 9 times, etc. The transmission rate went to zero,
while the error rate also went to zero. In the same way, one could repeat the codes in this
section, and the transmission rates would be slightly better than the earlier ones, though they
would still go to zero. What is wanted is a way to keep a high transmission rate while lowering
the error rate.
3.6 Shannon’s Random Codes.

Claude Shannon proved the existence of arbitrarily good codes in the presence of noise. The
codes he constructed are completely impractical, but they show the limits of what is possible.
Law SHANNON-1:
Over a noisy channel it is always possible to use a long
enough random code to signal arbitrarily close to the channel ca-
pacity with arbitrarily good reliability [known as Shannon’s Noisy
Coding Theorem].

The proof chooses an integer
for the blocksize of the information words —
must be very
large indeed. The information words are all
strings of s and s of length
. Corresponding
to each information word one chooses a codeword completely at random. The length of the
codewords must be greater than the blocksize divided by the channel capacity. Then it is
possible to prove Shannon’s result by choosing
large enough. (Notice that it is possible to
choose two identical random codewords. This will contribute another error, but as the blocksize
gets large, the probability of these errors will also tend to zero.)
The actual situation is just a little more complicated: a code table chosen at random might
be anything at all, and so might not perform well. Shannon’s proof of his result takes the
average over all possible codes. For large enough
, the average satisfies his theorem, so there
must exist an individual specific code table that satisfies his theorem. In practice, for large
a
random code is very likely to work.
These codes are impractical because the system requires keeping a table of all the random
code words. Assuming the table would fit into memory, encoding could be done efficiently.
However, for decoding one must find the code word in the table that most closely matches the
received code word (that has errors from the noise on the channel).

Exercise: Create a simulation of Shannon’s result for specific parameter values and for a
<= DDB
specific random code table (with codewords chosen at random). Use for the error
probability in the binary symmetric channel. Thus the channel capacity
will be
Now try a specific value for the blocksize
, say
6 E A B)>D
.
A B>D
, so that there will be information
2 6E2 8= $DDB B@<=?(DA

words and entries in the code/decode table. Shannon’s theory says that these random
code words must be at least

bits long. Try a longer length for
the code words, say
bits, and find the error rate of the simulation in this case. (We know
that for
large enough this error rate will be as small as we like in this case, but without some
estimates, we don’t know how it will work out for
6E >E
2
.) Try a shorter length, say for the
code word lengths. (We know that if the code word lengths are chosen less than

the error
rate can never tend to zero, no matter how big
is chosen.) Also try the value D
(or ) for B@
comparison.
[Ans: See an answer in the next chapter.]
4
Visualizing Channel Capacity
This section starts with work on an answer to the exercise at the end of the chapter on Coding
and Information Theory. The exercise asks for a simulation of Shannon’s random codes, so
the simulation starts with a random code table. This is not the same as Shannon’s proof of
his theorem, which uses the average of all possible code tables. In practice however, a single
specific large random table will almost certainly show the behavior expected by Shannon’s
theorem. (I don’t know if it’s possible to find an analytic solution to this simulation.)
The simulation is a challenge in several ways, since to get a reasonable code, a very large
blocksize is required by Shannon’s theorem. Large blocksizes require exponentially larger code
tables, and these quickly exhaust the memory of any actual system.
8=BE
Please keep in mind that all this discussion assumes a binary symmetric channel with prob-
8= DDB
ability , that is, on the average, one out of every four bits transmitted is reversed — an
extremely high error rate. The corresponding channel capacity is approximately .
E E E E =$=$=

Table 4.1 gives sample sizes for the random codeword tables, where the block sizes are just
arbitrarily set to multiples of , that is , , , , , , and where the codeword lengths
are taken to be at the channel capacity (which would be the minimum possible value for a good
code).
A
In practice, simulation runs need longer codeword lengths, perhaps up to twice the channel

E A
capacity values. I was able to run the simulation for blocksize = , using up to Megabytes

(
of main memory on an iMac. Hypothetical runs for blocksizes of or would take up to
E (
Gigabyte or Gigabytes of memory. (The memory requirements grow exponentially as the
blocksize grows: adding to the blocksize makes the memory requirements about times as
great.)
Blocksize Number Min Codeword Minimum Code Table

(bits) of Entries Length (bits) Size (in K bytes)
5 32 27 0.1 K
10 1 024 53 6K
15 32 768 80 327 K
20 1 048 576 106 13 894 K
25 33 554 432 132 553 648 K
30 1 073 741 842 159 21 206 401 K
Table 4.1 Capacity Simulation Parameters.

Initial results of simulation runs were discouraging and equivocal. It seemed that I would
need a far larger blocksize than my computer could run in order to get reasonable results. For
example, the results of runs answering the original questions in the exercise were:
6E >)( , D , and (

E)> A> 6E
For a blocksize of , runs with code word lengths of produce
respective error rates of , , and .
( A
6E
Thus at substantially under the channel capacity for codeword lengths ( bits or
of capacity), the error rate is still , indicating that the blocksize is not remotely close to a
value needed for reasonable accuracy. More discouraging was that these runs gave no hint of a
magic channel capacity value behind the scenes.
4.1 Visualizing Shannon’s Noisy Coding Theorem.

E E
D
I continued with simulation runs for block sizes of , , , and and for a variety of
codeword lengths (all divisible by for convenience of the simulation program). Eventually
I plotted all the data in a graph, and got the astonishing picture shown in Figure 4.1. (The
Postscript file that creates the graph can be found in the online material.)
The remarkable graph in Figure 4.1 shows how far away the simulation is from the good
results predicted by Shannon’s theorem. At the same time, the graph gives a clear idea of what
the results would be if one could keep increasing the blocksize: the graph would be increasingly
vertical as it crosses the vertical dashed line marking the channel capacity.
E
This specific graph shows the actual simulation results, with points connected by straight
lines. (No curve-fitting was carried out.) The black line (for blocksize = ), is accurate since

6E
each point represents trials, but there are only 14 points, so this graph has an angular
look. The graphs for blocksizes of
and are smoother (because more points are plotted),
but less accurate (smaller number of trials per point). Finally, the red graph (blocksize = )
is somewhat irregular because only
trials per point did not produce the accuracy of the
other graphs.
It is important to realize that the graph only illustrates what Shannon’s theorem proves. The
graph and these simulation results prove nothing, but just give an indication of what might be
true.
4.2 The Converse of the Noisy Coding Theorem.

The graph also illustrates another theorem known as the Converse of Shannon’s Noisy Coding
Theorem. Roughly speaking, the converse says that if one is signaling at a fixed rate more than
channel capacity (to the left of the vertical dashed line in the picture), and if the block size
gets arbitrarily large, then the error rate will get arbitrarily close to
. Stated another way,
at more than channel capacity, as the block size gets larger and larger, the error rate must get
closer and closer to . Contrast this with Shannon’s theorem, which says that if one signals
at less than channel capacity, and if the block size get arbitrarily large, then the error rate will
4. Visualizing Channel Capacity 33

get arbitrarily close to . In the limit as the block size tends to infinity, the graph will look like
a step function: at
to the left of the dashed vertical line and at to the right.
The proofs of both of Shannon’s theorems are found in books on information theory. In the
terms of this course they are very difficult and technical.
The Java simulation program in three files is found on page 179.
Figure 4.1. Simulation of Shannon’s Random Codes
(Error Probability = 0.25)
100
Blocksize = 5, Tablesize = 32, Trials = 10000000 @ 14 points
90 Blocksize = 20, Tablesize = 1048576, Trials = 10000 @ 30 points
Dashed vertical line: channel capacity = 0.18872 18755
(as codeword size, equals 1/0.18872 = 5.2988)
80
70
60
50
Error Rate (%)
40
30
20
10
0
0 2 4 6 8 10 12 14 16 18 20 22 24
Codeword Size (1 = Source Size)
Figure 4.1 Simulation of Shannon’s Random Codes
5
The Huffman Code for Compression
5.1 Lossless Compression.

Starting with a a file (that is, a message), one wants a compressed file smaller than the original
from which one can recover the original file exactly. This chapter focuses on lossless compres-
sion, meaning that not one bit of information is lost during the compression/decompression
process.
Claude Shannon proved the following significant result:
Law SHANNON-2:
In the absence of noise, it is always possible to encode or
transmit a message with a number of bits arbitrarily close to the
entropy of the message, but never less than the entropy [known
as Shannon’s Noiseless Coding Theorem].
To achieve this result, it may be necessary to lump together a large number of messages.
In contrast to the proof of Shannon’s Noisy Coding Theorem (discussed in the chapter on
coding theory), Shannon’s Noiseless Coding Theorem has a constructive proof given below as
a reasonable method for data compression, though the method is not used any more for actual
compression.
Intuitively, a random message has the largest entropy and allows no compression at all.
A message that is not random will have some “regularity” to it, some predictability, some
“patterns” in the sequence of its bits. Such patterns could be described in a more succinct way,
leading to compression.
These concepts provide a new way to describe random sequences: A finite sequence is
random if it has no succinct representation, that is, any program or algorithm that will generate
the sequence is at least as long as the sequence itself. This is the concept of algorithmic
information theory, invented by Chaitin and Kolmogoroff, which is beyond the scope of this
discussion.
Still speaking intuitively, the result of a good compression algorithm is a file that appears
random. (If it did not look random, it would be possible to compress it further.) Also, a good
compression algorithm will end up relatively close to the entropy, so one knows that no further
compression is possible.
Law COMPRESSION-1:
Just like Niven’s law (never fire a laser at a mirror), one
says: never compress a message that’s already compressed.
The joke is to take a large file, Shakespeare’s plays, say, and repeatedly compress the file
until the result is a single bit. Intuitively, one realizes that there really is information in the plays.
They could be compressed, but many bits would still be needed to represent the true information
in them. (Of course, the entropy of his plays gives the smallest size of any compressed version.)
Just as with inventors of perpetual motion machines, crackpots fixated on compression reg-
ularly announce fantastic compression schemes without knowledge of the limitations imposed
by information theory. As I write this (2002), a small company is claiming to compress any
random file and recover every bit of the original during decompression, assuming the original
file is “large enough”. Simply counting the number of possible files to compress and the num-
ber of their compressed forms easily shows that these claims are impossible, and the argument
requires nothing subtle from information theory. People making claims about perpetual motion
or faster than light travel are just suggesting a violation of the accepted laws of physics, some-
thing that might be true here or in another universe, but the compression crazies are suggesting
a violation of the laws of logic — impossible for reasonable people to imagine.
For one very simple example of compression, consider a file with many sequences of blank
characters. One needs a special character (or sequence of characters) to represent a sequence of
blanks. Use a Ascii blank itself for this purpose. This special character must always be followed
by an 8-bit number representing the number of blanks. Thus a single blank is represented by
two bytes: a normal blank, followed by a 1. If the file contains many long sequences of blanks,
each such sequence shorter than 256 blanks could be represented by two bytes. This might
provide a large compression. On the other hand, if the file mostly consists of isolated blanks,
the above technique will replace single blanks by two-byte sequences, so the “compression”
algorithm will output a larger file than the input.
A later section below presents Huffman’s compression algorithm, which in a sense is op-
timal. Huffman’s code provides an explicit solution to Shannon’s Noiseless Coding Theorem,
but Huffman’s algorithm has significant disadvantages. It usually needs an initial statistical
analysis of the file itself, and it usually requires transmitting a large decoding table along with
the file. For these and other reasons, Huffman’s code has been replaced with a large number of
other clever compression codes. The complete description is far beyond the scope of this book,
but the .gif images or the bit stream processed by a modem use very sophisticated algorithms
that adapt to the nature of the source file. These methods allow the receiving station to construct
a decode table “on the fly” as it carries out decompression.
Images with a .gif suffix use the LZW compression algorithm, which has an interesting his-
tory. Two researches named Lempel and Ziv came up with a remarkable compression algorithm
which they published in the scientific literature, though their companies also patented it: the
(surprise) Lempel-Zip method. Later an employee of Unisys named Welch made minor modifi-
cations to Lempel-Ziv to produce the LZW algorithm used in .gif images. Unisys patented the
algorithm and after its use became widespread started demanding payments for it. I personally
5. The Huffman Code for Compression 37
resent this situation because Unisys and even Welch had relatively little to do with the break-
through — Lempel and Ziv were the ones with the insight — yet Unisys wants money for its
minor modification of a standard algorithm.
5.2 Lossy Compression.

Lossy compression (as opposed to lossless) means that the process of compression followed
by decompression does not have to yield exactly the original file. One accepts a loss of qual-
ity. (Be careful not to spell the word as “lousy” or “loosey”, etc.) When the file is executable
computer code, this would be unacceptable — the lower-quality file would not likely be exe-
cutable any more. If the original file is a picture, however, the recovered file (after compres-
sion/decompression) may look nearly as good as the original.
Lossy compression is a huge area with many applications for our society. These include the
Jpeg standard for individual pictures and the Mpeg standard for motion pictures. Both these
standards are marvels of compression technology. Jpeg is used in modern digital cameras,
allowing the image to be saved in far less memory than the original representation of the color
of each pixel in the image. Mpeg is the basis for modern DVD movie players and for satellite
transmission of motion pictures.
5.3 Huffman Lossless Compression.

Variable Length and Prefix Codes: Elsewhere in this book codewords only occur that are
all the same length (for a given code). However the Huffman code in this section is a variable
length code. Another well-known but old-fashioned variable length code is the Morse code
widely used for radio and telegraph transmissions. Table 5.1 gives the code. The idea was
to use short codewords for the most commonly occurring letters and longer codewords for less
frequent letters.
Morse code presents an immediate decoding problem, since for example, an N is "-.",
but the codewords for B, C, D, K, X, and Y, also start with "-.". In fact, the code for C is
the same as the code for N repeated twice. Just given a sequence of dots and dashes, it is not
possible to uniquely break the sequence into letters. For this reason, Morse code requires an
extra “symbol”: a short pause between letters. (There is a longer pause between words.)
The Huffman code that is the subject here does not have this problem. It is a prefix code,
meaning that no codeword is a prefix of another codeword. Not only is it possible to separate

any string of s and s uniquely into codewords, but the decoding is very easy, since a unique
entry always matches the next part of the input. There are other codes that do not have the prefix
property, but that are nevertheless uniquely decodable. Such codes are not desirable because of
the difficulty of decoding.
The Huffman code starts with a sequence of symbols (a “file”) and computes the percent
frequency of each symbol.
Example 1. For example, if the sequence (or file) is aaaabbcd, then the frequency table is:
International Morse Code

Letter Morse Letter Morse Digit Morse
A .- N -. 0 -----
B -... O --- 1 .----
C -.-. P .--. 2 ..---
D -.. Q --.- 3 ...--
E . R .-. 4 ....-
F ..-. S ... 5 .....
G --. T - 6 -....
H .... U ..- 7 --...
I .. V ...- 8 ---..
J .--- W .-- 9 ----.
K -.- X -..-
L .-.. Y -.--
M -- Z --..
Table 5.1 The Morse Code.
Symbol: a, Weight: 0.5

Symbol: b, Weight: 0.25
Symbol: c, Weight: 0.125
Symbol: d, Weight: 0.125
The Huffman algorithm starts with a simple list of each symbol, regarding the list as a
sequence of single-element trees with the symbol and the frequency at the root node of each
tree (the only node in this case). Then the algorithm repeatedly combines the two least frequent
root nodes as left and right subtrees of a new root node with symbol and frequency the sum of
the two previous frequencies. (I use the symbol @ as the symbol for the root of the combined
tree.)
Thus, referring to Figure 5.1, the first step combines the single-node trees for letters c and
d, each with frequencies 0.125, into a single tree, with subtrees the trees above, and a root
node with frequency 0.125 + 0.125 = 0.25. Now this process is repeated with the new
node and the old node for the letter b with frequency 0.25. The final combination yields a
single tree with frequency 1.0.
If there are multiple choices for the “least frequent” root nodes, then make any choice. The
resulting Huffman trees and the resulting codes may not be the same, but the average code
lengths must always work out the same.
So step 1 in Figure 5.1combines c and d, yielding a new root node of frequency 0.25.
Then step 2 combines the result of step 1 with b. That result is combined in step 3 with a, to
give a single root node with frequency 1.0.
In the final part of the algorithm, one heads down the tree from the root as shown above,
building a code string as one goes, adding a 0 as one goes up and a 1 as one goes down.
+---d: 0.1250 (step 1)

|
+---+---@: 0.2500 (step 2)
| |
| +---c: 0.1250 (step 1)
|
+---+---@: 0.5000 (step 3)
| |
| +---b: 0.2500 (step 2)
|
---+---@: 1.0000
|
+---a: 0.5000 (step 3)
Figure 5.1 Huffman Tree: Entropy equal average code length.
+---d: 0.1250, 000 (step 1)

|
+---+---@: 0.2500, 00 (step 2)
| |
| +---c: 0.1250, 001 (step 1)
|
+---+---@: 0.5000, 0 (step 3)
| |
| +---b: 0.2500, 01 (step 2)
|
---+---@: 1.0000,
|
+---a: 0.5000, 1 (step 3)
Figure 5.2 Huffman Tree: Codewords.
The resulting binary codes for symbols are shown in bold in Figure 5.2(note the codes for
intermediate nodes also):
To encode using the result of the Huffman algorithm, one makes up a code table consisting
of each symbol followed by the corresponding codeword. To encode, look up each symbol
in the table, and fetch the corresponding codeword. (Encoding can be efficient if the table is
arranged so that binary search or hashing is possible.)
Symbol: a, Codeword: 1
Symbol: b, Codeword: 01
Symbol: c, Codeword: 001
Symbol: d, Codeword: 000
The same table could be used for decoding, by looking up successive sequences of code
symbols, but this would not be efficient. The process of decoding can be made simple and
efficient by using the above Huffman coding tree itself. Start at the root (left side) of the tree
+---d: 0.1250, 000 (step 1)

|
+---+---@: 0.2500, 00 (step 2)
| |
| +---c: 0.1250, 001 (step 1)
|
+---+---@: 0.5000, 0 (step 3)
| |
| +---b: 0.2500, 01 (step 2)
|
---+---@: 1.0000,
|
+---a: 0.5000, 1 (step 3)
Figure 5.3 Huffman Tree: Decoding.
and process the code symbols 0 and 1 one at a time. For a 0 head upward and for a 1 head
downward. When a leaf node (one with no subnodes) is reached, the symbol at that node is
the one being decoded. For example, in decoding the string 001, start at the root, head up
because of the first (leftmost) 0. Then head up again because of the second 0. Finally head
down because of the final 1. This is now a leaf node holding c, so that is the symbol decoded
from 001. The diagram in Figure 5.3shows the path through the tree in boldface:
The entropy of the four symbols above with the given probabilities is 1.75, and this is
exactly the same as the average code length, given by
0.5 * 1 + 0.25 * 2 + 0.125 * 3 + 0.125 * 3 = 1.75.
Huffman codes are always optimal (the best possible), but this particular code has average
code length equal to the entropy, and it is never possible to create a code with shorter average
length. Most Huffman codes have average code length greater than the entropy (unless all
frequencies are a fraction with numerator and denominator a power of 2). The next simple
example shows this.
Example 2. Start with the file aaaaaabc. Here is the frequency table, and the tree along with
the the code strings is in Figure 5.4:
Symbol: a, Weight: 0.75
Symbol: b, Weight: 0.125
Symbol: c, Weight: 0.125
In this case, the entropy is 1.061278 while the average code length is 1.25.
Product Codes: Now suppose one forms the “product” code of the code in Example 2 by con-
sidering all possible pairs of symbols and their respective probabilities, which are the products
of the probabilities for individual symbols:
Symbol: A (for aa), Weight: 0.5625 = 0.75 * 0.75
Symbol: B (for ab), Weight: 0.09375 = 0.75 * 0.125
Symbol: C (for ba), Weight: 0.09375 = 0.125 * 0.75
Symbol: D (for ac), Weight: 0.09375 = 0.75 * 0.125
Symbol: E (for ca), Weight: 0.09375 = 0.125 * 0.75
+---c: 0.1250, 00 (step 1)

|
+---+---@: 0.2500, 0 (step 2)
| |
| +---b: 0.1250, 01 (step 1)
|
---+---@: 1.0000,
|
+---a: 0.7500, 1 (step 2)
Figure 5.4 Huffman Tree: Entropy less than average code length.
Symbol: F (for bb), Weight: 0.015625 = 0.125 * 0.125

Symbol: G (for bc), Weight: 0.015625 = 0.125 * 0.125
Symbol: H (for cb), Weight: 0.015625 = 0.125 * 0.125
Symbol: I (for cc), Weight: 0.015625 = 0.125 * 0.125
<= EE>
Figure 5.5gives the corresponding Huffman tree, along with the code words:
<= 6E> E

With this product code, the entropy and average code length work out to be
and , but each new symbol (upper-case letter) represents two of the original symbols.

= BD E
Dividing by gives the original value for the entropy, but the average code length (per orig-
= > BD = E

inal symbol, not per new symbol) is , which is much closer to the entropy value of
= B A E

than the previous .
If one takes three symbols at a time, the average code length goes down to .
= > BD

Continuing with four at a time, five at a time, and so forth, it can be proved that the resulting
average code lengths get arbitrarily close to the entropy: . This, in turn, proves
Shannon’s Noiseless Coding Theorem, stated earlier.
Even though Huffman’s code is optimal (it yields the best possible code for a collection
of symbols and frequencies), the other adaptive algorithms (LZ or LZW) usually do a much
better job of compressing a file. How can this be, since Huffman is optimal? Suppose one
gathers statistics of frequencies of letters in English and creates an optimal Huffman code.
Such an analysis does not consider strings of letters in English, but only individual letters. The
Huffman product code for two letters at a time just assumes the frequencies are the products
of individual frequencies, but this is not true for actual English. For example, for pairs of
letters with “q” as the first letter, most pairs have probability 0, while the pair “qu” is about
as common as “q” by itself. It is possible to gather statistics about pairs of letters in English
and create a Huffman code for the pairs. One could proceed to statistics about longer strings
of letters and to corresponding Huffman codes. The resulting Huffman codes would eventually
perform better than the adaptive algorithms, but they would require an unacceptable amount of
statistical processing, and an unacceptably large code table. In contrast, the adaptive algorithms
do not need to transmit a code table, and they eventually adapt to long strings that occur over
and over.
A computer implementation of the Huffman algorithm appears on page 183. This software
generates printouts of the Huffman trees and other data reproduced above, but it does not read
or write actual binary data, nor does it transmit the dictionary (or the frequency table) along
+---D: 0.0938, 000 (step 5)

|
+---+---@: 0.1875, 00 (step 7)
| |
| +---C: 0.0938, 001 (step 5)
|
+---+---@: 0.4375, 0 (step 8)
| |
| | +---B: 0.0938, 010 (step 6)
| | |
| +---+---@: 0.2500, 01 (step 7)
| |
| | +---G: 0.0156, 011000 (step 2)
| | |
| | +---+---@: 0.0312, 01100 (step 3)
| | | |
| | | +---F: 0.0156, 011001 (step 2)
| | |
| | +---+---@: 0.0625, 0110 (step 4)
| | | |
| | | | +---I: 0.0156, 011010 (step 1)
| | | | |
| | | +---+---@: 0.0312, 01101 (step 3)
| | | |
| | | +---H: 0.0156, 011011 (step 1)
| | |
| +---+---@: 0.1562, 011 (step 6)
| |
| +---E: 0.0938, 0111 (step 4)
|
---+---@: 1.0000,
|
+---A: 0.5625, 1 (step 8)
Figure 5.5 Huffman Tree: Product code.

with the data, so that decoding is possible at the other end.
Exercises
1. Devise a simple example where there are different choices for the least trees and where
the Huffman algorithm yields different answers. Get an example where there are even
two different distributions for the lengths of the codewords. Verify that the average code
lengths are the same for the two examples. [Ans: See page 191 for an answer.]
2. After looking at the answer to the previous problem, see if you can create the simplest
possible example and argue that there is no simpler example.
3. Write a program that will translate to and from Morse code. In the coded text, use a blank
between the Morse codes for letters and three blanks between the codes for words. Only
insert newlines between words.
4. Expand the Huffman code implementation to handle binary files using techniques similar
to those of the Hamming code implementation. Comments at the end of the code indicate
how to handle these binary files.
6
The Hamming Code for Error Correction
6.1 Error correcting codes.

Codes that correct errors are essential to modern civilization and are used in devices from
modems to planetary satellites. The theory is mature, difficult, and mathematically oriented,
with tens of thousands of scholarly papers and books, but this chapter will only describe a
simple and elegant code, discovered in 1949.
6.2 Description of the Hamming Code.

Richard Hamming found a beautiful binary code that will correct any single error and will
detect any double error (two separate errors). The Hamming code has been used for computer
RAM, and is a good choice for randomly occurring errors. (If errors come in bursts, there are
other good codes that are particularly of interest to electrical engineers.) Unlike most other
error-correcting codes, this one is simple to understand.
The code uses extra redundant bits to check for errors, and performs the checks with special
check equations. A parity check equation of a sequence of bits just adds the bits of the sequence
and insists that the sum be even (for even parity) or odd (for odd parity). This chapter uses
even parity. Alternatively, one says that the sum is taken modulo (divide
by and take the
remainder), or one says that the sum is taken over the integers mod , .
A simple parity check will detect if there has been an error in one bit position, since even
parity will change to odd parity. (Any odd number of errors will show up as if there were just
1 error, and any even number of errors will look the same as no error.)
One has to force even parity by adding an extra parity bit and setting it either to or to to
make the overall parity come out even. It is important to realize that the extra parity check bit
participates in the check and is itself checked for errors, along with the other bits.
4
The Hamming code uses parity checks over portions of the positions in a block. Suppose
there are bits in consecutive positions from to

. The positions whose position number is

( D >
a power of are used as check bits, whose value must be determined from the data bits. Thus

the check bits are in positions , , , , , ..., up to the largest power of that is less than or
equal to the largest bit position. The remaining positions are reserved for data bits.
Each check bit has a corresponding check equation that covers a portion of all the bits, but
A ( E >

always includes the check bit itself. Consider the binary representation of the position numbers:
, , , , , , and so forth. If the position number
6. The Hamming Code for Error Correction 45
Position 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Bin Rep 1 10 11 100 101 110 111 1000 1001 1010 1011 1100 1101 1110 1111 10000 10001
Check:1 x x x x x x x x x
Check:2 x x x x x x x x
Check:16 x x
Table 6.1 Position of Parity Check Bits.
Position 1 2 3 4 5 6 7 8 9 10 11
Binary 1 10 11 100 101 110 111 1000 1001 1010 1011
Word 1 1 1 0 1 0 1 0 1 0 1
Check:1 1 1 1 1 1 1
Check:2 1 1 0 1 0 1
Check:4 0 1 0 1
Check:8 0 1 0 1
Table 6.2 Determining the Check Bit Values.

has a as its rightmost bit, then the check equation for check bit covers those positions. If

the position number has a as its next-to-rightmost bit, then the check equation for check bit

(
covers those positions. If the position number has a as its third-from-rightmost bit, then the
check equation for check bit covers those positions. Continue in this way through all check
bits. Table 6.1 summarizes this.
( D >
In detail, Table 6.1 shows the parity checks for the first 17 positions of the Hamming code.

(Check bits are in positions , , , , and , in bold italic in the table.)

( D
Table 6.2 assumes one starts with data bits (in plain type). The check equations
above are used to determine values for check bits in positions , , , and , to yield the word
11101010101 below, with check bits in bold italic here and in Table 6.2 .
Intuitively, the check equations allow one to “zero-in” on the position of a single error. For
example, suppose a single bit is transmitted in error. If the first check equation fails, then the
error must be in an odd position, and otherwise it must be in an even position. In other words, if
the first check fails, the position number of the bit in error must have its rightmost bit (in binary)
equal to one; otherwise it is zero. Similarly the second check gives the next-to-rightmost bit of
the position in error, and so forth.
Table 6.3 below gives the result of a single error in position (changed from a to a
). Three of the four parity checks fail, as shown below. Adding the position number of each
failing check gives the position number of the error bit, in this case.
The above discussion shows how to get single-error correction with the Hamming code.
Position 1 2 3 4 5 6 7 8 9 10 11 Result
Binary 1 10 11 100 101 110 111 1000 1001 1010 1011
Word 1 1 1 0 1 0 1 0 1 0 0 (err)
Check:1 1 1 1 1 1 0 1 fail
Check:2 1 1 0 1 0 0 2 fail
Check:4 0 1 0 1 - pass
Check:8 0 1 0 0 8 fail
Table 6.3 Correcting a 1-bit Error.
One can also get double-error detection by using a single extra check bit, which is in position
. (All other positions are handled as above.) The check equation in this case covers all bits,

including the new bit in position . Values for all the other check bits are calculated in any
order, but the overall check must come at the end, to determine the th check bit.
In case of a single error, this new check will fail. If only the new equation fails, but none

of the others, then the position in error is the new th check bit, so a single error of this new
bit can also be corrected. In case of two errors, the overall check (using position ) will pass,
but at least one of the other check equations must fail. This is how one detects a double error.
In this case there is not enough information present to say anything about the positions of the
two bits in error. Three or more errors at the same time can show up as no error, as two errors
detected, or as a single error that is “corrected” in the wrong position.

Notice that the Hamming code without the extra th check bit would react to a double
error by reversing some bogus position as if it were a single error. Thus the extra check bit
and the double error detection are very important for this code. Notice also that the check
bits themselves will also be corrected if one of them is transmitted in error (without any other
errors).
Law HAMMING-1:
The binary Hamming code is particularly useful because it
provides a good balance between error correction (1 error) and
error detection (2 errors).
6.3 Block sizes for the Hamming Code.

The Hamming code can accommodate any number of data bits, but it is interesting to list the
maximum size for each number of check bits. Table 6.4 below includes the overall check bit,
D > D
so that this is the full binary Hamming code, including double error detection.
For example, with bits or bytes, one gets
bits of data and uses bits for the
check bits. Thus an error-prone storage or transmission system would only need to devote out
Check Max Data Max Total

Bits Bits Size
3 1 4
4 4 8
5 11 16
6 26 32
7 57 64
8 120 128
Table 6.4 Sizes for the Binary Hamming Code.
of each >bytes or >8= E

to error correction/detection. Because of these pleasant properties,
this code has been used for RAM memory correction (before memory got so reliable).
6.4 A Java implementation of the Hamming Code.

The Hamming code lends itself to efficient implementation in hardware, and even a software
implementation just needs bit masks for the various parity checks and attention to detail. The
code given in this text in for reference, intended to be simple and understandable, rather than
efficient. The method here uses an array of bits (stored in int type). This means that actual bit
string input must be unpacked into the bit array, and then must be packed again at the end.
The implementation appears on page 193. The program allows any number of message
bits from 1 to 120, with corresponding total number of bits from 4 to 128. For example, if the
>2 6E = >>> >8= E

program uses the bold line in the table, there are 15 bytes of data and 1 byte devoted to check
bits. This gives an expansion factor of , using of the coded file for
check bits. The program does not just implement the lines in the table, but allows all values
less than or equal to 120 for the number of message bits.
7
Coping with Decimal Numbers
7.1 Decimal Error Detection and Error Correction.
theory, the binary number system is regarded as the finite field

The previous discussions have focused on coding theory applied to binary numbers. In coding
with two elements. The
theory can be extended to any finite field, but decimal numbers do not work because there is no
E E
finite field with 10 elements. It is possible to apply normal coding theory to the fields of order

and separately, and then handle a message modulo and modulo separately, combining
results at the end. It is also possible to convert a decimal number to binary and use binary error
E
detection and correction. For decimal numbers, neither of these approaches is satisfactory. In
the first case the extra error digits would be different numbers of binary and base digits,
which would not fit together and could not be combined into individual checkable decimal
digits. More importantly for both approaches, the possible errors in a message will be errors in
decimal digits. If the number is expressed in binary, a decimal error might change any number
of binary bits. If the number is represented using 4 bits per decimal digit, a single decimal error
could still change up to 4 bits. For these and other reasons, one needs new decimal-oriented
approaches to handle decimal numbers
Law DECIMAL-1:
One needs special methods for decimal-oriented applica-
tions; general binary coding theory is not useful.
7.2 Simple Error Detection Schemes.

In applications involving people, numbers with decimal digits are almost always used. (Called
the simian preference, because humans have 10 fingers – well, most of them anyway, and if
you count thumbs.) Humans and machines processing these numbers make decimal mistakes in
them. The most common mistakes that humans make are to alter a single digit or to transpose
two adjacent digits. It is also common to drop a digit or to add an extra digit. Machines can
misread a digit, though they would not transpose digits. Applications often add an extra check
digit to catch as many errors as possible. There is usually also a check equation which all digits,
including the check digit, must satisfy. On the average, one expects to detect about 90
7. Coping with Decimal Numbers 49
Suppose one has

4
digits in an application, such as a credit card number: %, , ...,
7% . A check digit will be added in position as . &
A Simple Check: The simplest check equation just adds all digits modulo (form the sum

and take the remainder on division by ) and expects to get . So the equation looks like:
& / % / / =$=$=6/ 7 % mod

where “mod” is given in Java or C by the operator %. Starting with the data digits, one needs to
&
choose the check digit so that the check equation will hold. This is easy to do by setting
& 4 % / / =$=$=6/ 7 % mod

The above simple scheme catches all errors in a single digit. In fact the value of the check
equation will tell the amount of the error but not which position is incorrect. This method
might be suitable for an optical scanning system that occasionally has trouble reading a digit
but knows the position in error. However, this scheme misses all adjacent transpositions, so it
is not desirable because transpositions are a common human error.
The U.S. Banking Check: A different and better check scheme is used by U.S. banks for an
8-digit processing number placed on bank checks. They add a check digit and use the check
equation:
& / A / % / A B / / B / mod
A / B /

It is essentially the same scheme to repeat the coefficients , A , and B
indefinitely. This
scheme also catches all single errors and it catches all adjacent transpositions of digits that
do not differ by 5. Thus 88.888% of adjacent transposition errors are caught (80 out of 90).
(Computer programs later in the book verify this fact; it is also easy to prove mathematically.)
The IBM Check: The most common check now in use, for example on credit card numbers, is
often called the “IBM check”. Here the check equation is:
%/ /

&/ / / =$=$= mod
where
means to multiply
by and add the decimal digits. In Java 2#a[i] =
E)(@@>
(2*a[i])/10 + (2*a[i])%10.
For example, if the account number is , then the check equation without the check
digit is:
KE / ( / @ / @ / > mod
/ / ( / / D / @ / / mod
/ ( / @ / @ / A mod > mod ><1
so that the check digit must equal 4 to make the check equation true. Actual credit cards
currently have 16 digits and place the check digit on the right, but they treat the other digits
as above, so that the first (leftmost) digit is acted on by , while the final 16th digit (the
( B B 6E@ )( (
rightmost digit, which is also the check digit) is just added in to the check sum. For example, if
a VISA card has number , then the rightmost is the check digit, chosen
so that the check equation will work out to zero:
( / / KB / / KB / / / /
/ E / @ / / / / / ( mod
D / / E / / E / / / /
/ E / @ / / ( / / ( / ( mod
@ @
This scheme detects all single-digit errors as well as all adjacent transpositions except for
and . Thus it catches 97.777% of all adjacent transposition errors (88 out of 90).
The ISBN Check: The ISBN number of a book uses a mod 11 check. The check equation is:
%/ A / (

&/ / =$=$=6/ 7 % mod 81

for

If
just keep repeating the weights from to . Actual ISBN numbers use

$( (>E)>@
@
and write the digits in reverse order. For example if the ISBN number is , then the
rightmost digit of is chosen so that the following check equation is true:
/ @ / D ( / B / > /
E ( / ( > / A E / > / @ mod
The check catches all single errors and all transpositions, whether adjacent or not. (If

then transpositions are caught if they are no more than 10 positions apart.) Unfortunately,

in this check the check “digit” has a value from to , and the ISBN number uses an

to represent . (I guess they were thinking of “ten” in Roman numerals.) Because the check

AB"(>> )(>
calculates numbers modulo and requires an extra symbol for the “digit” , it is not properly
speaking a decimal check at all. Here is an ISBN number with an in it: (the
ISBN for Franz Kafka’s Der Prozess).
All these checks have disadvantages. The IBM check seems best, except that it does miss

two transpositions. The ISBN check performs much better, but it has the serious disadvantage
of an extra inside every 11th number on the average.
Verhoeff’s Scheme: The next chapter presents a more interesting decimal error detecting
scheme developed by a Dutch mathematician Verhoeff: ¡a href=”verhoeff.html”¿Verhoeff’s
Check¡/a¿. This method has a section to itself because it is more challenging mathematically.
7.3 Error Detection Using More Than One Check Digit.

For a very low error rate, one could use two check digits and a modulo check, with weights @B

successive powers of . This check catches 100% of all errors on Verhoeff’s list (except for
additions or omissions), and 99.94% of adjacent double errors. Overall, it catches 99% of all
errors, compared with 90% for the single check digit schemes.
Keep in mind that the mod @B
check “digit” itself would be represented as two decimal
@@
digits. These two digits would be protected against all double errors. However, a double error
@B
would only be caught of the time if it involved adjacent digits where one is a check digit
(part of the mod check digit), and the other is a data digit.
For even better performance, one could use a modulo or a modulo @@B @@BA
check, again

@B @@B @@BA
with weights successive powers of , and with 3 or 4 check digits, respectively. (The numbers
, , and are each the largest prime less than the respective power of .)
7.4 Decimal Error Correction.

The Hamming code can be carried out modulo any prime number, in particular modulo ,
using the field
base %%
. This would allow single error correction. Such a check would work with
digits, but one would just treat the base
number as if it were base . However, any

check digits might have the additional value of , so there must be some way to represent this
value, such as (as in ISBN numbers) or (as in hexadecimal).
The Hamming mod 11 Code With Two Check Digits: With two mod check digits and
up to 9 data digits, the code would use the following two check equations:
/ A /.==$=6/ @ / %'& mod

% //
& % / ./ ==$=/ / %'& mod

Here the two check digits are in the first equation and in the second. One starts with
% &
data digits through , then determines the check digit so that the first check equation
%'& %
is true, and then determines the check digit so that the second check equation is true. To
&
correct any single error, get the amount of the error from the second check equation and the
position of the error from the first.
This system will correct any single error (including errors in the check digits themselves).
Unlike the binary Hamming code, this code will interpret many double errors as if they were
some other single error.
The Hamming mod 11 Code With Three Check Digits: Nine data digits are probably too
D
few for current applications, but the Hamming code will work fine with a third check digit,
allowing up to data digits. Again, such a system would correct any single error, including

a single error in one of the check digits. Keep in mind that all these mod
checks require the
%&
possibility of an (or something else) as the value for any of the check digits.
Here the digits go from
up to as high as
&
. The check digits are , , and . & % %%
(An actual implementation might collect these three digits at one end.) There are three check
equations:
/
% / / / A / =$=$= / @ / %'& /
&%% / % % / A % / =$=$=/ @ & / % /
=$== /
%%'& / %%% / %% / A %% / =$=$=/G@ %% / % & mod
/ % / / /.=$==/ / %'& /
&%% / % / % / % /.=$==/ & / % /
=$== /
%%'& / %%% / %% / %% /.=$==/ %% / % & mod
/ / / /.=$=$=/ / mod
& % %% % &

=$=$= '% & % $= == % % % &
The above equations explicitly show multipliers by to help keep the pattern clear. Start
with up to 118 data digits , , , , , , , . In either order, determine the
% %% &
D
check digits and using the first two equations. Then determine the check digit using
the third equation. If there are fewer than data digits, just set the remaining ones equal to
zero in the equations above (and leave them off in transmission and storage). As before, the
third equation gives the value of the error, and the first two equations give the location of the

2
error. If is the location of the error ( ), then the first equation gives , while the
second gives , and together these give .
Suppose this scheme were used for current credit card numbers. These use the IBM scheme
and have 15 data digits (expandable to any number of digits) and one check digit. One would
replace these with 15 data digits (expandable to 118 data digits) and 3 mod check digits.

Thus the new numbers would have two disadvantages compared with the old: two extra digits
62 8= BE

and the possibility of an in three of the digits. In exchange, the system would correct any
single error, and would have an extremely low error rate for random errors (
compared with the current error rate of ).
With the full 121 digits, this scheme interprets any two errors as if they were a single error
in some other position. However, some of these miscorrections try to put a ten, that is, an
D8=E D = E
into a position other than 0, 1 or 11, and so are recognized as a double error. Thus with 121
BE
digits, about of double errors are detected, or are undetected.
With only 18 digits, most double errors ( ) will be interpreted as an error in a position
greater than 18 and so will be detected as a double error. Also if a double error is interpreted

@8=?A
as a “correction” of a position other than 0, 1, or 11 to an , or if the amount of error comes
D( =?A 6E<=B

up as 0, these also represents a detected double error, detecting an additional , for a total
of of double errors detected, or undetected.
In summary, the three-digit Hamming mod 11 scheme would have an extremely low error
rate if used only for detection. It would allow the correction of any single error. However, for
a size such as 15 data digits, about6E of double errors would be erroneously interpreted as
a single error, compounding the problems. For these and other reasons, it does not seem likely
that anyone will want to use this code for practical decimal-oriented applications.
In contrast, the binary Hamming code (see the chapter on the Hamming code) corrects any
single error and detects any double error. In the mod Hamming code above, the overall
check gives the amount of the error. With the binary Hamming code, the corresponding check

is not needed, since if there is an error, its amount must be (that is, a changed to a or
vice versa). Moreover, any double error will still show up as an error according to the other
checks, but a double binary error appears as no error in the overall check. Thus the double
error detection works only in the special case of the base 2 Hamming code. Because of this the
binary Hamming code has often been implemented for hardware RAM memory. The absence
of double error detection in the Hamming mod code, and more importantly, the fact that
double errors can mask as a single correctable errors, are fatal flaws.
Law DECIMAL-2:
Hamming codes exist for prime bases other two, but be-
cause they do not support double error detection, and because
they may misinterpret a double error as a correctable single er-
ror, they are not useful.
7.5 Java Implementation of the Schemes.
✦ The U.S. Bank Scheme (all but 10 transpositions detected): see the program on page
203.
✦ The ”IBM” Scheme (all but 2 transpositions detected): see the program on page 206.
✦ The ISBN Scheme (all transpositions detected): see the program on page 209.
✦ The mod 97 Scheme (all transpositions detected): see the program on page 212.
✦ Hamming mod 11 Code, Test single error correction: see the program on page 215.
✦ Hamming mod 11 Code, Test handling of double errors: see the program on page
219.
8
Verhoeff’s Decimal Error Detection
In the past, researchers have given “proofs” that it is impossible for a check to detect all
adjacent transpositions (as well as all single errors). It is true that if one uses a simple sum of
digits with weights on the individual locations, then such a check is mathematically impossible.
However, the more general scheme in this section works.
8.1 Types of Decimal Errors.

In 1969, a Dutch mathematician named Verhoeff carried out a study of errors made by humans
in handling decimal numbers. He identified the following principal types:
✦ single errors:
changed to (60 to 95 percent of all errors)
✦ adjacent transpositions: changed to (10 to 20 percent)
✦ twin errors: changed to (0.5 to 1.5 percent)
✦ jump transpositions: changed to (0.5 to 1.5 percent)
✦ jump twin errors: changed to (below 1 percent)
✦ phonetic errors: changed to (0.5 to 1.5 percent; “phonetic” because in some
languages the two have similar pronunciations, as with thirty and thirteen)
✦ omitting or adding a digit: (10 to 20 percent)
8.2 The Dihedral Group D5.

@
Verhoeff had the clever idea to use some other method besides addition modulo for com-

bining the integers from to . Instead he used the operation of a group known as the dihedral
group , represented by the symmetries of a pentagon, which has ten elements that one can
@
name through . The discussion here will use for the group operation and for the identity
element. This is not a commutative group, so that
will not always equal . Table 8.1
gives the multiplication table for this group. (Each table entry gives the result of the bold table
entry on the left combined with the bold table entry across the top, written left to right. The
bold italic entries show results which are not commutative.)
The reader should realize that is a complex entity whose elements are somewhat arbi-
@
trarily mapped to the integers from to , and the group operation is not at all like ordinary
8. Verhoeff’s Decimal Error Detection 55
# 0 1 2 3 4 5 6 7 8 9
0 0 1 2 3 4 5 6 7 8 9
1 1 2 3 4 0 6 7 8 9 5
2 2 3 4 0 1 7 8 9 5 6
3 3 4 0 1 2 8 9 5 6 7
4 4 0 1 2 3 9 5 6 7 8
5 5 9 8 7 6 0 4 3 2 1
6 6 5 9 8 7 1 0 4 3 2
7 7 6 5 9 8 2 1 0 4 3
8 8 7 6 5 9 3 2 1 0 4
9 9 8 7 6 5 4 3 2 1 0
Table 8.1 Multiplication in the Dihedral Group .
@
@>
addition or multiplication. Keep in mind that the ten symbols through are the only symbols
@ @
in the group. When two are combined, you get another of them. There is no number , but

@ B *
only separate group elements and that could be combined in two ways: and

(
. There is no concept of order such as or in this group. The five numbers
through combine in the group just as they do with addition in the group , but the remaining
E > B D @
E
numbers are quite different, since each of , , , , and is its own inverse. (With ordinary
addition in
%'&
, only is its own inverse.)
Figure 8.1 at the end of this chapter shows a way to visualize this dihedral group. It is known
as the group of symmetries of a pentagon, meaning all rigid motions in the 2-dimensional plane
A (
that will transform a regular pentagon onto itself. First there are clockwise rotations by 0, 72,
144, 216, and 288 degrees. These rotations correspond to the group elements , , , , and ,
respectively. Any other possible rotation that would take a pentagon to itself is equivalent to one
of these, as for example a counter-clockwise rotation by 432 degrees is the same as element 4:
a clockwise rotation by 288 degrees. These 5 group elements are illustrated in the left column
of Figure 8.1.
The remaining 5 group elements can be visualized as each of the earlier rotations followed
by a reflection of the pentagon (flipping it over) along the vertical axis of symmetry. The right
E > B D @
column of Figure 8.1 shows the action at the left in each case followed by a reflection about
@
the vertical axis. These actions correspond to group elements , , , , and , as shown in the
figure. These actions show where the entries in Table 8.1 come from. For example, element
@
is a rotation by 288 degrees followed by a vertical reflection. If one wants to see the result of

@
, first do the action specified by , that is, a rotation by 144 degrees. Then follow that by
the action specified by : a further rotation by 288, for a rotation by 144 + 288 = 432, which is

> @ >
the same as a rotation by 72 (group element ), followed by the vertical reflection, which will
result finally in group element . This shows that
, as we also see from the table. All
other results of combining two group elements can be see to agree with the table entries in the
same way. The result of any combination of rotations (by any angle that is a multiple of 72
degrees) and of reflections (about any of the 10 axes of symmetry of the pentagon) must end
up equivalent to one of the 10 different group elements.
8.3 Verhoeff’s Scheme.

If one just used the check equation
& % =$=$=

7%
where is the group operation in , this would be much better than simply adding the
digits modulo 10, since in both cases single errors are caught, but in two-thirds of adjacent
transpositions are caught (60 out of 90, represented by the bold italic entries in the table above),
whereas ordinary addition catches no transpositions. This suggests that stirring things up a little
more would give the answer.
Verhoeff considered check equations of the form
& & % % =$=$=

7 % 7 %

where each is a permutation of the ten digits. He was able to get an excellent check in the

special case where is the ith iteration of a fixed permutation .
As the Java implementation will show, this check is not hard to program and is efficient, but
it employs several tables and could not be carried out by hand as all the earlier checks could.
This discussion will employ Java notation to keep the subscripts straight.
First the check needs the group operation, defined as a two-dimensional array op[i][j],
for i and j going from 0 to 9 giving the result of combining the two numbers in (the same
as the table above):
int[][] op= {
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9},
{1, 2, 3, 4, 0, 6, 7, 8, 9, 5},
{2, 3, 4, 0, 1, 7, 8, 9, 5, 6},
{3, 4, 0, 1, 2, 8, 9, 5, 6, 7},
{4, 0, 1, 2, 3, 9, 5, 6, 7, 8},
{5, 9, 8, 7, 6, 0, 4, 3, 2 ,1},
{6, 5, 9, 8, 7, 1, 0, 4, 3, 2},
{7, 6, 5, 9, 8, 2, 1, 0, 4, 3},
{8, 7, 6, 5, 9, 3, 2, 1, 0, 4},
{9, 8, 7, 6, 5, 4, 3, 2, 1, 0} };
Then comes an array inv, where inv[i] gives the inverse of each digit i in :
int[] inv = {0, 4, 3, 2, 1, 5, 6, 7, 8, 9};
Finally, the check requires another two-dimensional array giving the special permutation and
iterations of it:
int[][] F = new int[8][];
F[0] = new int[]{0, 1, 2, 3, 4, 5, 6, 7, 8, 9}; // identity perm
F[1] = new int[]{1, 5, 7, 6, 2, 8, 3, 0, 9, 4}; // "magic" perm
for (int i = 2; i < 8; i++) { // iterate for remaining perms
F[i] = new int[10];

for (int j = 0; j < 10; j++)
F[i][j] = F[i-1] [F[1][j]];
}
Now the check equation takes the form:
public static boolean doCheck(int[] a) {
int check = 0;
for (int i = 0; i < a.length; i++)
check = op[check][ F[i % 8][a[i]] ];
if (check != 0) return false;
else return true;
}
The check digit can be inserted in position a[0] using the array Inv, as is show in the
actual implementation later in this section.
Verhoeff’s check above catches all single errors and all adjacent transpositions. It also
@
catches 95.555% of twin errors, 94.222% of jump transpositions and jump twin errors, and

95.3125% percent of phonetic errors (assuming ranges from to ).
Law DECIMAL-3:
It is possible to use a single check digit to detect all single
errors and all adjacent transpositions, but this method is seldom
used.
I had earlier formulated the above law using the words “and nobody uses this method” at the
end. However, Professor Ralph-Hardo Schulz of the Freie Universität in Berlin pointed out that
Verhoeff’s method was used for serial numbers on German currency, before the introduction of
the Euro.
8.4 Java Implementation of Verhoeff’s Scheme.
✦ Use of the Dihedral Group (all but 30 transpositions detected): see the program on
page 223.
✦ Verhoeff’s Scheme (all transpositions detected): see the program on page 226.
58
Part III
Introduction to
Cryptography
60
9
Cryptograms and Terminology
9.1 Cryptograms.
Newspapers in the U.S. have long presented to their readers a special puzzle called a cryp-
togram. The puzzle has taken a quotation in capital letters and substituted another letter for
each given letter. The trick is to guess the substitutions and recover the original quotation. Here
is an example of a cryptogram:
ZFY TM ZGM LMGM ZA HF Z YZGJRBFI QRZBF
ATMQX TBXL WHFPNAMY ZRZGVA HP AXGNIIRM ZFY PRBILX,
TLMGM BIFHGZFX ZGVBMA WRZAL UO FBILX.
YHCMG UMZWL, VZXLMT ZGFHRY
It looks like complete gibberish, but if one knows, deduces, or guesses the translation
scheme, the key for uncovering the quotation, then it is understandable. In this case the key
is:
Alphabet: ABCDEFGHIJKLMNOPQRSTUVWXYZ
Translated to: ZUWYMPILBDJRVFHQSGAXNCTKOE
Given the quotation, the person making up this cryptogram would write Z for each A, U for
each B, W for each C, and so forth. Given the cryptogram as above, one just has to go backwards,
changing each Z back to A, and so forth. In this way, knowing the translation key, it is easy to
recover the original quotation:
AND WE ARE HERE AS ON A DARKLING PLAIN
SWEPT WITH CONFUSED ALARMS OF STRUGGLE AND FLIGHT,
WHERE IGNORANT ARMIES CLASH BY NIGHT.
DOVER BEACH, MATHEW ARNOLD
I have never solved one of these puzzles, but my parents often used to spend an hour or so
recovering such quotations. (I must try one sometime.) I remember that my mother would first
focus on a word that is a single letter as above, since this letter must be either an A or an I in
ordinary English. After trying I for awhile, assume that Z is an A. Then there is a three-letter
word ZFY that one now guesses starts with an A. This word appears twice, so one might guess
that it is AND. From this, the last word (last name of an author?) becomes A N D. There is
only one well-known author whose name looks like this, and the quotation is perhaps his most
famous one, so one would have solved the puzzle immediately.
As another approach, my mother would check the frequencies of all the letters. In the
scrambled quotation (leaving off the last line), they are far from uniform: Z:11, M:10, F:9,
G:8, B:7, A:7, etc. Now, E is the most frequent letter in English, and it is the next-to-most-
62 III. Introduction to Cryptography
frequent in this quotation. One can also look for words with double letters or with other unusual
features. With trial and error, and some luck, one soon has the quotation.
A Java program to produce cryptograms at random, using whatever quotation you wish
appears on page 229.
9.2 Terminology from Cryptography.

The “quotation” above is ordinarily called a message or plaintext in cryptography. The cryp-
togram is the ciphertext. The process of transforming the plaintext into ciphertext is encryption,
while the reverse process of recovering the plaintext from the ciphertext is decryption. The 26
letters used for encryption and decryption is called the key. The particular method of translating
plaintext to ciphertext is called a cryptosystem.
It is important to realize that a single key could transform an arbitrarily long piece of plain-
text. Thus instead of keeping a large message secret, one uses cryptography so that one need
only keep a short key secret. This leads to a law:
Law CRYPTOGRAPHY-1a:
Cryptography reduces the problem of keeping an arbitrar-
ily long message secret to the problem of keeping a short key
secret. What an impressive improvement!
Although the techniques of cryptography are wonderful and powerful, one also needs to
realize the limitations of these tools. There still remains something to keep secret, even if it is
short:
Law CRYPTOGRAPHY-1b:
Cryptography reduces the problem of keeping an arbitrar-
ily long message secret to the problem of keeping a short key
secret. This is little if any improvement, since the problem of
keeping something secret still remains.
Keeping keys secret and distributing them to users are fundamental difficult problems in
cryptography which this book will take up later.
9.3 Security From Cryptography.

Cryptography has many uses, but first and foremost it provides security for data storage and
transmission. The security role is so important that older books titled Network Security cov-
ered only cryptography. Times have changed, but cryptography is still an essential tool for
9. Cryptograms and Terminology 63
achieving security. Network “sniffers” work because packets are in the clear, unencrypted. In
fact, we make as little use of cryptography as we do because of a long-standing policy of the
U.S. government to suppress and discourage work in the area and uses of it, outside classified
military applications. If a transmission line needs security there are still only two basic options:
physical security, with fences, rasorwire, and guard dogs, or security using cryptography. (The
emerging field of quantum cryptography may yield a fundamentally different solution.) With
cryptography, it doesn’t matter if the line goes across a field or across the world.
Law CRYPTOGRAPHY-2:
Cryptography is the only means of securing the transmis-
sion of information where the cost is independent of the dis-
tance.
9.4 Cryptanalysis.
The early part of this section regarded a cryptogram as a special (simple) cryptographic code.
The process of recovering the original quotation is a process of breaking this code. This is
called cryptanalysis, meaning the analysis of a cryptosystem. In this case the cryptanalysis is
relatively easy. One simple change would make is harder: just realize that revealing where the
blanks (word boundaries) are gives a lot of information. A much more difficult cryptogram
would leave out blanks and other punctuation. For example, consider the cryptogram:
OHQUFOMFGFMFOBEHOQOMIVAHZJVOAHBUFJWUAWGKEHDPBFQOVOMLBEDBWMPZZVF
OHQDVAZGWUGFMFAZHEMOHWOMLAFBKVOBGXTHAZGWQENFMXFOKGLOWGFUOMHEVQ
One might also present this just broken into groups of five characters for convenience in han-
dling:
OHQUF OMFGF MFOBE HOQOM IVAHZ JVOAH BUFJW UAWGK EHDPB FQOVO
MLBED BWMPZ ZVFOH QDVAZ GWUGF MFAZH EMOHW OMLAF BKVOB GXTHA
ZGWQE NFMXF OKGLO WGFUO MHEVQ
Now there are no individual words to start working on, so it is a much more difficult cryp-
togram to break. However, this is an encoding of the same quotation, and there is the same
uneven distribution of letters to help decrypt the cryptogram. Eventually, using the letter distri-
butions and a dictionary, along with distributions of pairs of letters, one could get the quotation
back:
Translated to: OXKQFDZGACIVLHEJSMBWPNURTY
ANDWEAREHEREASONADARKLINGPLAINSWEPTWITHCONFUSEDALARMSOFSTRUGGLE
ANDFLIGHTWHEREIGNORANTARMIESCLASHBYNIGHTDOVERBEACHMATHEWARNOLD
Even here there are problems breaking the text into words, since it seems to start out with
AND WEAR E HE REASON ....
Notice that the uneven statistical distribution of symbols is still a strong point of attack
on this system. A much better system uses multiple ciphertext symbols to represent the more
common plaintext letters. This is called a homophonic code, and it can be arbitrarily hard to
cryptanalyze if one uses enough additional ciphertext symbols.
The cryptanalysis above assumed that the ciphertext (the cryptogram) was available, but
nothing else. However, often much more information is at hand, and good cryptosystems
must be resistant to analysis in these cases also. Often the cryptanalyst has both plaintext and
matching ciphertext: a known plaintext attack. In the case of cryptograms, the code would
be known for all letters in that particular plaintext, and this would effectively break the code
immediately unless the plaintext were very plain indeed. Sometimes the cryptanalyst can even
choose the plaintext and then view his own choice of plaintext along with the corresponding
ciphertext: a chosen plaintext attack.
Amateurs in cryptography sometimes think they should keep the method of encryption
secret, along with the particular key. This is a bad idea though, because sooner or later the
underlying method with be discovered or bought or leaked.
Law CRYPTANALYSIS-1:
The method or algorithm of a cryptosystem must not be
kept secret, but only the key. All security must reside in keeping
the key secret.
In most of computer science, an algorithm that only works once in a while is no reasonable
algorithm at all. The situation is reversed in cryptography because it is intolerable if ciphertext
can be decrypted even “once in a while”.
Ordinarily an algorithm that only occasionally works is
useless, but a cryptanalysis algorithm that occasionally works
makes the cryptosystem useless.
People naively think that as computers get faster, it gets easier to break a cryptosystem, but
this is actually backwards logic. The utility of cryptography depends on the asymptotic ease
of encryption and decryption compared with the asymptotic difficulty of cryptanalysis. Faster
machines simply increase this disparity.
The faster computers get, the more powerful cryptography
gets. [Radia Perlman]
9.5 Inventing Cryptosystems.

As I write this (2002), I just attended a talk and just visited a web site where in each case a
entirely new cryptosystem was put forward saying “the set of keys is so large that cryptanalysis
>
( = A @
by trying them all is entirely impossible.” No mention was made of attacks other than brute
force ones, not even known plaintext attacks. The cryptograms in this section have
DD
different keys — the equivalent of an -bit binary key. This is far too many
keys for a brute force attack, yet a cryptogram is easily broken with only ciphertext available,
and is trivial to break under a known plaintext attack.
While a large number of possible keys is a requirement for
a strong cryptosystem, this does not ensure strength.
Getting a new cryptosystem approved for reasonable use by individuals, companies, and
governments is now an involved process. It can even start with a proposal like one of the
systems mentioned above, but better is a method like that for the Advanced Encryption Standard
(AES), where a number of different systems were proposed and evaluated by a committee of
experts and by the larger community. Mathematical evaluations of the strength of a new system
are desirable, as well as the “test of time”: a long period during which many people worldwide
evaluate the system and try to break it. The AES has already undergone an enormous amount
of scrutiny, and this examination will continue indefinitely. No new cryptosystem should ever
be used for sensitive (unclassified) work, but only systems that have been thoroughly studied.
A cryptosystem should only be used after it has been
widely and publicly studied over a period of years.
For example, the public key cryptosystem proposed by Diffie and Hellman in 1976 (based
on the integer knapsack problem) was broken in its strongest form in 1985. On the other hand,
the RSA cryptosystem, proposed in 1978, still resists all attacks. (Most attacks are based on
factoring large composite integers, and no efficient algorithm for this problem has been found.)
10
Perfect Cryptography
The One-Time Pad
10.1 The Caesar Cipher.

People have used cryptography for thousands of years. For example, the Caesar Cipher, which
was used during the time of Julius Caesar, wraps the alphabet from A to Z into a circle. The
method employs a fixed shift, say of 3, to transform A to D, B to E, and so on until W to Z, X to A,
Y to B, and Z to C. Thus a message ATTACK becomes DWWDFN and appears incomprehensible
to someone intercepting the message. (Well, incomprehensible to someone not very smart.) At
the other end, one can reverse the transformation by stepping 3 letters in the opposite direction
to change DWWDFN back to ATTACK.
This example illustrates many concepts and terminology from cryptography. The original
message is also called the plaintext. The transformed message is also called the ciphertext or
the encrypted message, and the process of creating the ciphertext is encryption. The process of
getting the original message back is called decryption, using a decryption algorithm. Thus one
decrypts the ciphertext.
The basic method used, moving a fixed distance around the circle of letters, is the encryption
algorithm. In this case the decryption algorithm is essentially the same. The specific distance
moved, 3 in this case, is the key for this algorithm, and in this type of symmetric key system,
the key is the same for both encryption and decryption. Usually the basic algorithm is not
kept secret, but only the specific key. The idea is to reduce the problem of keeping an entire
message secure to the problem of keeping a single short key secure, following Law C1 in the
>
Introduction to Cryptography.

E
For this simple algorithm there are only possible keys: the shift distances of , , , etc.

E >
up to , although leaves the message unchanged, so a key equal to is not going to keep
)>
many secrets. If the key is greater than , just divide
by
and take the remainder. (Thus the
keys just form the integers modulo , the group described in the chapter Cryptographer’s
Favorites.)
E
If an interceptor of this message suspects the nature of the algorithm used, it is easy to try
each of the
keys (leaving out ) to see if any meaningful message results – a method of
breaking a code known as exhaustive search. In this case the search is short, though it still
might pose problems if the letters in the ciphertext are run together without blanks between
words.
The Caesar Cipher is just a special case of the cryptograms from the previous chapter, since
10. Perfect Cryptography: The One-Time Pad 67
with a shift of A for example, the cyprtogram key is:

Translated to: DEFGHIJKLMNOPQRSTUVWXYZABC
Here is a computer implementation of the Caesar cipher: see page 232.
10.2 The Beale Cipher.

The Beale Cipher is a just simple extension of the Caesar Cipher, but it is easy to use by hand
and it provides excellent security.
E
Consider the Caesar cipher of the previous section, and associate the letters A through Z

E A
with the numbers through , that is, A is associated with , B with , C with , and so on
until Z with . One can represent the previous shift of in the example by the letter D, so that
each letter specifies a shift. A special encryption method called the Beale cipher starts with
a standard text (the key in this case) like the U.S. Constitution (WE THE PEOPLE . . .)
and with the message to encrypt, say ATTACK. Write down the letters of the standard text on
one line, followed by the letters of the message on the next line. In each column, the upper
(
letter is interpreted as a shift to use in a Caesar cipher on the letter in the second row. Thus
below in the second column, the E in the first row means a shift of is applied to the letter T in
the second row, to get the letter X.
Standard text (key): WETHEP
Message: ATTACK
Encrypted message: WXMHGZ
The person receiving the encrypted message must know what the standard text is. Then this
receiver can reverse the above encryption by applying the shifts in the opposite direction to get
the original message back. This method will handle a message of any length by just using more
of the standard text. Notice that in this example the two Ts came out as different letters in the
encrypted message. For more security, one should not use a standard text as well known as the
one in this example. Instead the sender and receiver could agree on a page of a book they both
have with them as the start of their standard text.
In fact, the original historical Beale cipher consisted of three messages: one in the clear and
the other two encrypted. The first encrypted message used the start of the U.S. Constitution just
as above, and told of a buried treasure. The third message was to tell where to find the treasure,
but it has never been decrypted. In fact, if the standard text is not known, it can be very hard to
cryptanalyze a Beale cipher.
All the security of this system resides with the secrecy of the standard text. There are a
number of subtle pitfalls with this method, as with most of cryptography. For example, suppose
you make a trip to, ummmm, Karjackistan, and you want to communicate in secret with your
friend back home. You buy two copies of a cheap detective novel, and agree on a page as above.
The Karjackistan Secret Police might notice the novel you are carrying, and might digitize the
entire book and try all possible starting points within its text, as possible ways to decrypt your
transmissions. If that didn’t work, they could try taking every third letter from every starting
point, or try other more complex schemes.
Here is a computer implementation of the Beale cipher: see page 235.
10.3 Perfect Cryptography: The One-Time Pad.

It may be surprising to the reader that there exist simple “perfect” encryption methods, meaning
that there is a mathematical proof that cryptanalysis is impossible. The term “perfect” in cryp-
tography also means that after an opponent receives the ciphertext he has no more information
than before receiving the ciphertext.
The simplest of these perfect methods is called the one-time pad. Later discussion explains
why these perfect methods are not practical to use in modern communications. However, for the
practical methods there is always the possibility that a clever researcher or even a clever hacker
could break the method. Also cryptanalysts can break these other methods using brute-force
exhaustive searches. The only issue is how long it takes to break them. With current strong
cryptographic algorithms, the chances are that there are no short-cut ways to break the systems,
and current cryptanalysis requires decades or millennia or longer to break the algorithms by
exhaustive search. (The time to break depends on various factors including especially the
length of the cryptographic key.) To summarize, with the practical methods there is no absolute
¡em¿guarantee¡/em¿ of security, but experts expect them to remain unbroken. On the other
hand, the One-Time Pad is completely unbreakable.
The One-Time Pad is just a simple variation on the Beale Cipher. It starts with a random
sequence of letters for the standard text (which is the key in this case). Suppose for example one
uses RQBOPS as the standard text, assuming these are 6 letters chosen completely at random,
and suppose the message is the same. Then encryption uses the same method as with the Beale
Cipher, except that the standard text or key is not a quotation from English, but is a random
string of letters.
Standard text (random key): RQBOPS
Message: ATTACK
Encrypted message: RJUORC

So, for example, the third column uses the letter B, representing a rotation of , to transform
the plaintext letter T into the ciphertext letter U. The receiver must have the same random string
of letters around for decryption: RQBOPS in this case. As the important part of this discussion,
I want to show that this method is ¡em¿perfect¡/em¿ as long as the random standard text letters
are kept secret. Suppose the message is GIVEUP instead of ATTACK. If one had started with
random letters LBYKXN as the standard text, instead of the letters RQBOPS, then the encryption
would have taken the form:
Standard text (random key): LBYKXN
Message: GIVEUP
Encrypted message: RJUORC
The encrypted message (ciphertext) is the same as before, even though the message is
completely different. An opponent who intercepts the encrypted message but knows nothing
10. Perfect Cryptography: The One-Time Pad 69
about the random standard text gets no information about the original message, whether it might
be ATTACK or GIVEUP or any other six-letter message. Given any message at all, one could
construct a standard text so that the message is encrypted to yield the ciphertext RJUORC. An
opponent intercepting the ciphertext has no way to favor one message over another. It is in this
sense that the one-time pad is perfect.
In this century spies have often used one-time pads. The only requirement is text (the pad)
of random letters to use for encryption or decryption. (In fact, even now I would not want to
be found in a hostile country with a list of random-looking letters.) The party communicating
with the spy must have exactly the same text of random letters. This method requires the secure
exchange of pad characters: as many such characters as in the original message. In a sense the
pad behaves like the encryption key, except that here the key must be as long as the message.
But such a long key defeats a goal of cryptography: to reduce the secrecy of a long message to
the secrecy of a short key. If storage and transmission costs keep dropping, the one-time pad
might again become an attractive alternative.
Law PAD-1:
The one-time pad is a method of key transmission, not
message transmission. [Bob Blakeley]
During World War II the Germans used an intricate machine known as Enigma for encryp-
tion and decryption. As an important event of the war, British intelligence, with the help of
Alan Turing, the twentieth century’s greatest computer genius, managed to break this code. I
find it sobering to think that if the Germans had not been so confident in the security of their
machine but had used a one-time pad instead, they would have had the irritation of working
with pad characters, keeping track of them, and making sure that each ship and submarine had a
sufficient store of pad, but they would have been able to use a completely unbreakable system.
No one knows what the outcome might have been if the allies had not been able to break this
German code.
10.4 Random Characters For the One-Time Pad.

Later sections will dwell more on random number generation, but for now just note that the

one-time pad requires a truly random sequence of characters. If instead, one used a random
number generator to create the sequence of pad characters, such a generator might depend on a

single 32-bit integer seed for its starting value. Then there would be only different possible
pad sequences and a computer could quickly search through all of them. Thus if a random
number generator is used, it needs to have at least 128 bits of seed, and the seed must not be
derived solely from something like the current date and time. (Using the current time and date
would be terrible, allowing immediate cryptanalysis.)
Exercise: Write a program that will generate duplicate copies of sequences of random charac-
ters, for use in a one-time pad. [Ans: for programs that will generate a one-time pad, see
page 239. For a pair of wheels that makes the one-time pad easy to use, see page 242.]
11
Conventional Block Ciphers
11.1 Conventional Block Ciphers.

The word “conventional” is typically the same as “symmetric key”, or “single key”, or “clas-
sical”, indicating something from ancient history. But this is internet time, so in this case an
ancient theory is anything dated before 1976. After that came the cryptosystems variously de-
scribed as “asymmetric”, or “two-key”, or “public key” (actually mis-labeled, since this latter
is a two-key system where one key is public and the other private). Later sections will describe
asymmetric cryptosystems.
Block ciphers are in contrast to stream ciphers. A stream cipher handles one bit of key and
one bit of plaintext at a time, usually combining them with an exclusive-or to produce one bit
of ciphertext. Since the encryption cannot just depend on a single bit of key, the system must
have state or memory, so that what is done with one bit depends on what was done before.
Block ciphers take a block of plaintext, whose size depends on the cryptosystem, and use
a fixed key of some block length also depending on the cryptosystem, to produce a block of
ciphertext, usually the same length as the block of plaintext. Such encryption is “stand-alone”
and does not depend on what happened before. (Such a block cipher does not have state or
memory.) Having ciphertext block size the same as plaintext block size is important, because
then there is no data expansion with encryption.
>(
The identical block size of plaintext and ciphertext should not be too small or the system
D
would not be practical. In practice a convenient size is chosen, nowadays usually either or
)D >(
bits. The size of the key must be large enough to prevent brute-force attacks. Thus the
E>
ancient (25-year old) Data Encryption Standard (DES) had a -bit key and then bits in
>(
the original proposals. This was finally cut to bits on a transparently false pretext that eight
out of bits should be used for parity. It is now clear that the U.S. National Security Agency
>( E>
(NSA) wanted a key size that they could just barely break using a brute force attack and that no
one else could break. A -bit key requires 256 times as much work to break as does a -bit
D
one, but this is still obviously inadequate in today’s world. The new U.S. Advanced Encryption
Standard (AES) requires at least bits in a key, and this is now regarded as a minimum
( =B %
desirable size. There has been recent talk that this latter size will also soon give way to a brute-
force attack, but this is nonsense — DES is still not trivial to break, and the AES would be at

least or roughly harder, a very large number. (Note that this estimate is just for
@ E>
a brute-force attack — there might be easier ways to break any given system.) For additional
security, the AES also has key sizes of and bits available.
The section describing Claude Shannon’s noisy coding theorem used a random code for
11. Conventional Block Ciphers 71
error correction of data sent over a noisy channel. In cryptography, a random code with no
duplicate code words could be considered for a cryptographic code, but the code table would
need to be unacceptably large, and the decoding algorithm would be difficult to make efficient.
(Here the code table would take the place of the key.) Instead of using random ciphertexts,
which is not practical, one wants to have ciphertexts that appear to be random.
Encryption itself provides a function from each possible plaintext block to a ciphertext
block, with no duplicate values occurring. Similarly decryption gives the inverse mapping,
which would not be uniquely defined if there were duplicates. Encryption in this type of cryp-
tosystem is essentially a parameterized collection of encryption functions, one for each key
value.
The usual characterization of a block cipher assumes no memory or state from block to
block, that is, each plaintext block is always transformed to the same ciphertext block, assuming
the same key. The ciphers with state are stream ciphers. If the blocksize is small and there
is no state, one could just use attacks in the next section to try to determine the ciphertext
corresponding to each possible plaintext block.
Law BLOCKCIPHER-1:
In a block cipher, the blocksize must be large, as least 64
bits, unless the cipher has state.
Thus a cipher with small blocksize needs to actually be a stream cipher. The cipher block
chaining mode described later in this chapter is so important because it converts a block cipher
to a stream cipher.
11.2 Possible Attacks.

The basic method used for encryption should not be secret, but instead all security should
rely on the secrecy of the key. Thus opponents who know everything about the encryption
and decryption algorithms should still not be able to break the system unless they know the
particular key.
The key is now chosen large enough to eliminate brute-force attacks, but just a large key is
not enough to ensure security of a cryptosystem. The opponent may have additional information
about the cryptosystem. There are three simple types of attacks against a cipher (along with
other fancier ones not described here):
✦ Ciphertext only: here the opponent has access only to any number of ciphertexts.
This is the weakest assumption and would always be true. An opponent who cannot
even intercept encrypted messages obviously cannot determine the plaintexts.
✦ Known plaintext: This case assumes that an opponent can obtain plaintext/ciphertext
pairs. As an example, an embassy might first get permission for a press report by
sending the report in encrypted form on one day, and then actually releasing the
report the next.
✦ Chosen plaintext: This scenario assumes the opponents can ask for encryption of
plaintexts of their choosing, and see the resulting ciphertexts. This is the strongest
information normally assumed of opponents: that they essentially have access to an
“encryption box” with the key buried inaccessibly inside.
One always wants a cryptosystem to be secure against a chosen plaintext attack, so of course

the AES appears to resist this attack. Notice that a cryptosystem may often be strong (resistant
> ( = A @

to attacks) and yet not require as much work to break as a brute force attack. For example, the
cryptograms in an earlier section have keys of size or roughly . This is a very
large key space for brute-force search, but in fact a cryptogram is easily broken.
In addition to trying to break the cipher, an opponent can attempt various replay attacks:
retransmitting individual blocks or entire messages made up of many blocks. For example, on
one day an army might receive an encrypted RETREAT message. The next day, when they are
supposed to attack, the encrypted ATTACK message could be intercepted and replaced with the
previous day’s message. The army would decrypt the message and think they should again
retreat.
There are several ways to protect against these replay attacks. Additional data could be
included in each block, such as a sequence number, or the date. However, a better method,
explained in detail in the section on “modes of operation” below, makes each encrypted block
depend on all the previous blocks.
Other attacks include inserting, deleting or rearranging blocks of ciphertext, as well as
ciphertext searching: looking for a block that matches the block in another encrypted message,
without being able to decrypt the block.
11.3 Meet-In-the-Middle Attack on Double Encryption.
E>
From the beginning, critics of the DES’s short key were told that they could use double or triple
>D
DES encryption, thus using two or three -bit DES keys, and getting an effective key length of
%
%%
or bits. For example, double encryption uses two keys and , encrypting first with
the first key, and then encrypting the resulting ciphertext with the second key. A brute-force
attack on all pairs of keys would indeed require steps, but such a double encryption should
not be regarded as nearly as secure as a cryptosystem designed from scratch with a -bit key.
In fact, there is a special attack on such a system, called meet-in-the-middle.
In order to carry out an attack, one needs enough information to recognize when the attack is
successful. One possibility is that the plaintexts might have extra information (such as padding

with bits, or representing ASCII characters) that will allow their recognition. More common is
%
a known plaintext attack: Suppose one has several pairs of corresponding plaintext : ciphertext
from known plaintext information:
%

,
, etc. These correspond to double DES

E>
encryption using two unknown keys in succession: and
%
. The objective is to determine
these unknown -bit keys.
First calculate the ciphertexts
%
for all possible keys . These should be
stored as a (very large) hash table to allow efficient lookup. Then for each possible key ,
7 %
calculate

satisfies: % %

. Look up each
in the hash table. If an entry is found, then it

and

%

, for some keys and . This might represent
a false alarm, so one needs to check these two keys against another plaintext : ciphertext pair
Thus, instead of 3%%

as a second check. On the average, in steps, the desired pair of keys will be found.
steps (which is at present completely unrealistic), one can get by with
at most and
blocks of storage. Of course, these figures are also very large, but perhaps

not so far beyond what is possible. There is also a refinement of this attack called the time-

memory trade-off that uses steps of execution and blocks of storage, where /G
.
(See the Handbook of Applied Cryptography for details.) So even if
blocks of storage is not
possible, one can trade a smaller amount of storage for a larger amount of execution time.
There are clever ways to use block ciphers, as illustrated in the next section, that will
eliminate these meet-in-the-middle attacks.
11.4 Modes of Operation.

The discussion below assumes a fixed conventional (single key) block encryption scheme, such
as the Advanced Encryption Standard discussed in a later section. The methods work for any
such block cipher.
Electronic Codebook (ECB) Mode: The first method of using a block cipher is called The
Electronic Codebook (ECB) Mode . In this method, each block is encrypted independently of
each other block. This method obviously invites the replay of blocks mentioned earlier.

Cipher Block Chaining (CBC) Mode: This uses an Initialization Vector ( ) the size of one
block. The is exclusive-ored with the first message block before encryption to give the first
ciphertext block. Each subsequent message block is exclusive-ored with the previous ciphertext
block. The process is reversed on decryption. Figure 11.1illustrates the CBC mode (adapted
% ==$=
from the Handbook of Applied Cryptography):
In the image above, a sequence of plaintext blocks: , , ,
is being encrypted using

a key and block encryption algorithm . Step of the algorithm uses plaintext , key ,

7%
7%
and theciphertext
produced by the previous step. Step requires a special initialization
vector
&

. As shown, step of decryption uses the inverse decryption algorithm
and the same key , along with the ciphertext block
and the previous ciphertext block
.
This CBC mode has so many pleasant properties that no one should consider using ECB
7%
mode in its place.
Law BLOCKCIPHER-2:
Always use the cipher block chaining (CBC) mode instead
of the electronic code book (ECB) mode.
Properties of the CBC mode:

C0 = IV
Cj
Cj−1
n
Pj E −1 K
K E Cj−1
P
Cj
Encryption Decryption
CBC Mode
Figure 11.1 The Cipher Block Chaining Mode.

✦ What is transmitted: At each stage, one ciphertext block is transmitted. It must
also be arranged that the same secret key and the same initialization vector
are at both ends of the transmission, although the IV could be included in an initial
transmission.
✦ Initialization Vector (IV): This needs to be the same size as the plaintext and cipher-
text. It does not need to be kept secret, but an opponent should not be able to modify
it. For less security, the IV could be all zeros, making the first step of CBC the
same as ECB. For extra security, the IV could be kept secret along with the key, and
then an opponent will not be able to obtain even one pair of plaintext : ciphertext
corresponding to the given key.
✦ CBC converts a block cipher to a stream cipher: The CBC mode is essentially a
stream cipher that handles one block’s worth of bits at a time. The state or memory
in the system is the previous ciphertext.
✦ Each ciphertext block depends on the current plaintext and on all plaintext that
came before: If a single bit of the initialization vector of of the first plaintext block
is changed, then all ciphertext blocks will be randomly altered (50% of bits different
from what they were on the average).
✦ CBC is secure against the various attacks mentioned earlier: This includes all the
ways of fiddling with and searching for encrypted blocks.
✦ Use of CBC as a file checksum: The final ciphertext block depends on the entire
sequence of plaintexts, so it works as a checksum to verify the integrity of all the
plaintexts (that they are unaltered and are complete and in the correct order). As
D
with all checksums, there is a very small probability that the checksum will fail to
detect an error. Using a -bit block, for example, the probability of error is so
small that an advanced civilization could use this method for millennia and expect a
vanishingly small chance of a single undetected error.
✦ Recovery from errors in the transmission of ciphertext blocks: If one or more bits
of ciphertext block
are transmitted in error, the error will affect the recovery of
%
plaintext blocks and . The recovered block will be completely random-
ized (50% errors on the average), while plaintext block
%

will only have errors

in the same places where
has errors. All the remaining plaintexts will come out
free of error. Thus the CBC mode is self-synchronizing in the sense that it recovers
from bit transmission errors with only two recovered blocks affected.
Cipher Feedback (CFB) Mode: This is used for applications requiring that a portion of a
block be transmitted immediately. See the Handbook of Applied Cryptography for details.
76
Part IV
Public Key
Cryptography
78
12
Public Key Distribution Systems
The first ideas of public key cryptography can best be explained using a clever method
for two people to exchange a common secret key using only public communications. The
following example is not intended to be practical, but it illustrates the ideas. In fact, it would
work in practice, but there are better methods.
12.1 Merkle’s Puzzles.

Imagine that person A (“Alice”) wants to establish secure communication with a distant individ-
ual B (“Bob”). Alice and Bob have made no previous arrangements and can only communicate
over a public channel that anyone can listen in on. Another person (“Boris”) listens to all their
communications and has more computing power than they have. Also Alice and Bob have
no special secret methods not known to Boris. However, Boris can only listen in and cannot
change messages. In spite of all this, it is still possible for the two of them to establish se-
cure communication that Boris cannot understand. This surprising technique was developed by
R.C. Merkle before the introduction of true public key cryptography. It helps prepare students
of cryptography for even more surprising ideas.
E D
Alice and Bob must be able to create “puzzles” that are hard but not impossible to solve.
For example, they could agree that all but the low bits of a -bit AES key would be zero
bits. Then they could encrypt two blocks of information with this key. A brute-force breaking

of the “puzzle” would mean trying keys on the average. The encrypted information would
also start with a block of zero bits, so that they can tell when the puzzle is broken. Suppose it
takes Alice and Bob an hour of computer time to break one of these puzzles. It is important to
emphasize that Boris also knows exactly how the puzzles are constructed. Suppose Boris has a
$(( )D
much more powerful computer and can break a puzzle in just one minute.
$((
Bob creates of these puzzles, each containing a -bit random key in one block and

D
a sequence number identifying the puzzle (a number from to in binary with leading
zeros to make it bits long). Bob transmits all of these puzzles to Alice, in random order.
Alice chooses one puzzle at random and breaks it in one hour. She then keeps the random key
and sends the sequence number back to Bob. (Boris listens in to this sequence number, but it
does him no good because he doesn’t know which number goes with which puzzle.) Bob has
saved the content of the puzzles (or has generated them pseudo-randomly) so that he can look
up the sequence number and find the same random key that Alice has. Boris, their opponent,
(( 2 E
must break half of the puzzles on the average to recover the common key that Alice and Bob
E
are using. Even with his much faster machines, it still takes Boris minutes or days
on the average to break Alice and Bob’s system. Thus Alice and Bob get days of secret
80 IV. Public Key Cryptography
E
$((
communications. If they want more time, say days, then Bob just needs to send ten times as
many messages initially, that is, . If the disparity between their computing power and
that of the listeners is even greater, then again Bob must simply send more puzzles initially.
If Boris can actually alter messages (change or inject or delete them) as well as listening in,
then he might pretend to be Alice and communicate with Bob, or vice versa. In this case Alice
and Bob must be able to describe shared experiences or to communicate shared information to
one another to be sure they are not communicating with a stranger (Boris).
If Boris can intercept and change all messages between Alice and Bob, then he could answer
both Alice’s and Bob’s requests to communicate, and pretend to be each to the other. Once
Boris has secure communication established with both Alice and Bob, he could relay messages
back and forth between them, translating between the two cryptosystems. Then even if they
authenticated each other, he would still be listening in. This is called a man-in-the-middle
attack. In this extreme case, the method does not work. This shows the great care that must be
taken with cryptographic protocols. There are more complicated methods relying on a trusted
server that will foil this and other attacks
12.2 Commuting Ciphers.

Alice and Bob still want to communicate securely, without having set it up ahead of time, but
they would like a simpler and more mathematical system. The ground rules are the same as
above: Boris listens in, knows everything they know, and has more computing power.
If they each had their own secret commuting cipher, say Alice had and Bob had ,

then, using a common public integer , Alice could send

to Bob, and Bob could send
to Alice. Then Alice computes and Bob computes . Because
the two ciphers commute, these quantities are the same. The two people can use this common
value or a portion of it to make up a secret key for conventional cryptography. If Boris can’t
break the cipher, he has no way of knowing this secret value.
Actually, in this system, all that is needed is secret commuting one-way functions. Such
functions are discussed in a later section, though not commuting ones.
It is a fact that cryptosystems almost never commute. The next section describes the only
exception I know of.
12.3 Exponentiation and the Discrete Logarithm.

One can imagine a cryptosystem in which a fixed number is raised to a power modulo a

fixed prime , and the power is the plaintext . The result is the ciphertext . (There is a
cryptosystem called the Pohlig-Hellman System similar to this but a bit more complicated.) So

one fixes a (large) prime and an integer that is a generator for multiplication in the field .

(For this generator, see the section on Fermat’s Theorem in the “Favorites” chapter.) From a

message , calculate a ciphertext as described above:

mod
12. Public Key Distribution Systems 81

If the quantities above were ordinary real numbers, and the equation were , then

solving for would give
, the “logarithm base of ”. Because of this notation

from real numbers, one also refers to above as the “discrete logarithm of to base modulo
”. It turns out that there is no known efficient way to calculate discrete logarithms, even

knowing , , and . Using a brute force approach, one could just try all possible values of ,
but there are ways to do better than this, including a “meet-in-the-middle” approach similar to
J4
the attack of the same name on double encryption with block ciphers. There are algorithms that
4
are efficient if has no large prime factors. With the best known approaches, if the prime
is large and random, and if
has a large prime factor, then there is no efficient algorithm
to calculate discrete logarithms.
12.4 The Diffie-Hellman Key Distribution System.

With this system, Alice and Bob work with a large public prime , and a public generator .

mod .
Alice chooses a secret integer , while Bob chooses a secret . Alice sends mod to Bob,

8
mod , and these two quantities are
and Bob sends mod to Alice. Then in secret, Alice computes mod =
Similarly in secret, Bob computes mod =
the same, a common secret value that they can use for further communication. Boris listening
in cannot compute either or , and cannot discover this common secret.
Just as with the Merkle puzzles, if Boris can do more than listen in on the communications,
he could pretend to be Alice to Bob or vice versa, and he could even use a man-in-the-middle
attack to make Alice and Bob think they are communicating in secret when they are not. There
are more elaborate methods involving authenticated public keys that both Alice and Bob have,
and these methods allow them to establish a common secret key even in the presence of an
active opponent like Boris.
In order to set up a practical system as above, first choose a “large enough” random integer
. The size needed depends on how fast the computers have become and on whether there has
been progress in computing discrete logs; at present
should be at least )(
/ /
bits long. Then
test the numbers
,
,

, etc., until the next prime is found. It is not enough just
0/
to have a random prime, but the prime minus one must have a large prime factor. So just find

a larger prime with as a factor. For this purpose, test to see if it is prime. If it
is not a prime, start over again with another random
. Finally, one will have a prime with
4
, for a prime of about half the size of . In order to get a

the property that

generator, choose an at random, or start with . Check this to see if mod or
if mod
. When an doesn’t satisfy either of these, one knows the value is a generator,
and can be used for key distribution. (A random would also probably work, but not with
certainty.)
13
Public Key Cryptography
13.1 Beginnings of Public Key Cryptography.

In 1976 Diffie and Hellman started a revolution with their article introducing the concepts of
public key cryptography. Others discovered the same concepts earlier while doing classified
military applications, but the open publication caught many people’s imagination world-wide.
Instead of using a single key for encryption and decryption, the idea is to have one public
key for encryption (or one public encryption algorithm), a key that anyone can access and use.
This key will normally be available online from a trusted key server, in a way similar to numbers
in a phone book.
The key for decryption (or the algorithm for decryption), will be secret and known only to
the party for whom the message is encrypted. It must not be feasible to guess or deduce or
calculate the decryption key from a knowledge of the encryption key.
Each user has his or her own pair of encryption and decryption keys, distint from one
another.
13.2 Structure of a Public Key Cryptosystem.

Each separate user constructs a public encryption algorithm and a private decryption algo-
rithm . (Usually the algorithms are fixed, and the only information needed are keys supplied
to the algorithms.) Thus Alice has a pair and , and similarly Bob has and .
Each matching pair and of algorithms has the properties:
1. Encryption followed by decryption works:
, for any plaintext for
which the algorithm is defined.
2. Can encrypt efficiently: The algorithm can be calculated “efficiently”.
3. Can decrypt efficiently: The algorithm can be calculated “efficiently”.
4. Public and private keys stay that way: For an opponent (“Boris”) who knows , it
is still an “intractible” computation to discover .
The RSA public key cryptosystem also has the unusual and useful property that decryption
works the same as encryption.

5. Signing followed by verifying works: The set of messages is the same as the
set of ciphertexts
, for all , so that the decryption algorithm can be
13. Public Key Cryptography 83

applied to a message, resulting in what is called a signed message or a signature. If
,
that is,

is the signature corresponding to some plaintext , then
, for any message .
The word “efficient” above means that that calculation uses an acceptable amount of re-
sources, while “intractible” means roughly that this computation uses more resources than the
secrecy of the message is worth. (This varies with time, as computations get cheaper.)
Alice makes her public algorithm publicly available, by publishing it or by putting it online,
and so does every other user. In practice a public key server in needed to supply authenticated
public keys to users. Alice keeps her secret algorithm completely to herself, just as every other
user does. (No one else must know the secret algorithm.)
13.3 Modes of Operation for Public Key Cryptosystems.

Here the term “key” is used to mean the same as “algorithm” as explained above. Suppose
as above that Alice and Bob have public key and private key pairs:
1
1
for Alice and
for Bob. There must be some key distribution authority that keeps the public keys
and and makes them available to anyone requesting them. (Using signature techniques
discussed later, the authority can convince a requester that he or she is getting the correct keys.)
Bob sends a secret message to Alice: Bob gets Alice’s public key from the au-

thority. Bob calculates

key, so only Alice can compute

and sends to Alice. Only Alice knows the decryption
. Unless there is some failure in

the system, Alice knows that no one intercepting the transmission of will be able to recover

the message . However, anyone might have sent her this message, and even Bob might have
leaked the message to others. Notice that this mode does not require the special property 5 of
RSA, but is available for any public key cryptosystem.
Bob signs a message and sends it to Alice: This mode uses the special property 5.

above of the RSA cryptosystem. Other systems usually don’t have property 5., but it is still
possible to create digital signatures in a more complicated way as a later chapter will discuss.
Bob uses his secret decryption key on the message itself (rather than on ciphertext). For RSA,
the collection of all possible messages is the same as the collection of all possible ciphertexts:

any integer less than the fixed integer
used for that instance of RSA. Thus Bob calculates

. At the other end, Alice can retrieve Bob’s public key from the key authority
and use it to recover the message, by calculating . Anyone can
fetch Bob’s public key and do this calculation, so there is no secrecy here, but assuming the
system does not break down (that the key authority works and that the cryptosystem is not
leaked or stolen or broken), only Bob can have signed this message, so it must have originated
with him.
Bob can use this same method to broadcast a message intended for everyone, and anyone
can verify using Bob’s public key that the message can only have originated with him.
Bob signs a secret message and sends it to Alice: This can be done in two ways,
with Bob using his secret key and Alice’s public key . (Once again, this assumes RSA

with property 5 is used.) Calculate either , or . In either
case, Alice reverses the process, using her secret key and Bob’s public key . There is
one other slight problem with the RSA cryptosystem, because the maximum size of plaintext
or ciphertext is less than an integer
for Alice, and is less than a different integer
for Bob.
then

What happens then depends on which of these two integers is larger, for if
is the larger,
might be too large to be handled by using a single block, so one would have

the awkward business of breaking this into two blocks. However, in this case one can calculate
first, and this is definitely less than the block size of the key . In case the sizes are
reversed, just do the two steps above in the opposite order also, so there is no need for more
than one block even for a signed and secret message.
Notice that Alice knows the message must have originated with Bob, and that no one else
can read it, giving both authentication and secrecy.
Bob uses a hash function to sign an arbitrarily long message using only one block:
Here one just signs the hash code of the message. These matters will be discussed more thor-
oughly in the chapter on hash functions.
13.4 The Integer Knapsack Cryptosystem.

Diffie and Hellman proposed a specific public key cryptosystem in their 1976 paper: one using
an interesting problem known as the Integer Knapsack Problem. This particular approach has
been broken in most forms. This chapter includes it because it is a good illustration of how
public key cryptography works.
The knapsack problem is an example of an NP-complete problem — a class of problems
that are usable for various purposes, but thought to be intractible. A later chapter describes this
theory in much more detail. The knapsack problem occurs as a decision problem, that is, one
with just a Yes-No answer, and as an optimization problem, one with a specific answer.
Integer Knapsack Decision Problem: Given

positive integers 1
1$=$=$= 1

and a positive integer , is there a subset of the that adds up exactly
to ? (Yes or No answer.)
The sum has to add up to exactly — being close counts for nothing. The optimization
problem asks just which numbers must be added. Is is an interesting exercise to see that if one
had an efficient algorithm to answer the decision problem, then using this algorithm a relatively
small number of times, one could get another algorithm that would also answer the optimization
problem.
There are many different kinds of instances of the knapsack problem. Some of them are
easy to solve and some are hard. The trick with using this problem for public key cryptography
is to make breaking the problem look like a hard instance, but it is really a disguised hard
instance. If you know the trick, the disguise, you can transform it into an easy instance, and
use the secret key to decrypt.
13. Public Key Cryptography 85

One type of easy instance is one with relatively small numbers for the numbers . Then
standard dynamic programming algorithms will solve the problem. Another easy instance, the

one we will use here, is for the to form a superincreasing sequence, meaning that each value
is larger than the sum of all the values that came before.
14
The RSA Public Key Cryptosystem
14.1 History of the RSA Cryptosystem.

The history of RSA is still fascinating to me because I watched it unfold. In 1976, as discussed
in the previous chapter, Diffie and Hellman introduced the idea of a public key cryptosystem.
(Actually, the concept had been discovered earlier in classified work by British and American
military researchers, but no one knew this at the time.) Then a 1977 Scientific American article
by Martin Gardener talked about a new public key implementation by MIT researchers Rivest,
Shamir, and Adelman. This article caught my attention (along with many others), but did not
contain the details needed to fully understand the system. A year later the details were finally
published and the revolution in cryptography was in full motion. After more than twenty years
of research, RSA remains secure and has become the most popular public key cryptosystem.
Law RSA-1:
The RSA cryptosystem is the de facto world-wide standard
for public key encryption.
14.2 Description of the RSA Cryptosystem.

1
The RSA system is an asymmetric public key cryptosystem in the terms of the previous chapter.
Recall that this means that there are any number of pairs of algorithms both defined on
the same set of values. is the public encryption algorithm and is the private decryption
algorithm. These satisfy:

is the ciphertext corresponding

1. Encryption followed by decryption works: If

to some plaintext , then . (In other words: , for any message
.)

2. Can encrypt efficiently: For any message , there is an efficient algorithm to calculate
.

3. Can decrypt efficiently: For any message or ciphertext , there is an efficient algorithm
to calculate .
14. The RSA Public Key Cryptosystem 87
4. Public and private keys stay that way: From a knowledge of , there is no efficient way
to discover .

5. Signing followed by verifying works: The set of messages is the same as the set of
ciphertexts
, for all , so that the decryption algorithm can be applied to

a message, resulting in what is called a signed message or a signature. If
. (In other words:

is the signature corresponding to some plaintext , then
, for any message .)
As mentioned earlier, RSA is unique in having property 5, which makes signatures using it
particularly easy.
Users , ,

==$=
can create their own pairs ,
1
, of RSA key pairs. The 1 =$=$=
encryption algorithms are “published” or made available on a secure public key server, while
the decryption algorithms are kept secret from everyone except the originator. The previous
chapter has gone over how these can be used.
In RSA, the plaintexts and ciphertexts are just large positive integers, up to a certain size
depending on the specific key pair. The underlying algorithms are not secret but only certain
information used in them. The RSA system itself is constructed as follows:
Algorithm: RSA cryptosystem construction.
1. Choose random “large” prime integers and of roughly the same size, but
not too close together.
2. Calculate the product
(ordinary integer multiplication).
: less than
that has no factors in
4 4
3. Choose a random encryption exponent
common with either or
.
4. Calculate the (unique) decryption exponent satisfying
4 =
: 4
; mod
, for any message .
mod

5. The encryption function is
mod
, for any ciphertext .
6. The decryption function is

7. The public key (published) is the pair of integers
1: .

8. The private key (kept secret) is the triple of integers 1 <1 .
There is more to the story about each of the above items:
E
)(
1. At present, “large” means at least bits. For better security each prime should be at
least bits long. There are efficient algorithms for generating random numbers of a
given size that are almost certainly prime (see below).
2.
is then either at least )( or (D bits long.
: A
4 4 A
3. The encryption exponent can be just . If one is using this exponent, then the primes
must be such that
and
are not divisible by .
4. The decryption exponent must be calculated, and there are efficient algorithms to do this,
modulus for division,

L4 04
but they require a knowledge of and (see the chapter on favorite algorithms). The
, is the Euler phi function
of
, where this
is a function studied in number theory. One of the function’s properties is important in
proving that RSA works.
5. There are efficient algorithms for carrying out the modular exponentiation needed here
(see below).
6. The same efficient algorithm works here.
A
7. If it is known that is the encryption exponent, then only
needs to be published.

8. Only needs to be kept as the secret data for decryption (along with the public
and
:
). However, and can be efficiently calculated from the other numbers, and they are
needed anyway for the most efficient form of modular exponentiation. (See the RSA
implementation using the Chinese remainder theorem below.)
)( D
Some people are surprised that RSA just deals with large integers. So how does it represent
data? Suppose the value of
is at least
D
bits long. This is the same as bytes. In
principle then, one can just run bytes of Ascii text together and regard the whole as a single
RSA plaintext (a single large integer) to be encrypted or signed. In practice, the protocols will
demand additional data besides just the raw message, such as a timestamp, but there is room
for a lot of data in a single RSA block.
14.3 RSA Works: Decryption is the Inverse of Encryption.
To show that RSA decryption reverses what RSA encryption does, one only needs to show that:
H1

;
for any message , or specifically to show that
mod
H=
But recall that
: mod 4 4; , so that
; mod
mod

; mod

7 % 7 % mod
% mod
H=
The last line follows from the chapter on favorite algorithms which shows that the exponent
can be reduced modulo
L

4 4 =
14.4 Java Implementation of the Basic RSA System.

RSA uses arithmetic on integers at least )(
bits long. RSA has been implemented many times
in hardware, but if it is only used for key exchange, a software implementation is fast enough.
Any such implementation must start with routines to do extended precision arithmetic on the
large integers. Writing such routines is perhaps a reasonable project for an undergraduate CS
major as part of one course, with division causing the most grief. (Twenty years ago, I laid
out one weekend for such a project, but I ended up devoting more than a week to it.) Many
implementations are available, including the Java BigInteger class, and implementations in
symbolic algebra packages such as Maple or Mathematica.
This Java implementation of the RSA cryptosystem uses the Java BigInteger library
class. This arbitrary precision integer arithmetic class has all the methods one needs to im-
plement RSA without difficulty. In fact, it seems as if a number of specialized methods were
included just to make RSA implementation easy.
Here are additional comments about this particular implementation:
✦ Key generation:
✧ Using public keys of size )(

bits, it took about 15-60 seconds to gener-
ate two sets of keys on a Sun Ultra 10.
✧ The key generation has no unusual features. Primes and are chosen at

random, differing in length by 10-20 bits. (If the primes are too close to

, then factoring might be easier than it should be.) The primes are
4
and 4 A
A
also chosen so that do not have as a factor, because this
implementation uses as the encryption exponent.
✧ The only weak point with this key generation that I know of is with the
random number generation. For a good implementation, one would need
a special generator, with more bits for the seed. (The current generator
just uses the number of milliseconds since 1 Jan 1970, and that is clearly
insecure.)
A
✦ Encryption and Verification:
This uses an exponent of . The main known weakness here is that the message
must be bigger than the cube root of
, since otherwise the ciphertext will be ,
without any modular division. Smaller messages must be padded to make them long
enough.
✦ Decryption and Signing:

This can be sped up using the Chinese Remainder Theorem, as is shown in the next
subsection.
✦ Combination of Signing and Encrypting:
This common combination, used to keep the message secret and to authenticate its
source, is done in a simple way that checks the lengths of the public
values first,
using the longer one before the shorter one. Otherwise one might need to use two
RSA blocks in some cases.
)(
✦ The Test:
There is just a simple test of this software, though bits is a realistic size.
For the implementation code and the simple test see page 246.
14.5 Faster RSA Using the Chinese Remainder Theorem.
Here is an altered implementation of the RSA cryptosystem, using the the Chinese Remain-
der Theorem (CRT) to speed up decryption. Please refer first to the comments in the earlier
subsection and to other material about the RSA cryptosystem.
✦ Algorithm.
The algorithm presented here is described in items 14.71 and 14.75 in the Handbook
of Applied Cryptography, by Menezes, van Oorschot and Vanstone, CRC Press,

1996. If is ciphertext, then RSA decryption calculates
mod
, where
3=
Suppose one calculates
%

mod , and
mod

instead. The Chinese Remainder Theorem (and associated algorithm) allows one to
deduce mod from a knowledge of mod and mod .

Arithmetic mod should be done mod in an exponent, because 4

7 % mod (Fermat’s theorem).
Thus we can use the simpler calculation:


%
mod 7 % mod ,and
mod 7 % mod 3=
Finally, following the algorithm 14.71 referred to above, calculate
7 % mod , and
4
mod 3=

%
The final answer is:
mod

%/ =
(In calculating in my implementation, I had to check for a result less than , and I
had to add to the result in that case.)
✦ Security.
The CRT version of decryption requires the primes and , as well as the decryption

exponent , so this might seem to be an extra source of insecurity. However, it is
simple to factor the modulus
given the decryption exponent , so no security is
lost in using this method.
)(
✦ Performance.
Theory predicts that the CRT decryption should be 4 times as fast. I tried 600 -
8= 6EB
bit decryptions using a Sun Ultra 10 workstation. The average decryption time for
8= (>
the normal method was about seconds per decryption. With the CRT method
A8=I(
here, the average decryption time was about seconds per decryption, giving
a speedup by a factor of about . The more complicated algorithm has various
(
sources of extra overhead, so it is not surprising that the full speedup by a factor of
is not achieved.
A
✦ Summary.
If one uses an encryption and verifying exponent of as I am with this software,
A8=I(
then these operations are quite fast compared with decryption and signing (at least
100 times faster). A speedup by a factor of for decryption and signing is signif-
icant. The extra algorithmic complexity is minimal, so no one would want an RSA
algorithm without this speedup factor.
Law RSA-2:
RSA encryption should use exponent 3, making it hun-
dreds of time faster, and RSA decryption should use the Chinese
Remainder Theorem, making it four times as fast.
A
One does have to be careful with exponent in two ways: if the message is less than the cube

root of , then the encrypted message will be the same as the message, and if someone obtains
ciphertext for a message encrypted under several different public keys, it may be possible to
calculate the message.
The implementation code can be found on page 251.
>)(
Exercise: Write a “toy” implementation of RSA in the Java language, using the long type
( -bit integers) for the calculations. This should be a working implementation in every respect
except that the integers cannot be very large.
15
Rabin’s Version of RSA
15.1 Rabin’s Public Key Cryptosystem.

Michael Rabin discovered what I like to call a version of RSA, although it is more properly
regarded as a public key cryptosystem in its own right. During its early history, this system
was considered of theoretical, but not practical interest because of a “fatal flaw” (a quote from
Donald Knuth) that made it vulnerable to a chosen plaintext attack. However, there are ways
around the flaw, making this system a real competitor to RSA.
Law RABIN-1:
Rabin’s cryptosystem is a good alternative to the RSA
cryptosystem, though both depend on the difficulty of factoring
for their security.
15.2 Discrete Square Roots.

In the integers modulo
, using both addition and multiplication modulo
, if
in not a prime,
then not every non-zero element has a multiplicative inverse. But also of interest here are

elements that have a square root. The square root of an element is an element such that

. Some elements have several square roots, and some have none. In fact, number
theorists have been interested in these matters for hundreds of years; they even have a special
term for a number that has a square root: a quadratic residue. Thus this theory is not something
new invented just for cryptography.

In elementary algebra, one learns that positive numbers have two square roots: one positive
(
A @ ( E E A > A B E D @ @
and one irrational. In the same way, for the integers modulo a prime, non-zero
numbers that
are squares each have two square roots. For example, if
, , , , , , ( , and
then in
%% , ,
. Table 15.1 shows
,
( @ A
those numbers that have square roots:

A E > B D
Notice that , , and have their “ordinary” square roots of , , and , as well as an extra

square root in each case, while and each also have two square roots, and , , , , and
Rabin’s system uses

each have no square roots at all.

, where and are primes, just as with the RSA cryptosystem.
It turns out that the formulas are particularly simple in case ( A and ( A
(which
Numbers mod 11
Square Square Roots
1 1, 10
3 5, 6
4 2, 9
5 4, 7
9 3, 8
Table 15.1 Square Roots Modulo a Prime.
Numbers mod 21 = 3*7

Square Square Roots
1 1, 8, 13, 20
4 2, 5, 16, 19
7 7, 14
9 3, 18
15 6, 15
16 4, 10, 11, 17
18 9, 12
Table 15.2 Square Roots Modulo a Product of Two Primes.
A B
is true for every other prime on the average), so the rest of this chapter makes that assumption
about the primes used. The simplest such case has and . In this case Table 15.2
gives the square roots.
Here the “normal” situation is for a square to have four different square roots. However,

certain squares and square roots have either or as a divisor. In this case, each square has
two square roots (shown in bold italic above). Of course, all the numbers not appearing in the
B
left column don’t have a square root. A program that creates the above table appears on
page 256. This same section gives a table for and , again satisfying the special
/ 2 K4 /

Rabin property. In these tables, it looks as if there are a lot of bold italic entries, but in fact

such squares with or as a factor, while there are
/ / 4 A 2)(
there are
squares altogether. An actual Rabin instance will use very large
primes, so that only a vanishingly small number of them have the divisibility property, and the
chances of this happening at random can be ignored.
15.3 Rabin’s Cryptosystem.

A (
Each user chooses two primes and each equal to modulo , and forms the product
.
15. Rabin’s Version of RSA 95
Public key: the number

.

Private key: the numbers
Encryption: to encrypt a message
and .
, form the ciphertext
mod
.

%
Decryption: given ciphertext , use the formulas below to calculate the four square roots
modulo
of :
, and . One of the four is the original message , a second
J4
, ,
square root is
, and the other two roots are negatives of one another, but otherwise
random-looking. Somehow one needs to determine the original message from the other
three roots (see below).
In the special case in which both primes when divided by give remainder , there are ( A
simple formulas for the four roots:
Formulas for the four square roots of a square . Calculate in order:
/ , (extended GCD algorithm).
%
and , satisfying
$ mod .

$ mod .

%
/ mod
.
4 mod
.
Now the four square roots are

, , , and % . 4
. In case and
hence have or as a divisor, the formulas will only yield two square roots, each also with
4

or as a factor. For the large primes used in an instance of Rabin, there is a vanishingly small
chance of this happening. (Picture the chances that a -bit random prime number happens to E
divide evenly into a message!)
15.4 Cryptanalysis: the Complexity of Rabin’s Cryptosys-

tem.
The complexity of Rabin’s system (the difficulty of breaking it) is exactly equivalent to factoring
the number
. Suppose one has a Rabin encryption/decryption machine that hides the two
primes inside it. If one can factor
, then the system is broken immediately, since the above

formulas allow the roots to be calculated. Thus in this case one could construct the Rabin

machine. On the other hand, if one has access to a Rabin machine, then take any message ,

4
calculate , and submit to the Rabin machine. If the machine returns all four roots,

then and give no additional information, but either of the other two roots minus will

have one of or as a factor. (Take the greatest common divisor of it with
.)
The same proof that breaking Rabin is equivalent to factoring
provides what has been
called a “fatal flaw” in Rabin’s system. The above argument is just a chosen ciphertext attack.
It is not wise to allow an opponent to mount such an attack, but one would also not want a
cryptosystem vulnerable to the attack, which is the case with Rabin’s system. (However, see
the next section.)
15.5 Redundancy in the Message.

In order to distinguish the true message from the other three square roots returned, it is necessary
to put redundant information into the message, so that it can be identified except for an event
> ( >(
of vanishingly small probability. The Handbook of Applied Cryptography suggests replicating

the last bits of any message. Or one could use s as the last bits. In these or similar cases,
!7 >(
the Rabin machine would be programmed to return only messages with the proper redundancy,
and if is not a small enough margin of error, then just choose more than redundant
bits. Then the attack described above doesn’t work any more because the Rabin machine will
only return a decrypted message (a square root) with the proper redundancy. Thus the Rabin
machine returns at most one square root, and possibly none if someone is trying to cheat. (The
probability of having two square roots with the given redundancy is again vanishingly small.)
Breaking the new system is longer formally equivalent to factoring
, but it is hard to imagine
any cryptanalysis that wouldn’t also factor
.
Hugh Williams gave another variation of Rabin’s cryptosystem that avoids the “fatal flaw”
in a mathematically more elegant way.
15.6 A Simple Example.
E ) (
Here is an example with tiny values for the primes. Of course a real example would use primes
in the range from

, B
to
, and
BB 4 A
bits long, just as in the case of RSA.
B/ 4A
A >
Take . Then , so that and .
>A B BB
Suppose one uses -bit messages whose bits are then replicated to give bits, up to the

E %'& (!E %'&
number . Messages must be in the range from to , so this system of redundancy will work.
A A B ( 4 A B 4 A ( BB (!E

Start with data bits or . The replication gives or . Then mod

4 A B / ( BB >B
. Continuing the calculations, mod , and mod . Finally,
4 BB
mod and mod .
4 BB A >B (!E

These are two of the four square roots, and the remaining two are mod and
A (E
mod . In binary, the four square roots are , ,
, and . Only has the required redundancy, so this is the only
number that this modified Rabin machine will return.
Part V
Random Number
Generation
98
16
Random Number Generation
16.1 True Random Versus Pseudo-random Numbers.

Random numbers are very widely used in simulations, in statistical experiments, in the Monte
Carlo methods of numerical analysis, in other randomized algorithms, and especially in cryp-
tography. The connection with cryptography is very close, since any pseudo-random bit stream
along with exclusive-or provides a cryptosystem (though not necessarily a strong system), and
any good ciphertext should look like a pseudo-random bit stream (perhaps occurring in blocks).
This section focuses on random number generators used in simulation and numerical analysis,
but for use in cryptography the recommended random number generators are derived from
cryptosystems, both conventional and public key.
Law RNG-1:
Good ciphertext has the appearance of a true-random bit
stream.
From the beginning (where “beginning” is the 1940s, the start of the computer age) there
was interest in so-called “true” random numbers, that is, numbers generated by a random pro-
cess in the world. Physical events such as the radioactive decay of particles are unpredictable
except for their behavior averaged over time, and so could be used as a source of random num-
bers, but these events have been difficult to utilize and have been disappointing in practice.
More promising recently are possibilities from quantum theory, but such matters are outside
the scope of this discussion.
By far the most common source of random numbers is some deterministic process, such
as a software algorithm. These provide “random-looking” numbers, but the numbers are not
really random — there is always an exact algorithm for specifying them. This is the reason that
researchers now describe such numbers using the word “pseudo”, which means “false”. These
are not true random numbers, but for most applications they can be just as useful. Sometimes
they can be more useful, as for example when one wants to repeat a simulation with exactly the
same random or pseudo-random numbers.
Law RNG-2:
Anyone who uses software to produce random numbers is
in a “state of sin”. [John von Neumann]
100 V. Random Number Generation
At first one might think that the best way to get random-looking numbers is to use a “ran-
dom” algorithm – one that does crazy operations, everything imaginable, in every possible or-
der. Donald Knuth tried out such an algorithm as an example, and showed that its performance
was no good at all. In its first run, Knuth’s “random” algorithm almost immediately converged
to a fixed point. Knuth was arguing that one should use science and great care in generating
pseudo-random numbers.
Law RNG-3:
One should not use a random method to generate random
numbers. [Donald Knuth]
An early suggested source of pseudo-random numbers was an equation which was much
later to become a part of modern “chaos” theory. The next chapter describes a generator derived
from this equation.
Another early idea for a source of random numbers was to use the bits or digits in the
expansion of a transcendental number such as , the ratio of the circumference of a circle to its
diameter.
3.14159 26535 89793 23846 26433 83279 50288 41972 ... (decimal)
3.11037 55242 10264 30215 14230 63050 56006 70163 21122 ... (octal)
It has long been conjectured that this is a very good source of pseudo-random numbers, a
conjecture that has still not been proved. In 1852 an English mathematician named William
B
Shanks published 527 digits of , and then in 1873 another 180 digits for a total of 707. These
numbers were studied statistically, and an interesting excess of the number was observed in
the last 180 digits. In 1945 von Neumann wanted to study statistical properties of the sequence
of digits and used one of the early computers to calculate several thousand. Fortunately for
Shanks his triumph was not spoiled during his lifetime, but his last 180 digits were in error
and his last 20 years of effort were wasted. Also there was no “excess of 7s”. The number
has now been calculated to many billions of places, but the calculation of its digits or bits is
too hard to provide a good source of random numbers. The later digits are harder to calculate
than earlier ones, although a recent clever algorithm allows calculation of the
th binary (or
hexadecimal) digit without calculating the previous ones.
Later work focused on a particularly simple approach using a congruence equation, as
described below.
16.2 Linear Congruence Generators.

This approach uses a linear congruence equation of the form:
%

/ mod

&
where all terms are integers, is the multiplier, (usually taken to be ) is the increment, and
is the modulus. An initial seed is . Each successive term is transformed into the
16. Random Number Generation 101
m k
% (Beware! RANDU)
%
%
%
%

Table 16.1 Parameters for Linear Congruence Generators.
4
next, so that a function to return random numbers has the unusual property of automatically
cycling itself to the next number. The pseudo-random terms are in the range from to .

To get (more-or-less) uniformly distributed floating point numbers between and , just do
a floating point division
by . Assuming that , the quality of the numbers produced

4
depends heavily on and .
This type of generator can produce at most different numbers before it starts to repeat.

4
To get this behavior, one can start with a prime number for and use a generator for so that

>
all numbers will be produced in a repeating cycle, starting with whatever the seed is.
<% > E )AB

The old generator provided by the C standard library used -bit integers, and so had a
maximum possible cycle length of — a ridiculously small cycle, making the
A B>B
generator useless except for toy applications. The C Standard Library still allows this function

A
to return numbers in the range from to , although a larger range is now also possible.
<% / A >EEA@ %
When -bit machines first became popular in the 1960s, the multiplier for RANDU, the
most common generator at that time, was , taken modulo . This multiplier
%4
gave extremely poor performance and was eventually replaced by better ones. The most com-

B >D B
mon replacements used the fact that is a prime and searched for a good muliplier. The
multiplier frequently used (starting in 1969) was
%4 $(B)(DA>)(>
and the constant was taken to
be zero. This generator is quite efficient, and has a cycle length of . The
multiplier was chosen so that various statistical properties of the sequence would be similar to
the results for a true random sequence. In the 1970s when I first started using this sequence the
cycle length seemed quite long – now it seems short since I have frequently run experiments

with hundreds of billions of trials.
%4 >D B
Knuth in his chapter on conventional random number generators approves the values
and above as “adequate”, but he has suggestions for better values, such as
those given in Table 16.1 (except for RANDU in the first line).
Knuth suggests other more complicated generators, including one that combines the first
two table entries above:
( D B 7 % mod % 4 1

(
> @ 7 % mod % 4 "(@ 1

4C mod % 4 1
where independent seeds are needed for and , and the sequence of the

& & make up the
output random numbers. The period is nearly the square of the component generators. Knuth
also recommends:
7 % 4CA $( 6E@ >@ 7 mod % 4 1
B $D D DA
which has very good performance and whose period is the square of . Of course two inde-
pendent seeds and are needed to start the sequence off with .
& %
16.3 Other Distributions.
Random numbers other than the uniform distribution are sometimes needed. The two most
common of these are the normal distribution and the exponential distribution. The easiest
way to generate numbers according to these distributions is to transform uniformly distributed
numbers to the given distribution.
The formula for the exponential distribution is especially simple:
4 ;

If is uniformly distributed on the interval from 0 to 1, then will be exponentially distributed
with mean 1. For random events which occur once every unit of time on the average, the times
between such events will be distributed according to this exponential distribution.
Similarly, there is a more complicated formula giving normally distributed numbers:
% 4 ; %

4 ; %
If
%
and are independent and uniformly distributed on the interval from 0 to 1, then %
and will be independent and normally distributed with mean 0 and variance 1. Note that a
generator based on these formulas will produce normal random numbers two-at-a-time.
There are other equivalent transformations to the normal distribution that are more effi-
cient (see Donald Knuth, Seminumerical Algorithms, 3rd Ed., pages 122–132), but the above
formulas should serve all but the most demanding needs. In fact Java has a library method
returning normally distributed numbers (nextGaussian()), and this method uses one of the
more efficient transformations than the equations given here
16.4 Commentary.
Knuth has other suggestions for efficient random number generators of high quality, where
“quality” is measured by a variety of statistical tests that compare the output of the pseudo-
random generator with true random numbers. If for a given test the comparison says the two sets
of numbers look similar, then one says the generator “passes” this particular test. A generator
that passes all the popular tests that people can devise is of high quality.
However, even generators of high quality are mostly not usable in cryptography. For ex-
ample, given several successive numbers of a linear congruence generator, it is possible to
compute the modulus and the multiplier with reasonable efficiency. One could make the gener-
ator more complex in order to resist this attack, but there would still be no proof or assurance
of the difficulty of “reverse engineering” the generator. Instead, if generators are needed in
cryptographic applications, one is usually created using a conventional cipher such as the Ad-
vanced Encryption Standard or using a public key cipher such as RSA or one of its variants.
The AES-based generator will be efficient and will satisfy most practical requirements, but the
RSA-based systems, while extremely slow compared to the others, have a very strong property
of being cryptographically secure, a term that means the generator will pass all possible effi-
cient statistical tests. These matters will be defined and discussed in the chapter after the next
one.
16.5 Java Implementations of these Generators.

Each of the generators in the previous table is implemented with the Java program on page
259. For simplicity this program uses the Java BigInteger class for all the implementations.
In the resulting code one does not need to worry about overflow, but the generators run very
much slower than if they were carefully tuned to the available hardware. However even the
inefficient implementation will generate millions of random numbers in just a few minutes on
current personal computers, and this will be fast enough for most applications.
In case faster generators are needed, various sources show how to use the 32-bit hardware
units directly. Knuth also presents a practical additive generator implemented in C that is very
fast. (See Donald Knuth, Seminumerical Algorithms, 3rd Ed., page 286.)
For a Java implementation of Knuth’s two-seed generator of the previous section, along
with transformations to the other distributions, see page 262, which gives an applet displaying
the three distributions.
The implementation above uses Java long type to avoid integer overflow. On conven-
tional machines without 64-bit integers (for example, if programming in C or C++), even the
implementation of a simple generator such as the very common
%

>D B mod % 4
poses problems because the multiply step overflows a 32-bit integer. This generator was usually
coded in assembly languages on IBM 360++ machines, where the ready access to all 64 bits of
a product makes implementation easy. On machines with only 32-bit integers, one can break
the integers into pieces during the calculations.
Java/C/C++ function: rand2
// rand2: version using ints. Works on all hardware, by
// breaking up numbers to avoid overflow.
int seed2 = 11111;
double rand2() {
int a = 16807,
m = 2147483647,
q = 127773,
r = 2836;
int hi, lo, test;
hi = seed2/q;
lo = seed2 - q*hi;
test = a*lo - r*hi;
if (test > 0) seed2 = test;
else seed2 = test + m;
return((double)seed2/(double)m);
}
This function, in exactly the form given above, works in Java, C, and C++.
In another approach, one can use the double type, which includes exact 52-bit integer
arithmetic as a special case. If the multiplier is small enough to not overflow a 52-bit integer,
then everything can be done using doubles. (In C the operator % does not work for doubles,
while it does in Java.) Here is the C versions of this function. (For this to work, you may need
to include a special library such as math.h to get the function floor.h.)
C/C++ function: rand1
// rand1: version using doubles. Works on all hardware.
double seed1 = 11111.0;
double rand1() {
double a = 16807.0,
m = 2147483647.0;
double q;
seed1 = a*seed1;
q = floor(seed1/m);
seed1 = seed1 - q*m;
return(seed1/m);
}
This particular generator once represented the minimum standard for a random number gener-
ator. I suggest that one now ought to use Knuth’s double generator as the minimum standard,
shown here in C:
Java function: rand1
// rand: version using doubles. Works on all hardware.
// seed1 = 48271*seed1 mod 2ˆ31 - 1
// seed2 = 40691*seed1 mod 2ˆ31 - 249
// seed = (seed1 - seed2) mod 2ˆ31 -1
double seed;
double rand() {
double a1 = 48271.0, a2 = 40692.0,
m = 2147483647.0, m2 = 2147483399;
double q1, q2;
double q, diff;
seed1 = a1*seed1; seed2 = a2*seed2;
q1 = floor(seed1/m); q2 = floor(seed2/m2);
seed1 = seed1 - q1*m; seed2 = seed2 - q2*m2;
// now combine results
if ((diff = seed1 - seed2) < 0.0) diff = diff + m;
q = floor(diff/m);
seed = diff - q*m; return(seed/m);
}
To convert this to Java, one just needs to write Math.floor in place of floor. In the past
such a generator might be slow because of all the floating point operations, including 4 floating
point divides, but now extremely fast floating point hardware is commonplace.
Exercise: Use BigInteger to implement Knuth’s two more complex generators described
above.
17
Random Numbers from Chaos Theory
Many interesting results have come from the field known as “chaos theory”, but I know of
only two published methods for obtaining pseudo-random numbers from this theory. Wolfram
proposed a discrete method based on cellular automata, and I made my own proposal based on
a common equation from chaos theory called the “logistic equation”. This method is unusual
because it works directly with floating point numbers, producing uniformly distributed values.
17.1 The Logistic equation.

The logistic equation is the iterative equation
( 4 1 , , =
It is historically interesting as an early proposed source of pseudo-random numbers. Ulam
and von Neumann suggested its use in 1947. The equation was mentioned again in 1949 by
von Neumann, and much later in 1969 by Knuth, but it was never used for random number
generation.
&
1 <1$=$==
To use the equation to produce random numbers, start with a real number
7 %
and define
, for
applying the function . The resulting sequence & 1 % 1 1 1=$=$=

. In this way one gets a sequence of numbers by just repeatedly
will all lie between and ,
but they will not be uniformly distributed in the interval. However, the numbers have a precise
algebraic distribution that one can transform to the uniform distribution as shown below.
Another function closely related to the logistic equation is the “tent” function defined by
4 / 3 1, 6 , 2 ,

, 1
1

2
In contrast with , this function does directly yield numbers that are uniformly distributed. The
function can be transformed to using the equation

2 =
The inverse transformation (following equation) will transform numbers produced by to
those produced by , in other words, to the uniform distribution.
7% 2
=
Thus the sequence
17. Random Numbers from Chaos Theory 107
7 % , where

7 % , for
81 1 <1A<1$=$=$=

numbers.
gives uniformly distributed
Using the notation , , and so forth, the sequence
takes the form
7 % , for
81 1 31A81$=$==
This sequence will be uniformly distributed for “almost all” starting values .
17.2 Behavior in Infinite Precision.
The above equations give uniformly distributed sequences of real numbers if one could use
7 %
precision” real numbers, that is, mathematical reals. Even in this case,
what are called “infinite
the sequence
in the sequence

does not at all behave like a true random sequence. A cycle occurs
in case it repeats after finitely many iterations. There are infinitely
A2)(
many finite cycles, even though “almost all” real numbers will not belong to such a cycle. For
example, is transformed to itself by . Starting with a random initial value, one would
A2)(
avoid a finite cycle with probability , but even the influence of the short cycles will have a bad
A2)(
effect, producing non-randomness. For example, if one starts with an value very close to ,
successive values will also be close to (though each new value not as close), so even this
theoretical sequence is definitely not always random-looking.
17.3 Behavior in Finite Precision.

Using actual computer floating point hardware that implements the IEEE float or double
type (32 bits, about 7 digits of precision, or 64 bits, about 16 digits of precision), the behavior
of these functions is quite different from the theoretical behavior in infinite precision. The tent
function converges almost immediately to zero, since each iteration adds another low-order
bit. The actual logistic equation acts a little better, but still has a relatively short initial run
( @(5A E
followed by a relatively short cycle. Table 17.1 gives results of experiments run to determine
the cycle structure. For a float one can try out all possible starting values to see
the cycle lengths that occur and their percent occurrence, as shown in the table. Notice that

@A
after an initial run of a few thousand values, the function falls into the cycle of length (the

B
cycle that maps to itself) of the time. The second part of Table 17.1 gives results of
trials for doubles, using random starting points and recording the percentages.
At this point, the logistic equation would seem useless for random number generation, since
it has non-random behavior and often falls into a stationary point. However, I came up with
two separate ideas that together make this approach useful.
Ordinary Logistic Equation Remapped Logistic Equation

Cycle Structure, float Cycle Structure, float
Cycle Percent Average Cycle Percent Average
length occurrence initial run length occurrence initial run
1 93.0% 2034 13753 89.9% 4745
930 5.6% 340 3023 5.4% 1150
431 1.0% 251 2928 3.4% 670
106 0.35% 244 1552 0.66% 355
205 0.1% 83 814 0.6% 266
5 0.002% 31 9 0.035% 191
4 0.0004% 7 1 0.00017% 14
3 0.00005% 2 3 0.000024% 1.5
1 0.00002% 0 1 0.000012% 1
Cycle Structure, double Cycle Structure, double
5 638 349 69.4% 54 000 000 112 467 844 80.5% 105 000 000
1 15.2% 10 000 000 61 607 666 5.7% 23 000 000
14 632 801 11.3% 8 500 000 35 599 847 4.3% 19 000 000
10 210 156 1.5% 5 900 000 1 983 078 3.6% 39 000 000
2 625 633 1.3% 3 800 000 4 148 061 3.3% 60 000 000
2 441 806 1.2% 5 200 000 15 023 672 2.5% 19 000 000
1 311 627 0.028% 240 000 12 431 135 0.084% 5 500 000
960 057 0.014% 200 000 705 986 0.084% 670 000

Table 17.1 Cycles of the logistic equation (boldface = the cycle ).
17.4 The Re-mapped Logistic Equation.

The logistic equation yields numbers very close to (on the positive side) and very close to .

Available floating point numbers “pile up” near , but there is no similar behavior near . I was

4
able to restructure the equation so that values occurring near are re-mapped to the negative

side of . The following definition does this, mapping the interval from to to itself (now

called to distinguish from the similar function that maps the interval from to to itself):
4 1 , 1
4 4 1 , 1
where 4 62
<= @ D@A DD A)(!E )(!B)> .
In infinite precision, this re-mapped equation behaves exactly like the original, but with

E
floating point numbers there is no longer any convergence to the cycle of length . Besides this

D ADD0> @
cycle, the other cycles have lengths about or times as long. The previous table also gives
results of experiments on the remapped equation, similar to those before, trying all
possible starting values for a float and over starting values for a double.
A slightly different function will again transform these numbers to the uniform distribu-
tion from to :

2 2 , for , 1
2 4 2 / 8=E , for 4 8 1
This equation does much better, but it is still not remotely good enough. The final piece
of the puzzle is to combine multiple logistic equations into a lattice and couple them weakly
together.
17.5 The Logistic Lattice.

Researchers in chaos theory often use a lattice of numbers to which the logistic equations are
applied. Adjacent lattice nodes affect one another to a small degree. This model is a greatly
simplified version of a fluid undergoing turbulence. In the 1-dimensional case, the nodes are an

array of numbers:
& 1 % 1 1$= = = 1 7 %
The equation applied to each node at the
th stage, producing the
L/ st stage is:
% / 7 % 4 / % 1 , H1

where is a special viscosity constant, and the calculation of 4 and / is done modulo

, that is, by dividing by and taking the remainder. In effect, this wraps the linear array of
nodes into a circle. Combining terms gives:
%

4 /
7 % / % 1 ,

H1
In two dimensions, the equations take the form:
% 4F( , / , 7 % / % / 7 % / % 1

H1 H1

for
Here again arithmetic with both subscripts is carried out modulo . In both the 1- and
2-dimensional versions above, the numbers should be doubles, and the constant should be
!7 %
small, so as not to disturb the uniform distribution and to promote more turbulence. I have used
B A HA
for this constant. It is necessary to iterate the equations before outputting a value; I have
used
iterations. The sizes should be at least size in one dimension and size in two
dimensions.
If the initial values are symmetric about some axis, then the lattice will repeat as if there

were just single logistic equations, so it would be best to use another random number generator
to initialize the lattice values. The individual nodes are probably be independent of one another,
so that this will produce or
random reals at a time, depending on whether it is 1- or 2-
dimensional.
If one is worried about a symmetric set of values coming up (a highly improbable occur-
rence), one could use a variation of the equations that is not symmetric, such as:
%
4 ( / =
7 % / 8=?@
% /

=
% / 8?= D
% 1
, 7 H1 , H1

As a way of visualizing the 2-dimensional lattice of size A A (with @ nodes), if the left and
right sides are pasted together, the lattice would form a vertical hollow cylinder. Then if the
top and bottom sides are pasted together, it would form a donut-shaped object (called a torus
by mathematicians). (The picture is similar to the old “Pac Man” games, where the Pac Man
would exit on one side and immediately come in from the other side.)
The pseudo-random number generator based on this lattice seems to be a very good source
of random numbers, but from the nature of this theory, it is not possible to prove results about
the probable length of cycles or about the quality of its random behavior. It seems likely that
for almost all starting values (nine doubles), the generator will not cycle for a very long
time. It has been tested for random initial values and did not cycle for billions of iterations.
The numbers produced gave good results when subjected to statistical tests. Nevertheless, the
“perfect” generators of the previous section are to be preferred to this one.
17.6 Java Implementation of this Generator.

In the Java implementation it was necessary to iterate each step enough times so that each
node would fill completely with noise and so that any possible symmetries would completely
A -A
disappear. Java code to determine these parameters appears on page 266. The random number
generator itself, for the lattice, appears in the same section.
18
Statistical Tests and Perfect Generators
18.1 Maurer’s universal statistical test.

Maurer’s universal statistical test can be found on page 270.
18.2 The Blum-Blum-Shub perfect generator.
The Blum-Blum-Shub perfect generator appears on page 272.

Part VI
The Advanced
Encryption Standard
114
19
The Advanced Encryption Standard
19.1 Overview.
D
> D $@ E>
The new U.S. Advanced Encryption Standard (AES) is a block cipher with block size of
> )( A D
bits, or bytes. Keys for the cipher come in one of three lengths: , , or bits, which
( A
is , , or bytes. The algorithm is oriented toward bytes ( bits), but there is also emphasis
on what the AES specification calls words, which are arrays of bytes (or bits, the size of
an int in Java). My Java implementation presented here tends to de-emphasize the words.

The main mathematical difficulty with the algorithm is that it uses arithmetic over the field
. Even the field itself only poses real difficulties with multiplication of field elements
and with multiplicative inverses. These topics are covered in Section 2 below.
Otherwise, the AES algorithm is just an annoying number of details to orchestrate, but not
really difficult. Section 3 below covers the S-Boxes, while Section 4 shows how the keys are
handled. Section 5 covers the remainder of the encryption algorithm, and Section 6 covers
decryption.
I have implemented the algorithm following the AES specification and using B. Gladman’s
commentary. I haven’t worried about efficiency, but mostly have just tried to produce a clear,

simple, working Java program. One exception is to give Gladman’s more efficient imple-
mentation of multiplication in because it is interesting (see Section 2). Gladman has
produced an optimized C implementation and has a lot to say on the subject of efficient imple-
mentation, especially with methods using tables.
Law AES-1:
Conventional block ciphers are always ugly, complicated,
inelegant brutes, and the AES is no exception.
In order to create such a cryptosystem, one must remember that anything done by encryption
must be undone during decryption, using the same key since it is a conventional (symmetric
key) system. Thus the focus is on various invertible operations. One standard technique in
using the key is to derive a string somehow from the key, and use xor to combine it with the
emerging ciphertext. Later the same xor reverses this. Otherwise there are “mixing” operations
that move data around, and “translation” (or “substitution”) operations that replace one piece
of data with another. This last operation is usually carried out on small portions of ciphertext
using so-called “S-boxes”, which define replacement strings. One set of mixing, replacements,
116 VI. The Advanced Encryption Standard
Key Sizes versus Rounds

Key Plaintext Number of
Block Size Block Size Rounds
(Nk words) (Nb words) (Nr)
AES-128 4 4 10
AES-192 6 4 12
AES-256 8 4 14
Table 19.1 Parameters for the AES.
and exclusive-or with a string derived from the key is called a round. Then there will typically
be a number of rounds. The AES uses different numbers of rounds for the different key sizes
according to Table 19.1 above. This table uses a variable Nb for the plaintext block size, but
it is always 4 words. Originally the AES was going to support different block sizes, but they
settled on just one size. However, the AES people (at the NIST) recommend keeping this as a
named constant in case a change is ever wanted.
Remember that a word is 4 bytes or 32 bits. The names Nk, Nb, and Nr are standard for
the AES. In general, I try to use the names in the AES specification, even when they do not
conform to Java conventions.
The particular form of this type of algorithm, with its rounds of mixing and substitution
and exclusive-or with the key, was introduced with the official release of the Data Encryption
Standard (DES) in 1977 and with work preceding the release. The DES has a block size of
64 bits and a very small key size of 56 bits. From the beginning the key size of the DES was
controversial, having been reduced at the last minute from 64 bits. This size seemed at the
edge of what the National Security Agency (but no one else) could crack. Now it is easy to
break, and completely insecure. The AES, with its minimum of 128 bits for a key should not
be breakable by brute force attacks for a very long time, even with great advances in computer
hardware.
19.2 Outline of the AES Algorithm.

Here is the AES algorithm is outline form, using Java syntax for the pseudo-code, and much of
the AES standard notation:
Constants: int Nb = 4; // but it might change someday
int Nr = 10, 12, or 14; // rounds, for Nk = 4, 6, or 8
Inputs: array in of 4*Nb bytes // input plaintext
array out of 4*Nb bytes // output ciphertext
array w of 4*Nb*(Nr+1) bytes // expanded key
Internal work array:
state, 2-dim array of 4*Nb bytes, 4 rows and Nb cols
Algorithm:
19. The Advanced Encryption Standard (AES) 117
void Cipher(byte[] in, byte[] out, byte[] w) {

byte[][] state = new byte[4][Nb];
state = in; // actual component-wise copy
AddRoundKey(state, w, 0, Nb - 1); // see Section 4 below
for (int round = 1; round < Nr; round++)
SubBytes(state); // see Section 3 below
ShiftRows(state); // see Section 5 below
MixColumns(state); // see Section 5 below
AddRoundKey(state, w,
round*Nb, (round+1)*Nb - 1); // Section 4
}
SubBytes(state); // see Section 3 below
ShiftRows(state); // see Section 5 below
AddRoundKey(state, w, Nr*Nb, (Nr+1)*Nb - 1); // Section 4
out = state; // component-wise copy
}
Let’s go down the items in the above pseudo-code in order:
Multiplication in GF(256): this is not mentioned explicitly above, but the indi-
vidual functions use it frequently. It is described in Section 2 below.
Nb: Right now, this is always 4, the constant number of 32-bit words that make up a
block for encryption and decryption.
Nr: the number of rounds or main iterations of the algorithm. The possible values depend
on the three different possible key sizes the earlier table showed.
in: the input block of 128 bits of plaintext, arranged as 4*Nb = 16 bytes, numbered 0
to 15 and arranged in sequence.
out: the output block of 128 bits of ciphertext, arranged the same as in.
state: the internal array that is worked on by the AES algorithm. It is arranged as a 4
by Nb 2-dimensional array of bytes (that is, 4 by 4).
w: the expanded key. The initial key is of size 4*Nk bytes (see table earlier), and this
is expanded to the array w of 4*Nb*(Nr+1) = 16*(Nr+1) bytes for input to the en-
cryption algorithm. Each round uses 4*Nb bytes, and each portion of w is used only
once. (There are Nr+1 full rounds, and an extra use before and after in partial rounds,
for a total of Nr+1 uses.) This function for expanding the key is described in Section 4
below.
SubBytes(state): this takes each byte of the state and independently looks it up
in an “S-box” table to substitute a different byte for it. (The same S-box table is also used
in the key expansion.) Section 3 shows how the S-box table is defined and constructed.
ShiftRows(state): this simply moves around the rows of the state array. See
Section 5 below.
MixColumns(state): this does a much more complicated mix of the columns of the
state array. See Section 5 below.
AddRoundKey(state, w, param1, param2): this takes the 4*Nb*(Nr+1)
bytes of the expanded key, w, and does an exclusive-or of successive portions of the
expanded key with the changing state array. The integer values param1 and param2
take on different values during execution of the algorithm, and they give the inclusive
range of columns of the expanded key that are used. My implementation of this function
doesn’t use these parameters, because each round just uses the next 4*Nb bytes of w.
The details of this function appear in Section 4 below.
Well, that’s pretty much it. Now the remaining sections just have to fill in a large number
of missing details. Section 6 gives the inverse algorithm for decryption, but this is not a big
problem, since the parameters and functions are either identical or similar.
20
The Finite Field GF(256)
20.1 Finite Fields.
/ /
A field is an algebraic object with two operations: addition and multiplication, represented by
and , although they will not necessarily be ordinary addition and multiplication. Using ,

4
all the elements of the field must form a commutative group, with identity denoted by and the

7%
inverse of denoted by . Using , all the elements of the field except must form another

"/ /
commutative group with identity denoted and inverse of denoted by . (The element
has no inverse under .) Finally, the distributive identity must hold: ,

for all field elements , , and .
There are a number of different infinite fields, including the rational numbers (fractions),
the real numbers (all decimal expansions), and the complex numbers. Cryptography focuses

prime integer and any integer
on finite fields. It turns out that for any
greater than or equal

to , there is a unique field with
elements in it, denoted . (The “GF” stands for
“Galois Field”, named after the brilliant young French mathematician who discovered them.)
Here “unique” means that any two fields with the same number of elements must be essentially
the same, except perhaps for giving the elements of the field different names.

In case
is equal to , the field is just the integers mod , in which addition and multipli-
cation are just the ordinary versions followed by taking the remainder on division by . The
a non-zero element in , finding
7%
only difficult part of this field is finding the multiplicative inverse of an element, that is, given
. This is the same as finding a such that .
This calculation can be done with the extended Euclidean algorithm, as is explained elsewhere
in these notes.
20.2 The Finite Field GF(2n).

The case in which
is greater than one is much more difficult to describe. In cryptography, one

D
almost always takes to be in this case. This section just treats the special case of and
, that is, , because this is the field used by the new U.S. Advanced Encryption
Standard (AES).
The AES works primarily with bytes (8 bits), represented from the right as:
%&=
The 8-bit elements of the field are regarded as polynomials with coefficients in the field
:
/ / / / / / % % / &=
The field elements will be denoted by their sequence of bits, using two hex digits.
20.3 Addition in GF(2n).

To
/
add two field elements, just add the corresponding polynomial coefficients using addition in

. Here addition is modulo , so that , and addition, subtraction and exclusive-or
are all the same. The identity element is just zero: (in bits) or (hex).
20.4 Multiplication in GF(2n).
Multiplication is this field is more complicated and harder to understand, but it can be im-
plemented very efficiently in hardware and software. The first step in multiplying two field
/
elements is to multiply their corresponding polynomials just as in beginning algebra (except

$(
that the coefficients are only or , and makes the calculation easier, since many
terms just drop out). The result would be an up to degree polynomial — too big to fit into
one byte. A finite field now makes use of a fixed degree eight irreducible polynomial (a poly-
nomial that cannot be factored into the product of two simpler polynomials). For the AES the
polynomial used is the following (other polynomials could have been used):
H
/
/ / /
(hex).
The intermediate product of the two polynomials must be divided by . The remainder
from this division is the desired product.
This sounds hard, but is easier to do by hand than it might seem (though error-prone). To
/ /
/ /
make it easier to write the polynomials down, adopt the convention that instead of

D(A
just write the exponents of each non-zero term. (Remember that terms are either
BE)( >( > EA
zero or have a as coefficient.) So write the following for : .
Now try to take the product (which is the same as in hex-
adecimal). First do the multiplication, remembering that in the sum below only an odd number
of like powered terms results in a final term:
(7 5 4 2 1) * (6 4 1 0) gives (one term at a time)
(7 5 4 2 1) * (6) = (13 11 10 8 7)
(7 5 4 2 1) * (4) = (11 9 8 6 5)
(7 5 4 2 1) * (1) = (8 6 5 3 2)
(7 5 4 2 1) * (0) = + 7 5 4 2 1)
----------------------------------
(13 10 9 8 5 4 3 1)
20. The Finite Field GF(256) 121
The final answer requires the remainder on division by

, or
D(A
. This is like
ordinary polynomial division, though easier because of the simpler arithmetic.
(13 10 9 8 5 4 3 1)
(8 4 3 1 0) * (5) = (13 9 8 6 5)
--------------------------------
(10 6 4 3 1)
(8 4 3 1 0) * (2) = (10 6 5 3 2)
--------------------------------
(5 4 2 1)

(the remainder)
Here the first element of the quotient is and the second element of the quotient is . Thus
the final result says that > EA A>
in the field. (When I did the calculations above,
I made two separate mistakes, but checked my work with techniques below.)
20.5 Improved Multiplication in GF(2n).

The above calculations could be converted to a program, but there is a better way. One does
D B
the calculations working from the low order terms, and repeatedly multiplying by . If the
result is of degree , just add (the same as subtract) BE)( >(
to get degree . Again this can be
illustrated using the above notation and the same example operands:
i powers of r: r * (i) Simplified Result Final Sum
========================================================================
0 (7 5 4 2 1 ) (7 5 4 2 1 )
------------------------------------------------------------------------
1 (7 5 4 2 1 )*(1) = (8 6 5 3 2)
+(8 4 3 1 0)
---------------------
(6 5 4 2 1 0) + (6 5 4 2 1 0)
------------------
(7 6 0)
------------------------------------------------------------------------
2 (6 5 4 2 1 0)*(1) = (7 6 5 3 2 1)
------------------------------------------------------------------------
3 (7 6 5 3 2 1) *(1) = (8 7 6 4 3 2)
+(8 4 3 1 0)
---------------------
(7 6 2 1 0)
------------------------------------------------------------------------
4 (7 6 2 1 0)*(1) = (8 7 3 2 1)
+(8 4 3 1 0)
--------------------- (7 6 0)
(7 4 2 0) + (7 4 2 0)
-------------------
(6 4 2)
------------------------------------------------------------------------
5 (7 4 2 0)*(1) = (8 5 3 1)
+(8 4 3 1 0)
---------------------
(5 4 0)
(6 4 2)
6 (5 4 0)*(1) = (6 5 1) + (6 5 1)
-----------------
(5 4 2 1)
The final answer is the same as before. Here is an algorithm (almost) in Java that realizes
the above calculations:
public byte FFMul(unsigned byte a, unsigned byte b) {
unsigned byte aa = a, bb = b, r = 0, t;
while (aa != 0) {
if ((aa & 1) != 0)
r = r ˆ bb;
t = bb & 0x80;
bb = bb << 1;
if (t != 0)
bb = bb ˆ 0x1b;
aa = aa >> 1;
}
return r;
}
Unfortunately, Java has no unsigned byte type, and the logical operations produce a
32-bit integer. Finally, one ought to be able to use Java’s “right shift with zero fill” operator
>>>, but it doesn’t work as it is supposed to. See Appendix B for a discussion of the problems
encountered in converting the above “Java” program to actual Java. Later examples below show
the code for this function.
20.6 Using Logarithms to Multiply.

When I was young (a long time ago) there were no pocket calculators. We had to do without
modern conveniences like Gatorade. Mathematical calculations, if not done with a slide rule,
were carried out by hand. Often we used printed tables of logarithms to turn multiplications into
easier additions. (In these “elder” days, believe it or not, the printed tables included tables of the
A8=I( B
logarithms of trig functions of angles, so that you got the log directly for further calculations.)
As a simple example, suppose one wanted the area of a circle of radius cm. Use the
famous formula (“pie are square, cake are round”), so one needs to calculate
)A8=?( B )A8=?( B A<= $( >8=
WeA<=?( would
B =?A>@B > and A8= $( > =?(@B 6E> . Now add two copies of the first
look up the logarithm (base 10) of each number in the printed table:
number and one of the second:
=A>@B >
/ =A>@B > /.=I(@B 6E)> 8A = A>EDD8=
Finally, take the “anti-log” (that is, take to the power A<= A>EDD ) to get the final answer:
6B "( = . This works because

/ / =
The actual use of log tables was much more horrible than the above might indicate. In case
you want to find out how it really worked, look at Appendix A, but have an air sickness bag
handy.
In a similar way, in finite fields one can replace the harder multiplication by the easier
addition, at the cost of looking up “logarithms” and “anti-logarithms.”
20.7 Generators in Fields.

First must come the concept of a generator of a finite field. Generators also play a role is certain
simple but common random number generators, as is detailed in another section. A generator
is an element
whose successive powers take on every element except the zero. For example, in
the field
% E $A
, try successive powers of several elements, looking for a generator:
Try powers of , taken modulo :
E % E ,
E A E A ,
E A E $A D ,
E A D E
A ,
E
so successive powers of just take on the values , E , D , , and repeat, so that E is not a
( A
generator.
Now try powers of , taken modulo :
(<% ( ,
( A > A A ,
( A A (
A ,
( A ( $A @ ,
( A @ (
A ,
( A ( $A ,
( A , @ , , , and
(
so successive powers make a longer cycle, but still not all elements: , ,
A
repeat, so is also not a generator.

Finally try successive powers of , taken modulo :
!% ,
A ( A ( ,
A D A D ,
A D
A A,
Table of “Exponential” Values

E(rs) s
0 1 2 3 4 5 6 7 8 9 a b c d e f
0 01 03 05 0f 11 33 55 ff 1a 2e 72 96 a1 f8 13 35
1 5f e1 38 48 d8 73 95 a4 f7 02 06 0a 1e 22 66 aa
2 e5 34 5c e4 37 59 eb 26 6a be d9 70 90 ab e6 31
3 53 f5 04 0c 14 3c 44 cc 4f d1 68 b8 d3 6e b2 cd
4 4c d4 67 a9 e0 3b 4d d7 62 a6 f1 08 18 28 78 88
5 83 9e b9 d0 6b bd dc 7f 81 98 b3 ce 49 db 76 9a
6 b5 c4 57 f9 10 30 50 f0 0b 1d 27 69 bb d6 61 a3
r 7 fe 19 2b 7d 87 92 ad ec 2f 71 93 ae e9 20 60 a0
8 fb 16 3a 4e d2 6d b7 c2 5d e7 32 56 fa 15 3f 41
9 c3 5e e2 3d 47 c9 40 c0 5b ed 2c 74 9c bf da 75
a 9f ba d5 64 ac ef 2a 7e 82 9d bc df 7a 8e 89 80
b 9b b6 c1 58 e8 23 65 af ea 25 6f b1 c8 43 c5 54
c fc 1f 21 63 a5 f4 07 09 1b 2d 77 99 b0 cb 46 ca
d 45 cf 4a de 79 8b 86 91 a8 e3 3e 42 c6 51 f3 0e
e 12 36 5a ee 29 7b 8d 8c 8f 8a 85 94 a7 f2 0d 17
f 39 4b dd 7c 84 97 a2 fd 1c 24 6c b4 c7 52 f6 01
Table 20.1 Table of Powers of 0x03, the “exponentials”.
A A
A >,
A >
A , ,
A $A
A $A @,
A @
A E,
A E
A B ,,
A $A
A A B
A ,
( D A > @ E B
> B
so successive powers take on all non-zero elements: , , , , , , , , , , , , and

repeat, so finally is a generator. (Wheh! In fact,
the generators are , , , and .)

E B ( E B @
However
is not a generator for the field , so it doesn’t work to just try . (The genera-
tors in are , , , , , , , , , and .)
A /
The above random search shows that generators are hard to discover and are not intuitive.

EE
It turns out that , which is the same as as a polynomial, is the simplest generator for
. Its powers take on all non-zero values of the field. In fact Table 20.1 , a table
of “exponentials” or “anti-logs”, gives each possible power. (The table is really just a simple
A9
linear table, not really 2-dimensional, but it has been arranged so that the two hex digits are on
different axes.) Here

is the field element given by , where these are hex numbers,
and the initial “ ” is left off for simplicity.
Similarly, Table 20.2 is a table of “logarithms”, where the entry

is the field element

Table of “Logarithm” Values

L(rs) s
0 1 2 3 4 5 6 7 8 9 a b c d e f
0 – 00 19 01 32 02 1a c6 4b c7 1b 68 33 ee df 03
1 64 04 e0 0e 34 8d 81 ef 4c 71 08 c8 f8 69 1c c1
2 7d c2 1d b5 f9 b9 27 6a 4d e4 a6 72 9a c9 09 78
3 65 2f 8a 05 21 0f e1 24 12 f0 82 45 35 93 da 8e
4 96 8f db bd 36 d0 ce 94 13 5c d2 f1 40 46 83 38
5 66 dd fd 30 bf 06 8b 62 b3 25 e2 98 22 88 91 10
6 7e 6e 48 c3 a3 b6 1e 42 3a 6b 28 54 fa 85 3d ba
r 7 2b 79 0a 15 9b 9f 5e ca 4e d4 ac e5 f3 73 a7 57
8 af 58 a8 50 f4 ea d6 74 4f ae e9 d5 e7 e6 ad e8
9 2c d7 75 7a eb 16 0b f5 59 cb 5f b0 9c a9 51 a0
a 7f 0c f6 6f 17 c4 49 ec d8 43 1f 2d a4 76 7b b7
b cc bb 3e 5a fb 60 b1 86 3b 52 a1 6c aa 55 29 9d
c 97 b2 87 90 61 be dc fc bc 95 cf cd 37 3f 5b d1
d 53 39 84 3c 41 a2 6d 47 14 2a 9e 5d 56 f2 d3 ab
e 44 11 92 d9 23 20 2e 89 b4 7c b8 26 77 99 e3 a5
f 67 4a ed de c5 31 fe 18 0d 63 8c 80 c0 f7 70 07
Table 20.2 Table of Inverse Powers of 0x03, the “logarithms”.
that satisfies A

?9
, where these are hex numbers, and again the initial “ ” is left off.
These tables were created using the multiply function in the previous section. A Java pro-
> EA
gram that directly outputs the HTML source to make the tables appears on page 273.
EA A
As an example, suppose one wants the product

above, leaving off the “ ”). Use the table above to look up

and : > EA
(the same product as in the examples
and

>
. This means that
> EA A % A &

A % & A ; % =

, just subtract EE as shown. This works ;

powers of A repeat after EE iterations. Now use the table to look up A % , which is the
If the sum above gets bigger than because the
answer: A> .
Thus these tables give a much simpler and faster algorithm for multiplication:
public byte FFMulFast(unsigned byte a, unsigned byte b){
int t = 0;;
if (a == 0 || b == 0) return 0;
t = L[a] + L[b];
if (t > 255) t = t - 255;
return E[t];
}
Multiplicative Inverses of Each Element

inv(rs) s
0 1 2 3 4 5 6 7 8 9 a b c d e f
0 – 01 8d f6 cb 52 7b d1 e8 4f 29 c0 b0 e1 e5 c7
1 74 b4 aa 4b 99 2b 60 5f 58 3f fd cc ff 40 ee b2
2 3a 6e 5a f1 55 4d a8 c9 c1 0a 98 15 30 44 a2 c2
3 2c 45 92 6c f3 39 66 42 f2 35 20 6f 77 bb 59 19
4 1d fe 37 67 2d 31 f5 69 a7 64 ab 13 54 25 e9 09
5 ed 5c 05 ca 4c 24 87 bf 18 3e 22 f0 51 ec 61 17
6 16 5e af d3 49 a6 36 43 f4 47 91 df 33 93 21 3b
r 7 79 b7 97 85 10 b5 ba 3c b6 70 d0 06 a1 fa 81 82
8 83 7e 7f 80 96 73 be 56 9b 9e 95 d9 f7 02 b9 a4
9 de 6a 32 6d d8 8a 84 72 2a 14 9f 88 f9 dc 89 9a
a fb 7c 2e c3 8f b8 65 48 26 c8 12 4a ce e7 d2 62
b 0c e0 1f ef 11 75 78 71 a5 8e 76 3d bd bc 86 57
c 0b 28 2f a3 da d4 e4 0f a9 27 53 04 1b fc ac e6
d 7a 07 ae 63 c5 db e2 ea 94 8b c4 d5 9d f8 90 6b
e b1 0d d6 eb c6 0e cf ad 08 4e d7 e3 5d 50 1e b3
f 5b 23 38 34 68 46 03 8c dd 9c 7d a0 cd 1a 41 1c
Table 20.3 Table of Inverses of Each Field Element.
As before, this is Java as if it had an unsigned byte type, which it doesn’t. The actual
Java code requires some short, ugly additions. (See Unsigned bytes in Java in Appendix B to
convert the above “Java” program to actual Java.)
This section has presented two algorithms for multiplying field elements, a slow one and a
fast one. As a check, here is a program that compares the results of all 65536 possible products
to see that the two methods agree (which they do): see Compare multiplications on page 275.
20.8 The Multiplicative Inverse of Each Field Element.

Later work with the AES will also require the multiplicative inverse of each field element except
, which has no inverse. This inverse is easy to calculate, given the tables above. If is the

A
(leaving off the “ ”), then the inverse of is
9 79
> > >
generator , so that for example, to
, so the inverse of
7
find the inverse of , look up in the “log” table to see that is

, and from the “exponential” table, this is . Table 20.3 gives the result of
carrying out the above procedure for each non-zero field element.
The next chapter has a table generating program that will calculate and print HTML source
for the above table.
21
Advanced Encryption Standard: S-Boxes
21.1 S-Boxes and the SubBytes Transformation.

Many different block ciphers use a special substitution called an “S-box”. The AES also has
these S-boxes, which it terms the “SubBytes Transformation”. S-boxes provide an invertible
(reversible) transformation of segments of plaintext during encryption, with the reverse during
E>
decryption. With the AES it is a single simple function applied over and over again to each byte
during stages of the encryption, returning a byte. Each of the possible byte values is trans-
E>
formed to another byte value with the SubBytes transformation, which is a full permutation,
meaning that every element gets changed, and all possible elements are represented as the
result of a change, so that no two different bytes are changed to the same byte. The SubBytes

transformation changes a single entry as follows (here stands for the th bit of a byte value
).
byte SubBytesEntry(byte b) {
byte c = 0x63;
if (b != 0)} // leave 0 unchanged {\timesbf
b = multiplicativeInverse(b);
for (i = 0; i < 8; i++)
b[i] = b[i] ˆ b[(i+4)%8] ˆ b[(i+5)%8] ˆ
b[(i+6)%8] ˆ b[(i+7)%8] ˆ c[i];
return b;
E>
}
In practice, the values given by the transformation of each byte should be precomputed
and stored in a table. Because the table is computed only once before the start of encryption,
there is less need to optimize its performance. Here is a copy of the table. This and the next
table were printed using the Java program on page 277.)
21.2 The Inverse SubBytes Transformation.

The table of the inverse SubBytes transformation could be generated using the inverse of the
formula for SubBytes given above (a similar function). However, it is even easier to just directly
invert the previous table, without recalculating.
21.3 The Function SubBytes(state).

S-Box Values
S(rs) s
0 1 2 3 4 5 6 7 8 9 a b c d e f
0 63 7c 77 7b f2 6b 6f c5 30 01 67 2b fe d7 ab 76
1 ca 82 c9 7d fa 59 47 f0 ad d4 a2 af 9c a4 72 c0
2 b7 fd 93 26 36 3f f7 cc 34 a5 e5 f1 71 d8 31 15
3 04 c7 23 c3 18 96 05 9a 07 12 80 e2 eb 27 b2 75
4 09 83 2c 1a 1b 6e 5a a0 52 3b d6 b3 29 e3 2f 84
5 53 d1 00 ed 20 fc b1 5b 6a cb be 39 4a 4c 58 cf
6 d0 ef aa fb 43 4d 33 85 45 f9 02 7f 50 3c 9f a8
r 7 51 a3 40 8f 92 9d 38 f5 bc b6 da 21 10 ff f3 d2
8 cd 0c 13 ec 5f 97 44 17 c4 a7 7e 3d 64 5d 19 73
9 60 81 4f dc 22 2a 90 88 46 ee b8 14 de 5e 0b db
a e0 32 3a 0a 49 06 24 5c c2 d3 ac 62 91 95 e4 79
b e7 c8 37 6d 8d d5 4e a9 6c 56 f4 ea 65 7a ae 08
c ba 78 25 2e 1c a6 b4 c6 e8 dd 74 1f 4b bd 8b 8a
d 70 3e b5 66 48 03 f6 0e 61 35 57 b9 86 c1 1d 9e
e e1 f8 98 11 69 d9 8e 94 9b 1e 87 e9 ce 55 28 df
f 8c a1 89 0d bf e6 42 68 41 99 2d 0f b0 54 bb 16
Table 21.1 S-Box Values.
Inverse S-Box Values

iS(rs) s
0 1 2 3 4 5 6 7 8 9 a b c d e f
0 52 09 6a d5 30 36 a5 38 bf 40 a3 9e 81 f3 d7 fb
1 7c e3 39 82 9b 2f ff 87 34 8e 43 44 c4 de e9 cb
2 54 7b 94 32 a6 c2 23 3d ee 4c 95 0b 42 fa c3 4e
3 08 2e a1 66 28 d9 24 b2 76 5b a2 49 6d 8b d1 25
4 72 f8 f6 64 86 68 98 16 d4 a4 5c cc 5d 65 b6 92
5 6c 70 48 50 fd ed b9 da 5e 15 46 57 a7 8d 9d 84
6 90 d8 ab 00 8c bc d3 0a f7 e4 58 05 b8 b3 45 06
r 7 d0 2c 1e 8f ca 3f 0f 02 c1 af bd 03 01 13 8a 6b
8 3a 91 11 41 4f 67 dc ea 97 f2 cf ce f0 b4 e6 73
9 96 ac 74 22 e7 ad 35 85 e2 f9 37 e8 1c 75 df 6e
a 47 f1 1a 71 1d 29 c5 89 6f b7 62 0e aa 18 be 1b
b fc 56 3e 4b c6 d2 79 20 9a db c0 fe 78 cd 5a f4
c 1f dd a8 33 88 07 c7 31 b1 12 10 59 27 80 ec 5f
d 60 51 7f a9 19 b5 4a 0d 2d e5 7a 9f 93 c9 9c ef
e a0 e0 3b 4d ae 2a f5 b0 c8 eb bb 3c 83 53 99 61
f 17 2b 04 7e ba 77 d6 26 e1 69 14 63 55 21 0c 7d
Table 21.2 Inverse S-Box Values.

21. Advanced Encryption Standard: S-Boxes 129
The Java pseudo-code for this part is now very simple, using the Sbox array defined above:
void SubBytes(byte[][] state) {
for (int row = 0; row < 4; row++)
for (int col = 0; col < Nb; col++)
state[row][col] = Sbox[state[row][col]];
}
22
AES Key Expansion
22.1 Overview of Key Expansion.

In a simple cipher, one might exclusive-or the key with the plaintext. Such a step is easily
reversed by another exclusive-or of the same key with the ciphertext. In the case of the AES,
there are a number of rounds, each needing its own key, so the actual key is “stretched out” and
transformed to give portions of key for each round. This is the key expansion that is the topic
of this section.
The key expansion routine, as part of the overall AES algorithm, takes an input key (denoted
key below) of 4*Nk bytes, or Nk 32-bit words. Nk has value either 4, 6, or 8. The output is
an expanded key (denoted w below) of 4*Nb*(Nr+1) bytes, where Nb is always 4 and Nr is
the number of rounds in the algorithm, with Nr equal 10 in case Nk is 4, Nr equal 12 in case
Nk is 6, and Nr equal 14 in case Nk is 8.
The key expansion routine below states most of the actions in terms of words or 4-byte
units, since the AES specification itself emphasizes words, but my implementation uses bytes
exclusively.
Inputs: int Nk = 4, 6, or 8; // the number of words in the key
array key of 4*Nk bytes or Nk words // input key
Output: array w of Nb*(Nr+1) words or 4*Nb*(Nr+1) bytes // expanded key
Algorithm:
void KeyExpansion(byte[] key, word[] w, int Nw) {

int Nr = Nk + 6;
w = new byte[4*Nb*(Nr+1)];
int temp;
int i = 0;
while ( i < Nk) {
w[i] = word(key[4*i], key[4*i+1], key[4*i+2], key[4*i+3]);
i++;
}
i = Nk;
while(i < Nb*(Nr+1)) {
temp = w[i-1];
if (i % Nk == 0)
temp = SubWord(RotWord(temp)) ˆ Rcon[i/Nk];
else if (Nk > 6 && (i%Nk) == 4)
temp = SubWord(temp);
w[i] = w[i-Nk] ˆ temp;
22. AES Key Expansion 131
Expanded Key Sizes in Words

Key Length Number of Rounds Exp. Key Size
(Nk words) (Nr) (Nb(Nr+1) words)
4 10 44
6 12 52
8 14 60
Table 22.1 Expanded Key Sizes.
Powers of x = 0x02
i 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
01 02 04 08 10 20 40 80 1b 36 6c d8 ab 4d 9a
Table 22.2 Parameters for Linear Congruence Generators.
i++;
}
Discussion of items in the above pseudo-code in order:
The constant Nb = 4: This was mentioned earlier. Nb is the number of words in an
AES block, and right now it is always 4.
The key, key: the input key consists of Nk words, or 4*Nk bytes.
The expanded key, w: This consists of Nb*(Nk+1) words, or 4*Nb*(Nk+1) bytes.
The range of sizes are in Table 22.1 .
&1 %11 %11 1

RotWord(): This does the following simple cyclic permutation of a word: change
to . & 7 %1 81 <1 . Table

Rcon[i]: This is defined as the word:
powers of :
22.2 contains values of
Notice that in the algorithm for key expansion, the first reference to Rcon is
&
Rcon[i/Nk], where i has value Nk, so that the smallest index to Rcon is 0, and
this uses .
SubWord(): This just applies the S-box value used in SubBytes to each of the 4 bytes
in the argument.
22.2 Use of Key Expansion in the AES Algorithm.

The function KeyExpansion() merely supplies a much expanded (and transformed) key
for use by the AddRoundKey() function in the main AES algorithm (see Section 1). This
does a byte-wise exclusive-or of 4*Nb = 16 bytes at a time of the key with the 4*Nb =
16 bytes of the state. Successive segments of 4*Nb = 16 bytes of the expanded key are
exclusive-ored in before the rounds of the algorithm, during each round, and at the end of
the rounds. In the end, there are Nr rounds, but Nr+1 exclusive-ors of parts of the expanded
key. Since none of the expanded key is used more than once, this means that algorithm needs
4*Nb*(Nr+1) = 16*(Nr+1) bytes of expanded key, and this is just the amount provided
by the KeyExpansion() function.
23
AES Encryption
23.1 Final Parts of the AES Algorithm.

The previous 4 subsections have covered most of the details needed to complete the AES algo-
rithm. One still needs a description and code for the following routines:
ShiftRows()
MixColumns()
AddRoundKey()
One also needs to organize a number of minor details to get a complete working Java
program.
In the first two parts, the AES is moving around and “stirring up” data in the 4-by-4 array
of bytes named state.
23.2 Implementation of ShiftRows().

The action of shifting rows is particularly simple, just performing left circular shifts of rows 1,
2, and 3, by amounts of 1, 2, and 3 bytes. Row 0 is not changed. The actual Java code below
does this.
void ShiftRows(byte[][] state) {
byte[] t = new byte[4];
for (int r = 1; r < 4; r++) {
for (int c = 0; c < Nb; c++)
t[c] = state[r][(c + r)%Nb];
for (int c = 0; c < Nb; c++)
state[r][c] = t[c];
}
}
23.3 Implementation of MixColumns().

The action of mixing columns works on the columns of the state array, but it is much more
complicated that the shift columns action. As described in the AES specification, it treats each
column as a four-term polynomial with coefficients in the field

. All this is similar to the

description of the field itself, except with an extra layer of complexity. These polynomials are
added and multiplied just using the operations of the field on the coefficients, except
dividing by the polynomial /

that the result of a multiplication, which is a polynomial of degree up to 6, must be reduced by

and taking the remainder.
The columns are each multiplied by the fixed polynomial:

A / / / 1
where
A
stands for the field element A
. This sounds horrible, but mathematical manip-
ulations can reduce everything to the following simple algorithm, where multiplication in the
field is represented below by . The principle change needed to convert this to actual Java is
to replace with a call to FFMul(). (Gladman gives a shorter but more obscure version of
this code.)
void MixColumns(byte[][] s) {
byte[] sp = new byte[4];
for (int c = 0; c < 4; c++) {
sp[0] = (0x02 # s[0][c]) ˆ (0x03 # s[1][c]) ˆ
s[2][c] ˆ s[3][c];
sp[1] = s[0][c] ˆ (0x02 # s[1][c]) ˆ
(0x03 # s[2][c]) ˆ s[3][c];
sp[2] = s[0][c] ˆ s[1][c] ˆ
(0x02 # s[2][c]) ˆ (0x03 # s[3][c]);
sp[3] = (0x03 # s[0][c]) ˆ s[1][c] ˆ
s[2][c] ˆ (0x02 # s[3][c]);
for (int i = 0; i < 4; i++) s[i][c] = sp[i];
}
}
23.4 Implementation of AddRoundKey().

As described before, portions of the expanded key w are exclusive-ored onto the state matrix
Nr+1 times (once for each round plus one more time). There are 4*Nb bytes of state, and since
each byte of the expanded key is used exactly once, the expanded key size of 4*Nb*(Nr+1)
bytes is just right. The expanded key is used, byte by byte, from lowest to highest index, so there
is no need to count the bytes as they are used from w, but just use them up and move on, as the
following near-Java code shows. This code assumes the key has already been expanded into the
array w, and it assumes a global counter wCount initialized to 0. The function AddRoundKey
uses up 4*Nb = 16 bytes of expanded key every time it is called.
void AddRoundKey(byte[][] state) {
for (int c = 0; c < Nb; c++)
for (int r = 0; r < 4; r++)
state[r][c] = state[r][c] ˆ w[wCount++];
}
23. AES Encryption 135
23.5 Java Implementation of Encryption Using AES.

My Java implementation of AES encryption uses six classes:
AESencrypt, which provides the principle functions for AES encryption,
Tables, which gives values from computed tables and various untility functions.
GetBytes, which reads plaintexts and keys,
Copy, which gives two simple copy functions needed by AES,
Print, which prints 1- and 2-dimensional arrays of bytes for debugging, and
AEStest, which which is a driver for the tests.
A combined listing of all the ecryption classes is found on page 282.

See the next chapter for test runs of the program, where encryption is followed by decryp-
tion.
24
AES Decryption
24.1 Modifications to AES for Decryption.

The following functions need minor (or more major) revision for decryption:
Cipher(), changed to InvCipher(), which is the main decryption outline. It is of

course very similar to the Cipher() function, except that many of the subfunctions are
themselves inverses, and the order of functions within a round is different.
ShiftRows(), changed to InvShiftRows() – just minor changes.
MixColumns(), changed to InvMixColumns() – the inverse function, similar but

with different constants in it.
AddRoundKey(), changed to InvAddRoundKey() – just works backwards along the
expanded key.
As before, one also needs to organize a number of minor details to get a complete working
Java program.
24.2 Implementation of InvCipher().

Here is Java pseudo-code for the inverse cipher. The various steps must be carried out in reverse
order. These are arranged into rounds as with encryption, but the functions in each round are
in a slightly different order than the order used in encryption. The AES specification has also
supplied an equivalent inverse cipher in which the individual parts of each round are in the
same order as with encryption. This might make a hardware implementation easier, but I have
not used it here.
int Nr = 10}, 12}, or 14; }// rounds, for Nk = 4, 6, or 8
Inputs: array in of 4*Nb bytes // input ciphertext
array out of 4*Nb bytes // output plaintext
array w of 4*Nb*(Nr+1) bytes // expanded key
Internal work array:
state, 2-dim array of 4*Nb bytes, 4 rows and Nb cols
Algorithm:
24. AES Decryption 137
void InvCipher(byte[] in, byte[] out, byte[] w) {

byte[][] state = new byte[4][Nb];
state = in; // actual component-wise copy
AddRoundKey(state, w, Nr*Nb, (Nr+1)*Nb - 1);
for (int round = Nr-1; round >= 1; round--) {
InvShiftRows(state);
InvSubBytes(state);
AddRoundKey(state, w, round*Nb, (round+1)*Nb-1);
InvMixColumns(state);
}
InvShiftRows(state);
InvSubBytes(state);
AddRoundKey(state, w, 0, Nb - 1);
out = state; // component-wise copy
24.3 Implementation of InvShiftRows().

This just does the inverse of ShiftRows: doing a left circular shift of rows 1, 2, and 3, by
amounts of 1, 2, and 3 bytes. The actual Java code below does this.
void InvShiftRows(byte[][] state) {
for (int r = 1; r < 4; r++) {
for (int c = 0; c < Nb; c++)
t[(c + r)%Nb] = state[r][c];
for (int c = 0; c < Nb; c++)
state[r][c] = t[c];
}
}
24.4 Implementation of InvMixColumns().

The MixColumns() function was carefully constructed so that it has an inverse. I will add in

the theory of this here (or elsewhere) later. For now, it suffices to say that the function multiplied
each column by the inverse polynomial of :
7 % /
/ @ / : 1
The resulting function, when simplified, takes the following form in Java pseudo-code,
where as before # indicates multiplication in the field:
void InvMixColumns(byte[][] s) {
byte[] sp = new byte[4];
for (int c = 0; c < 4; c++) {
sp[0] = (0x0e # s[0][c]) ˆ (0x0b # s[1][c]) ˆ
(0x0d # s[2][c]) ˆ (0x09 # s[3][c]);
sp[1] = (0x09 # s[0][c]) ˆ (0x0e # s[1][c]) ˆ
(0x0b # s[2][c]) ˆ (0x0d # s[3][c]);
sp[2] = (0x0d # s[0][c]) ˆ (0x09 # s[1][c]) ˆ
(0x0e # s[2][c]) ˆ (0x0b # s[3][c]);

sp[3] = (0x0b # s[0][c]) ˆ (0x0d # s[1][c]) ˆ
(0x09 # s[2][c]) ˆ (0x0e # s[3][c]);
for (int i = 0; i < 4; i++) s[i][c] = sp[i];
}
}
24.5 Implementation of InvAddRoundKey().

Since the AES specification uses a parameterized AddRoundKey() function, it is its own
inverse, using the parameters in the opposite order. My implementation just lets AddRound-
Key() exclusive-or in another 16 bytes every time it is called, so I need a slightly different
function, where wCount is initialized to 4*Nb*(Nr+1):
void InvAddRoundKey(byte[][] state) {
for (int c = Nb - 1; c >= 0; c--)
for (int r = 3; r >= 0 ; r--)
state[r][c] = state[r][c] ˆ w[--wCount];
}
24.6 Java Implementation of Decryption Using AES.

As before, it’s a matter of putting it all together, with a number of details to make the Java
work correctly. My Java implementation uses the old Tables, GetBytes, Copy, and Print
classes along with the new classes:
AESdecrypt, which provides the principle functions for AES decryption, and
AESinvTest, to test decryption.
A combined listing of all the decryption classes appears on page 290.

Test runs of the program, where encryption is followed by decryption, appear on page
293.
Part VII
Identification and Key
Distribution
140
25
Passwords
25.1 Passwords.
Passwords are at present the most common method for verifying the identity of a user. This
is a flawed method; systems continue to use passwords because of their ease of use and ease
of implementation. Among many problems are the successful guessing of user’s passwords,
and the intercepting of them or uncovering them online. To prevent guessing and for additional
security, the NSA recommends using a random 8-letter password that is regularly changed.
Since such a stream of passwords is almost impossible to remember (certainly for me), the
hapless user is forced to write these passwords down, adding to the insecurity. Thus passwords
need to be protected by cryptographic techniques, whether they are stored or transmitted.
Several simple techniques can help make the old-fashioned form of passwords easier to
memorize. First, the system can present a user with a list of possible random passwords from
which to choose. With such a choice, there may be one password that is easier for a given user
to remember. Second, the most common passwords are limited to 8 characters, and experience
has shown that users have a hard time picking such a short password that turns out to be secure.
If the system allows passwords of arbitrary length (fairly common now), then users can employ
pass phrases: a phrase or sentence that is not going to be in dictionaries yet is easy for the given
user to remember. My favorite pass phrase is “Dexter’s mother’s bread,” but I won’t be able to
use it any more.
Personal physical characteristics form the basis for a number of identification methods now
in use. The characteristics or biometrics range from fingerprints to iris patterns, from voice to
hand geometry, among many examples. These techniques are outside the scope of this book.
The remaining two sections study two uses of one-way functions to help secure passwords.
25.2 Password Files.

A simple system password scheme would just have a secret file holding each user’s account
name and the corresponding password. There are several problems with this method: if some-
one manages to read this file, they can immediately pretend to be any of the users listed. Also,
someone might find out about a user’s likely passwords from passwords used in the past.
For the reasons above and others, early Unix systems protected passwords with a one-way
function (described in an earlier chapter). Along with the account name, the one-way function
applied to the password is stored. Thus given a user (Alice, of course), with account name
and password , and given a fixed one-way function , the system would store
and
142 VII. Identification and Key Distribution

as a table entry in the password file, with similar entries for other users. When Alice
supplies her password to the system, the software computes of her password and compares
this result with the table entry. In this way the systems administrators themselves will not know
the passwords of users and will not be able to impersonate a user.
In early Unix systems it was a matter of pride to make the password file world readable.
A user would try to guess other’s passwords by trying a guess : first calculate and

then compare this with all table entries. There were many values of to try, such as entries

in a dictionary, common names, special entries that are often used as passwords, all short
passwords, and all the above possibilities with special characters at the beginning or the end.
These “cracker” programs have matured to the point where they can always find at least some
passwords if there are quite a few users in the system. Now the password file is no longer
public, but someone with root privileges can still get to it, and it sometimes leaks out in other
ways.
To make the attack in the previous paragraph harder (that attack is essentially the same as
ciphertext searching), systems can first choose the one-way function to be more execution-
time intensive. This only slows down the searches by a linear factor. Another approach uses
an additional random table entry, called a salt. Suppose for example that each password table
entry has another random -bit field (the salt), different for each password. When Alice first
puts her password into the system (or changes it), she supplies . The system chooses the salt

and calculates 1
, where is fixed up to handle two inputs instead of one.
The password file entry for Alice now contains , , and . With this change, an attack on

a single user is the same, but the attack of the previous paragraph on all users at the same time
now takes either an extra factor of time equal to either or the number of users, whichever is

smaller. Without the salt, an attacker could check if “Dexter” were the password of any user

by calculating “Dexter” and doing a fast search of the password file for this entry. With the
salt, to check if Alice is using “Dexter” for example, the attacker must retrieve Alice’s salt

and calculate “Dexter”

1
. Each user requires a different calculation, so this simple device
greatly slows down the dictionary attack.
25.3 Lamport’s Scheme.

The schemes mentioned above are useless if an opponent can intercept Alice’s communication
with the computer when she is trying to log in. She would be sending her password in the clear
(without encryption). There are schemes to use encryption to protect the password transmission,
but in 1981 Lamport developed a simple scheme that is proof against eavesdroppers.
With Lamport’s scheme, Alice first decides how many times she wants to log in without
redoing the system, say,
times. She chooses a random secret initial value , and uses a
&
1 1$=$=$=1 1
one-way function to calculate
%'&&& 1

% means , , %'and &&&

where
& & & & so forth. Alice wants to
%'&&&

authenticate herself repeatedly with the computer system

. She must transfer to
25. Passwords 143
in some secure way, possibly by just personally dropping it off. This value is stored in the

@ @@
computer system along with a counter value of . When Alice wants to authenticate herself
to
, she sends and to
. Now
calculates
'% &&&
and compares it
%'&&&

with the value already stored, authenticating Alice the first time. Now
replaces the
old values with and @@@
. Only Alice knows any value except
%'&&&
, so only Alice can

calculate and supply to
. (Alice could keep all the around, or she could calculate any
&
desired one of them from scratch starting with .) The next time Alice wants to authenticate
@@D
herself to
, she will supply and
.
Even if someone hijacked Alice’s communication with
, intercepting and pretending
to be her, they could only do it once, since they could not calculate the previous value . Al-

ice’s use of these passwords continues for up to
authentications, at which time Alice must
&
restart the whole system with a new initial value . This system is proof against eavesdrop-
ping, and even a more active opponent can only impersonate Alice once without intercepting
another of Alice’s authentications. If the authentication ever fails for Alice, she will not retry

the same , but will use 7 % >E"(
the next time.
In case of a communication breakdown, the system
may have (say) and
>E
stored,

and may receive from Alice and
% , for example. In this case
knows to calculate
>E

%
before comparing with the stored entry, and of course
will store and
as the new entry.
%
26
Zero-Knowledge Protocols
26.1 The Classes NP and NP-complete.
26.2 Zero-Knowledge Proofs.
26.3 Hamiltonian Cycles.

An NP-complete problem known as the Hamiltonian Cycle Problem gives a good illustration
of a simple zero-knowledge proof. The problem starts with an undirected graph
1
,

where is a finite set of vertices, is a set of ordered pairs of vertices forming edges, and
& 1 % 1 1$=$=$= 1 7 %
1 1
if
, , H4
is an edge, so is

1 %
. A path in is a sequence of vertices

such for each
, is an edge. A path is a cycle (ends where it starts)
7% &

if . A path is simple (doesn’t cross itself) if no vertex appears more than once. A
path is complete if it includes every vertex. The Hamiltonian Cycle Problem (HC) asks if a
given graph has a simple complete cycle. It turns out that HC is an NP-complete problem, so
in general no efficient algorithm is known.
If one had an efficient algorithm to solve HC, then one would also have an efficient algo-
rithm to actually obtain the Hamitonian cycle itself. First check the entire graph to see if there
is such a cycle. Then try deleting each edge in turn, checking again if there is still a Hamilin-
tonian cycle, until only the edges of a Hamiltonian cycle remain. (There may be more than one
such cycle, but this method will find one of them.)
For a given graph, even a large one, it may be easy to decide this problem, but there is no
known efficient algorithm to solve the general problem as the number of vertices increases.
Consider now the specific simple graph in Figure 26.1. The graph illustrated in this figure
is the same as the vertices and edges of a dodecahedron (a 12-sided regular polyhedron with
each side a pentagon). All that is present is a wire framework of the edges, and the framework
has been opened up from one side and squashed onto the plane.
This graph is not complicated, but it still takes most people at least a minute or two to
find one of the Hamiltonian cycles in the graph. Try to do it now, without peeking ahead (and
without writing in the book). Once you have found a cycle, read on. The cycle is shown later
in Figure 26.2 as a thicker set of lines.
26. Zero-Knowledge Protocols 145
15 19
5
6 14
7 13
1 4
8 12
2 3
16 18
9 11
10
17
Figure 26.1 Vertices and Edges of a Dodecahedron.

Dodecahedron Graph
Vertex Edges Cycle
0 1, 4, 5 1
1 0, 2, 7 7
2 1, 3, 9 3
3 2, 4, 11 4
4 0, 3, 13 0
5 0, 6, 14 14
6 5, 7, 15 5
7 1, 6, 8 8
8 7, 9, 16 16
9 2, 8, 10 2
10 9, 11, 17 9
11 3, 10, 12 10
12 11, 13, 18 11
13 4, 12, 14 12
14 5, 13, 19 13
15 6, 16, 19 6
16 8, 15, 17 17
17 10, 16, 18 18
18 12, 17, 19 19
19 14, 15, 18 15
Table 26.1 Dodecahedron Graph.
In computer algorithms, graphs are often represented as a list of vertices, and for each
4 @
vertex, a list of the vertices which together with the first one make up an edge. Suppose that

the vertices are numbered from to
, or in this case, to . Table 26.1 gives information
describing the graph in this way. The rightmost column also shows a Hamiltonian cycle by
giving, for each vertex, the next vertex in the cycle.
Now on to a zero-knowledge proof by Alice to Bob that she has a Hamiltonian cycle of this
graph while revealing nothing about the cycle itself.
Alice must carry out a number of steps in the process of a probablistic proof. After
almost certain that Alice has a Hamiltonian cycle for the graph, with only a
7
stages, Bob can be
probability of that she does not have a cycle, but is cheating. At each stage Alice chooses
a new (true) random permutation of the vertices in the graph. In the example below, we assume
she has chosen the rearrangement of vertices given in Table 26.2 .
Alice rearranges the table by sorting the newly renumbered vertices into order again. For
each vertex, the list of vertices at the other end of an edge must be made all the same length,
using extra dummy entries. (In the case here all the lists are already the same length.) Finally
15 19
5
6 14
7 13
1 4
8 12
2 3
16 18
9 11
10
17
Figure 26.2 Hamiltonian Cycle on a Dodecahedron.
Original vertex: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Changed to: 12 17 16 19 8 13 7 3 10 5 18 11 2 9 15 6 0 14 4 1
Table 26.2 Permutation of Vertices.

Permuted Dodecahedron Graph

Vertex Edges Cycle Permutation
0 14, 6, 10 14 16
1 6, 4, 15 6 19
2 11, 4, 9 11 12
3 10, 7, 17 10 7
4 14, 1, 2 1 18
5 16, 10, 18 16 9
6 0, 1, 7 7 15
7 3, 13, 6 13 6
8 12, 9, 19 12 4
9 15, 2, 8 2 13
10 5, 3, 0 0 8
11 19, 2, 18 18 11
12 17, 8, 13 17 0
13 12, 7, 15 15 5
14 18, 4, 0 4 17
15 1, 13, 9 9 14
16 19, 17, 5 19 2
17 12, 16, 3 3 1
18 11, 5, 14 5 10
19 16, 8, 11 8 3
Table 26.3 Permuted Dodecahedron Graph.
she must randomize the order of these lists. Table 26.3 shows the results of all these changes.
Alice needs a means to selectively reveal parts of this last table, but she has to commit to the
entire table without being able to cheat. This is easy to do with any cryptosystem, such as the
AES, for example. For each item in the table needing concealment, use a separate encryption
with a separate key. (Or similar items can be handled with a single key, as long as an extra
random string is included to foil ciphertext comparisons.) Alice sends all the information in the
table, with each item encrypted. Notice that if Alice encrypts the vertex number 17 with some
AES key, it is not feasible for her to find another key that decrypts the same ciphertext to some
other number, say 16, assuming there are a large number of redundant bits in the ciphertext.
So Alice sends the entire table in permuted and encrypted form to Bob. Bob then asks Alice
to show him one of two parts of the table:
the original graph, or

the Hamiltonian cycle.
In the second case, Alice sends the keys needed to reveal only the parts of the table in boldface.
(The vertex numbers are implicit, starting at 0.) This shows each succesive vertex in the cycle,
and shows that the vertex is actually on an edge. All this is assuming that the hidden graph is
really the correct one. If Alice wanted to cheat, it would be easy for her to make up numbers
so that it would look like a Hamiltonian cycle was present.
In the second case, Alice sends the keys needed to reveal all parts of the table except for the
column labeled “cycle”. The extra column labeled “permutation” is there to make it easy for
Bob to verify that the permuted graph is still the same as the original. (There are no efficient
algorithms to show that two graphs are really the same.)
If Alice doesn’t know a cycle, she can cheat and answer either question correctly, but she
has to commit herself to an answer before knowing what part Bob will ask for. In the first case
she reveals nothing at all about how the Hamiltonian cycle is situated in the graph, but only
that it exists. In the second case she reveals nothing at all about the Hamiltonian cycle. At each
stage of this probablistic proof, Alice chooses a new permutation of the vertices, rearranges
the table again, and chooses new keys for encrypting. At each step, if she is cheating, she will
caught half the time on the average.
Alice could conceivably use HC to establish her identity. First she constructs a large,

complex graph , making sure at each stage of the construction that has a Hamiltonian

cycle. Alice securely transfers the graph to a trusted server . Then if Alice wanted to prove
her identity to Bob, she could have Bob retrieve the graph from , and then she could prove
(probablistically, in zero knowledge) that she has a Hamiltonian cycle. Even an eavesdropper
would not learn the Hamiltonian cycle. This particular method has two separate flaws as a way
for Alice to identify herself:
The method is subject to replay attacks, where Boris just sends someone else exactly the
same messages Alice transmits as a way to “prove” that he is Alice.
It is necessary for Alice to create a “hard” instance of HC, that is, an instance where
known algorithms take a very long time to find a Hamiltonian cycle.
Alice could handle the first flaw by adding a timestamp to the part of the graph that she
encrypts. The second flaw above is the difficult one, since finding hard instances of an NP-
complete problem is often a difficult computation itself, and that especially seems to be the
case with HC. The next chapter shows how to use factoring as the hard problem, and factoring
has been studied much more extensively than HC.
26.4 Other Zero-Knowledge Proofs.

27
Identification Schemes
27.1 Use of Public Keys.
27.2 Zero-Knowledge Schemes.

This section considers protocols for proving identity, starting with the Fiat-Shamir Identifica-
tion Protocol, a zero-knowledge proof. The Handbook of Applied Cryptography lists a number
of possible attacks on identification protocols (Section 10.6), including impersonation, replay,
interleaving, reflection, and forced delay. The impersonation attack is the standard attempt to
try to break an identification system, and will be discussed below. A replay attack is also stan-
dard, and is kept from working by the nature of the challenge-response protocol considered in
this section.
The interleaving attack can be very powerful, since there are so many possibilities when
two of more protocols are carried out in a concurrent fashion. To give just one example of this
difficulty, consider the following strategy for an low-ranked chess player to improve his ranking
at online or correspondence chess. The player (call
him signs up for two games with two
high-ranked players (call them and
), where plays black and
plays white. Player
waits for
to make the first move as white. Then plays this move in his game with . Then
waits for ’s response as black, and makes this response to
’s move. Continuing in this
way, is guaranteed either a win and a loss or two draws against the good players, and either
outcome will improve ’s ranking.
There is another attack which I call the repetitive divide-and-conquer strategy. This is
illustrated with a simple scam used by gamblers. Suppose every Sunday there are sports games
which people bet on. Suppose the mark (call him ), gets a letter early in the week predicting
the outcome of the game on the next Sunday. doesn’t think much about it, but notices that the
prediction is correct. The next week, another prediction comes in the mail that also is correct.
This continues for ten weeks, each time a correct prediction of the game outcome arriving days
before the game. At this point receives another letter, offering to predict an eleventh game
for $10000 dollars. The letter guarantees that the prediction will be correct, and suggests that
will be able to make far more than this amount of money by betting on the game. The final
prediction might or might not be correct, but was set up. In fact, the letter writer started
with (say) 4096 letters, half predicting Sunday’s game one way, and half predicting it the other
way. After the first game, roughly 2048 of the recipients got a wrong prediction, but they get
27. Identification Schemes 151
no further letters, because the letter writer ignores them, focusing on the (roughly) 2048 people
who got a correct first prediction. At the end of 10 weeks, and after writing approximately 8184
letters, one would expect an average of 4 individuals to have received 10 correct predictions,
including above.
Here is the protocol:

Protocol: Fiat-Schamir Identification. (Alice) proves knowledge of a secret to (Bob),

using a trusted center .
1. Setup for : The trusted center chooses large random primes and and forms the
product

. (These should fit the requirements for the RSA cryptosystem.)

publishes
for all users, and keeps and secret, or just destroys them.

, , 4
2. Setup for each user: Each user chooses a random secret satisfying
which has no factors in common with
. Then calculates mod
and lets store

this value in public. (Each user has his own unique secret value and each user stores a
unique square of this value with . Notice that no secrecy is needed at , but users need
to rely on the integrity of information supplied by . If someone wants to calculate the

secret just knowing the public value , they would have to take the discrete square root
of , and this problem is computationally equivalent to factoring
, as was discussing in
the chapter on Rabin’s cryptosystem.)
, , 4
3. Witness step: chooses a random number satisfying
committment, calculates

mod
, and sends to .
to use as a
4. Challenge step:

chooses a random challenge bit and sends it to : .

5. Response step: In case , sends : back to

. In case : , calculates
mod
, and sends this back to .

:
6. Verification step: verifies the response from by calculating mod
. If

:
, then should be equal to the that was first sent, that is, is a square root of the

value , a calculation that is only efficiently

possible for . If , then should be
equal to mod
mod
, and must verify this equality.
At each iteration of the above protocol, the chances that is not who she claims to be and
62 &
is in fact cheating goes down by a factor of 2. One should repeat the protocol at least 50 times,
to reduce the probability of successfully cheating to less than .
As part of the analysis
of this protocol, suppose someone else, say
(Carlos), wished to
pretend to be to . We are not assuming that
can intercept an ongoing protocol between

and , but we do assume that
can communicate with , claiming to be . Of course
can get ’s public value from , but

does not know ’s secret .
could try to guess

what ’s challenge bit might be:

If
guesses that will respond with a challenge of

, then
would only need to :
send a random mod
just like , and respond to the challenge with the original
. would accept this round of the protocol.
If
guesses that

will respond with a challenge of :

, then
must provide a

response so that when calculates mod
, this is the same as mod
, where
is the initial value sent by
to , and is ’s public value that anyone (including
)

can look up. Notice that
doesn’t need to send the square of a random value at the first

step, but only needs to send a value such that later
can send another value , with the
property that when calculates mod
and
same. In other words, it needs to be true that mod

mod
, these will work out to be the
mod
, or that 7%
7%
,
where is the inverse of using multiplication mod
. ( has such an inverse because
, and therefore have no factors in common with
.) Thus the initial that
should
send in this case should be the square of some random times the inverse of . If
sends this value and if challenges with the bit : , then

will be able to cheat
that time. The problem for
is that if he sends such a value for the initial
successfully
: 7%
, and if challenges with
value, but will not be able to calculate the square root of =
,
should respond with the square root of the original
, and so
will fail
at this step.
In summary, a cheating

could succeed in either part of the protocol if he knew ahead of
time which challenge bit would send back.
can choose the initial value to succeed either
way, but no single value succeeds both ways for
. Thus
’s cheating is detected half of the
time at each iteration of the protocol.
28
Threshold Schemes
28.1 Introduction to Threshold Schemes.

When all security depends on a single secret key, this key becomes a point of attack and a
source of vulnerability. It would be better to have several people share the responsibility of
>
maintaining secret keys, and the schemes in this chapter are just what is needed.
Suppose there are executives in a company that maintains a locked room with company
(
secrets inside. When the company was incorporated, the founders wanted to make sure that
only a majority (that is, at least ) of the executives could together gain access to the room. For
this purpose, the founders created a special steel door with a sliding rod on its outside. The
rod slides through holes in two flanges on the door and through a hole in another flange beside
>
the door. The rod in turn has a ring welded onto it so that it cannot be removed from the two
flanges on the door. Between the door’s flanges, there are special padlocks around the rod.
The dimensions are chosen so that if any four of the six padlocks are removed, the rod will slide
just far enough so that the door will open. If fewer than four padlocks are removed, the rod will
still barely catch, and the door is still locked. (See the figure for an illustration of this Rube
Goldberg contraption.) Each executive is given a key to one of the locks, and then everyone can
( >
be sure that only four of more of the executives of the company, working together, can open the
1
door. (There is nothing special about the numbers and in this example.)

In general, a
threshold scheme starts with a secret , with
users who can share the
,
Each user is given a share of the secret: , for , ,
secret, and with a threshold value of of the users needed to recover the secret, where

.

. Then the threshold scheme must
4
somehow arrange that any of these shares can be used to produce (or compute) the secret,
whereas any (81>
or fewer will not allow the secret to be recovered. The door with its rod and
locks above give a simple example of a (or “four out of six”) threshold scheme, except
that its crude mechanical nature is limiting.
28.2 (t, t) Threshold Schemes.

Suppose one wants to hide information in computer storage, say a secret book , regarding it
as a string of bits. Given a source of true random numbers, this is easy to do. Create a true
%
%
random file

with as many bits in it as the information to be secured. Let
%
.
%
Now keep the files and separate from one another, and the book is safe. To recover the
book, just calculate . An opponent who knows only or
%
has no information
about the book . This is obvious for
%
, since it was true random and had nothing to do
with . The file was defined using , but since it is the exclusive-or of a true random bit
Thus by themselves, each of and

string, it also has all the properties of a true random bit string, assuming it is taken in isolation.
% is true random and each gives no information about
the book . However the two are not independent of one another. The two together include
the complete information in . If were some illegal book, then just possessing or
%
an example of a
<1
could not possibly be illegal, and only possessing them both together could be illegal. This is
threshold scheme. This scheme gives perfect security: just knowing one
1
or the other share gives no information at all about the book.
threshold scheme, starting with the book to be kept
4 % 1 % 1$==$=1 7 4 %
% F=$==
This method extends easily to a
secret, and with random files
Then the s work as shares, where again any
. Define

of them give no information about the
. % 7%
book (all books of the same size are possible and equally likely), but the exclusive-or of all
of them gives immediately.
28.3 Threshold Schemes Using Planes in Space.

This section describes one method for obtaining a threshold scheme without giving a full im-
plementation. The purpose is to increase intuition about this concept.
<1
A line in the plane is uniquely determined by points on it. This property allows the

threshold scheme. Suppose the secret is a real number. Associate

< 1 / 1 8 1
construction of a
in the plane. Choose a non-vertical line through at 1

the secret with the point

random, of the form . Each user gets the point as his
1
share. From the information that any one user has, any value for is possible, because the
8 1
line might go through and
two points uniquely determine the line , and setting

for any . However, if two users get together, their

in the equation for the line gives
immediately.
A A81
Similarly a plane in -dimentional space is uniquely determined by any three points on it.
Just as in the previous paragraph, this leads to a
threshold scheme. The same thing can be
% 1 1 1$=$== 1

7 % /=$=$= / % % /
done in -dimensional space, consisting of all ordered lists of numbers: .
4 7 %
Then a random linear equation of the form specifies a
hyperplane in this space. As above, with only points on this hyperplane, any value of
is still possible, but with points there will be equations in unknowns (the coefficients
along with ), which can be solved for all unknowns including the crucial .
The scheme in this section can be made to work, but I will not finish it, instead turning
to another approach. The method of this section would need to switch from floating point
numbers, and would still need to decide what picking an equation “at random” might mean.
28.4 Shamir’s (t, n) Threshold Scheme.
4
Adi Shamir developed a threshold scheme based on polynomial interpolation. The scheme is
based on the fact that a polynomial function of degree is uniquely determined by any
28. Threshold Schemes 155
points on it.
A
4 A /
Example. Suppose

and the “secret” is . Choose a “random” polynomial
1 1 A81 1 (81> 1 6 2 3 1 A 2)( 1$=$=$=
. Then , the secret. Points on the graph of this function yield shares, for
% / /
example: Using only the first three shares and assuming an
equation of the form , one gets three linear equations in three unknowns
by plugging in the values from the three shares.
% / /
@ / A /
> > %% /0( /
These equations are easy to solve, for , 4 A , and , and so just from the
shares, the secret is seen to be .
%
Here is the general case of Shamir’s 1
threshold scheme:
integers and , ,
.
1. Start with a secret , a desired number of shares
, and a threshold , where all three are

2. Choose
a prime bigger than both and
. Everything will be done over the finite field
7 % 7 %/ =$== / / % /
.
4
4
3. Choose a random polynomial of degree :
by choosing the coefficients

uniformly and at random from the interval from to
inclusive, that is, from .

shares as points 1
4. Compute

on the graph of . (The -coordinates do not have
to be consecutive integers, but no -coordinate can be zero, since that would immediately
reveal the secret.) These shares are distributed securely to each of the
users.
5. If any users get together with their shares, they know distinct points on the polyno-
mial’s graph, and so the users can compute the polynomial’s coefficients, including the
constant term, which is the secret.
There are a number of ways to calculate the polynomial from the shares in the situation

above, but perhaps the most convenient way is to use the Lagrange interpolation formula: 1 , ,
A polynomial

of degree is uniquely determined by points,

assuming that the are all distinct. The polynomial is given by the formula:
for ,
4
=

4

% %

Here the Greek (sigma) means to add up terms obtained by setting . 1 31$=$=$=1
Similarly the Greek (pi) means to multiply terms following it together, where takes on

values from to , but leaving off in each case. The computations are all done in , that is,

4
4
modulo . In the case of the in the denominator, one needs to use the multiplicative
inverse of in .
1 4
, ,
It is easy to see that this formula works, because it is a degree

polynomial in that
agrees with each of the points for
A
, and this polynomial must be unique. To
get a feeling for the equation, look at a special case, say, . Then the equation is
4 4 4 % 4 4 % 4
% % 4 % 4 4 % 4 4 % 4 =
/ /
Now evaluate , that is, set % throughout. The multipliers of % both reduce to ,

%
is a times each of and , so this reduces to just . Similarly,
and
there
while
. The general formula behaves in exactly the same way. %
It is not necessary to determine the entire random polynomial , but only , which is
the secret . By substituting above, the equations become
1 where

%

4
=
%

This is a particularly easy way to calculate the secret from the shares.
28.5 Properties of Shamir’s Scheme.

Shamir’s approach to thresholds has a number of pleasant properties (see the Handbook of
Applied Cryptography):
4
4
1. It is perfect is the sense that the secret can be computed from shares, but even
shares gives no information about the secret. In other words, given shares, all
possible values for the secret are still equally probable.
2. One can calculate new shares and distribute them to new users along with the ones already
passed out.
3. One can distribute more than one share to a user and in this way give that user more
power over the secret.
4. In case the secret is too large for the convenient computer arithmetic of a given imple-
mentation, the secret can be broken into two or more blocks, and security will still be
perfect. Thus there is no reason for large integers and extended precision arithmetic in
this example.
Law THRESHOLD-1:
Shamir’s (t, n) threshold scheme gives perfect security for
a shared secret, since t users can recover the secret, while t – 1
or fewer users still have no information about the secret.
28.6 Java Implementation of Shamir’s Scheme.

An implementation of Shamir’s scheme appears on page 299. Here is the output of a run of the
software, where the parameters on the command line are s t n p, in that order:
% java ThresholdTest 88 3 5 97
New (3,5) threshold scheme, with p = 97 and s = 88
Function f(x) = 88*xˆ0 + 53*xˆ1 + 43*xˆ2
All 5 Output Shares: (1,87) (2,75) (3,52) (4,18) (5,70)
Recover secret from t = 3 shares, with p = 97

All 3 Input Shares: (4,18) (2,75) (3,52)
C[0] = 2/(2-4) ( or 96) 3/(3-4) ( or 94) = 3
C[1] = 4/(4-2) ( or 2) 3/(3-2) ( or 3) = 6
C[2] = 4/(4-3) ( or 4) 2/(2-3) ( or 95) = 89
Secret = 3*18 + 6*75 + 89*52 = 88
A
Here is another larger run. (By an unlikely coincidence, the values of the polynomial below are

symmetric going down from and up from . Something like this will always be true, but it is
surprising that it appears here for such small values of .)

All 3 Input Shares: (4,1453) (2,1396) (3,1453)
C[0] = 2/(2-4) ( or 1998) 3/(3-4) ( or 1996) = 3
C[1] = 4/(4-2) ( or 2) 3/(3-2) ( or 3) = 6
C[2] = 4/(4-3) ( or 4) 2/(2-3) ( or 1997) = 1991
Secret = 3*1453 + 6*1396 + 1991*1453 = 1111
Yet another run. This would overflow with 32-bit Java int type, but works because the Java
64-bit long type is used.

All 3 Input Shares: (4,217355) (2,7586850) (3,6783370)
C[0] = 2/(2-4) ( or 10316016) 3/(3-4) ( or 10316014) = 3
C[1] = 4/(4-2) ( or 2) 3/(3-2) ( or 3) = 6
158 . Identification and Key Distribution
C[2] = 4/(4-3) ( or 4) 2/(2-3) ( or 10316015) = 10316009

Secret = 3*217355 + 6*7586850 + 10316009*6783370 = 2222222
Java
Programs
160
Program I.1.a
Demonstration of Xor
Referred to from page 4.
The code below shows how xor can be used to interchange data elements.
Java class: Xor
// Xor.java: test xor function ˆ for interchanges
public class Xor {
// main function to try out Base class

public static void main (String[] args) {
int a = 123456789, b = -987654321;
printThem(a, b);
// interchange a and b
a = aˆb;
b = aˆb;
a = aˆb;
printThem(a, b);
a = 234234234; b = -789789789;
printThem(a, b);
// interchange a and b
a ˆ= b;
b ˆ= a;
a ˆ= b;
printThem(a, b);
} // end of main
private static void printThem(int a, int b) {

System.out.println("a: " + a + ", \tb: " + b);
}
}
Here is the output of a run:

% java Xor
a: 123456789, b: -987654321
a: -987654321, b: 123456789
a: 234234234, b: -789789789
a: -789789789, b: 234234234
Program I.1.b
Formulas for Logarithms
Java supplies a function to calculate natural logs, base : 3= B $D D D D(!E@ (!E
. To calculate
;
62
logs to other bases, you need to multiply by a fixed constant: for a log base multiply by
.
Java class: Logs
// Logs.java: try out logarithm formulas
public class Logs {
// main function to try out Logs class

System.out.println("log base 2 of 1024 = " + log2(1024));
System.out.println("log 2 = " + Math.log(2));
System.out.println("1/log 2 = " + 1/Math.log(2));
} // end of main
// log2: Logarithm base 2

public static double log2(double d) {
return Math.log(d)/Math.log(2.0);
}

}
}
Here is the output of a run:

% java Logs
log base 2 of 1024 = 10.0
log base 10 of 1000 = 2.9999999999999996
log 2 = 0.6931471805599453
1/log 2 = 1.4426950408889634
log base 10 of 2 = 0.30102999566398114
Program I.1.c
Fermat’s Theorem Illustrated

7%

Recall that Fermat’s theorem says that given a prime and a non-zero number , mod

is always equal to . Here is a table for illustrating this theorem. Notice below that the

value is always by the time the power gets to , but sometimes the value gets to earlier.

The initial run up to the value is shown in boldface in the table. A value of for which the
%
whole row is bold is called a generator. In this case 2, 6, 7, and 8 are generators.
%'&
11 2 2 4 8 5 10 9 7 3 6 1
11 3 3 9 5 4 1 3 9 5 4 1
11 4 4 5 9 3 1 4 5 9 3 1
11 5 5 3 4 9 1 5 3 4 9 1
11 6 6 3 7 9 10 5 8 4 2 1
11 7 7 5 2 3 10 4 6 9 8 1
11 8 8 9 6 4 10 3 2 5 7 1
11 9 9 4 3 5 1 9 4 3 5 1
11 10 10 1 10 1 10 1 10 1 10 1
Java code to produce the table above and the one below.
Java class: Fermat
// Fermat.java: given a prime integer p, calculate powers of a
// fixed element a mod p. Output html table
public class Fermat {
// main function to do all the work

long p = (Long.parseLong(args[0])); // the fixed prime base
System.out.println("<table border nosave >");
System.out.println("<tr><th>p</th><th>a</th><th></th>");
for (int col = 1; col < p; col++)
System.out.print("<th>a<sup>" + col + "</sup></th>");
System.out.println("</tr><tr colspan=" + (p+2) + "></tr>");
for (long row = 2; row < p; row++) {
System.out.print("<tr align=right><td>" + p);
System.out.print("</td><td>" + row + "</td><td></td>");
boolean firstCycle = true;
for (long col = 1; col < p; col++) {
if (firstCycle)
System.out.print("<td><b><font color=FF0000>" +
pow(row, col, p) + "</font></b></td>");
else
System.out.print("<td>" + pow(row, col, p) + "</td>");
164 Program I.1.c
if (firstCycle)
if (pow(row, col, p) == 1) firstCycle = false;
}
System.out.println("</tr>");
}
System.out.println("</table>");
} // end of main
// pow: calculate xˆy mod p, without overflowing

// (Algorithm from Gries, The Science of Programming, p. 240
public static long pow(long x, long y, long p) {
long z = 1;
while (y > 0) {
while (y%2 == 0) {
x = (x*x) % p;
y = y/2;
}
z = (z*x) % p;
y = y - 1;
}
return z;
}
}
Here is a larger table with p = 23. There are 10 generators.

23 2 2 4 8 16 9 18 13 3 6 12 1 2 4 8 16 9 18 13 3 6 12 1
23 3 3 9 4 12 13 16 2 6 18 8 1 3 9 4 12 13 16 2 6 18 8 1
23 4 4 16 18 3 12 2 8 9 13 6 1 4 16 18 3 12 2 8 9 13 6 1
23 5 5 2 10 4 20 8 17 16 11 9 22 18 21 13 19 3 15 6 7 12 14 1
23 6 6 13 9 8 2 12 3 18 16 4 1 6 13 9 8 2 12 3 18 16 4 1
23 7 7 3 21 9 17 4 5 12 15 13 22 16 20 2 14 6 19 18 11 8 10 1
23 8 8 18 6 2 16 13 12 4 9 3 1 8 18 6 2 16 13 12 4 9 3 1
23 9 9 12 16 6 8 3 4 13 2 18 1 9 12 16 6 8 3 4 13 2 18 1
23 10 10 8 11 18 19 6 14 2 20 16 22 13 15 12 5 4 17 9 21 3 7 1
23 11 11 6 20 13 5 9 7 8 19 2 22 12 17 3 10 18 14 16 15 4 21 1
23 12 12 6 3 13 18 9 16 8 4 2 1 12 6 3 13 18 9 16 8 4 2 1
23 13 13 8 12 18 4 6 9 2 3 16 1 13 8 12 18 4 6 9 2 3 16 1
23 14 14 12 7 6 15 3 19 13 21 18 22 9 11 16 17 8 20 4 10 2 5 1
23 15 15 18 17 2 7 13 11 4 14 3 22 8 5 6 21 16 10 12 19 9 20 1
23 16 16 3 2 9 6 4 18 12 8 13 1 16 3 2 9 6 4 18 12 8 13 1
23 17 17 13 14 8 21 12 20 18 7 4 22 6 10 9 15 2 11 3 5 16 19 1
23 18 18 2 13 4 3 8 6 16 12 9 1 18 2 13 4 3 8 6 16 12 9 1
23 19 19 16 5 3 11 2 15 9 10 6 22 4 7 18 20 12 21 8 14 13 17 1
23 20 20 9 19 12 10 16 21 6 5 8 22 3 14 4 11 13 7 2 17 18 15 1
23 21 21 4 15 16 14 18 10 3 17 12 22 2 19 8 7 9 5 13 20 6 11 1
23 22 22 1 22 1 22 1 22 1 22 1 22 1 22 1 22 1 22 1 22 1 22 1
Program I.2.a
Basic GCD Algorithm
This class gives (in bold below) two versions of the simple greatest common division algorithm:
the first recursive and the second iterative.
Java class: GCD
// GCD: Greatest Common Divisor
public class GCD {

if (y == 0) return x;
return gcd1(y, x % y);
}

while (y != 0) {
long r = x % y;
x = y; y = r;
}
return x;
}
public static void main(String[] args) {

long x = Long.parseLong(args[0]);
long y = Long.parseLong(args[1]);
long z = GCD.gcd1(x, y);
System.out.println("Method 1: gcd(" + x + ", " +
y + ") = " + z);
z = GCD.gcd2(x, y);
System.out.println("Method 2: gcd(" + x + ", " +
y + ") = " + z);
}
}
Here are several runs of this program:

% java GCD 819 462
Method 1: gcd(819, 462) = 21
Method 2: gcd(819, 462) = 21
% java GCD 40902 24140
Method 1: gcd(40902, 24140) = 34
Method 2: gcd(40902, 24140) = 34
Program I.2.b
Extended GCD Algorithm
This extended greatest common divison algorithm is the version in Knuth’s Seminumerical
Algorithms, Third Edition. When the algorithm finishes, x*u[0] + y*u[1] = u[2] =
gcd(x, y).
Java class: ExtGCDsimple
// ExtGCDsimple: Extended GCD
public class ExtGCDsimple {

while (v[2] != 0) {
long q = u[2]/v[2];
for (int i = 0; i < 3; i++) {
t[i] = u[i] -v[i]*q; u[i] = v[i]; v[i] = t[i];
}
}
return u;
}

u = ExtGCDsimple.GCD(x, y);
System.out.println("gcd(" + x + ", " + y + ") = " + u[2]);
System.out.println("(" + u[0] + ")*" + x + " + " +
"(" + u[1] + ")*" + y + " = " + u[2]);
}
}
Here are several runs of this program:

% java ExtGCDsimple 819 462
gcd(819, 462) = 21
(-9)*819 + (16)*462 = 21
% java ExtGCDsimple 40902 24140
gcd(40902, 24140) = 34
(337)*40902 + (-571)*24140 = 34
Program I.2.c
Extended GCD Algorithm (debug version)
The long (debug oriented) version of this program was discussed in the text of the book.
Java class: ExtGCD
// ExtGCD: Extended GCD (long version)
public class ExtGCD {
public static long[] GCD(long x, long y) { // assume not 0 or neg

long[] v = new long[3];
long[] t = new long[3];
// at all stages, if w is any of the 3 vectors u, v or t, then
// x*w[0] + y*w[1] = w[2] (this is verified by "check" below)
// u = (1, 0, u); v = (0, 1, v);
u[0] = 1; u[1] = 0; u[2] = x; v[0] = 0; v[1] = 1; v[2] = y;
System.out.println("q\tu[0]\tu[1]\tu[2]\tv[0]\tv[1]\tv[2]");
while (v[2] != 0) {
long q = u[2]/v[2];
// t = u - v*q;
t[0] = u[0] -v[0]*q; t[1] = u[1] -v[1]*q; t[2] = u[2] -v[2]*q;
check(x, y, t);
// u = v;
u[0] = v[0]; u[1] = v[1]; u[2] = v[2]; check(x, y, u);
// v = t;
v[0] = t[0]; v[1] = t[1]; v[2] = t[2]; check(x, y, v);
System.out.println(q + "\t"+ u[0] + "\t" + u[1] + "\t" + u[2] +
"\t"+ v[0] + "\t" + v[1] + "\t" + v[2]);
}
return u;
}
public static void check(long x, long y, long[] w) {

if (x*w[0] + y*w[1] != w[2]) {
System.out.println("*** Check fails: " + x + " " + y);
System.exit(1);
}
}

u = ExtGCD.GCD(x, y);
System.out.println("\ngcd(" + x + ", " + y + ") = " + u[2]);
168 Program I.2.c
System.out.println("(" + u[0] + ")*" + x + " + " +

"(" + u[1] + ")*" + y + " = " + u[2]);
}
}
Here is a sample run (with a few extra tabs inserted by hand):

% java ExtGCD 123456789 987654321
q u[0] u[1] u[2] v[0] v[1] v[2]
0 0 1 987654321 1 0 123456789
8 1 0 123456789 -8 1 9
13717421 -8 1 9 109739369 -13717421 0
gcd(123456789, 987654321) = 9
(-8)*123456789 + (1)*987654321 = 9
% java ExtGCD 1122334455667788 99887766554433

q u[0] u[1] u[2] v[0] v[1] v[2]
11 0 1 99887766554433 1 -11 23569023569025

4 1 -11 23569023569025 -4 45 5611672278333
4 -4 45 5611672278333 17 -191 1122334455693
4 17 -191 1122334455693 -72 809 1122334455561
1 -72 809 1122334455561 89 -1000 132
8502533754 89 -1000 132 -756725504178 8502533754809 33
4 -756725504178 8502533754809 33 3026902016801 -34010135020236 0
gcd(1122334455667788, 99887766554433) = 33
(-756725504178)*1122334455667788 + (8502533754809)*99887766554433 = 33
% java ExtGCD 384736948574637 128475948374657

q u[0] u[1] u[2] v[0] v[1] v[2]
2 0 1 128475948374657 1 -2 127785051825323
1 1 -2 127785051825323 -1 3 690896549334
184 -1 3 690896549334 185 -554 660086747867
1 185 -554 660086747867 -186 557 30809801467
21 -186 557 30809801467 4091 -12251 13080917060
2 4091 -12251 13080917060 -8368 25059 4647967347
2 -8368 25059 4647967347 20827 -62369 3784982366
1 20827 -62369 3784982366 -29195 87428 862984981
4 -29195 87428 862984981 137607 -412081 333042442
2 137607 -412081 333042442 -304409 911590 196900097
1 -304409 911590 196900097 442016 -1323671 136142345
1 442016 -1323671 136142345 -746425 2235261 60757752
2 -746425 2235261 60757752 1934866 -5794193 14626841
4 1934866 -5794193 14626841 -8485889 25412033 2250388
6 -8485889 25412033 2250388 52850200 -158266391 1124513
2 52850200 -158266391 1124513 -114186289 341944815 1362
825 -114186289 341944815 1362 94256538625 -282262738766 863
1 94256538625 -282262738766 863 -94370724914 282604683581 499
1 -94370724914 282604683581 499 188627263539 -564867422347 364
1 188627263539 -564867422347 364 -282997988453 847472105928 135
2 -282997988453 847472105928 135 754623240445 -2259811634203 94
1 754623240445 -2259811634203 94 -1037621228898 3107283740131 41
2 -1037621228898 3107283740131 41 2829865698241 -8474379114465 12
3 2829865698241 -8474379114465 12 -9527218323621 28530421083526 5
2 -9527218323621 28530421083526 5 21884302345483 -65535221281517 2
2 21884302345483 -65535221281517 2 -53295823014587 159600863646560 1
2 -53295823014587 159600863646560 1 28475948374657 -384736948574637 0
gcd(384736948574637, 128475948374657) = 1
(-53295823014587)*384736948574637 + (159600863646560)*128475948374657 = 1
Program I.2.d
Testing Two Exponential Algorithms
The code below the two algorithms for carrying out integer exponentiation that were presented
in the text. Each function has additional debug statements to provide extra output.
Java class: Exp
// Exp: test Java versions of two exponentiation algorithms
public class Exp {
// exp1: uses binary representation of the exponent.

// Works on binary bits from most significant to least.
// Variable y only present to give loop invariant:
// xˆy = z, and y gives the leading bits of Y.
public static long exp1(long x, long Y[], int k) {
long y = 0, z = 1;
int round = 0;
dump1("Initial. ", x, y, z);
for (int i = k; i >= 0; i--) {
y = 2*y;
z = z*z;
dump1("Round: " + (round) + ", ", x, y, z);
if (Y[i] == 1) {
y++;
z = z*x;
dump1("Round: " + (round++) + ", ", x, y, z);
}
}
return z;
}
// dump1: function to spit out debug information

private static void dump1(String s, long x, long y, long z) {
System.out.println(s + "x: " + x + ",\ty: " + y +
",\tz: " + z + ",\t(xˆy): " + (exp(x, y)));
}
// exp2: uses binary rep of exponent, without constructing it.

// Works on binary bits from least significant to most.
// Loop invariant is: z*xˆy = XˆY
public static long exp2(long X, long Y) {
long x = X, y = Y, z = 1;
dump2("Initial. ", x, y, z);
int round = 1;
while (y > 0) {
while (y%2 == 0) {
x = x*x;
y = y/2;
dump2("Round: " + (round) + ", ", x, y, z);
170 Program I.2.d
}
z = z*x;
y = y - 1;}
dump2("Round: " + (round++) + ", ", x, y, z);
{\em{}
return z;
}
// exp: extra copy of exp2 function without debug code

public static long exp(long X, long Y) {
long x = X, y = Y, z = 1;
while (y > 0) {
while (y%2 == 0) {
x = x*x;
y = y/2;
}
z = z*x;
y = y - 1;
}
return z;
}
// dump2: function to spit out debug information

private static void dump2(String s, long x, long y, long z) {
System.out.println(s + "x: " + x + ",\ty: " + y +
",\tz: " + z + ",\tz*(xˆy): " + (z*exp(x, y)));
}

// Convert y to array Y of bits
long Y[] = new long[50];
int k = 0;
long yt = y;
while (yt > 0) {
Y[k++] = yt % 2;
yt = yt/2;
}
k--;
System.out.println("Try first exponentiation algorithm ...");
long z1 = Exp.exp1(x, Y, k);
System.out.println("Method 1: exp1(" + x + ", " + y +
") = " + z1 + "\n");
System.out.println("Try second exponentiation algorithm ...");
long z2 = Exp.exp2(x, y);
System.out.println("Method 2: exp2(" + x + ", " + y +
") = " + z2);
}
}
Here are results of test runs, with a few extra blanks to improve readability:
% java Exp 3 12
Try first exponentiation algorithm ...
Initial. x: 3, y: 0, z: 1, (xˆy): 1
Round: 1, x: 3, y: 0, z: 1, (xˆy): 1
Round: 1, x: 3, y: 1, z: 3, (xˆy): 3
Round: 2, x: 3, y: 2, z: 9, (xˆy): 9
Round: 2, x: 3, y: 3, z: 27, (xˆy): 27
Round: 3, x: 3, y: 6, z: 729, (xˆy): 729
Round: 3, x: 3, y: 12, z: 531441, (xˆy): 531441
Method 1: exp1(3, 12) = 531441
Try second exponentiation algorithm ...

Initial. x: 3, y: 12, z: 1, z*(xˆy): 531441
Round: 1, x: 9, y: 6, z: 1, z*(xˆy): 531441
Round: 1, x: 81, y: 3, z: 1, z*(xˆy): 531441
Round: 1, x: 81, y: 2, z: 81, z*(xˆy): 531441
Round: 2, x: 6561, y: 1, z: 81, z*(xˆy): 531441
Round: 2, x: 6561, y: 0, z: 531441, z*(xˆy): 531441
Method 2: exp2(3, 12) = 531441
% java Exp 2 23
Try first exponentiation algorithm ...
Initial. x: 2, y: 0, z: 1, (xˆy): 1
Round: 1, x: 2, y: 0, z: 1, (xˆy): 1
Round: 1, x: 2, y: 1, z: 2, (xˆy): 2
Round: 2, x: 2, y: 2, z: 4, (xˆy): 4
Round: 2, x: 2, y: 4, z: 16, (xˆy): 16
Round: 2, x: 2, y: 5, z: 32, (xˆy): 32
Round: 3, x: 2, y: 10, z: 1024, (xˆy): 1024
Round: 3, x: 2, y: 11, z: 2048, (xˆy): 2048
Round: 4, x: 2, y: 22, z: 4194304, (xˆy): 4194304
Round: 4, x: 2, y: 23, z: 8388608, (xˆy): 8388608
Method 1: exp1(2, 23) = 8388608
Try second exponentiation algorithm ...

Initial. x: 2, y: 23, z: 1, z*(xˆy): 8388608
Round: 1, x: 2, y: 22, z: 2, z*(xˆy): 8388608
Round: 2, x: 4, y: 11, z: 2, z*(xˆy): 8388608
Round: 2, x: 4, y: 10, z: 8, z*(xˆy): 8388608
Round: 3, x: 16, y: 5, z: 8, z*(xˆy): 8388608
Round: 3, x: 16, y: 4, z: 128, z*(xˆy): 8388608
Round: 4, x: 256, y: 2, z: 128, z*(xˆy): 8388608
Round: 4, x: 65536, y: 1, z: 128, z*(xˆy): 8388608
Round: 4, x: 65536, y: 0, z: 8388608, z*(xˆy): 8388608
Method 2: exp2(2, 23) = 8388608
Program II.3.a
Formula for Channal Capacity
Program with the simple formula for channel capacity:

Java class: Capacity
// Capacity.java: calculate channel capacity, binary symmetric channel
// p: the channel probability for a binary symmetric channel
public class Capacity {
// main function to do calculation

double p = Double.parseDouble(args[0]); // channel probability
System.out.println("Probability: " + p +
", Capacity: " + capacity(p));
} // end of main
// capacity: the capacity of the binary symmetric channel

private static double capacity(double p) {
if (p == 0 || p == 1) return 1;
return 1 + p*log2(p) + (1 - p)*log2(1 - p);
}

}
}
Typical output:
% javac Capacity.java
% java Capacity 0.3
Probability: 0.3, Capacity: 0.11870910076930735
% java Capacity 0.999
% java Capacity 0.5
Probability: 0.5001, Capacity: 2.8853901046232977E-8
Probability: 0.51, Capacity: 2.8855824719009604E-4
Program II.3.b
Table of Channal Capacities
Here is the program to print an HTML table of channel capacities. The resulting table (when
interpreted by an HTML browser) will print a table that looks much like the one in the text.
Java class: CapacityTable
// CapacityTable.java: print table of capacities
import java.text.DecimalFormat;
public class CapacityTable {
static final int TABLE_SIZE = 20;
static DecimalFormat twoDigits = new DecimalFormat("0.00");
static DecimalFormat fifteenDigits = new
DecimalFormat("0.000000000000000");

double p; // channel probability
System.out.println("<table border>");
System.out.println("<tr><td><b>Probability</b></td>");
System.out.println("<td><b>Channel Capacity</b></td></tr>");
System.out.println("<tr><td></td><td></td></tr>");
for (int i = 0; i <= TABLE_SIZE/2; i++) {
p = (double)i/TABLE_SIZE;
System.out.print("<tr><td>" + twoDigits.format(p));
System.out.print(" or " + twoDigits.format(1-p));
System.out.println("</td><td>" +
fifteenDigits.format(capacity(p)) + "</td></tr>");
}
} // end of main

if (p == 0 || p == 1) return 1;
}

}
}
Here is the output, as an HTML table:

<table border>
<tr><td><b>Probability</b></td>
174 Program II.3.b
<td><b>Channel Capacity</b></td></tr>
<tr><td></td><td></td></tr>
<tr><td>0.00 or 1.00</td><td>1.000000000000000</td></tr>
<tr><td>0.05 or 0.95</td><td>0.713603042884044</td></tr>
<tr><td>0.10 or 0.90</td><td>0.531004406410719</td></tr>
<tr><td>0.15 or 0.85</td><td>0.390159695283600</td></tr>
<tr><td>0.20 or 0.80</td><td>0.278071905112638</td></tr>
<tr><td>0.25 or 0.75</td><td>0.188721875540867</td></tr>
<tr><td>0.30 or 0.70</td><td>0.118709100769307</td></tr>
<tr><td>0.35 or 0.65</td><td>0.065931944624509</td></tr>
<tr><td>0.40 or 0.60</td><td>0.029049405545331</td></tr>
<tr><td>0.45 or 0.55</td><td>0.007225546012192</td></tr>
<tr><td>0.50 or 0.50</td><td>0.000000000000000</td></tr>
</table>
Program II.3.c
Inverse of the Channal Capacity formula
Here is a Java program that prints a table of channel capacities and corresponding channel
probabilities. (The function calculating the inverse of the channel capacity function is given in
boldface.)
Java class: CapacityInverse
// CapacityInverse.java: print table of inverse capacities
public class CapacityInverse {
static final int TABLE_SIZE = 20;
static DecimalFormat eightDigits =
new DecimalFormat("0.00000000");
double p; // channel probability
double c; // channel capacity
System.out.println("<table border><tr align=center>");
System.out.println("<td><b>Channel<br>Capacity</b></td>");
System.out.println("<td><b>Probability<br>p</b></td>");
System.out.println("<td><b>Probability<br>1-p</b></td></tr>");
for (int i = 0; i <= TABLE_SIZE; i++) {
c = (double)i/TABLE_SIZE;
System.out.print("<tr><td>" + c);
if ((int)(10*c) == 10*c) System.out.print("0");
System.out.print("</td><td>" +
eightDigits.format(capacityInverse(c)) + "</td>");
System.out.println("</td><td>" +
eightDigits.format(1 - capacityInverse(c)) +
"</td></tr>");
}
} // end of main

if (p == 0 || p == 1) return 1;
}
// capacityInverse: the inverse of the capacity function,

// uses simple bisection method
private static double capacityInverse(double c) {
if (c < 0 || c > 1) return -1;
double lo = 0, hi = 0.5, mid, cLo, cHi, cMid;
176 Program II.3.c
do {
mid = (lo + hi)/2;
cLo = capacity(lo);
cHi = capacity(hi);
cMid = capacity(mid);
if (c > cMid) hi = mid;
else lo = mid;
} while (hi - lo > 1.0E-15);
return mid;
}

}
}
Here is the table printed by the above program (roughly as it would look in a browser):
Channel Probability Probability

Capacity p 1-p
0.00 0.50000000 0.50000000
0.05 0.36912775 0.63087225
0.10 0.31601935 0.68398065
0.15 0.27604089 0.72395911
0.20 0.24300385 0.75699615
0.25 0.21450174 0.78549826
0.30 0.18929771 0.81070229
0.35 0.16665701 0.83334299
0.40 0.14610240 0.85389760
0.45 0.12730481 0.87269519
0.50 0.11002786 0.88997214
0.55 0.09409724 0.90590276
0.60 0.07938260 0.92061740
0.65 0.06578671 0.93421329
0.70 0.05323904 0.94676096
0.75 0.04169269 0.95830731
0.80 0.03112446 0.96887554
0.85 0.02153963 0.97846037
0.90 0.01298686 0.98701314
0.95 0.00560717 0.99439283
1.00 0.00000000 1.00000000
Program II.3.d
Table of Repetition Codes
Here is a Java program that will generate the entire table, for any input probability p:
Java class: RepetitionTable
// RepetitionTable.java: given p, calculate a table of duplicates
public class RepetitionTable {

double p = Double.parseDouble(args[0]); // channel probability
int[] values = {1, 3, 5, 7, 9, 11, 25, 49, 99, 199};
System.out.println("<table BORDER NOSAVE >");
System.out.println("<tr><td align=center><b>Number of</b>");
System.out.println(" <br><b>Duplicates</b></td>");
System.out.println(" <td align=center><b>Transmission</b>");
System.out.println(" <br><b>Rate</b></td>");
System.out.println(" <td align=center><b>Error</b>");
System.out.println(" <br><b>Rate</b></td>");
System.out.println(" <td align=center><b>Success</b>");
System.out.println(" <br><b>Rate</b></td></tr>");
for (int len = 0; len < values.length; len++) {
int n = values[len];
double result = 0;
for (long i = n; i > n/2; i--)
result += comb(n,i)*Math.pow(p,i)*Math.pow(1-p,n-i);
System.out.println("<tr><td>" + n + "</td><td>" +
100.0/n + "\%</td><td>" + (100.0 - 100.0*result) +
"\%</td><td>" + 100.0*result + "\%</td></tr>" );
}
} // end of main
// comb(n, i): # of combinations of n things taken i at a time

private static double comb(long n, long i) {
double result = 1.0;
if (i < n/2) i = n-i;
for (long j = n; j > i; j--)
result *= ((double)j/(j-i));
return result;
}
}
Here is the table generated for p = 2/3:

178 Program II.3.d

1 100.0% 33.3333333333334% 66.6666666666666%
3 33.333333333333336% 25.925925925926023% 74.07407407407398%
5 20.0% 20.987654320987772% 79.01234567901223%
7 14.285714285714286% 17.329675354366827% 82.67032464563317%
9 11.11111111111111% 14.484580602550523% 85.51541939744948%
11 9.090909090909092% 12.208504801097504% 87.7914951989025%
25 4.0% 4.151367840779045% 95.84863215922095%
49 2.0408163265306123% 0.7872479136560173% 99.21275208634398%
99 1.0101010101010102% 0.030913686260717554% 99.96908631373928%
199 0.5025125628140703% 6.250990635692233E-5% 99.99993749009364%
Here is the same table “cleaned up” a bit by hand:
1 100.0% 33.3% 66.7%
3 33.3% 25.9% 74.1%
5 20.0% 20.99% 79.01%
7 14.3% 17.33% 82.67%
9 11.1% 14.48% 85.52%
11 9.1% 12.21% 87.79%
25 4.0% 4.15% 95.85%
49 2.0% 0.787% 99.213%
99 1.0% 0.0309% 99.9691%
199 0.5% 0.0000625% 99.9999375%
Program II.4.a
The Simulation Program
Here is the Java simulation program in three files. The program uses the blocksize (variable
N in file Shannon.java, accessible as a command line argument). The program calculates
2**N as the size of the code table (variable expN in file Shannon.java). The length of each
codeword in bytes is also a variable (CWS in file Shannon.java) accessible as a command
line argument. Thus the number of bits in each codeword is 8*CWS. The main data struc-
ture is the coding table: expN entries each of size CWS bytes. Each entry is the class Word,
and the table itself is of class Table. This coding table is allocated inside Table.java, and
each entry is allocated inside Word.java and filled with random bits. The simulation
is repeated simSize many times (another command line argument inside Shannon.java).
At each iteration, a random index in the coding table is chosen (length N bits), and the corre-
sponding codeword (length CWS bytes) is fetched from the table. The codeword is ”perturbed”
by reversing each bit with probability 1 - p = 0.25, where p is a variable inside Shan-
non.java. The table is then checked for the closest match to this new perturbed word. Here
”closest” means to check each entry to see the number of bit positions in which it differs from
the perturbed word. The program focuses on the word or words in the table that differ from the
perturbed word in the smallest number of bit positions. If there is more than 1 ”closest match”,
this is regarded as an error, as is the case in which the closest match is a word different from
the original unperturbed word. (In case of more than one closest match, one could choose a
word at random, but this program does not do that.) The error rate is simply the percent of
errors compared with all trials. The program uses a reasonably clever and efficient method
for comparing codewords (as bit strings). They are compared byte-by-byte. To compare two
bytes, say b1 and b1, in function countDiffs inside file Table.java, the function first
calculates b = b1 b̂2 (the bit-wise exclusive-or). A 1 bit in b represents a difference in the
two byte values, so one needs only to count the number of 1s in the byte b. This is done with a
table lookup in the array c, declared in Word.java, but used in Table.java. The variable
b ranges from -128 to 127 inclusive, so it is necessary to access c[b+128] and to create c
to give the correct answers when used in this way. The array of Strings s (inside Word.java)
gives the bit representation of each value of b, but this was only used for debugging.
Java class: Word
// Word.java: an array of CWS (codeword size) bytes
import java.util.Random;
public class Word {
public static int[] c = {
// number of 1 bits in 2s complement value (use value+128)
// used in class Table
1,2,2,3,2,3,3,4, 2,3,3,4,3,4,4,5, 2,3,3,4,3,4,4,5, 3,4,4,5,4,5,5,6,
2,3,3,4,3,4,4,5, 3,4,4,5,4,5,5,6, 3,4,4,5,4,5,5,6, 4,5,5,6,5,6,6,7,
2,3,3,4,3,4,4,5, 3,4,4,5,4,5,5,6, 3,4,4,5,4,5,5,6, 4,5,5,6,5,6,6,7,
3,4,4,5,4,5,5,6, 4,5,5,6,5,6,6,7, 4,5,5,6,5,6,6,7, 5,6,6,7,6,7,7,8,
180 Program II.4.a
0,1,1,2,1,2,2,3, 1,2,2,3,2,3,3,4, 1,2,2,3,2,3,3,4, 2,3,3,4,3,4,4,5,

1,2,2,3,2,3,3,4, 2,3,3,4,3,4,4,5, 2,3,3,4,3,4,4,5, 3,4,4,5,4,5,5,6,
1,2,2,3,2,3,3,4, 2,3,3,4,3,4,4,5, 2,3,3,4,3,4,4,5, 3,4,4,5,4,5,5,6,
2,3,3,4,3,4,4,5, 3,4,4,5,4,5,5,6, 3,4,4,5,4,5,5,6, 4,5,5,6,5,6,6,7};
public byte[] w; // the only data field in this class
// Word: construct and fill bytes with random values

public Word(Random ranNumGen) {
w = new byte[Shannon.CWS]; // allocate CWS bytes
for (int j = 0; j < Shannon.CWS; j++)
w[j] = (byte)(256*ranNumGen.nextDouble() - 128);
}
// Word: construct and copy input Word u into new class

public Word(Random ranNumGen, Word u) {
w = new byte[Shannon.CWS];
for (int j = 0; j < Shannon.CWS; j++)
w[j] = u.w[j];
}
}
Java class: Table

//Table.java: the code table for Shannon’s random code
public class Table {
public Word[] t; // the only data field in this class
// Table: constructor. Allocate expN = 2**N random words

public Table(Random ranNumGen) {
t = new Word[Shannon.expN];
for (int i = 0; i < Shannon.expN; i++)
t[i] = new Word(ranNumGen);
}
// search: search Table t for an input word w

public int search (Word w) {
int comp;
int minComp = Shannon.CWS*8 + 1;
int minCompCount = -100000000;
int index = -200000000;
for (int i = 0; i < Shannon.expN; i++) {
comp = compare(t[i], w); // count bits that differ
if (comp == minComp) // an old minimum
minCompCount++;
if (comp < minComp) { // a new minimum
index = i;
minComp = comp;
minCompCount = 1;
}
}
if (minCompCount == 1) return index; // unique minimum
else return -minCompCount; // several different minimums
4. Visualizing Channel Capacity 181
// compare: return count of differences of bits of input words

private int compare(Word u, Word v) {
int diffs = 0;
for (int i = 0; i < Shannon.CWS; i++)
diffs += countDiffs(u.w[i], v.w[i]);
return diffs;
}
// countDiffs: return count of differences of bits of input bytes

private int countDiffs(byte b1, byte b2) {
byte b = (byte)(b1ˆb2); // xor gives 1 where bytes differ
return Word.c[b+128]; // table lookup gives # of 1 bits
}
// getWord: fetch a word at a given index: part of simulation

public Word getWord(int index) {
return t[index];
}
// printTable: print the whole table, debug only

public void printTable() {
for (int i = 0; i < Shannon.expN; i++) {
System.out.print("Entry " + i + ": ");
t[i].printWord();
}
}
}
Java class: Shannon

// Shannon.java: a simulation of Shannon’s random coding
import java.util.Random; // use fancy rng for reproducability
public class Shannon {
public static final double P = 0.75; // prob of no error
public static int N; // blocksize, from command line
public static int expN; // = 2**N, table size, calculated from N
public static final double C = capacity(P);
public static int CWS; // the codeword size, bytes, from cmd line
private static Random ranNumGen = new Random(); // diff each time
public static double log2(double d) { // for log2 in Java

}
public static double capacity(double p) { // channel capacity

if (p == 0 || p == 1) return 1;
}
public static int randInt(int i) { // rand int, between 0 and i-1

return (int)(ranNumGen.nextDouble()*i);
}
182 Program II.4.a
// perturb: alter bits of input word, each time with prob 1-P
public static Word perturb(Word v) {
Word u = new Word(ranNumGen, v);
int[] mask = {1, 2, 4, 8, 16, 32, 64, -128};
for (int i = 0; i < Shannon.CWS; i++)
for (int j = 0; j < 8; j++)
if (ranNumGen.nextDouble() > Shannon.P) {
u.w[i] = (byte)(mask[j]ˆu.w[i]);
}
return u;
}

int simSize = Integer.parseInt(args[0]); // # of trials
N = Integer.parseInt(args[1]); // block size
CWS = Integer.parseInt(args[2]); // codeword size
expN = 1;
for (int i = 0; i < N; i++)
expN = expN*2; // expN = 2**N, table size in Table.java
System.out.println("simSize: " + simSize +
", Blocksize: " + Shannon.N +
", Codeword size (bytes): " + Shannon.CWS +
", expN: " + Shannon.expN);
// count matches and two kinds of mismatches
int numMatch = 0, numNonMatch = 0, numMultiMatch = 0;
Table tab = new Table(ranNumGen); // the coding table
for (int k = 0; k < simSize; k++) {
int ind = randInt(Shannon.expN); // index of rand code word
Word w = tab.getWord(ind); // w is the random code word
Word u = perturb(w); // u is w with random noise added
int ind2 = tab.search(u); // closest match, perturbed code word
if (ind2 == ind) numMatch++;
else if (ind2 >= 0) { // matched wrong code word, not one sent
numNonMatch++;
}
else if (ind2 < 0) numMultiMatch++; // multiple matches
if (k%500 == 499) {
System.out.print("Error Rate: " +
(k+1 - numMatch)/(double)(k+1));
System.out.println(", Match: " + numMatch +
", Non-Match: " + numNonMatch +
", Multiples: " + numMultiMatch);
}
} // for
System.out.print("Error Rate: " +
(simSize - numMatch)/(double)simSize);
System.out.println(", Match: " + numMatch +
", Non-Match: " + numNonMatch +
", Multiples: " + numMultiMatch);
}
}
Program II.5.a
The Huffman Algorithm
Here is a Huffman code program in 6 files, coded in Java. The program is for demonstration
purposes, and needs additional code to perform practical file compression, as is detailed at the
end of this section. The program below either reads a file directly from standard input, or if the
file name is on the command line, it uses that for the input. The program analyzes the input
file to get the symbol frequencies, and then calculates the code words for each symbol. It also
creates the output coded file. However, this file is a string of 0 and 1 ascii characters, not
binary numbers. The code also produces a human-readable version of the Huffman decoding
tree, as well as the entropy of the file and the average code length of the resulting Huffman
code.
The encode algorithm (function encode inside Huffman.java) just uses sequential
search, although the corresponding decode algorithm makes efficient use of the Huffman tree.
The priority queue (implemented in the file PQueue.java) just uses a simple list and sequen-
tial search, whereas a good priority queue should be implemented with a heap.
Java class: Entry
// Entry.java: entry in the code frequency table
class Entry {
public char symb; // character to be encoded
public double weight; // prob of occurrence of the character
public String rep; // string giving 0-1 Huffman codeword for char
}
Java class: Table
// Table.java: Huffman code frequency table
import java.io.*;
class Table {
public final int MAXT = 100; // maximum # of different symbols
public int currTableSize; // current size as table constructed
public Entry[] tab; // the table array, not allocated
private Reader in; // internal file name for input stream
String file = ""; // the whole input file as a String
private boolean fileOpen = false; // is the file open yet?
private String fileName; // name of input file, if present
private int totalChars = 0; // total number of chars read
char markerChar = ’@’; // sentinal at end of file
// Table: constructor, input parameter: input file name or null

public Table(String f) {
fileName = f;
currTableSize = 0;
tab = new Entry[MAXT];
}
// getNextChar: fetches next char. Also opens input file

184 Program II.5.a
private char getNextChar() {

char ch = ’ ’; // = ’ ’ to keep compiler happy
if (!fileOpen) {
fileOpen = true;
if (fileName == null)
in = new InputStreamReader(System.in);
else {
try {
in = new FileReader(fileName);
} catch (IOException e) {
System.out.println("Exception opening " + fileName);
}
}
}
try {
ch = (char)in.read();
System.out.println("Exception reading character");
}
return ch;
}
// buildTable: fetch each character and build the Table

public void buildTable() {
char ch = getNextChar();
while (ch != 65535 && ch != markerChar) { // EOF or sentinal #
totalChars++;
file += ch;
int i = lookUp(ch);
if (i == -1) { // new entry
tab[currTableSize] = new Entry();
tab[currTableSize].symb = ch;
tab[currTableSize].weight = 1.0;
tab[currTableSize].rep = "";
currTableSize++;
}
else { // existing entry
tab[i].weight += 1.0;
}
// System.out.print(ch); // for debug
ch = getNextChar();
} // while
// finish calculating the weights
for (int j = 0; j < currTableSize; j++)
tab[j].weight /= (double)totalChars;
}
// lookUp: loop up the next char in the Table tab

public int lookUp(char ch) {
for (int j = 0; j < currTableSize; j++)
if (tab[j].symb == ch) return j;
return -1;
}

public double log2(double d) {
return Math.log(d) / Math.log(2.0);
}
// entropy: calculate entropy of the Table

public double entropy() {
double res = 0.0;
for (int i = 0; i < currTableSize; i++)
res += tab[i].weight * log2(1.0/tab[i].weight);
return res;
}
// aveCodeLen: calculate average code length

public double aveCodeLen() {
double res = 0.0;
for (int i = 0; i < currTableSize; i++)
res += tab[i].weight * tab[i].rep.length();
return res;
}
}
Java class: TreeNode
// TreeNode.java: node in the Huffman tree, used for encode/decode
class TreeNode {
public double weight; // probability of symb occurring
public char symb; // the symbol to be encoded
public String rep; // string of 0’s and 1’s, huffman code word
public TreeNode left, right; // tree pointeres
public int step; // step # in construction (for displaying tree)
}
Java class: ListNode
// ListNode.java: node in linked list of trees, initially root nodes
class ListNode {
public TreeNode hufftree;
public ListNode next;
}
Java class: PQueue
// PQueue.java: implement a priority queue as a linked list of trees
// Initialize it as a linked list of singleton trees
class PQueue {
ListNode list = null; // this points to the main list
// insert: insert new entry into the list

public void insert(TreeNode t) {
ListNode l = new ListNode();
l.hufftree = t;
l.next = list;
list = l;
}
// buildList: create the initial list with singleton trees

public void buildList(Entry[] tab, int n) {
186 Program II.5.a
int i;
TreeNode tNode;
for (i = 0; i < n; i++) {
tNode = new TreeNode();
tNode.weight = tab[i].weight;
tNode.left = tNode.right = null;
tNode.symb = tab[i].symb;
tNode.rep = "";
insert(tNode);
}
}
// least: Remove and return from the list that tree with greatest
// root weight; sort of a pain in the ass to write
public TreeNode least() {
ListNode l, oldl, minl = null, oldminl = null; // for compiler
double minw = 1000000;
oldl = list;
l = list;
while (l != null) {
if (l.hufftree.weight < minw) {
minw = l.hufftree.weight;
oldminl = oldl;
minl = l;
}
oldl = l;
l = l.next;
}
if (minl == oldminl) {
list = list.next;
return minl.hufftree;
}
oldminl.next = minl.next;
return minl.hufftree;
}
}
Java class: Huffman

// Huffman.java: the Huffman tree algorithm
class Huffman {
public TreeNode tree; // the decoding tree
public Table t; // the frequency and encoding table
public PQueue p; // priority queue for building the Huffman tree
private int depth; // depth variable for debug printing of tree
String encodedFile, decodedFile; // files as Strings
char markerChar = ’@’; // sentinal at end of file
public DecimalFormat fourDigits = new DecimalFormat("0.0000");
// Huffman: constructor, does all the work

public Huffman(String fileName) {
t = new Table(fileName);
t.buildTable();
p = new PQueue();
p.buildList(t.tab, t.currTableSize);
tree = huffman(t.currTableSize);
insertRep(tree, t.tab, t.currTableSize, "");
displayTree(tree);
t.dumpTable();
encodedFile = encode(t.file);
System.out.println("Entropy: " + t.entropy() +
", Ave. Code Length: " + t.aveCodeLen());
}
// encode: translate the input file to binary Huffman file

public String encode(String file) {
String returnFile = ""; // encoded file to return (as a String)
for (int i = 0; i < file.length(); i++) {
int loc = t.lookUp(file.charAt(i));
if (loc == -1) {
System.out.println("Error in encode: can’t find: " +
file.charAt(i));
System.exit(0);
}
returnFile += t.tab[loc].rep;
}
return returnFile;
}
// decode: translate the binary file (as a string) back to chars

public String decode(String file) {
String returnFile = ""; // decoded file to return (as a String)
TreeNode treeRef; // local tree variable for chasing into tree
int i = 0; // index in the Huffman String
while (i < file.length()) { // keep going to end of String
treeRef = tree; // start at root of tree
while (true) {
if (treeRef.symb != markerChar) { // at a leaf node
returnFile += treeRef.symb;
break;
}
else if (file.charAt(i) == ’0’) { // go left with ’0’
treeRef = treeRef.left;
i++;
}
else { // go right with ’1’
treeRef = treeRef.right;
i++;
}
} // while (true)
} // while
return returnFile;
}
// huffman: construct the Huffman tree, for decoding

public TreeNode huffman(int n) {
188 Program II.5.a
int i;
TreeNode tree = null; // = null for compiler
for (i = 0; i < n-1; i++) {
tree = new TreeNode();
tree.left = p.least();
tree.left.step = i + 1; // just for displaying tree
tree.right = p.least();
tree.right.step = i + 1; // just for displaying tree
tree.weight = tree.left.weight +
tree.right.weight;
tree.symb = markerChar; // must not use ’@’ in input file
tree.rep = "";
p.insert(tree);
}
return tree;
}
// displayTree: print out tree, with initial and final comments

public void displayTree(TreeNode tree) {
System.out.println("\nDisplay of Huffman coding tree\n");
depth = 0;
displayTreeRecurs(tree);
}
// displayTreeRecurs: need recursive function for inorder traveral

public void displayTreeRecurs(TreeNode tree) {
depth++; // depth of recursion
String s = "";
if (tree != null) {
s = display(tree.rep + "0");
System.out.println(s);
displayTreeRecurs(tree.left);
s = display(tree.rep);
System.out.print(s + "+---");
if (depth != 1) {
if (tree.symb == markerChar) System.out.print("+---");
}
System.out.print(tree.symb + ": " +
fourDigits.format(tree.weight) + ", " + tree.rep);
if (depth != 1)
System.out.println(" (step " + tree.step + ")");
else System.out.println();
displayTreeRecurs(tree.right);
s = display(tree.rep + "1");
}
depth--;
}
// display: output blanks and verical lines to display tree

// (tricky use of rep string to display correctly)
private String display(String rep) {
String s = " ";
for (int i = 0; i < rep.length() - 1; i++) { // initial chars

if (rep.charAt(i) != rep.charAt(i+1) ) s += "|";
else s += " ";
s += " ";
}
return s;
}
// insertRep: tricky function to use Huffman tree to create rep

public void insertRep(TreeNode tree, Entry tab[], int n, String repr) {
//recursive function to insert Huffman codewords at each node.
// this could just insert at the leaves.
String s1, s2;
tree.rep = repr;
if ((tree.left) == null && (tree.right) == null) {
for (int i = 0; i < n; i++)
if (tree.symb == tab[i].symb)
tab[i].rep = tree.rep;
return;
}
s1 = repr; s1 += "0";
insertRep(tree.left, tab, n, s1); // recursive call to left
s2 = repr; s2 += "1";
insertRep(tree.right, tab, n, s2); // recursive call to right
}
// main: doesn’t do much; just feeds in input file name

Huffman huff;
// pass an input file name if present on command line
if (args.length > 0) huff = new Huffman(args[0]);
else huff = new Huffman(null);
}
}
Here is an input file:

% cat Testit.txt
aaaaabbbbbcccccccccccccccdddddddddddddddddeeeeeeeeeeeeeeeeee
ffffffffffffffffffffffffffffffffffffffff#
Here is the output with various debug dumps:
% java Huffman Testit.txt
Display of Huffman coding tree
+---f: 0.4000, 0 (step 5)
|
+---@: 1.0000,
|
| +---b: 0.0500, 1000 (step 1)
| |
| +---+---@: 0.1000, 100 (step 2)
| | |
| | +---a: 0.0500, 1001 (step 1)
190 Program II.5.a
| |
| +---+---@: 0.2500, 10 (step 4)
| | |
| | +---c: 0.1500, 101 (step 2)
| |
+---+---@: 0.6000, 1 (step 5)
|
| +---d: 0.1700, 110 (step 3)
| |
+---+---@: 0.3500, 11 (step 4)
|
+---e: 0.1800, 111 (step 3)
Dump of Table ----->

Size: 6
Entry 0. Symbol: a, Weight: 0.05, Representation: 1001
Entry 1. Symbol: b, Weight: 0.05, Representation: 1000
Entry 2. Symbol: c, Weight: 0.15, Representation: 101
Entry 3. Symbol: d, Weight: 0.17, Representation: 110
Entry 4. Symbol: e, Weight: 0.18, Representation: 111
Entry 5. Symbol: f, Weight: 0.4, Representation: 0
----> End Dump of Table
Entropy: 2.251403369717592, Ave. Code Length: 2.3
Input file (as a String):

ffffffffffffffffffffffffffffffffffffffff
Encoded file (as a String):

100110011001100110011000100010001000100010110110110110110110
110110110110110110110110111011011011011011011011011011011011
011011011011011011111111111111111111111111111111111111111111
11111111110000000000000000000000000000000000000000
Decoded file (as a String):

ffffffffffffffffffffffffffffffffffffffff
A Program to Produce Actual Compressed Binary Huffman Files: The changes and ad-
ditions involve the methods encode and decode in the class Huffman. First, the input file
would be read twice, once to build the frequency table, and a second time to translate to the
Huffman code. Next, the method needs to write the frequency table in some form. (It is also
possible to write the actual Huffman tree in coded form.) Then encode would write bits in a
way similar to the Hamming algorithm in the next section.
The decode method would first read the frequency table and build an the same Huffman
tree used inside encode. Then the method would read bits, again in a way similar to that of
the Hamming algorithm. The final translation to the original symbols would be the same.
Program II.5.b
Two Distinct Huffman Codes
There are simple examples of Huffman codes, where there are two structurally distinct Huffman
trees for the same set of symbols and frequencies. Because the Huffman code is optimal, these
codes must have the same average code length. Here is one simple example: a:9, b:5,
c:4, d:3, and e:3. There are clearly two distinct ways to construct the tree, resulting in
Huffman codes for the symbols with a different set of lengths for the codewords. In both cases
the average code length is 2.25 bits per symbol.
Here are two different sets of lengths:
+---a: 0.3750, 0 (step 4)
|
---+---@: 1.0000,
|
| +---e: 0.1250, 100 (step 1)
| |
| +---+---@: 0.2500, 10 (step 3)
| | |
| | +---d: 0.1250, 101 (step 1)
| |
+---+---@: 0.6250, 1 (step 4)
|
| +---c: 0.1667, 110 (step 2)
| |
+---+---@: 0.3750, 11 (step 3)
|
+---b: 0.2083, 111 (step 2)

+---c: 0.1667, 00 (step 2)

|
+---+---@: 0.3750, 0 (step 4)
| |
| +---b: 0.2083, 01 (step 2)
|
---+---@: 1.0000,
|
| +---e: 0.1250, 100 (step 1)
| |
| +---+---@: 0.2500, 10 (step 3)
192 Program II.5.b
| | |
| | +---d: 0.1250, 101 (step 1)
| |
+---+---@: 0.6250, 1 (step 4)
|
+---a: 0.3750, 11 (step 3)


Program II.6.a
The Hamming Algorithm
This section presents an implementation of the binary Hamming code. The Java source was de-
signed for simplicity and ease of understanding, rather than for efficiency. The basic Hamming
algorithm is implemented using arrays of “bits” (0 or 1 stored in an int), with from 1 to 120
message bits and from 4 to 128 bits in the codeword. In order to read and write files of bytes,
it is necessary to unpack each byte so that the Hamming routines can work on it, and then to
pack the result for writing. (A more efficient implementation would not use arrays of bits in
this way and so would not need the packing and unpacking.)
The complete array-based implementation of the Hamming code is in the Java class Ham-
ming. This is a straightforward implementation. Here are comments about individual features:
The constructor Hamming: this just builds the necessary masks, described next.
The 2-dimensional array m: nine 128-bit masks used to decide which bits to use in each
parity check: m[0], m[1], m[2], m[3], m[4], m[5], m[6], m[7]. The mask m[i]
gives the bits to check for the check bit in position pow(2,i). Thus m[0] is all 1 bits,
m[1] has odd-numbered bits 1, m[2] has alternating pairs of 0’s and 1’s, and so forth.
The mask m0 has a 1 bit in those positions used for check bits: all powers of 2.
The encodeMessage method: Takes an input array of message bits (where the length
of the array gives the number of bits), and produces an output array of bits that represents
the Hamming codeword obtained by inserting extra check bits. Again the length of the
codeword is given by the length of the array.
The insertMessage method: Called by encodeMessage, this inserts message bits
into each non-check bit postion.
The insertCheckBits method: Called by encodeMessage, this inserts the proper
check bit values.
The decodeMessage method: This first checks for errors (checkErrors), then cor-
rects a single error if one is found, and finally extracts the message without check bits
and returns it. In case of a detected double error, a null is returned.
Expanding the implementation size: This is easy to do. For example, to double the
maximum size, change MAX CHK LEN from 8 to 9, MAX RES LEN from 128 to 256, and
add 128 to end of list defining checkPos. (The class HammingDecode accesses the
instatiation of Hamming and builds another array based on these sizes.)
Making use of the class: The following code show how to use the class Hamming, as it
is used in the classes HammingEncode and HammingDecode below:
194 Program II.6.a
Using class Hamming

Hamming ham = new Hamming();
int[] mess, res;
// create mess and fill it with one block’s worth of message bits
res = ham.encodeMessage(mess); // res allocated inside ham
// create res and fill it with one block’s worth of message and check bits
mess = ham.decodeMessage(res); // mess allocated inside ham
Here is Java code for the class Hamming, with a few extra debug lines reporting the number of
errors corrected.
Java class: Hamming
// Hamming: implement Hamming code
// Uses arrays of "bits": mess = uncoded input, res = coded result
public class Hamming {
public final int MAX_CHK_LEN = 8; // max number of check digits
public final int MAX_RES_LEN = 128; // 2ˆ(MAX_CHK_LEN - 1)
public int[] checkPos = {0,1,2,4,8,16,32,64}; // positions to check
public final int MAX_MESS_LEN = MAX_RES_LEN - MAX_CHK_LEN; // 120
private int[][] m = new int[MAX_CHK_LEN][MAX_RES_LEN]; // check masks
private int[] m0 = new int[MAX_RES_LEN]; // mask for message insertion
private int[] buf = new int[MAX_RES_LEN]; // buffer for coded messages
public int errCount; // ****** extra counter for debugging ******
// Hamming: constructor to create masks

public Hamming() {
for (int i = 0; i < MAX_CHK_LEN; i++)
for (int j = 0; j < MAX_RES_LEN; j++)
if (i == 0) m[i][j] = 1;
else m[i][j] = (j >> (i - 1))%2;
for (int i = 0; i < MAX_RES_LEN; i++) m0[i] = 1;
for (int i = 0; i < MAX_CHK_LEN; i++) m0[checkPos[i]] = 0;
// encodeMessge: insert message bits and then set check bits

public int[] encodeMessage(int[] mes) {
int res[] = insertMessage(mes);
insertCheckBits(res);
return res;
}
// insertMessage: put message bits into non-check positions

public int[] insertMessage(int[] mess) {
for (int i = 0; i < MAX_RES_LEN; i++) buf[i] = 0;
int loc = 0, i = 0;
while (i < mess.length) {
if (m0[loc] == 1) buf[loc] = mess[i++];
if (loc >= MAX_RES_LEN) System.exit(1000 + i);
loc++;
}
int[] res = new int[loc];
for (int j = 0; j < loc; j++) res[j] = buf[j];

return res;
}
// insertCheckBits: add the parity check bits

public void insertCheckBits(int[] res) {
for (int i = MAX_CHK_LEN - 1; i >= 0; i--) {
int checkRes = 0; // holds sum of bits for this parity check
for (int j = 0; j < res.length; j++)
if (m[i][j] == 1) checkRes += res[j];
if (checkPos[i] < res.length)
res[checkPos[i]] = checkRes%2;
}
}
// decodeMessage: correct errors and extract message bits

public int[] decodeMessage(int[] res) {
int errCode = checkErrors(res);
if (errCode >= 0) { // single error in position errCode
res[errCode] ˆ= 1; // correct single error
errCount++; // ******* extra count for debugging ********
}
if (errCode >= -1) return extractMessage(res); // no errors left
return null; // errCode == -2 means DOUBLE ERROR
}
// extractMessage: get back message bits from non-check positions

public int[] extractMessage(int[] res) {
for (int i = 0; i < MAX_RES_LEN; i++) buf[i] = 0;
int loc = 0, i = 0;
while (i < res.length) {
if (m0[i] == 1) buf[loc++] = res[i];
if (loc >= MAX_RES_LEN) System.exit(2000 + i);
i++;
}
int[] mess = new int[loc];
for (int j = 0; j < loc; j++) mess[j] = buf[j];
return mess;
}
// checkErrors: do error check, return position of error

// return -1 for no error, return -2 for double error
public int checkErrors(int[] res) {
int[] checkRes = new int[MAX_CHK_LEN];
int errorPos = 0;
for (int i = 0; i < MAX_CHK_LEN; i++) {
checkRes[i] = 0;
for (int j = 0; j < res.length; j++)
if (m[i][j] == 1) checkRes[i] += res[j];
checkRes[i] %= 2;
}
for (int i = 1; i < MAX_CHK_LEN; i++)
if (checkRes[i] == 1) errorPos += checkPos[i];
196 Program II.6.a
if (errorPos == 0 && checkRes[0] == 0) return -1; // no error

if (errorPos == 0 && checkRes[0] == 1) return 0; // error at 0
if (errorPos > 0 && checkRes[0] == 1) return errorPos; // error
if (errorPos > 0 && checkRes[0] == 0) return -2; // double error
return 999;
}
}
The final three classes in this section implement the encoding and decoding of an arbitrary
binary file using the Hamming code. The method here encodes anywhere from 1 bit to 120
bits at a time. This requires 4 to 128 bits in the Hamming codeword. The test below shows
the results of encoding a 3116-byte binary PDF file into files of varying sizes depending on the
message length. The coded file is then decoded to get the original back.
The coded file starts with a byte giving the number of message bits in each codeword. This
file does not usually have a number of bits divisible by 8, so the last byte of the file indicates
how many bits of the next-to-the-last byte are part of the coded result. As part of the debugging,
single errors were simulated at some random bit position in each codeword before the decoding
step.
The class HammingEncode reads bytes of a source file (using encodeHammingBit).
Each bit of each byte is sent of another function (encodeBit), which accumulates them until
there is a block the size of the desired message size. This block is transformed to a coded
Hamming block when an instance of Hamming adds check bits. Then the resulting block is
sent one bit at a time to a function writeBit which accumulates them until it has 8 to write
as a byte.
The class HammingDecode reads bytes of a source file (using decodeHammingBit).
The first byte of the source file gives the message size for the particular Hamming code used
in the file. The last byte of the file gives the number of bits used in the next-to-the-last byte, so
it necessary to read three bytes ahead during processing. After the first byte, each bit of each
byte is sent of another function (decodeBit), which accumulates them until there is a block
the size of the desired codeword size. This block is transformed to a message block when an
instance of Hamming removes check bits. Then the resulting block is sent one bit at a time to
a function writeBit (different from the previous writeBit) which accumulates them until
it has 8 to write as a byte.
Java class: HammingEncode
// HammingEncode: encode an input file with the Hamming code
import java.io.*;
public class HammingEncode {
Hamming ham = new Hamming(); // Hamming code implementation
InputStream in; // input file
OutputStream out; // output file
int currPos = 0, currPosRest = 0; // keep track of bit positions
int bRest = 0; // last byte
int messLen; // bit length of original message
int[] mess, messRest, res; // message, remaining message, code result
int[] bMask = {0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80}; // bit masks
public HammingEncode(int messL, InputStream infile, OutputStream outfile) {

messLen = messL;
in = infile;
out = outfile;
mess = new int[messLen];
}
// encodeHammingBit: read bytes, pass bits to encodeBit.

// Bits of a byte numbered from 0 as least significant
public void encodeHammingBit() {
writeByte(messLen); // write message length
try {
int b; // input byte
int bit; // output bit
while ((b = in.read()) != -1) {
for (int dum = 0; dum < 8; dum++) {
bit = b%2;
encodeBit(bit);
b = b >> 1;
}
} // end of while
System.err.println("Error reading input file");
System.exit(-1);
} // end try
encodeBit(-1);
}
// encodeBit: pass bit at a time to growing Hamming array

private void encodeBit(int bit) {
if (bit == -1) { // no more bits; finish up
if (currPos == 0) {
writeBit(-1); // send finish up message to writeBit
return;
}
messRest = new int[currPos];
for (int i = 0; i < currPos; i++)
messRest[i] = mess[i];
res = ham.encodeMessage(messRest);
// *** for debugging, insert random error into half of the blocks
if (Math.random() < 0.5) { // insert half the time
int errPos = (int)(Math.random()*res.length); // random positioon
res[errPos] ˆ= 1; // insert the actual bit error
}
for (int i = 0; i < res.length; i++)
writeBit(res[i]);
writeBit(-1); // send finish up message to writeBit
return;
}
mess[currPos++] = bit;
if (currPos == messLen) {
res = ham.encodeMessage(mess);
// *** for debugging, insert random error into half of the blocks
198 Program II.6.a
if (Math.random() < 0.5) { // insert half the time

int errPos = (int)(Math.random()*res.length); // random position
res[errPos] ˆ= 1;
}
for (int i = 0; i < res.length; i++)
writeBit(res[i]);
currPos = 0; // reset position for next block
}
}
// writeBit: accumulate 8 bits, then write the byte

private void writeBit(int bit) {
if (bit == -1) { // received finish up message
writeByte(bRest); // write last partial byte
writeByte(currPosRest); // how many bits count in last byte?
return;
}
if (bit == 1) bRest |= bMask[currPosRest]; // insert bit
currPosRest++;
if (currPosRest == 8) {
writeByte(bRest);
currPosRest = 0;
bRest = 0;
}
}
// writeByte: accumulate bits, then write byte

private void writeByte(int b) {
try {
out.write(b);
System.err.print("Error writing file");
System.exit(-1);
}
}
}
Java class: HammingDecode

// HammingDecode: decode an input file coded with the Hamming code
import java.io.*;
public class HammingDecode {
Hamming ham = new Hamming(); // Hamming code implementation
int currPos = 0, currPosRest = 0; // bit positions in bytes
int bRest = 0; // last byte
int resLen, messLen; // bit length of: coded result, original message
int[] mess, res, resRest; // message, coded result, remaining result
int[] bMask = {0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80}; // bit mask
int[] codeLen; // codeLen[i] = j means if messLen == i, then resLen == j
// HammingDecode: assign files and build codeLen table

public HammingDecode(InputStream infile, OutputStream outfile) {
codeLen = new int[ham.MAX_MESS_LEN + 1];

codeLen[0] = 0;
int next = 4, j = 3;
for (int i = 1; i <= ham.MAX_MESS_LEN; i++) {
codeLen[i] = next;
if (j < ham.checkPos.length && next == ham.checkPos[j]) {
j++; next++;
}
next++;
}
in = infile;
out = outfile;
}
// decodeHammingBit: read bytes, pass bits to instance of Hamming

// Bits of a byte numbered from 0 as least significant
public void decodeHammingBit() {
try {
int b; // input byte
int bNext; // next byte (read ahead)
int bEnd; // next byte after bNext
int bit; // output bit
messLen = in.read(); // initial byte holds message length
resLen = codeLen[messLen]; // deduce coded result length
res = new int[resLen];
bNext = in.read(); // read ahead because last byte contains
bEnd = in.read(); // the number of bits in next-to-last byte
while (true) {
b = bNext;
bNext = bEnd;
bEnd = in.read();
if (bEnd == -1) { // end-of-file
// bNext give # of bits of b to use
for (int dum = 0; dum < bNext; dum++) {
bit = b%2;
decodeBit(bit);
b = b >> 1;
}
decodeBit(-1); // send end-of-file message
return;
}
for (int dum = 0; dum < 8; dum++) {
bit = b%2;
decodeBit(bit);
b = b >> 1;
}
} // end of while
System.err.println("Error reading input file");
System.exit(-1);
} // end try
}
200 Program II.6.a
// decodeBit: decode and send off bits

private void decodeBit(int bit) {
if (bit == -1) { // no more bits; finish up
if (currPos == 0) { // ******** temp debug output ********
System.out.println(ham.errCount + " errors detected");
return; // no leftovers
}
resRest = new int[currPos];
for (int i = 0; i < currPos; i++)
resRest[i] = res[i];
mess = ham.decodeMessage(resRest);
if (mess == null) { // double error
System.out.println("Double error detected");
System.exit(-1);
}
for (int i = 0; i < mess.length; i++)
writeBit(mess[i]);
writeBit(-1); // end-of-file message
System.out.println(ham.errCount + " errors detected"); // temp output
return;
}
res[currPos++] = bit;
if (currPos == resLen) {
mess = ham.decodeMessage(res);
if (mess == null) { // double error
System.out.println("Double error detected");
System.exit(-1);
}
for (int i = 0; i < mess.length; i++)
writeBit(mess[i]);
currPos = 0;
}
}
// writeBit: accumulate bits until ready to write a byte

private void writeBit(int bit) {
if (bit == -1) { // no more bits
if (currPosRest == 0) return;
writeByte(bRest);
return;
}
if (bit == 1) bRest |= bMask[currPosRest]; // insert bit
currPosRest++;
if (currPosRest == 8) {
writeByte(bRest);
currPosRest = 0;
bRest = 0;
}
}
// writeByte: actually write the byte

private void writeByte(int b) {
try {
out.write(b);
System.exit(-1);
}
}
}
Java class: HammingFiles

// HammingFiles: encode or decode files with Hamming code
import java.io.*;
public class HammingFiles {
static InputStream in; // input file args[1] or args[2]
static OutputStream out; // output file args[2] or args[3]
public static void openFiles(String infile, String outfile) {

try {
in = new FileInputStream(infile);
out = new FileOutputStream(outfile);
System.err.println("Error opening files");
System.exit(-1);
}
}

if (args[0].equals("-encode")) {
int messLen = Integer.parseInt(args[1]);
if (messLen > 120) {
System.err.println("Error: Message length > 120");
System.exit(-1);
}
openFiles(args[2], args[3]);
HammingEncode hamEncode = new HammingEncode(messLen, in, out);
hamEncode.encodeHammingBit();
}
else if (args[0].equals("-decode")) {
HammingDecode hamDecode = new HammingDecode(in, out);
hamDecode.decodeHammingBit();
}
else System.err.println("Usage: java HammingBit " +
"(-encode messageLen | -decode) infile outfile");
}
}
Finally, here is a sample debug run of this software with a 3116-byte input binary file (PDF
format): utsa.pdf. Extra errors are inserted half the time.
% java HammingFiles -encode 53 utsa.pdf utsa53.code
% java HammingFiles -decode utsa53.code utsa2_53.pdf
212 errors detected
202 Program II.6.a
% wc utsa53.code (coded file, using 53 bits per message)

15 95 3531 utsa53.code
% wc utsa2_120.pdf (recovered file using Hamming code)
53 175 3116 utsa2_53.pdf
Here are different sizes of the coded file when different message lengths are used for the coding.
These are not experimental results, but the sizes are strictly dictated by the properties of the
Hamming code. Notice that message lengths 4 and 5 result in exactly the same sizes for the
final encoded files.
Hamming-encoded File
Message Codeword File Coded
size size size file size
1 4 3116 12467
2 6 3116 9351
3 7 3116 7273
4 8 3116 6235
5 10 3116 6235
10 15 3116 4677
20 26 3116 4054
30 37 3116 3846
40 47 3116 3664
50 57 3116 3555
100 108 3116 3368
120 128 3116 3327
Program II.7.a
U.S. Banking Scheme
Here is the simple scheme used by U.S. banks, involving successive weights of 3, 7 and 1,
repeated.
Java class: ErrorDetection
// ErrorDetection.java: base class for single-digit error detection
public class ErrorDetection {
public static void printArray(int[] a) {

for (int i = 0; i < a.length; i++) {
if (a[i] == 10) System.out.print("X ");
else System.out.print(a[i]);
if (i%5 == 0) System.out.print(" ");
}
System.out.println();
}
public static void printUnchecked(int[] a) {

System.out.print("? ");
System.out.print(a[i]);
}
}
}
Java class: BanksErrorDetection
// BanksErrorDetection.java: Implement scheme used by US banks
public class BanksErrorDetection extends ErrorDetection {
public static int insertCheck(int[] a) {

int check = 0;
{\timesbf if (i%3 == 1)
check = (check + 3*a[i])%10;
else if (i%3 == 2)
check = (check + 7*a[i])%10;
else
check = (check + a[i])%10;}
if (check == 0) a[0] = 0;
else a[0] = -check + 10;
return a[0];
}

int check = 0;
204 Program II.7.a

if (i%3 == 1) check = (check + 3*a[i])%10;
else if (i%3 == 2) check = (check + 7*a[i])%10;
else check = (check + a[i])%10;
else return true;
}
// main function
int[] a = new int[9];
boolean checkFlag = false;
a[i] = (int)(Math.random() * 10.0);
printUnchecked(a);
BanksErrorDetection.insertCheck(a);
printArray(a);
System.out.println(BanksErrorDetection.doCheck(a));
a[4] = (a[4] + 1)%10;
BanksErrorDetection.printArray(a);
System.out.println(BanksErrorDetection.doCheck(a));
// test all adjacent transpositions

System.out.println("\nUS Banks, error detection scheme");
System.out.println("\nTest all adjacent transpositions ...");
for (int pos = 4; pos < 7; pos++)
for (int p1 = 0; p1 < 10; p1++)
for (int p2 = 0; p2 < 10; p2++) {
if (p1 != p2) {
a[pos] = p1; a[pos+1] = p2;
BanksErrorDetection.insertCheck(a);
// interchange
a[pos] ˆ= a[pos+1];
a[pos+1] ˆ= a[pos];
if (BanksErrorDetection.doCheck(a)) {
System.out.println("Warning: Interchange of " +
p1 + " and " + p2 + " not detected");
checkFlag = true;
}
}
}
if (checkFlag)
System.out.println("At least one transposition undetected");
else
System.out.println("All transpositions detected");
} // end of main
}
Here is the output, showing a simple test, and a test of all adjacent interchanges. Here if digits
differing by 5 are interchanged, the error goes undetected. I have tested interchanges for each
pair of the three weights, so the 10 missed transpositions are repeated 3 times below.
? 95967 315
1 95967 315
true
1 95977 315
false
US Banks, error detection scheme
Test all adjacent transpositions ...

Warning: Interchange of 0 and 5 not detected
At least one transposition undetected
Program II.7.b
IBM Scheme
Here is the ”IBM” scheme used for credit card numbers.


if (a[i] == 10)
System.out.print("X ");
else
}
}

}
}
}
Java class: IbmErrorDetection
// IbmErrorDetection.java: Implement "IBM" decimal error detection
public class IbmErrorDetection extends ErrorDetection{
{\timesbf private static int sharp(int d) {

return (2*d)/10 + (2*d)%10;
}}

int check = 0;
check = (check + a[i])%10;
else
check = (check + sharp(a[i]))%10;}
if (check == 0) a[0] = 0;
return a[0];
}

int check = 0;
else
check = (check + sharp(a[i]))%10;}
else return true;
}
// main function
printUnchecked(a);
IbmErrorDetection.insertCheck(a);
printArray(a);
System.out.println(IbmErrorDetection.doCheck(a));
a[4] = (a[4] + 1)%10;
IbmErrorDetection.printArray(a);
System.out.println(IbmErrorDetection.doCheck(a));

for (int p1 = 0; p1 < 10; p1++)
for (int p2 = 0; p2 < 10; p2++) {
if (p1 != p2) {
a[8] = p1; a[9] = p2;
IbmErrorDetection.insertCheck(a);
// interchange
a[8] ˆ= a[9];
a[9] ˆ= a[8];
a[8] ˆ= a[9];
if (IbmErrorDetection.doCheck(a)) {
checkFlag = true;
}
}
}
if (checkFlag)
else
} // end of main
}
208 Program II.7.b
Here is the output, showing a simple test, and a test of all adjacent interchanges. Interchange
errors not caught are 09 and 90.
? 31623 91033 1003
7 31623 91033 1003
true
7 31633 91033 1003
false

Program II.7.c
ISBN mod 11 Scheme
Here is the ISBN mod 11 scheme used for U.S. book publishing numbers.

if (a[i] == 10)
else
}
}

}
}
}
Java class: ISBNErrorDetection
// ISBNErrorDetection.java: Implement mod 11 check used by ISBN
public class ISBNErrorDetection extends ErrorDetection {

int check = 0;
check = (check + (i%10 + 1)*a[i])%11;
if (check == 0) a[0] = 0;
return a[0];
}

int check = 0;
check = (check + (i%10 + 1)*a[i])%11;
else return true;
}
210 Program II.7.c
// main function
ISBNErrorDetection.printUnchecked(a);
ISBNErrorDetection.insertCheck(a);
ISBNErrorDetection.printArray(a);
System.out.println(ISBNErrorDetection.doCheck(a));
a[4] = (a[4] + 3)%10;
ISBNErrorDetection.printArray(a);
System.out.println(ISBNErrorDetection.doCheck(a));

System.out.println("\nISBN error detection scheme");
for (int p1 = 0; p1 < 10; p1++)
for (int p2 = 0; p2 < 10; p2++) {
if (p1 != p2) {
a[pos] = p1; a[pos+1] = p2;
ISBNErrorDetection.insertCheck(a);
// interchange
a[pos+1] ˆ= a[pos];
if (ISBNErrorDetection.doCheck(a)) {
checkFlag = true;
}
}
}
if (checkFlag)
else
} // end of main
}
Here is the output, first showing a simple test. I tweaked the test until the check ”digit” was an
”X”. Next is a test of all adjacent interchanges. Here all interchanges are caught.
? 11696 554
X 11696 554
true
X 11626 554
false
ISBN error detection scheme

All transpositions detected
Program II.7.d
Mod 97 Scheme
Here is the mod 97 scheme used for extra error detection. The check below tests its performance
for adjacent double error detection. One expects this check to catch apprximately 99% of all
random errors. However, it catches 99.94% of all adjacent double errors (except for possible
adjacent double errors involving one of the two decimals representing the check ”digit” and the
first actual data digit).

if (a[i] == 10)
else
}
}

}
}
}
Java class: Mod97ErrorDetection
// Mod97ErrorDetection.java: Impl the mod 97 check for error detection
// uses successive powers of 10 (modulo 97) for the weights
public class Mod97ErrorDetection extends ErrorDetection {

int check = 0;
int weight = 10;
check = (check + weight*a[i])%97;
weight = (weight*10)%97;
}
if (check == 0) a[0] = 0;
return a[0];

int check = 0;
int weight = 1;
check = (check + weight*a[i])%97;
weight = (weight*10)%97;
}
else return true;
}
// main function
// no need for a random start
// try all adjacent double errors

int errorCount = 0;
int totalCount = 0;
// try each successive position (all the same)
// try every pair of digits for the initial pair
for (int p1 = 0; p1 < 10; p1++)
for (int p2 = 0; p2 < 10; p2++) {
// insert the initial pair
a[pos] = p1; a[pos+1] = p2;
// do the check and insert mod 97 check "digit"
Mod97ErrorDetection.insertCheck(a);
// try every pair of digits for the double error
for (int n1 = 0; n1 < 10; n1++)
for (int n2 = 0; n2 < 10; n2++)
// only try if an actual change
if (n1 != p1 || n2 != p2) {
totalCount++;
// insert new air as an error
a[pos] = n1; a[pos+1] = n2;
// check if the change is not detected
if (Mod97ErrorDetection.doCheck(a)) {
System.out.println("Error, old digits: " +
p1 + p2 + ", new digits: " + n1 + n2 +
". Position: " + pos);
errorCount++;
}
}
}
System.out.println("Adjacent double errors undetected: " +
errorCount + ", out of " + totalCount + ", or " +
((double)errorCount/totalCount*100) + "%");
214 Program II.7.d
} // end of main
}
Here is the output, showing that there are only 6 kinds of adjacent double errors that remain
undetected. For example, "10" changed to "89". Here in the weight equation, "10" is an
additional 1 + 0*10 = 1 (along with extra power of 10 weight), while "89" is an additional
8 + 9*10 = 98 (along with the same extra power of 10 weight). When the 98 is reduced
modulo 97, it also becomes 1, so that the new equation has the same value as the old. The
error rate is 0.060606...%, or the equation catches 99.94% of all adjacent double errors.
Error, old digits: 00, new digits: 79. Position: 1
(similar entries for Position = 2, 3, ..., 98)
Adjacent double errors undetected: 588, out of 970200,

or 0.06060606060606061%
Program II.7.e
Hamming mod 11 Scheme
Here is a test of the Hamming mod 11 Error Correcting code, using 3 check digits and 121
digits altogether. The test starts with a random initial word. It first inserts the proper check
digits in positions 0, 1 and 11. The test then makes a random change in each position and
corrects the change, so there are 121 changes (some are 0 added — no change). In repeated
runs, the code has always corrected the proper position to the old value.
Java class: H11EC
// H11EC.java: Implement the mod 11 Hamming code
public class H11EC {
public static int[] inv = {0, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1};
// Using the sum1 check sum, if have error e and check1 result c,
// then pos[e][c] gives the position in error (modulo 11),
// using the first check equation.
// If the error is e and check11 result is c,
// then pos[e][c] gives the value position/11.
public static int[][] pos = {
{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10},
{0, 6, 1, 7, 2, 8, 3, 9, 4, 10, 5},
{0, 4, 8, 1, 5, 9, 2, 6, 10, 3, 7},
{0, 3, 6, 9, 1, 4, 7, 10, 2, 5, 8},
{0, 9, 7, 5, 3, 1, 10, 8, 6, 4, 2},
{0, 2, 4, 6, 8, 10, 1, 3, 5, 7, 9},
{0, 8, 5, 2, 10, 7, 4, 1, 9, 6, 3},
{0, 7, 3, 10, 6, 2, 9, 5, 1, 8, 4},
{0, 5, 10, 4, 9, 3, 8, 2, 7, 1, 6},
{0, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1}};

if (a[i] == 10)
else
}
}

System.out.print("? "); // position 0
System.out.print("?"); // position 1
216 Program II.7.e
if (i == 11) System.out.print("?"); // position 11

}
}
public static void insertCheck(int[] a) {

a[1] = inv[sum1NoCheck(a)];
if (!doCheck(a))
System.out.println("Failure in insertCheck");
}

int error = sum0(a); // amount of error
int check1 = sum1(a);
if (error == 0 && check1 == 0 && check11 == 0) return true;
if (error == 0) return false; // a double error
int position = pos[error][check11]*11 + pos[error][check1];
if (position >= a.length) {
System.out.println("doCheck: position: " + position +
", error: " + error + ", check1: " + check1 +
", check11: " + check11);
System.exit(0);
}
a[position] = (a[position] - error + 11)%11;

System.out.println("Position " + position +
" corrected to " + a[position]);
return true;
}
public static int sum0(int[] a) {

int check = 0;
return check;
}
public static int sum0NoCheck(int[] a) {

int check = 0;
return check;
}

int check = 0;
check = (check + (i%11)*a[i])%11;
return check;
}

int check = 0;
check = (check + (i%11)*a[i])%11;
return check;
}

int check = 0;
check = (check + ((i/11)%11)*a[i])%11;
return check;
}

int check = 0;
check = (check + ((i/11)%11)*a[i])%11;
return check;
}
// main function
if (i != 11) a[i] = (int)(Math.random() * 10.0);
H11EC.insertCheck(a);
int oldValue = a[i];
a[i] = (a[i] + (int)(Math.random() * 10.0))%10;
System.out.print("Position: " + i + " changed from " +
oldValue + " to " + a[i] + "; ");
if (oldValue == a[i]) System.out.println();
H11EC.doCheck(a);
if (a[i] != oldValue)
System.out.println("**************************");
}
} // end of main
}
Here is the output from a run:

% java H11EC
Position: 0 changed from 10 to 3; Position 0 corrected to 10
Position: 2 changed from 1 to 1;
218 Program II.7.e
. . . (many lines omitted) . . .

Program II.7.f
Hamming mod 11 Scheme, Double Errors
Here is a Java program that simulates random double errors. There are two input parameters
on the command line: the number of simulation trials to run, and the number of digits to use.
The program identifies various outcomes:
Outcome % for 18 digits % for 121 digits

Positions 1 and 11 give location 9.19% 10.00%
of error, but position 0 gives
amount of error as 0.
The check tries to correct an er- 75.01% 0.00%
ror with location out of range of
the number.
Check tries to do single error 1.30% 8.51%
correction with a ten (X) some-
where besides positions 0, 1, or
11
Misses error: miscorrects as 14.44% 81.48%
if there were a single error
Thus, with 18 digits, only 14.4% of double errors get ”corrected” as if there were a single error,
while with 121 digits 81.5% are miscorrected in this way.
Java class: H11ED
// H11ED.java: Implement the mod 11 Hamming code
public class H11ED {
public static int[] inv = {0, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1};
public static int totalTrials, misCorrected = 0,

errorZero = 0, toTen = 0, subscriptRange = 0, allZero = 0;
public static int arraySize;
public static int[][] pos = {
{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10},
{0, 6, 1, 7, 2, 8, 3, 9, 4, 10, 5},
{0, 4, 8, 1, 5, 9, 2, 6, 10, 3, 7},
{0, 3, 6, 9, 1, 4, 7, 10, 2, 5, 8},
{0, 9, 7, 5, 3, 1, 10, 8, 6, 4, 2},
{0, 2, 4, 6, 8, 10, 1, 3, 5, 7, 9},
{0, 8, 5, 2, 10, 7, 4, 1, 9, 6, 3},
{0, 7, 3, 10, 6, 2, 9, 5, 1, 8, 4},
{0, 5, 10, 4, 9, 3, 8, 2, 7, 1, 6},
{0, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1}};
220 Program II.7.f

if (a[i] == 10)
else
}
}

System.out.print("? "); // position 0
System.out.print("?"); // position 1
if (i == 11) System.out.print("?"); // position 11
}
}
public static void insertCheck(int[] a) {

}

int error = sum0(a); // amount of error
if (error == 0 && check1 == 0 && check11 == 0) {
allZero++;
return true;
}
if (error == 0) {
// System.out.println("Double error: check 0 is zero");
errorZero++;
return false; // a double error
}
int position = pos[error][check11]*11 + pos[error][check1];
if (position >= a.length) {
subscriptRange++;
return false;
}
a[position] = (a[position] - error + 11)%11;
if (a[position] == 10 && (position != 0 && position != 1 &&
position != 11)) {
toTen++;
return false;
}
misCorrected++;
return true;
}

int check = 0;
return check;
}

int check = 0;
return check;
}

int check = 0;
check = (check + (i%11)*a[i])%11;
return check;
}

int check = 0;
check = (check + (i%11)*a[i])%11;
return check;
}

int check = 0;
check = (check + ((i/11)%11)*a[i])%11;
return check;
}

int check = 0;
check = (check + ((i/11)%11)*a[i])%11;
return check;
}
// main function
totalTrials = Integer.parseInt(args[0]); // total num of trials
arraySize = Integer.parseInt(args[1]); // size of array
int[] a = new int[arraySize];
int loc1, loc2;
222 Program II.7.f
// start with random word

if (i != 11) a[i] = (int)(Math.random() * 10.0);
H11ED.insertCheck(a);
for (int i = 0; i < totalTrials; i++) {
// try random pair of errors, choose 2 distinct random ints
loc1 = (int)(Math.random() * a.length);
do {
loc2 = (int)(Math.random() * a.length);
} while (loc1 == loc2);
a[loc1] = (a[loc1] + (int)(Math.random() * 9.0 + 1.0))%10;
a[loc2] = (a[loc2] + (int)(Math.random() * 9.0 + 1.0))%10;
H11ED.doCheck(a);
}
if (totalTrials != (misCorrected + errorZero + toTen +
subscriptRange + allZero))
System.out.println("Count Off");
System.out.println("Total: " + totalTrials +
", errorZero: " + errorZero +
", toTen: " + toTen +
", subscript: " + subscriptRange +
", misCorrected: " + misCorrected +
", allZero: " + allZero);
} // end of main
}
Here are the results of two runs, the first using 18 digits (15 data digits) and the second using
the maximum of 121 digits (118 data digits). In each case a large number of double errors were
deliberately introduced. In the first case, all but 14.9% of these double errors were detected.
In the second case only 18.5% of double errors were detected.
% myjava H11ED 10000000 18
Total: 10000000, errorZero: 919248, toTen: 130462, subscript: 7501457,
misCorrected: 1444074, allZero: 4759
% myjava H11ED 1000000 121
Total: 1000000, errorZero: 100002, toTen: 85080, subscript: 0,
misCorrected: 814827, allZero: 91
Program II.8.a
Use of the Dihedral Group
Here is the a scheme using the dihedral group without any special permutations. Notice that
2/3 of all adjacent transpositions are detected (60 out of 90).

if (a[i] == 10)
else
}
}

}
}
}
Java class: DihedralErrorDetection
// DihedralErrorDetection.java: the dihedral group
// for decimal error detection
public class DihedralErrorDetection extends ErrorDetection {
private static int[][] op= {

{0, 1, 2, 3, 4, 5, 6, 7, 8, 9},
{1, 2, 3, 4, 0, 6, 7, 8, 9, 5},
{2, 3, 4, 0, 1, 7, 8, 9, 5, 6},
{3, 4, 0, 1, 2, 8, 9, 5, 6, 7},
{4, 0, 1, 2, 3, 9, 5, 6, 7, 8},
{5, 9, 8, 7, 6, 0, 4, 3, 2 ,1},
{6, 5, 9, 8, 7, 1, 0, 4, 3, 2},
{7, 6, 5, 9, 8, 2, 1, 0, 4, 3},
{8, 7, 6, 5, 9, 3, 2, 1, 0, 4},
{9, 8, 7, 6, 5, 4, 3, 2, 1, 0} };
private static int[] inv = {0, 4, 3, 2, 1, 5, 6, 7, 8, 9};

224 Program II.8.a

int check = 0;
check = op[check][ a[i] ];
a[0] = inv[check];
return a[0];
}

int check = 0;
check = op[check][ a[i] ];
else return true;
}
// main function
DihedralErrorDetection.printUnchecked(a);
DihedralErrorDetection.insertCheck(a);
DihedralErrorDetection.printArray(a);
System.out.println(DihedralErrorDetection.doCheck(a));
a[4] = (a[4] + 1)%10;
printArray(a);
System.out.println(DihedralErrorDetection.doCheck(a));

System.out.println("\nThe straight dihedral group");
System.out.println("\nTest all adjacent transpositions");
for (int p1 = 0; p1 < 10; p1++)
for (int p2 = 0; p2 < 10; p2++) {
if (p1 != p2) {
a[8] = p1; a[9] = p2;
DihedralErrorDetection.insertCheck(a);
// interchange
a[8] ˆ= a[9];
a[9] ˆ= a[8];
a[8] ˆ= a[9];
if (DihedralErrorDetection.doCheck(a)) {
checkFlag = true;
}
}
}
if (checkFlag)
else
} // end of main
}
Here is the output, showing a simple test, and a test of all adjacent interchanges. Notice that 30
(out of 90) adjacent transpositions go undetected.
? 49588 58802 3606
8 49588 58802 3606
true
8 49598 58802 3606
false
The straight dihedral group
Test all adjacent transpositions

Program II.8.b
Verhoeff’s Scheme
Here is Verhoeff’s scheme using the dihedral group using special permutations. Notice that all
adjacent transpositions are detected.

if (a[i] == 10)
else
}
}

}
}
}
Java class: VerhoeffErrorDetection
// VerhoeffErrorDetection.java: Verhoeff’s decimal error detection
public class VerhoeffErrorDetection extends ErrorDetection {
private static int[][] op= {

{0, 1, 2, 3, 4, 5, 6, 7, 8, 9},
{1, 2, 3, 4, 0, 6, 7, 8, 9, 5},
{2, 3, 4, 0, 1, 7, 8, 9, 5, 6},
{3, 4, 0, 1, 2, 8, 9, 5, 6, 7},
{4, 0, 1, 2, 3, 9, 5, 6, 7, 8},
{5, 9, 8, 7, 6, 0, 4, 3, 2 ,1},
{6, 5, 9, 8, 7, 1, 0, 4, 3, 2},
{7, 6, 5, 9, 8, 2, 1, 0, 4, 3},
{8, 7, 6, 5, 9, 3, 2, 1, 0, 4},
{9, 8, 7, 6, 5, 4, 3, 2, 1, 0} };
private static int[] inv = {0, 4, 3, 2, 1, 5, 6, 7, 8, 9};

private static int[][] F = new int[8][];
public VerhoeffErrorDetection() { // identity and "magic" perms

F[0] = new int[]{0, 1, 2, 3, 4, 5, 6, 7, 8, 9}; // identity
F[1] = new int[]{1, 5, 7, 6, 2, 8, 3, 0, 9, 4}; // "magic"
for (int i = 2; i < 8; i++) {
F[i] = new int[10];
for (int j = 0; j < 10; j++)
F[i][j] = F[i-1] [F[1][j]];
}
}

int check = 0;
a[0] = inv[check];
return a[0];
}

int check = 0;
else return true;
}
// main function
VerhoeffErrorDetection v = new VerhoeffErrorDetection();
VerhoeffErrorDetection.printUnchecked(a);
VerhoeffErrorDetection.insertCheck(a);
VerhoeffErrorDetection.printArray(a);
System.out.println(VerhoeffErrorDetection.doCheck(a));
a[4] = (a[4] + 1)%10;
VerhoeffErrorDetection.printArray(a);
System.out.println(VerhoeffErrorDetection.doCheck(a));

System.out.println("\nTest all adjacent transpositions");
for (int p1 = 0; p1 < 10; p1++)
for (int p2 = 0; p2 < 10; p2++) {
if (p1 != p2) {
a[8] = p1; a[9] = p2;
VerhoeffErrorDetection.insertCheck(a);
// interchange
a[8] ˆ= a[9];
a[9] ˆ= a[8];
228 Program III.8.b
a[8] ˆ= a[9];
if (VerhoeffErrorDetection.doCheck(a)) {
checkFlag = true;
}
}
}
if (checkFlag)
else
} // end of main
}
Here is the output, showing a simple test, and a test of all adjacent interchanges. All interchange
errors are detected.
? 75787 12372 9429
1 75787 12372 9429
true
1 75797 12372 9429
false
Test all adjacent transpositions

All transpositions detected
Program III.9.a
Cryptogram Program
Here is a Java program to create standard cryptograms, as they are found in newspapers. The
program will read the quotation to be scrambled into a cryptogram from the standard input. In
Unix this file can just be directed into the program, as shown in the commands below. Each
time it is executed, the program will create a new and unique translation table to create the
cryptogram. The resulting table and cryptogram itself are output on the standard output file,
which might be redirected into a named file.
Java class: Cryptogram
// Cryptogram: create a cryptogram as in a newspaper
import java.io.*;
public class Cryptogram {
private char[] alf = new char[26]; // translation vector

public Cryptogram() {
for (int i = 0; i < alf.length; i++) alf[i] = (char)(’A’ + i);
randomize();
}
private int rand(int r, int s) { // r <= rand <= s
return (int)((s - r + 1)*Math.random() + r);
}
private void randomize() {
for (int i = 0; i < alf.length - 1; i++) {
// Note: for a random permutation, replace "i+1" by "i" below
// However, we want no letter to remain in its original spot
int ind = rand(i+1, alf.length - 1);
char t = alf[i];
alf[i] = alf[ind];
alf[ind] = t;
}
}
public void printArray() {
System.out.print("Alphabet: ");
for (int i = 0; i < alf.length; i++)
System.out.print((char)(’A’ + i));
System.out.print("Translated to: ");
for (int i = 0; i < alf.length; i++)
System.out.print(alf[i]);
System.out.println("\n");
}
// getNextChar: fetch next char.
public char getNextChar() {
try {
ch = (char)System.in.read();
230 Program III.9.a
}
return ch;
}
public void createCryptogram() {
char ch;
while ((byte)(ch = getNextChar()) != -1) {
if (Character.isUpperCase(ch)) ch = alf[ch - ’A’];
System.out.print(ch);
}
}
// main: for cryptogram program
Cryptogram crypto = new Cryptogram();
crypto.printArray();
crypto.createCryptogram();
}
}
Here is a run of the program, first showing the quotation to be translated, and then the translated
version, that is, the cryptogram:
% cat quote.text
AND WE ARE HERE AS ON A DARKLING PLAIN
SWEPT WITH CONFUSED ALARMS OF STRUGGLE AND FLIGHT,
WHERE IGNORANT ARMIES CLASH BY NIGHT.
DOVER BEACH, MATHEW ARNOLD
% java Cryptogram < quote.text
Translated to: ZUWYMPILBDJRVFHQSGAXNCTKOE
ZFY TM ZGM LMGM ZA HF Z YZGJRBFI QRZBF

ATMQX TBXL WHFPNAMY ZRZGVA HP AXGNIIRM ZFY PRBILX,
TLMGM BIFHGZFX ZGVBMA WRZAL UO FBILX.
YHCMG UMZWL, VZXLMT ZGFHRY
Now suppose one wants to have nothing but the letters in the cryptogram (no spaces, newlines,
or other punctuation). This is the same as the other program, except that the createCryp-
togram() has become:
public void createCryptogram()
char ch;
while ((byte)(ch = getNextChar()) != -1)
if (Character.isUpperCase(ch))
ch = alf[ch - ’A’];
System.out.print(ch);
Here is the output of this program:

% java Cryptogram2 < quote.text
Translated to: OXKQFDZGACIVLHEJSMBWPNURTY
OHQUFOMFGFMFOBEHOQOMIVAHZJVOAHBUFJWUAWGKEHDPBFQOVOMLBEDBWMPZZVF
OHQDVAZGWUGFMFAZHEMOHWOMLAFBKVOBGXTHAZGWQENFMXFOKGLOWGFUOMHEVQ
Here is what the message looks like after decrypting:
ANDWEAREHEREASONADARKLINGPLAINSWEPTWITHCONFUSEDALARMSOFSTRUGGLE
ANDFLIGHTWHEREIGNORANTARMIESCLASHBYNIGHTDOVERBEACHMATHEWARNOLD
Program III.10.a
Caesar Cipher
Here is a Java implementation of the Caesar cipher. The program reads the input message from
the standard input and outputs the ciphertext on the standard output. The key is an integer in
the range from 0 to 25 inclusive, and one must use the same key for encryption and decryption.
The program uses a function rotate to translate lowercase characters in a circle by a distance
of key.
Java class: Caesar
// Caesar.java: implement the Caesar cipher
// This carries out a simple rotation of lower-case letters, and
// does nothing to all other characters, making the decryption
// process even easier, because caps and punctuation marks survive
// unchanged.
// Usage: java Caesar (-d | -e) key
// Above, option "-d" is for decryption, "-e" is for encryption
import java.io.*;
public class Caesar {
private Reader in; // standard input stream for message
private int key; // (en|de)cryption key
// Caesar: constructor, opens standard input, passes key

public Caesar(int k) {
// open file
key = k;
}
// (en|de)crypt: just feed in opposite parameters
public void encrypt() { translate(key); }
public void decrypt() { translate(-key); }
// translate: input message, translate

private void translate(int k) {
char c;
while ((byte)(c = getNextChar()) != -1) {
if (Character.isLowerCase(c)) {
c = rotate(c, k);
}
System.out.print(c);
}
}
// getNextChar: fetches next char.

public char getNextChar() {
try {
10. The One-Time Pad 233
}
return ch;
}
// rotate: translate using rotation, version with table lookup
public char rotate(char c, int key) { // c must be lowercase
String s = "abcdefghijklmnopqrstuvwxyz";
int i = 0;
while (i < 26) {
// extra +26 below because key might be negative
if (c == s.charAt(i)) return s.charAt((i + key + 26)%26);
i++;
}
return c;
}
// main: check command, (en|de)crypt, feed in key value

if (args.length != 2) {
System.out.println("Usage: java Caesar (-d | -e) key");
System.exit(1);
}
Caesar cipher = new Caesar(Integer.parseInt(args[1]));
if (args[0].equals("-e")) cipher.encrypt();
else if (args[0].equals("-d")) cipher.decrypt();
else {
System.out.println("Usage: java Caesar (-d | -e) key");
System.exit(1);
}
}
}
Here is the result of an initial run of the program. First is the message file (a quotation for Ec-
clesiastes), followed by encryption by the key 3, and then by encryption followed by decryption
(both using the same key), showing that the original message results. An simple analysis of
the ciphertext would show a distribution of letters that would immediately lead to breaking the
code. Notice also that the plaintext and ciphertext both end in double letters (ll and oo).
% cat message.text
i returned, and saw under the sun, that the race is not to the swift,
nor the battle to the strong, neither yet bread to the wise, nor yet
riches to men of understanding, nor yet favour to men of skill; but
time and chance happeneth to them all.
% java Caesar -e 3 < message.text

l uhwxuqhg, dqg vdz xqghu wkh vxq, wkdw wkh udfh lv qrw wr wkh vzliw,
qru wkh edwwoh wr wkh vwurqj, qhlwkhu bhw euhdg wr wkh zlvh, qru bhw
ulfkhv wr phq ri xqghuvwdqglqj, qru bhw idyrxu wr phq ri vnloo; exw
wlph dqg fkdqfh kdsshqhwk wr wkhp doo.
% java Caesar -e 3 < message.text | java Caesar -d 3

234 Program III.10.a
Notice that the ciphertext and decrypted plaintext both have all the original punctuation charac-
ters, making it even easier to break the system. A more reasonable system would drop all such
punctuation characters from the ciphertext. The one must break the decrypted ciphertext into
separate words — a task that is not hard for English. Here is a run of the program that has been
altered to discard all characters except for lower-case letters. The final version of the message
shows the words run together.
% java Caesar2 -e 3 < message.text
luhwxuqhgdqgvdzxqghuwkhvxqwkdwwkhudfhlvqrwwrwkhvzliwqruwkhedwwo
hwrwkhvwurqjqhlwkhubhweuhdgwrwkhzlvhqrubhwulfkhvwrphqrixqghuvwd
qglqjqrubhwidyrxuwrphqrivnlooexwwlphdqgfkdqfhkdsshqhwkwrwkhpdoo
% java Caesar2 -e 3 < message.text | java Caesar2 -d 3

ireturnedandsawunderthesunthattheraceisnottotheswiftnorthebattl
etothestrongneitheryetbreadtothewisenoryetrichestomenofundersta
ndingnoryetfavourtomenofskillbuttimeandchancehappenethtothemall
Program III.10.b
Beale Cipher
Here is a Java implementation of the Beale cipher. As with the Caesar cipher, the program reads
the input message from the standard input and outputs the ciphertext on the standard output.
However, this program also reads a file to use as the key: key1.text in the first run and
key1.text in the second run. The first key file is very simple — just the letter d repeated
over and over. This shows that the Beale cipher includes the Caesar cipher as a special case.
The second key file is just another quotation (from B.F. Skinner). The program Beale.java
only uses successive lowercase letters from a key file (and discards the other letters).
Java class: Beale
// Beale.java: implement the Beale cipher
// Usage: java Beale (-d | -e) keyFile
// The program reads a separate file (keyFile) for the key or pad.
// The message is just the standard input.
// Caps and punctuation marks in the message remain unchanged.
// Only lowercase letters are used in the key file/
import java.io.*;
public class Beale {
private Reader messIn; // System.in for message
private Reader keyIn; // keyFile for key file
// Beale: constructor -- just open files

public Beale(String keyFile) {
messIn = new InputStreamReader(System.in);
try {
keyIn = new FileReader(keyFile);
System.out.println("Exception opening keyFile");
}
}
// (en|de)crypt: just feed in opposite parameters

public void encrypt() { translate(1); }
public void decrypt() { translate(-1); }
// translate: read keyFile and input, translate

private void translate(int direction) {
char c, key_c;
while ((byte)(c = getNextChar(messIn)) != -1) {
if (Character.isLowerCase(c)) {
// fetch lowercase letter from key file
while (!Character.isLowerCase(key_c = getNextChar(keyIn)))
;
c = rotate(c, ((direction*(key_c - ’a’)) + 26)%26);
}
System.out.print(c);
236 Program III.10.b
}
}
// getNextChar: fetches next char. Also opens input file

public char getNextChar(Reader in) {
try {
}
return ch;
}
// rotate: translate using rotation -- simpler version

// This just uses arithmetic on char types,
public char rotate(char c, int key) {
int res = ((c - ’a’) + key + 26)%26 + ’a’;
return (char)res;
}
// main: check command, (en|de)crypt, feed in keyFile

System.out.println("Usage: java Beale (-d | -e) keyFile");
System.exit(1);
}
Beale cipher = new Beale(args[1]);
if (args[0].equals("-e")) cipher.encrypt();
else if (args[0].equals("-d")) cipher.decrypt();
else {
System.out.println("Usage: java Beale (-d | -e) keyFile");
System.exit(1);
}
}
}
Here are the results of a run of the program where the key file consists of all letters ”d”, so that
is does the same rotation by 3 as the previous example of the Caesar cipher:
% cat message.text
% cat key1.text
dddddddddddddddddddddddddddddddddddddddddddddd
% java Beale -e key1.text < message.text

l uhwxuqhg, dqg vdz xqghu wkh vxq, wkdw wkh udfh lv qrw wr wkh vzliw,
qru wkh edwwoh wr wkh vwurqj, qhlwkhu bhw euhdg wr wkh zlvh, qru bhw
ulfkhv wr phq ri xqghuvwdqglqj, qru bhw idyrxu wr phq ri vnloo; exw
wlph dqg fkdqfh kdsshqhwk wr wkhp doo.
% java Beale -e key1.text < message.text | java Beale -d key.text

Notice that the ciphertext and decrpted plaintext has all the original punctuation characters,
making it easier to break the system. Here is a run of a variation of the program that discards
all characters except for lower-case letters:
% java Beale2 -e key1.text < message.text
luhwxuqhgdqgvdzxqghuwkhvxqwkdwwkhudfhlvqrwwrwkhvzliwqruwkhedwwo
hwrwkhvwurqjqhlwkhubhweuhdgwrwkhzlvhqrubhwulfkhvwrphqrixqghuvwd
qglqjqrubhwidyrxuwrphqrivnlooexwwlphdqgfkdqfhkdsshqhwkwrwkhpdoo
% java Beale2 -e key1.text < message.text | java Beale2 -d key1.text

Finally, here is a real run with an actual non-trivial key file: key2.text. Recall that without
knowing the text of the key file, this would be very difficult to cryptanalyze (break). For
example, as before, the message ends with double letters: ll. However, this time the ciphertext
ends with two different letters: gp.
% cat key2.text
A Golden Age, whether of art or music or science or peace or
plenty, is out of reach of our economic and governmental techniques.
something may be done by accident, as it has from time to time in the
past, but not by deliberate intent. At this very moment enormous
numbers of intelligent men and women of good will are trying to build a
better world. But problems are born faster than they can be solved.
% java Beale -e key2.text < message.text

w chxhxrak, egk wrk znuxf kty kcp, hysv blr teqv xw nqx hf isi fpgnl,
bik hmv favazj hi klg ggfavi, nrlzvzv prf fexao ms vor eymi, fgf kim
yqpnqs rp qhb bj vldgtaweawifo, gvr qjk tmowgv mc fmz sn fdppa; bmm
ucfr oge akeykf lrpiivrml gh malu sgp.
% java Beale -e key2.text < message.text | java Beale -d key2.text

Here as before is a run that discards the punctuation characters:
% java Beale2 -e key2.text < message.text
wchxhxrakegkwrkznuxfktykcphysvblrteqvxwnqxhfisifpgnlbikhmvfavaz
238 Program III.10.b
jhiklgggfavinrlzvzvprffexaomsvoreymifgfkimyqpnqsrpqhbbjvldgtawe
awifogvrqjktmowgvmcfmzsnfdppabmmucfrogeakeykflrpiivrmlghmalusgp
% java Beale2 -e key2.text < message.text | java Beale2 -d key2.text

Program III.10.c
Generating a One-Time Pad
Here is a Java program which, when executed, generates a unique Postscript file that will
print two copies of 1000 random letters for a one-time pad. This program is just for simple
demonstration purposes, and would not be suitable for applications because of weaknesses in
the random number generator and its seed. A program for actual use would need a more secure
generator with at least a 128-bit seed made up from various random inputs.
Java class: Pad
// Pad.java: generate Postscript code to print a one-time pad
public class Pad {

static DecimalFormat twoDigits = new DecimalFormat("00");
static char[] let =
{’A’,’B’,’C’,’D’,’E’,’F’,’G’,’H’,’I’,’J’,’K’,’L’,’M’,
’N’,’O’,’P’,’Q’,’R’,’S’,’T’,’U’,’V’,’W’,’X’,’Y’,’Z’};
static int xCoord = 0, yCoord = 0, lineCount = 0;

System.out.println("%!PS-Adobe-2.0");
System.out.println("/Courier-Bold findfont 14 scalefont setfont");
System.out.println("/onepad {");
for (int i = 0; i < 20; i++) {
System.out.println("0 " + yCoord + " moveto");
System.out.print("(" + twoDigits.format(lineCount) + " ");
for (int j = 0; j < 50; j++) {
System.out.print(oneLet());
if (j%5 == 4) System.out.print(" ");
if (j%10 == 9) System.out.print(" ");
}
System.out.println(") show");
yCoord -= 15;
if (lineCount == 9) System.out.println();
if (lineCount%5 == 4) yCoord -= 15;
lineCount++;
}
System.out.println("}def");
System.out.println("gsave 30 750 translate onepad grestore");
System.out.println("gsave 30 360 translate onepad grestore");
System.out.println("10 390 moveto");
System.out.print ("(============t=e=a=r===h=e=r=e=======");
System.out.println("=======t=e=a=r===h=e=r=e==========) show");
System.out.println("showpage");
} // end of main
private static char oneLet() { return let[(int)(Math.random()*26)];}

240 Program III.10.c
Here is typical output of the program, a Postscript file. (It will have different random characters
in it each time it is generated by executing the Java program.)
%!PS-Adobe-2.0
/Courier-Bold findfont 14 scalefont setfont
/onepad {
0 0 moveto
(00 XLCWT HZZTC HUTXA GQAUN FXCUI QFBVW DKAPS SXKHK XBLLP LTHFO ) show
0 -15 moveto
(01 XUMVS SRMGB SPSDI UAFYO CQYHQ CYSHU UCATL HLDKZ XWFGR LRMOL ) show
0 -30 moveto
(02 DXCBR FMARY MOLUR CDJVT MACWT IRFVP DZBBY ZTFZG HCUUL YYKPF ) show
0 -45 moveto
(03 QSWIH HNWRC FQQHM ZEKJP URILH THNSY TLMWA CWYJV AWOWY IAELZ ) show
0 -60 moveto
(04 KJKJE NGKDD NYMYP NRNHK KRIOX TIFSY TGYRS HZLTQ MCEBA YRBTU ) show
0 -90 moveto
(05 CZLFW HYSTM YGDBN GVAZE RFCQY YFHLC OGLHY OEBNB QDQIS EHIBB ) show
0 -105 moveto
(06 ZIWQC TYERS UZGSS EGOVX KGXJV AONYN HXZCR RLUSN SXXBR EJKQH ) show
0 -120 moveto
(07 OIJMF LCZHA CGMYK LPLMU CIMHZ WBVHD JXNZJ CJLSN PJVFL HENCU ) show
0 -135 moveto
(08 ZOHEW JRWPC BHFRZ MEYBW SHFPQ DYQFH ARDQI UAHOC OAQKR LZPBF ) show
0 -150 moveto
(09 YAKMA YJVYY HSAZB SMILL WMHAJ KMYNO YHTXY GIISP CVYQC OMCIP ) show
0 -180 moveto
(10 NPNVY VACCU YWOFW GLHTO TOUPS LRLOV SZWBL WYNJO NOHCZ SZSZS ) show
0 -195 moveto
(11 UBSOT FSRSM BLDEC IXQDZ STPAJ YOMBO CGIAD HUTHZ JAURL MCRNY ) show
0 -210 moveto
(12 VPVSF RSXSQ UMOMQ KGJCY BIJXE QQDFF BHOGG RCTXO KJRMC QINMZ ) show
0 -225 moveto
(13 BEFIF UIEVB UGWZJ ENOLD CTNBD JWKCR JTIKX DIRYK WUZPH HGTAU ) show
0 -240 moveto
(14 IPAOV IHDQA VENRD RCFLR ZRTOW NIHWY TGWQX WWRHY QSBMD IWSMQ ) show
0 -270 moveto
(15 IYCMO WUPZM OFPJW ZUUDB QZIHP RNQTT UMTDS ANATC DANSQ YUYGT ) show
0 -285 moveto
(16 XYCZU AKGWK ZPEMS TAQLE KKIYG UANUN ZNTPP NZVTX KPZCD BUYMT ) show
0 -300 moveto
(17 DKZSJ SIMPP EAFSS KNBOO OATKY UKMYP NZPGS ELIHP JWWAQ ILRBV ) show
0 -315 moveto
(18 RSBTM XROFG MPGJT HIEIZ XRSNN FIRYQ AFGGR BYVQY RXCFV CBBUO ) show
0 -330 moveto
(19 ZTRPC HHVJO BPXHL UGNLP CLVUN DTGZI NTCBM EDBRG ZQUOY PFZHO ) show
}def
gsave 30 750 translate onepad grestore
gsave 30 360 translate onepad grestore
10 390 moveto
(============t=e=a=r===h=e=r=e==============t=e=a=r===h=e=r=e==========) show
showpage
Here is the page that will be generated by the above Postscript program (shrunk to 60% of
normal size).
00 XLCWT HZZTC HUTXA GQAUN FXCUI QFBVW DKAPS SXKHK XBLLP LTHFO
01 XUMVS SRMGB SPSDI UAFYO CQYHQ CYSHU UCATL HLDKZ XWFGR LRMOL
02 DXCBR FMARY MOLUR CDJVT MACWT IRFVP DZBBY ZTFZG HCUUL YYKPF
03 QSWIH HNWRC FQQHM ZEKJP URILH THNSY TLMWA CWYJV AWOWY IAELZ
04 KJKJE NGKDD NYMYP NRNHK KRIOX TIFSY TGYRS HZLTQ MCEBA YRBTU
05 CZLFW HYSTM YGDBN GVAZE RFCQY YFHLC OGLHY OEBNB QDQIS EHIBB
06 ZIWQC TYERS UZGSS EGOVX KGXJV AONYN HXZCR RLUSN SXXBR EJKQH
07 OIJMF LCZHA CGMYK LPLMU CIMHZ WBVHD JXNZJ CJLSN PJVFL HENCU
08 ZOHEW JRWPC BHFRZ MEYBW SHFPQ DYQFH ARDQI UAHOC OAQKR LZPBF
09 YAKMA YJVYY HSAZB SMILL WMHAJ KMYNO YHTXY GIISP CVYQC OMCIP
10 NPNVY VACCU YWOFW GLHTO TOUPS LRLOV SZWBL WYNJO NOHCZ SZSZS
11 UBSOT FSRSM BLDEC IXQDZ STPAJ YOMBO CGIAD HUTHZ JAURL MCRNY
12 VPVSF RSXSQ UMOMQ KGJCY BIJXE QQDFF BHOGG RCTXO KJRMC QINMZ
13 BEFIF UIEVB UGWZJ ENOLD CTNBD JWKCR JTIKX DIRYK WUZPH HGTAU
14 IPAOV IHDQA VENRD RCFLR ZRTOW NIHWY TGWQX WWRHY QSBMD IWSMQ
15 IYCMO WUPZM OFPJW ZUUDB QZIHP RNQTT UMTDS ANATC DANSQ YUYGT
16 XYCZU AKGWK ZPEMS TAQLE KKIYG UANUN ZNTPP NZVTX KPZCD BUYMT
17 DKZSJ SIMPP EAFSS KNBOO OATKY UKMYP NZPGS ELIHP JWWAQ ILRBV
18 RSBTM XROFG MPGJT HIEIZ XRSNN FIRYQ AFGGR BYVQY RXCFV CBBUO
19 ZTRPC HHVJO BPXHL UGNLP CLVUN DTGZI NTCBM EDBRG ZQUOY PFZHO
============t=e=a=r===h=e=r=e==============t=e=a=r===h=e=r=e==========
00 XLCWT HZZTC HUTXA GQAUN FXCUI QFBVW DKAPS SXKHK XBLLP LTHFO
01 XUMVS SRMGB SPSDI UAFYO CQYHQ CYSHU UCATL HLDKZ XWFGR LRMOL
02 DXCBR FMARY MOLUR CDJVT MACWT IRFVP DZBBY ZTFZG HCUUL YYKPF
03 QSWIH HNWRC FQQHM ZEKJP URILH THNSY TLMWA CWYJV AWOWY IAELZ
04 KJKJE NGKDD NYMYP NRNHK KRIOX TIFSY TGYRS HZLTQ MCEBA YRBTU
05 CZLFW HYSTM YGDBN GVAZE RFCQY YFHLC OGLHY OEBNB QDQIS EHIBB
06 ZIWQC TYERS UZGSS EGOVX KGXJV AONYN HXZCR RLUSN SXXBR EJKQH
07 OIJMF LCZHA CGMYK LPLMU CIMHZ WBVHD JXNZJ CJLSN PJVFL HENCU
08 ZOHEW JRWPC BHFRZ MEYBW SHFPQ DYQFH ARDQI UAHOC OAQKR LZPBF
09 YAKMA YJVYY HSAZB SMILL WMHAJ KMYNO YHTXY GIISP CVYQC OMCIP
10 NPNVY VACCU YWOFW GLHTO TOUPS LRLOV SZWBL WYNJO NOHCZ SZSZS
11 UBSOT FSRSM BLDEC IXQDZ STPAJ YOMBO CGIAD HUTHZ JAURL MCRNY
12 VPVSF RSXSQ UMOMQ KGJCY BIJXE QQDFF BHOGG RCTXO KJRMC QINMZ
13 BEFIF UIEVB UGWZJ ENOLD CTNBD JWKCR JTIKX DIRYK WUZPH HGTAU
14 IPAOV IHDQA VENRD RCFLR ZRTOW NIHWY TGWQX WWRHY QSBMD IWSMQ
15 IYCMO WUPZM OFPJW ZUUDB QZIHP RNQTT UMTDS ANATC DANSQ YUYGT
16 XYCZU AKGWK ZPEMS TAQLE KKIYG UANUN ZNTPP NZVTX KPZCD BUYMT
17 DKZSJ SIMPP EAFSS KNBOO OATKY UKMYP NZPGS ELIHP JWWAQ ILRBV
18 RSBTM XROFG MPGJT HIEIZ XRSNN FIRYQ AFGGR BYVQY RXCFV CBBUO
19 ZTRPC HHVJO BPXHL UGNLP CLVUN DTGZI NTCBM EDBRG ZQUOY PFZHO
Program III.10.d
Circles for a One-Time Pad
Here are two circles for use in creating a rotating tool for making use of one-time pad characters.
Each image has been shrunk by 60%.
Outer letters belong to the ciphertext or to the pad.
25 0 1
24 A B 2
Z
Y C 3
23 D
X
22
4
W
E
21
5
V
F
20
6
G
U
Attach smaller circle here.

19
H
T
7
S
I
18
8
R
J
17
Q K
10
16 P L
O N M 11
15
14 13 12
Place arrow of inner circle on pad character.

To encrypt, follow inner circle plaintext letter to outer circle ciphertext letter.
To decrypt, follow outer circle ciphertext letter to inner circle plaintext letter.
Copyright  2003 by Neal R. Wagner. All rights reserved.
Z A B
C
Y 25 0 1
24 2 D
X 3
23
9 20 21 W
E
22
4
Inner letters are
T U V
F
5
plaintext characters
G
7 8 6
H I
R 18 1
Set arrow on pad character

S
7
9
1
J
10
11 16 K
1 5 Q
L 12 13 14 P
M N O
(Cut along dashed line.)

244 Program III.10.d
Here is Postscript source to create the larger circle (at full size):
%!PS-Adobe-2.0
/r 360 26 div def
/inch { 72 mul} def
/Tempstr 2 string def
/radius 225 def
/circleofLetters {
[(A) (B) (C) (D) (E) (F) (G) (H) (I) (J)
(K) (L) (M) (N) (O) (P) (Q) (R) (S) (T)
(U) (V) (W) (X) (Y) (Z)]
/ary exch def % the array of letters
0 1 25{ % from 0 to 25 in steps of 1

/ind exch def % the for loop value
/rind ind 0 eq {0} {26 ind sub} ifelse def
gsave
ind r mul neg rotate % rotate by (for loop value)*360/26
/Helvetica-Bold findfont 23 scalefont setfont
ary ind get stringwidth pop 2 div neg radius 5 sub moveto
ary ind get show
% convert ind to string, store in Tempstr, using cvs
ind Tempstr cvs stringwidth pop 2 div neg radius 25 add moveto
ind Tempstr cvs show
3 setlinewidth
0 radius 30 sub moveto 0 20 rlineto stroke
grestore
} for
}def
/circles {
3 setlinewidth
newpath 0 0 radius 20 sub 0 360 arc stroke
newpath 0 0 6 0 360 arc stroke
} def
40 80 moveto (Place arrow of inner circle on pad character.) show
40 60 moveto (To encrypt, follow inner circle plaintext) show
( letter to outer circle ciphertext letter.) show
/Helvetica-BoldOblique findfont 15 scalefont setfont
40 40 moveto (To decrypt, follow outer circle ciphertext) show
( letterto inner circle plaintext letter.) show
/Helvetica-BoldOblique findfont 15 scalefont setfont
130 700 moveto (Outer letters belong to the ciphertext or to the pad.) show
8.5 inch 2 div 11 inch 2 div translate

-90 20 moveto (Attach smaller circle here.) show
circleofLetters circles showpage
Here is Postscript source to create the smaller circle (at full size):
%!PS-Adobe-2.0
/r 360 26 div def
/inch { 72 mul} def
/Tempstr 2 string def
/radius 120 def
/circleofLetters {
[(A) (B) (C) (D) (E) (F) (G) (H) (I) (J)
(K) (L) (M) (N) (O) (P) (Q) (R) (S) (T)
(U) (V) (W) (X) (Y) (Z)]
/ary exch def % the array of letters
0 1 25{ % from 0 to 25 in steps of 1

/ind exch def % the for loop value
gsave
ind r mul neg rotate % rotate by (for loop value)*360/26
ary ind get stringwidth pop 2 div neg radius 30 add moveto
ary ind get show
% convert ind to string, store in Tempstr, using cvs
ind Tempstr cvs stringwidth pop 2 div neg radius 10 add moveto
ind Tempstr cvs show
3 setlinewidth
0 radius 50 add moveto 0 20 rlineto stroke
grestore
} for
}def
/circles {
3 setlinewidth
0 0 radius 60 add 0 360 arc stroke
0 0 6 0 360 arc stroke
1 setlinewidth
0 0 radius 80 add 0 360 arc [5 3] 0 setdash stroke
} def
/arrowhead {
newpath 0 radius 78 add moveto
-5 -12 rlineto 10 0 rlineto closepath fill
} def
8.5 inch 2 div 11 inch 2 div translate

circleofLetters circles arrowhead
-45 50 moveto (Inner letters are) show
-60 35 moveto (plaintext characters) show
-80 -50 moveto (Set arrow on pad character) show
-75 radius 100 add neg moveto ((Cut along dashed line.))show
showpage
Program IV.14.a
RSA Implementation
This Java implementation of the basic RSA cryptosystem uses the Java BigInteger library
class. This is just a “skeleton” implementation that creates keys from scratch and uses them,
but does not save keys to a file for repeated use, or fetch such keys from the file. For further
comments about the implementation, see the chapter on the RSA cryptosystem.
This code implements RSA using 3 Java classes:
:
RSAPublicKey: The data and methods needed for RSA public keys, with the modu-
lus
and exponent , along with a username to keep the keys straight. The important
methods are encryption and verification.

RSAPrivateKey: This extends the previous class to add the primes and , and the
:
decryption exponent as data members. Important methods include decryption and sign-
ing, along with key generation.
RSATest: A class to test out the system with realistic key sizes (1024 bits).
Java class: RSAPublicKey

// RSAPublicKey: RSA public key
import java.math.*; // for BigInteger
public class RSAPublicKey {
public BigInteger n; // public modulus
public BigInteger e = new BigInteger("3"); // encryption exponent
public String userName; // attach name to each public/private key pair
public RSAPublicKey(String name) {

userName = name;
}
// setN: to give n a value in case only have public key

public void setN(BigInteger newN) {
n = newN;
}
// getN: provide n
public BigInteger getN() {
return n;
}
// RSAEncrypt: just raise m to power e (3) mod n

public BigInteger RSAEncrypt(BigInteger m) {
return m.modPow(e, n);
}
14. The RSA Cryptosystem 247
// RSAVerify: same as encryption, since RSA is symmetric

public BigInteger RSAVerify(BigInteger s) {
return s.modPow(e, n);
}
}
Java class: RSAPrivateKey

// RSAPrivateKey: RSA private key
import java.util.*; // for Random
public class RSAPrivateKey extends RSAPublicKey{
private final BigInteger TWO = new BigInteger("2");
private final BigInteger THREE = new BigInteger("3");
private BigInteger p; // first prime

private BigInteger q; // second prime
private BigInteger d; // decryption exponent
public RSAPrivateKey(int size, Random rnd, String name) {

super(name); generateKeyPair(size, rnd);
}
public void generateKeyPair(int size, Random rnd) { // size = n (bits)

// want sizes of primes close, not too close (10-20 bits).
int size1 = size/2;
int size2 = size1;
int offset1 = (int)(5.0*(rnd.nextDouble()) + 5.0);
int offset2 = -offset1;
if (rnd.nextDouble() < 0.5) {
offset1 = -offset1; offset2 = -offset2;
}
size1 += offset1; size2 += offset2;
// generate two random primes, so that p*q = n has size bits
BigInteger p1 = new BigInteger(size1, rnd); // random int
p = nextPrime(p1);
BigInteger pM1 = p.subtract(BigInteger.ONE);
BigInteger q1 = new BigInteger(size2, rnd);
q = nextPrime(q1);
BigInteger qM1 = q.subtract(BigInteger.ONE);
n = p.multiply(q);
BigInteger phiN = pM1.multiply(qM1); // (p-1)*(q-1)
BigInteger e = THREE;
d = e.modInverse(phiN);
}
// nextPrime: next prime p after x, with p-1 and 3 relatively prime

public BigInteger nextPrime(BigInteger x) {
if ((x.remainder(TWO)).equals(BigInteger.ZERO))
x = x.add(BigInteger.ONE);
while(true) {
BigInteger xM1 = x.subtract(BigInteger.ONE);
if (!(xM1.remainder(THREE)).equals(BigInteger.ZERO))
if (x.isProbablePrime(10)) break;
248 Program IV.14.a
x = x.add(TWO);
}
return x;
}
// RSADecrypt: decryption function

public BigInteger RSADecrypt(BigInteger c) {
return c.modPow(d, n);
}
// RSASign: same as decryption for RSA (since it is a symmetric PKC)

public BigInteger RSASign(BigInteger m) {
return m.modPow(d, n);
}
public BigInteger RSASignAndEncrypt(BigInteger m, RSAPublicKey other) {

// two ways to go, depending on sizes of n and other.getN()
if (n.compareTo(other.getN()) > 0)
return RSASign(other.RSAEncrypt(m));
else
return other.RSAEncrypt(RSASign(m));
}
public BigInteger RSADecryptAndVerify(BigInteger c,

RSAPrivateKey other) {
return other.RSAVerify(RSADecrypt(c));
else
return RSADecrypt(other.RSAVerify(c));
}
}
Java class: RSATest

// RSATest: Test RSA Implementation
public class RSATest {

Random rnd = new Random();
BigInteger m, m1, m2, m3, c, s, s1;
RSAPrivateKey alice = new RSAPrivateKey(1024, rnd, "Alice");
RSAPrivateKey bob = new RSAPrivateKey(1024, rnd, "Bob ");
m = new BigInteger(
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210");
System.out.println("Message m:\n" + m + "\n");
System.out.println("ALICE ENCRYPTS m FOR BOB; BOB DECRYPTS IT:");
c = bob.RSAEncrypt(m); // Using Bob’s public key

System.out.println("Message encrypted with Bob’s public key:\n" +
c + "\n");
m1 = bob.RSADecrypt(c); // Using Bob’s private key
System.out.println("Original message back, decrypted:\n" +
m1 + "\n");
System.out.println("ALICE SIGNS m FOR BOB; BOB VERIFIES SIGNATURE:");

s = alice.RSASign(m); // Using Alice’s private key
System.out.println("Message signed with Alice’s private key:\n" +
c + "\n");
m2 = alice.RSAVerify(s); // Using Alice’s public key
System.out.println("Original message back, verified:\n" +
m2 + "\n");
System.out.println("BOB SIGNS AND ENCRYPTS m FOR ALICE;" +

"\n ALICE VERIFIES SIGNATURE AND DECRYPTS:");
c = bob.RSASignAndEncrypt(m, alice);
System.out.println("Message signed and encrypted," +
"\n using Bob’s secret key and Alice’s public key:\n" +
c + "\n");
m3 = alice.RSADecryptAndVerify(c, bob);
System.out.println("Original message back, verified and decrypted," +
"\n using Alice’s secret key and Bob’s public key:\n" + m1);
}
}
A Test Run.
Here is a run of the above test class, showing simple encryption, signing, and a combination of
signing and encryption. Unix commands appear in boldface.
% javac RSAPublicKey.java
% javac RSAPrivateKey.java
% javac RSATest.java
% java RSATest
Message m:
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
ALICE ENCRYPTS m FOR BOB; BOB DECRYPTS IT:

Message encrypted with Bob’s public key:
623387565362752740557713183298294394842904981992063743592594
444564441837460636112777656456876530809960075397677835706720
975503107091512394844823973429619412227989318053859609889705
833638590603829414072912488421444136560245226367742777088035
320797857638669726447023121838563030894198617138062884887534
61181488
Original message back, decrypted:

123456789098765432101234567890987654321012345678909876543210
250 Program IV.14.a
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
ALICE SIGNS m FOR BOB; BOB VERIFIES SIGNATURE:

Message signed with Alice’s private key:
439372186570975468769351997719781598373182012124463139482307
489690210537347252337014410355961412993510456692671294912453
273016133512641221457438226428152346246137898433600050671846
820367818956782439911588622179202993280665145767078425158675
477487815532190472078890000508679901413377886884336511130898
31991525
Original message back, verified:

123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
BOB SIGNS AND ENCRYPTS m FOR ALICE;

ALICE VERIFIES SIGNATURE AND DECRYPTS:
Message signed and encrypted,
using Bob’s secret key and Alice’s public key:
273343686041287035582131939498270198283482482925968756815127
460868394303184668630498664328401815999198789180360679068712
158591543810756483853639934216530189187930766930230410896090
625811526914278154412722949212590885102373509772635346723555
053689737302508347955040075638919222996974205568230648971866
74601320
Original message back, verified and decrypted,

using Alice’s secret key and Bob’s public key:
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
Program IV.14.b
Faster RSA, Using Chinese Remainder
Theorem
Here is an altered implementation of the RSA cryptosystem, using the the Chinese Remainder
Theorem (CRT) to speed up decryption. Please refer first to the basic RSA as described in the
main RSA section and then to the description of this version of RSA. Additions and changes
related to the faster implementation are highlighted in boldface.
Java class: RSAPublicKey
// RSAPublicKey: RSA public key
// See the listing in the previous section for this class.
Java class: RSAPrivateKeyFast
// RSAPrivateKeyFast: RSA private key, using fast CRT algorithm
public class RSAPrivateKeyFast extends RSAPublicKey{
private final BigInteger TWO = new BigInteger("2");
private final BigInteger THREE = new BigInteger("3");
private BigInteger p; // first prime

private BigInteger q; // second prime
private BigInteger d; // decryption exponent
private BigInteger p1, pM1, q1, qM1, phiN; // for key generation
private BigInteger dp, dq, c2; // for fast decryption
public RSAPrivateKeyFast(int size, Random rnd, String name) {

super(name); generateKeyPair(size, rnd);
}
public void generateKeyPair(int size, Random rnd) { // size = n

// want sizes of primes close, not too close, 10-20 bits apart.
int size1 = size/2;
int size2 = size1;
int offset1 = (int)(5.0*(rnd.nextDouble()) + 5.0);
int offset2 = -offset1;
if (rnd.nextDouble() < 0.5) {
offset1 = -offset1; offset2 = -offset2;
}
size1 += offset1; size2 += offset2;
// generate two random primes, so that p*q = n has size bits
p1 = new BigInteger(size1, rnd); // random int
p = nextPrime(p1);
pM1 = p.subtract(BigInteger.ONE);
q1 = new BigInteger(size2, rnd);
252 Program IV.14.b
q = nextPrime(q1);
qM1 = q.subtract(BigInteger.ONE);
n = p.multiply(q);
phiN = pM1.multiply(qM1); // (p-1)*(q-1)
d = e.modInverse(phiN);
// remaining stuff needed for fast CRT decryption
dp = d.remainder(pM1);
dq = d.remainder(qM1);
c2 = p.modInverse(q);
}
// nextPrime: next prime p after x, with p-1 and 3 rel prime

public BigInteger nextPrime(BigInteger x) {
if ((x.remainder(TWO)).equals(BigInteger.ZERO))
x = x.add(BigInteger.ONE);
while(true) {
BigInteger xM1 = x.subtract(BigInteger.ONE);
if (!(xM1.remainder(THREE)).equals(BigInteger.ZERO))
if (x.isProbablePrime(10)) break;
x = x.add(TWO);
}
return x;
}
// RSADecrypt: decryption function, <b>fast CRT version</b>

public BigInteger RSADecrypt(BigInteger c) {
// See 14.71 and 14.75 in Handbook of Applied Cryptography,
// by Menezes, van Oorschot and Vanstone
BigInteger cDp = c.modPow(dp, p);
BigInteger cDq = c.modPow(dq, q);
BigInteger u = ((cDq.subtract(cDp)).multiply(c2)).remainder(q);
if (u.compareTo(BigInteger.ZERO) < 0) u = u.add(q);
return cDp.add(u.multiply(p));
}
// RSASign: same as decryption for RSA (since it is a symmetric PKC)

public BigInteger RSASign(BigInteger m) {
// return m.modPow(d, n);
return RSADecrypt(m); // use fast CRT version
}
public BigInteger RSASignAndEncrypt(BigInteger m,

RSAPublicKey other) {
return RSASign(other.RSAEncrypt(m));
else
return other.RSAEncrypt(RSASign(m));
}
public BigInteger RSADecryptAndVerify(BigInteger c,

RSAPrivateKeyFast other) {
return other.RSAVerify(RSADecrypt(c));
else
return RSADecrypt(other.RSAVerify(c));
}
}
Java class: RSATestFast
// RSATestFast: Test Fast RSA Implementation
public class RSATestFast {
BigInteger m, m1, m2, m3, c, s, s1;
RSAPrivateKeyFast alice = new RSAPrivateKeyFast(1024, rnd, "Alice");
RSAPrivateKeyFast bob = new RSAPrivateKeyFast(1024, rnd, "Bob ");
m = new BigInteger(
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210" +
"1234567890987654321012345678909876543210");
System.out.println("Message m:\n" + m + "\n");
System.out.println("ALICE ENCRYPTS m FOR BOB; BOB DECRYPTS IT:");
c = bob.RSAEncrypt(m); // Using Bob’s public key
System.out.println("Message encrypted with Bob’s public key:\n" +
c + "\n");
m1 = bob.RSADecrypt(c); // Using Bob’s private key
System.out.println("Original message back, decrypted:\n" + m1 +"\n");
System.out.println("ALICE SIGNS m FOR BOB; BOB VERIFIES SIGNATURE:");

s = alice.RSASign(m); // Using Alice’s private key
System.out.println("Message signed with Alice’s private key:\n" +
s + "\n");
m2 = alice.RSAVerify(s); // Using Alice’s public key
System.out.println("Original message back, verified:\n" + m2 + "\n");
System.out.println("BOB SIGNS AND ENCRYPTS m FOR ALICE;" +

"\n ALICE VERIFIES SIGNATURE AND DECRYPTS:");
c = bob.RSASignAndEncrypt(m, alice);
System.out.println("Message signed and encrypted," +
"\n using Bob’s secret key and Alice’s public key:\n" + c +"\n");
m3 = alice.RSADecryptAndVerify(c, bob);
System.out.println("Original message back, verified and decrypted," +
"\n using Alice’s secret key and Bob’s public key:\n" + m1);
}
}
Here is a run of the above test class, showing simple encryption, signing, and a combination of
254 Program IV.14.b
signing and encryption. Unix commands appear in boldface.

% javac RSAPublicKey.java
% javac RSAPrivateKey.java
% javac RSATest.java
% java RSATest
Message m:
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
ALICE ENCRYPTS m FOR BOB; BOB DECRYPTS IT:

Message encrypted with Bob’s public key:
543405813676648078057012762872599683813667674133659925377335
760556755516424469233387398561035220096421942902314004442496
355392009986359056374479092883194576861821720618133177330634
484625941715294402963142587566926665244387837038418691448876
173245292324151150663861262596533907168126172311922973506760
70135287
Original message back, decrypted:

123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
ALICE SIGNS m FOR BOB; BOB VERIFIES SIGNATURE:

Message signed with Alice’s private key:
239990627092163586938360727219071875855725965597290038843626
784334056744376101809741282946428993573655987183640986372900
356678910437032277772334474986578993935720568974198358713462
782149869678768897151584050391219800123956436243445248715199
025995371266867400947136422789069497185692715034294109803570
5104040
Original message back, verified:

123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
BOB SIGNS AND ENCRYPTS m FOR ALICE;

ALICE VERIFIES SIGNATURE AND DECRYPTS:
Message signed and encrypted,
using Bob’s secret key and Alice’s public key:
555680954489228451633956186412450975924427391258695282224282
350607993390891939181686306232760912706003539593775370490376
870445903174464182907612502285232696602221467528497111242219
800301035480234847470533403244513111604794010697819018320289
165817224833283798363570908599851775688615057167242160604046
11712970
Original message back, verified and decrypted,

using Alice’s secret key and Bob’s public key:
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
123456789098765432101234567890987654321012345678909876543210
Program IV.15.a
Square Roots mod n = p*q
Here is a program that prints a table of square roots modulo a product of two primes Rabin’s
cryptosystem makes use of square roots modulo n = p*q, where p and q are primes both
equal to 3 modulo 4. The program below, however, works with any two primes and produces
a table of all square roots. Those square roots with a factor in common with either p or q are
shown in the table in bold italic.
Java class: SquareTable
// SquareTable.java: create table of squares and square roots
public class SquareTable {
// Link: to allow linked lists of square roots

private class Link {
int entry; // square root entry
Link next; // next link
}
private Link[] table; // table entry for each possible square

private int p, q, n; // primes and product
// SquareTable: constructor, passes in the primes

public SquareTable(int pIn, int qIn) {
p = pIn; q = qIn; n = p*q;
table = new Link[n];
}
// buildTable: construct the table: array of linked lists

public void buildTable() {
for (int i = 1; i < n; i++) {
int iSqr = i*i % n;
Link link = new Link();
Link ptr;
if (table[iSqr] == null) {
table[iSqr] = link;
}
else {
ptr = table[iSqr];
while (ptr.next != null)
ptr = ptr.next;
ptr.next = link;
}
link.entry = i;
link.next = null;
}
}
15. Rabin’s Version of RSA 257
// printTable: print the table in HTML form

public void printTable() {
boolean div; // for entries with either prime as divisor
System.out.println("<table border>");
System.out.println("<tr><th colspan=2>Numbers mod " + n +
" = " + p + "*" + q + "</th><tr>");
System.out.println("<tr><th>Square</th>");
System.out.println("<th>Square Roots</th></tr>");
for (int j = 1; j < n; j++) {

div = false;
if (j%p == 0 || j%q == 0) div = true;
if (table[j] != null) {
System.out.print("<tr><td>");
if (div)
System.out.print("<font color=FF0000><i>" +
j + "</i></font>");
else
System.out.print(j);
System.out.print("</td><td>");
if (div)
System.out.print("<font color=FF0000><i>");
Link loc = table[j];
while (loc != null) {
System.out.print(loc.entry);
if (loc.next != null)
System.out.print(", ");
loc = loc.next;
}
if (div)
System.out.print("</i></font>");
System.out.println("</td></tr>");
}
}
}
// main: feed in primes p and q from command line

SquareTable squareTable = new SquareTable(
Integer.parseInt(args[0]), Integer.parseInt(args[1]));
squareTable.buildTable();
squareTable.printTable();
}
}
Here is a table with p = 7, q = 11, and n = p*q = 77. Notice that the bold italic entries
all have either 7 or 11 as a factor. Notice also the symmetry: if s is a square root, then so is
n - s.
258 Program V.15.a
Numbers mod 77 = 7*11

Square Square Roots
1 1, 34, 43, 76
4 2, 9, 68, 75
9 3, 25, 52, 74
11 33, 44
14 28, 49
15 13, 20, 57, 64
16 4, 18, 59, 73
22 22, 55
23 10, 32, 45, 67
25 5, 16, 61, 72
36 6, 27, 50, 71
37 24, 31, 46, 53
42 14, 63
44 11, 66
49 7, 70
53 19, 30, 47, 58
56 21, 56
58 17, 38, 39, 60
60 26, 37, 40, 51
64 8, 36, 41, 69
67 12, 23, 54, 65
70 35, 42
71 15, 29, 48, 62
Program V.16.a
Linear Congruence Random Number
Generators
This section contains an implementation of several simple linear congruence random number
generators, using the Java BigInteger class. This makes the generators slow, but eliminates
any overflow problems. In practice this method is fast enough for most applications.
The class Congruence below implements a specific generator with input multiplier, modulus,
and seed. The the class Generators creates 8 instances of generators with different values
for the multiplier and modulus. Notice that generator number 0 is the infamous “RANDU”
which should not be used.
Java class: Congruence
// Congruence: linear congruence generators, all using BigInteger
public class Congruence {
public BigInteger k; // multiplier
public BigInteger m; // modulus
public BigInteger s; // seed
private BigInteger x; // next generator value
private int rBits;
private int twoToRBits; // 2ˆrBits
private BigInteger bigTwoToRBits;
// Congruence: constructor starts with multiplier, modulus and seed

public Congruence(String ks, String ms, String ss) {
k = new BigInteger(ks);
m = new BigInteger(ms);
s = new BigInteger(ss);
x = s;
// System.out.println("k: " + k + ", m: " + m + ", s: " + s);
}
// nextValue: cycle to the next BigInteger value and return it

public BigInteger nextValue() {
x = (k.multiply(x)).mod(m);
return x;
}
// doubleValue: return x/m as a double

public double doubleValue() {
return (x.doubleValue()) / (m.doubleValue());
}
}
Java class: Generators
// Generators.java: a variety of random number generators
260 Program V.16.a
public class Generators {

public Congruence cong[] = new Congruence[8]; // linear congruence RNGs
// Generators: construct 8 linear congruence generators at once

public Generators(String seed) {
cong[0] = new Congruence("65539",
"2147483648", seed); // m = 2ˆ31
"2147483647", seed); // m = 2ˆ31 - 1
"2147483399", seed); // m = 2ˆ31 - 249
"2147483647", seed); // m = 2ˆ31 - 1
"2147483647", seed); // m = 2ˆ31 - 1
"4294967296", seed); // m = 2ˆ32
"281474976710656", seed); // m = 2ˆ48
"18446744073709551616", seed); // m = 2ˆ64
}
// nextValue: return next value of generator number i

public BigInteger nextValue(int i) {
return cong[i].nextValue();
}
// doubleValue: return double corresponding to value of gen i

public double doubleValue(int i) {
return cong[i].doubleValue();
}
// stuff below is just to demonstrate each generator

int iCong = Integer.parseInt(args[0]); // iCong = specific RNG
String seed = args[1]; // seed for the RNG
Generators gen = new Generators(seed);
System.out.println("Generator: " + iCong +
", k: " + gen.cong[iCong].k +
", m: " + gen.cong[iCong].m +
", s: " + gen.cong[iCong].s);
for (int i = 0; i < 4; i++) {
BigInteger x = gen.nextValue(iCong);
System.out.println(x + " \t" + gen.doubleValue(iCong));
}
}
}
Here is brief output from each generator. More thorough testing appears in a later section.
16. Linear Congruence Random Number Generators 261
% java Generators 0 11111

Generator: 0, k: 65539, m: 2147483648, s: 11111
728203829 0.33909633243456483
74155679 0.03453142894431949
333550557 0.15532158175483346
1333902231 0.6211466300301254
Generator: 1, k: 16807, m: 2147483647, s: 11111
186742577 0.08695878884147797
1108883372 0.5163640587201175
1139744538 0.5307349090141873
132318926 0.06161580144502958
Generator: 2, k: 40692, m: 2147483399, s: 11111
452128812 0.21053890903675387
535338671 0.24928652358816208
2077084275 0.9672178494917436
61700458 0.028731518031166862
Generator: 3, k: 48271, m: 2147483647, s: 11111
536339081 0.24975234700820984
1708414366 0.7955424332970485
1350332739 0.6287976818293323
1487990525 0.6928995836958752
Generator: 4, k: 62089911, m: 2147483647, s: 11111
538750434 0.2508752207508661
276008834 0.128526629008598
2053582492 0.9562738672626548
665642739 0.30996405487412776
Generator: 5, k: 69069, m: 4294967296, s: 11111
767425659 0.17868021014146507
1131441535 0.26343426085077226
605430195 0.1409627019893378
656544599 0.15286370157264173
Generator: 6, k: 31167285, m: 281474976710656, s: 11111
346299703635 0.001230303694068624
63576637476655 0.22586958961545278
245552537051579 0.8723778572473115
85713711193271 0.3045162742170895
Generator: 7, k: 6364136223846793005, m: 18446744073709551616, s: 11111
5547548633005734427 0.3007332139936904
943960907862842303 0.05117222335231413
3205834541165224339 0.17378863870802033
1510653287195956183 0.08189267879251122
Program V.16.b
Exponential and Normal Distributions
This section demonstrates transformations from the uniform distribution to the exponential and
the normal distributions. The version of uniform distribution implemented is Knuth’s method
with 2 seeds. An applet plots all three distributions, using 1000, 10000, 100000, and 1000000
points. Each plot uses 500 intervals over the range, although the exponential and normal distri-
butions ignore the few points outside the displayed range. So the pictures at the end give 500
vertical lines that represent approximately 2, 20, 200, and 2000 trials. In each case the area
covered by all the lines is approximately the same.
Java class: DistPlot
// DistPlot: plot 3 random distributions: exponential, uniform, normal
import java.applet.*;
import java.awt.*;
import java.awt.event.*;
public class DistPlot extends Applet implements ActionListener{
int[] expC = new int[1000]; // counter for nextExpDist
int[] unifC = new int[500]; // counter for nextUniformDist
int[] normC = new int[500]; // counter for nextNormalDist
double scale; // scale factor for displaying distributions
int xStart = 50, yStart = 150; int xSide = 500;

boolean firstTime = true; // to paint axes the first time through
Button next0, next1, next2, next3; // buttons
int iter; // number of random points to plot
public void init() {

setBackground(Color.white);
next0 = new Button("1000"); next1 = new Button("10000");
next2 = new Button("100000"); next3 = new Button("1000000");
next0.addActionListener(this); next1.addActionListener(this);
next2.addActionListener(this); next3.addActionListener(this);
add(next0); add(next1); add(next2); add(next3);
}
public void paint(Graphics g) {

// for exponential distribution
g.drawString("Graph showing exponential distribution " +
"with average 1", xStart, 50);
g.drawString("Iterations: " + iter, xStart + 350, 50);
g.drawLine(xStart, yStart, xStart + xSide, yStart);
for (int i = 0; i <= 5; i++)
g.drawLine(xStart + i*xSide/5, yStart,
xStart + i*xSide/5, yStart + 10);
for (int i = 0; i <= 5; i++)
g.drawString(i*1 + "", xStart + i*xSide/5 - 1, yStart + 23);
// for uniform distribution

g.drawString("Graph showing uniform distribution " +
"from 0 to 1", xStart, 200);
g.drawLine(xStart, yStart+150, xStart + xSide, yStart+150);
for (int i = 0; i <= 10; i++)
g.drawLine(xStart + i*xSide/10, yStart+150,
xStart + i*xSide/10, yStart + 10 +150);
for (int i = 0; i <= 9; i++)
g.drawString("0." + i + "", xStart + i*xSide/10 - 1,
yStart + 23 +150);
g.drawString("1", xStart + 10*xSide/10 - 1,
yStart + 23 +150);
// for normal distribution

g.drawString("Graph showing normal distribution " +
"from 0 to 1", xStart, 350);
g.drawLine(xStart, yStart+300, xStart + xSide, yStart+300);
for (int i = 0; i <= 10; i++)
g.drawLine(xStart + i*xSide/10, yStart+300,
xStart + i*xSide/10, yStart + 10 +300);
for (int i = 0; i <= 10; i++)
g.drawString(i-5 + "", xStart + i*xSide/10 - 1,
yStart + 23 +300);
firstTime = false;
if (!firstTime) {
for(int dummy = 0; dummy < iter; dummy++) {
int i = (int)(100.0*nextExpDist());
if (i < 1000) expC[i]++;
else expC[999]++;
unifC[(int)(500.0*nextUniformDist())]++;
double r = nextNormalDist();
int j = (int)( 50.0*r) + 250;
if (r < 0.0) j = (int)( 50.0*r) - 1 + 250;
if (j < 0) normC[0]++;
else if (j >= 500) normC[499]++;
else normC[j]++;
}
// exponential distribution
g.setColor(new Color(255, 0, 0)); // red
for (int i = 0; i < 500; i++)
g.drawLine(xStart + i, yStart - 1,
xStart + i, yStart - (int)(expC[i]/scale));
// uniform distribution
for (int i = 0; i < 500; i++)
g.drawLine(xStart + i, yStart - 1 + 150,
xStart + i, yStart - (int)(unifC[i]/scale) + 150);
// normal distribution
for (int i = 0; i < 500; i++)
g.drawLine(xStart + i, yStart - 1 + 300,
xStart + i, yStart - (int)(normC[i]/scale) + 300);
}
}
264 Program V.16.b
// class variables used by nextUniformDist

long seed1 = (int)(100000000.0*Math.random());
long seed2 = (int)(100000000.0*Math.random());
long seed3;
// nextUniform: uniformly dist on the interval from 0 to 1
private double nextUniformDist() {
long m = 2147483647;
long k1 = 2718283;
long k2 = 314159269;
seed3 = (k1*seed1 - k2*seed2)%m;
if (seed3 < 0) seed3 += m;
seed1 = seed2; seed2 = seed3;
return (double)seed2 / (double)m;
}
// nextExpDist: exponentially dist with average interarrival time = 1

private double nextExpDist() {
return 1.0*(-Math.log(nextUniformDist()));
}
// class variables used by nextNormalDist

boolean nextNormal = true;
double saveNormal;
// nextNormalDist: mean 0 and variance 1
// x1 = sqrt(-2*log(u1))*cos(2*Pi*u2)
// x2 = sqrt(-2*log(u1))*sin(2*Pi*u2)
private double nextNormalDist() {
double u1, u2, x1, x2;
if (nextNormal) {
u1 = nextUniformDist(); u2 = nextUniformDist();
double temp = Math.sqrt(-2.0*Math.log(u1));
x1 = temp*Math.cos(2.0*Math.PI*u2);
x2 = temp*Math.sin(2.0*Math.PI*u2);
saveNormal = x2;
nextNormal = false;
return x1;
}
else {
nextNormal = true;
return saveNormal;
}
}
public void actionPerformed(ActionEvent e) {

for (int i = 0; i < 500; i++) {
expC[i] = 0; expC[i + 500] = 0; unifC[i] = 0; normC[i] = 0;
}
if (e.getSource() == next0) {
iter = 1000; scale = 0.1;
}
else if (e.getSource() == next1) {
iter = 10000; scale = 1.0;
}
iter = 100000; scale = 10.0;
}
iter = 1000000; scale = 100.0;
}
repaint();
}
}
The output of the above program has not been included here because it takes up a lot of memory.
Program V.17.a
The logistic Lattice as a RNG
The program in this section uses the logistic lattice in two dimensions to return pseudo-random
37 % >
doubles. Based on experiments described later in this section, this program sets the constants
NU to and NSTEP to . NSTEP is half the number of iterations, so the logistic equation
will be iterated times at each node. With this value of NU, the given number of iterations
are sufficient to assure that each node will fill completely with noise and become independent
@
of each other node by the time the code supplies more random numbers. The code below uses
all 9 doubles at each stage, thereby increasing efficiency by a factor of . (Each node will be
independent of the other nodes.)
Java class: Chaotic2D
// Chaotic2D.java: a random number generator based on chaos theory
public class Chaotic2D {
private final int NMAX = 3; // size of lattice
private final double BETA = 0.292893218813452476; // magic number in f
private double NU; // = 1.0e-13, // viscosity constant in step
private final double TWO_DIV_PI = 0.636619772367581343; // 2/Pi in S
private int NSTEP; // = 60, // half # of steps to iterate
private int flag; // flag used in nextBlock
private double[][] t; // seed array, where the work occurs
private double[][] tn; // extra copy of seed array
private double[][] tret; // array for returning values (these are
// transformed to the uniform distribution)
private Random random; // extra RNG to initialize seed array t
// mod: need instead of %, because % can yield negative result

private int mod(int i, int j) { // If i is negative, then i%j
int k = i%j; // may be negative. In general,
if (k < 0) k = k + j; // result is machine dependent.
return(k); // Check your own architecture.
}
// Chaotic2D: constructor -- allocate, use seed and auxiliary RNG

// to intialize array t, which serves as the real seed (9 doubles)
public Chaotic2D(long seed) {
t = new double[NMAX][NMAX];
tn = new double[NMAX][NMAX];
tret = new double[NMAX][NMAX];
NU = 1.0e-13; NSTEP = 60; flag = -1;
random = new Random(seed);
for (int i = 0; i < NMAX; i++)
for (int j = 0; j < NMAX; j++)
t[i][j] = 2.0*random.nextDouble() - 1.0;
}
17. Random Numbers From Chaos Theory 267
private double f(double x) { // Remapped logistic equation

double temp = Math.abs(x);
if (temp <= BETA) return(2.0*temp*(2.0-temp));
else return(-2.0*(1.0-temp)*(1.0-temp));
}
private void step(double[][] t, double[][] tn) { // Coupled map lattice

t[i][j] = f(t[i][j]);
tn[i][j] = (1.0 - 4.0*NU)*t[i][j] +
NU*(1.1*t[mod(i-1, NMAX)][j] + // 1.1, 0.9, 1.2, 0.8
0.9*t[mod(i+1, NMAX)][j] + // added to prevent
1.2*t[i][mod(j-1, NMAX)] + // falling into stable
0.8*t[i][mod(j+1, NMAX)]); // configurations
}
private double S(double x) { // Change distribution to uniform

if (x >= 0) return(TWO_DIV_PI*Math.asin(Math.sqrt(x/2)));
else return(TWO_DIV_PI*Math.asin(Math.sqrt(-x/2)) + 0.5);
}
private void chaoticUniform() { // the generator itself

for (int i = 0; i < NSTEP; i++) { // Iterate step 2*NSTEP times
step(t, tn);
step(tn, t);
}
tret[i][j] = S(t[i][j]);
}
public double nextRandom() { // call chaoticUniform once every 9 times

double r = 0; // keep compiler happy
if (flag == -1) {
chaoticUniform(); // called only once every 9 times
flag = 8;
}
int xf = flag/3, yf = flag%3;
r = tret[xf][yf];
flag--;
return r;
}

long seed = Long.parseLong(args[0]); // seed for RNG, up to 18 digits
Chaotic2D chaotic2D = new Chaotic2D(seed);
for (int i = 0; i < 10; i++)
System.out.println(chaotic2D.nextRandom());
}
268 Program V.17.a
Below is simple output showing 10 random numbers coming from this generator. The next
section shows a test of this generator using 256 000 random numbers. Notice that the seed is
a Java long or up to 18 decimal digits. The real seed inside the generator is the array of 9
doubles, and it would be possible to use this directly as the seed, rather than the method here
of seeding using the Java Random class as an auxiliary generator.
% java Chaotic2D 999999999999999999
0.52692498814042
0.9980968834981016
0.5445878089596392
0.7249430391767984
0.07845814949526263
0.875011892397908
0.007317912434666028
0.497471911776451
0.8005000406206588
0.5236179345954219
7 %
Next is an experiment showing how the array of nodes evolves as the generator iterates. This

7%
experiment shows the results of three parallel runs, with the viscosity constant equal to ,
, and
. Simple changes were made to the code to allow these different values and to
print values of all 9 nodes.
8= E 8=?(@@@@@@@ @@@@@

If all 9 initial numbers are the same, all node values will remain the same. In the run shown
8 = 7 % 37 %
below, all nodes were set to except for one which was set to . The
The value
resulting output is shown for three values of in three columns below: , , and .
leaves all nodes independent of one another, that is, just 9 independent logistic
equations. Iterations 2, 48, 74, 100, and 120 are shown.
Notice that after 120 iterations, all node values in the middle and right columns have drifted
completely apart. This is the reason that 120 iterations are used in the generator itself.
<= E
8=BE
Notice also that the initial value of is transformed to itself by the remapped logistic equa-
8=?DAAAAAA =$=$=
tion. (This corresponds to in the original equation.) This value is then transformed to
by the transformation that yields numbers uniformly distributed in the inter-
val from to .
The results of coupled values perturbing nearby nodes is shown in bold below:
v = 0 v = 10ˆ-12 v = 10ˆ-13
Iteration number: 2
0.8333333333331865 0.8333333333331865 0.8333333333331865
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
17. Random Numbers From Chaos Theory 269
0.8333333333333334 0.8333333333333334 0.8333333333333334

0.8333333333333334 0.8333333333333334 0.8333333333333334
Iteration number: 48
0.5040834897791248 0.5036049050782517 0.5021691550116464
0.8333333333333334 0.8333333331818866 0.8333333333219761
0.8333333333333334 0.8333333332366295 0.83333333332515
0.8333333333333334 0.8333333332015176 0.8333333333229862
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333332189119 0.8333333333241839
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.8333333333333334 0.8333333333333334 0.8333333333333334
0.3602326768156076 0.5833674744417872 0.47153769384353167
0.8333333333333334 0.8231749738724833 0.8325716798549819
0.8333333333333334 0.826847020511059 0.8327845046032647
0.8333333333333334 0.8244919644092912 0.8326394144500628
0.8333333333333334 0.8333333333331181 0.8333333333333328
0.8333333333333334 0.8333333333331748 0.833333333333333
0.8333333333333334 0.8256584292117882 0.8327197098392694
0.8333333333333334 0.8333333333331479 0.8333333333333328
0.8333333333333334 0.8333333333332305 0.8333333333333331

0.28322543483494855 0.9957436720669621 0.5329869100296993
0.8333333333333334 0.8698208441903008 0.3663600043966403
0.8333333333333334 0.24828271001134866 0.06072814757025873
0.8333333333333334 0.6086436328075263 0.7253693182396282
0.8333333333333334 0.8333152337340284 0.8333332244057435
0.8333333333333334 0.8333207796242934 0.8333332571545267
0.8333333333333334 0.7364311648949555 0.2422766762266617
0.8333333333333334 0.8333179900269688 0.8333332360776878
0.8333333333333334 0.833324925779414 0.8333332795639763

0.8935574910647093 0.5833052795925358 0.7821714186572044
0.8333333333333334 0.7615175283592651 0.30797021553244913
0.8333333333333334 0.7909250549646865 0.07806647066456404
0.8333333333333334 0.6059141053682064 0.6417576242558025
0.8333333333333334 0.35452800150669156 0.719114512983271
0.8333333333333334 0.16981560934264034 0.7534540927550613
0.8333333333333334 0.045160386927817195 0.4919490670536787
0.8333333333333334 0.7447109018158442 0.7313534352471185
0.8333333333333334 0.9826254952427942 0.7769521043864014
Program V.18.a
Maurer’s Universal Test
Java class: Maurer

// Maurer.java: implement Maurer’s "Universal" Randomness Test
public class Maurer {

private Generators gen;
private double[] mu = {0, 0.7326495, 1.5374383, 2.4016068, 3.3112247,
4.2534266, 5.2177052, 6.1962507, 7.1836656,
8.1764248, 9.1723243, 10.170032, 11.168765,
12.168070, 13.167693, 14.167488, 15.167379};
private double[] sigma2 = {0, 0.690, 1.338, 1.901, 2.358,
2.705, 2.954, 3.125, 3.238,
3.311, 3.356, 3.384, 3.401,
3.410, 3.416, 3.419, 3.421};
private int L; // number of bits in a block

private int V; // 2ˆL, size of T
private int Q; // number of initial runs to fill T
private int K; // number of production runs
private int[] T;
// Maurer: constructor for this test
// pBits: number of bits in primes p and q. n = p*q has 2*pBits bits
// rSize: number of bits to return each time
public Maurer(int rBits) {
L = rBits;
V = (int)(Math.pow(2, L) + 0.000001);
Q = 10*V;
K = 1000*V;
T = new int[V];
gen = new Generators(rBits); // fetch rBits at a time
double cLK = 0.7 - (0.8/(double)L) + (1.6 + (12.8/(double)L))*
Math.pow((double)K, -4.0/(double)L);
double sig2 = cLK*cLK*sigma2[L]/(double)K;
double sig = Math.sqrt(sig2);
System.out.println("\n*** Maurer’s Universal Test ***");
System.out.println("L: " + L + ", V: " + V +
", Q: " + Q + ", K: " + K);
System.out.println("Variance: " + sig2 + ", sigma: " + sig);
}
public double doTest() {

for (int i = 0; i < 100; i++) {
gen.nextBlock();
// System.out.print(bbs.nextBlum() + " ");
18. Statistical Tests and Perfect Generators 271
// if (i%20 == 19) System.out.println();

}
int bi;
for (int j = 0; j < V; j++) T[j] = 0;
for (int i = 1; i <= Q; i++) {
bi = gen.nextBlock();
T[bi] = i;
}
double sum = 0;
for (int i = Q+1; i <= Q + K; i++) {
bi = gen.nextBlock();
sum += log2((double)(i - T[bi]));
T[bi] = i;
if (i % 1000 == 0) System.out.print("*");
}
double Xu = (double)sum/(double)K;
return Xu;
}
private double log2(double x) {

return Math.log(x)/Math.log(2.0);
}

int rBits = Integer.parseInt(args[0]); // rBits = bits returned
Maurer maurer = new Maurer(rBits);
System.out.println(maurer.doTest());
}
}
Program V.18.b
The Blum-Blum-Shub Perfect Generator
Program VI.20.a
Generate Multiplication Tables
The following Java program uses the slow multipy function to generate two tables
needed for fast multiplication: a table of all powers of the generator 0x03, and the inverse
table. (The tables in the main section had a few extra frills inserted by hand.)
Java class: FFMultTables
// FFMultTables: create the arrays E and L, write html versions
public class FFMultTables {
public byte[] E = new byte[256];
public byte[] L = new byte[256];
private String[] dig = {"0","1","2","3","4","5","6","7",
"8","9","a","b","c","d","e","f"};
public byte FFMul(byte a, byte b) {
byte aa = a, bb = b, r = 0, t;
while (aa != 0) {
if ((aa & 1) != 0)
r = (byte)(r ˆ bb);
t = (byte)(bb & 0x80);
bb = (byte)(bb << 1);
if (t != 0)
bb = (byte)(bb ˆ 0x1b);
aa = (byte)((aa & 0xff) >> 1);
}
return r;
}
public String hex(byte a) {

return dig[(a & 0xff) >> 4] + dig[a & 0x0f];
}
public String hex(int a) {

return dig[a];
}
public void loadE() {

byte x = (byte)0x01;
int index = 0;
E[index++] = (byte)0x01;
for (int i = 0; i < 255; i++) {
byte y = FFMul(x, (byte)0x03);
E[index++] = y;
// System.out.print(hex(y) + " ");
x = y;
}
}
public void loadL() {

274 Program VI.20.a
int index;
for (int i = 0; i < 255; i++) {
L[E[i] & 0xff] = (byte)i;
}
}
public void printE() {

System.out.print("<table border><tr><td></td>");
for (int i = 0; i < 16; i++)
System.out.print("<th>" + hex(i) + "</th>");
for (int i = 0; i < 256; i++) {
if (i%16 == 0)
System.out.print("<tr><th> " + hex(i/16) +
" </th>");
System.out.print("<td> " + hex(E[i]) +
" </td>");
if (i%16 == 15) System.out.println("</tr>");
}
}
public void printL() {

System.out.print("<table border><tr><td></td>");
for (int i = 0; i < 16; i++)
for (int i = 0; i < 256; i++) {
if (i%16 == 0)
" </th>");
if (i == 0)
System.out.print("<td> </td>");
else
System.out.print("<td> " + hex(L[i]) +
" </td>");
}
}

FFMultTables ffm = new FFMultTables();
ffm.loadE();
ffm.loadL();
ffm.printL();
ffm.printE();
}
}
Program VI.20.b
Compare Multiplication Results
There have been two algorithms for multiplying field elements, a slow one and a fast one. As
a check, the following program compares the results of all 65536 possible products to see that
the two methods agree (which they do):
Java class: FFMultTest
// FFMultTest: test two ways to multiply, all 65536 products
public class FFMultTest {
public FFMultTest() {
loadE();
loadL();
}
public byte[] E = new byte[256]; // powers of 0x03

public byte[] L = new byte[256]; // inverse of E
"8","9","a","b","c","d","e","f"};
{\timesbf // FFMulFast: fast multiply using table lookup

public byte FFMulFast(byte a, byte b){
int t = 0;;
if (a == 0 || b == 0) return 0;
t = (L[(a & 0xff)] & 0xff) + (L[(b & 0xff)] & 0xff);
if (t > 255) t = t - 255;
return E[(t & 0xff)];
}
// FFMul: slow multiply, using shifting

while (aa != 0) {
if ((aa & 1) != 0)
t = (byte)(bb & 0x80);
bb = (byte)(bb << 1);
if (t != 0)
aa = (byte)((aa & 0xff) >> 1);
}
return r;
}}
// hex: print a byte as two hex digits

}
276 Program VI.20.b
// loadE: create and load the E table

int index = 0;
for (int i = 0; i < 255; i++) {
E[index++] = y;
// System.out.print(hex(y) + " ");
x = y;
}
}
// loadL: load the L table using the E table

int index;
for (int i = 0; i < 255; i++) {
}
}
// testMul: go through all possible products of two bytes

public void testMul() {
byte a = 0;
for(int i = 0; i < 256; i++) {
byte b = 0;
for(int j = 0; j < 256; j++) {
byte x = FFMul(a, b);
byte y = FFMulFast(a, b);
if (x != y) {
System.out.println("a: " + hex(a) + ", b: " +
hex(b) + ", x: " + hex(x) + ", y: " + hex(y));
System.exit(1);
}
b++;
}
a++;
}
}

FFMultTest ffmult = new FFMultTest();
ffmult.testMul();
}
}
Program VI.21.a
Generate AES Tables
Here is a Java program that will generate a number of 256-byte tables needed for the Advanced
Encryption Standard:
Java class: Tables
// Tables: construct and print 256-byte tables needed for AES
public class Tables {
public Tables() {
loadE();
loadL();
loadInv();
loadS();
loadInvS();
loadPowX();
}
public byte[] E = new byte[256]; // "exp" table (base 0x03)

public byte[] L = new byte[256]; // "Log" table (base 0x03)
public byte[] S = new byte[256]; // SubBytes table
public byte[] invS = new byte[256]; // inv of SubBytes table
public byte[] inv = new byte[256]; // multi inverse table
public byte[] powX = new byte[15]; // powers of x = 0x02
"8","9","a","b","c","d","e","f"};
// FFMulFast: fast multiply using table lookup

int t = 0;;
if (a == 0 || b == 0) return 0;
if (t > 255) t = t - 255;
}

while (aa != 0) {
if ((aa & 1) != 0)
t = (byte)(bb & 0x80);
bb = (byte)(bb << 1);
if (t != 0)
aa = (byte)((aa & 0xff) >> 1);
}
return r;
278 Program VI.21.a

}
// hex: print a single digit (for tables)

public String hex(int a) {
return dig[a];
}

int index = 0;
for (int i = 0; i < 255; i++) {
E[index++] = y;
x = y;
}
}

int index;
for (int i = 0; i < 255; i++) {
}
}
// loadS: load in the table S

public void loadS() {
int index;
for (int i = 0; i < 256; i++)
S[i] = (byte)(subBytes((byte)(i & 0xff)) & 0xff);
}
// loadInv: load in the table inv

public void loadInv() {
int index;
for (int i = 0; i < 256; i++)
inv[i] = (byte)(FFInv((byte)(i & 0xff)) & 0xff);
}
// loadInvS: load the invS table using the S table

public void loadInvS() {
int index;
for (int i = 0; i < 256; i++) {
invS[S[i] & 0xff] = (byte)i;
}
}
21. The S-Boxes 279
// loadPowX: load the powX table using multiplication

public void loadPowX() {
int index;
byte xp = x;
powX[0] = 1; powX[1] = x;
for (int i = 2; i < 15; i++) {
xp = FFMulFast(xp, x);
powX[i] = xp;
}
}
// FFInv: the multiplicative inverse of a byte value

public byte FFInv(byte b) {
byte e = L[b & 0xff];
return E[0xff - (e & 0xff)];
}
// ithBIt: return the ith bit of a byte

public int ithBit(byte b, int i) {
int m[] = {0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80};
return (b & m[i]) >> i;
}
// subBytes: the subBytes function

public int subBytes(byte b) {
byte inB = b;
int res = 0;
if (b != 0) // if b == 0, leave it alone
b = (byte)(FFInv(b) & 0xff);
byte c = (byte)0x63;
for (int i = 0; i < 8; i++) {
int temp = 0;
temp = ithBit(b, i) ˆ ithBit(b, (i+4)%8) ˆ
ithBit(b, (i+5)%8) ˆ ithBit(b, (i+6)%8) ˆ
ithBit(b, (i+7)%8) ˆ ithBit(c, i);
res = res | (temp << i);
}
return res;
}
// printTable: print a 256-byte table

public void printTable(byte[] S, String name) {
System.out.print("<table border>");
System.out.print("<tr><th colspan=2 rowspan=2>" +
name + "(rs)</th>");
System.out.print("<th colspan=16>s</th></tr><tr>");
for (int i = 0; i < 16; i++)
System.out.println("</tr><tr><th rowspan=17>r</th></tr>");
for (int i = 0; i < 256; i++) {
if (i%16 == 0)
280 Program VI.21.a
" </th>");
System.out.print("<td> " + hex(S[i]) +
" </td>");
}
}
// printL: print the L table

public void printL() {
printTable(L, "L");
}
// printE: print the E table

public void printE() {
printTable(E, "E");
}
// printS: print the S table

public void printS() {
printTable(S, "S");
}
// printInv: print the inv table

public void printInv() {
printTable(inv, "inv");
}
// printInvS: print the invS table

public void printInvS() {
printTable(invS, "iS");
}
// printpowX: print the powX table

public void printPowX() {
System.out.print("<table border><tr><th colspan=17>");
System.out.print("Powers of x =
0x02</th></tr><tr><th>i</th><th></th>");
for (int i = 0; i < 15; i++)
System.out.print("<th>" + i + "</th>");
System.out.println("</tr><tr><th>x<sup>i</sup></th><th></th>");
for (int i = 0; i < 15; i++)
System.out.print("<td>" + hex(powX[i]) + "</td>");
System.out.println("</tr></table>");
}

Tables sB = new Tables();
// sB.printL();
// sB.printE();
// sB.printS();
// sB.printInvS();
22. The S-Boxes 281
// sB.printInv();
sB.printPowX();
}
}
Program VI.23.a
AES Encryption
Encryption in the AES uses 6 classes: 2 principal ones, 3 utility ones, and a main driver. The
results of testing both encryption and decryption appear after the next section on decryption.
The class AESencrypt provides all the principle functions for the AES encryption al-
gorithm.
The class Tables gives access to computed tables and utility functions.
The class GetBytes just reads bytes represented as Ascii hex characters (not in binary).
The class Copy copies arrays back and forth for the AES.
The class Print prints 1-and 2-dimensional arrays of bytes for debugging.
The class AEStest is a driver for testing encryption.
Java class: AESencrypt

// AESencrypt: AES encryption
public class AESencrypt {
private final int Nb = 4; // words in a block, always 4 for now
private int Nk; // key length in words
private int Nr; // number of rounds, = Nk + 6
private int wCount; // position in w for RoundKey (= 0 each encrypt)
private AEStables tab; // all the tables needed for AES
private byte[] w; // the expanded key
// AESencrypt: constructor for class. Mainly expands key

public AESencrypt(byte[] key, int NkIn) {
Nk = NkIn; // words in a key, = 4, or 6, or 8
Nr = Nk + 6; // corresponding number of rounds
tab = new AEStables(); // class to give values of various functions
w = new byte[4*Nb*(Nr+1)]; // room for expanded key
KeyExpansion(key, w); // length of w depends on Nr
}
// Cipher: actual AES encrytion

public void Cipher(byte[] in, byte[] out) {
wCount = 0; // count bytes in expanded key throughout encryption
byte[][] state = new byte[4][Nb]; // the state array
Copy.copy(state, in); // actual component-wise copy
AddRoundKey(state); // xor with expanded key
for (int round = 1; round < Nr; round++) {
Print.printArray("Start round " + round + ":", state);
SubBytes(state); // S-box substitution
ShiftRows(state); // mix up rows

MixColumns(state); // complicated mix of columns
}
Print.printArray("Start round " + Nr + ":", state);
SubBytes(state); // S-box substitution
ShiftRows(state); // mix up rows
Copy.copy(out, state);
}
// KeyExpansion: expand key, byte-oriented code, but tracks words

private void KeyExpansion(byte[] key, byte[] w) {
byte[] temp = new byte[4];
// first just copy key to w
int j = 0;
while (j < 4*Nk) {
w[j] = key[j++];
}
// here j == 4*Nk;
int i;
while(j < 4*Nb*(Nr+1)) {
i = j/4; // j is always multiple of 4 here
// handle everything word-at-a time, 4 bytes at a time
for (int iTemp = 0; iTemp < 4; iTemp++)
temp[iTemp] = w[j-4+iTemp];
if (i % Nk == 0) {
byte ttemp, tRcon;
byte oldtemp0 = temp[0];
for (int iTemp = 0; iTemp < 4; iTemp++) {
if (iTemp == 3) ttemp = oldtemp0;
else ttemp = temp[iTemp+1];
if (iTemp == 0) tRcon = tab.Rcon(i/Nk);
else tRcon = 0;
temp[iTemp] = (byte)(tab.SBox(ttemp) ˆ tRcon);
}
}
else if (Nk > 6 && (i%Nk) == 4) {
temp[iTemp] = tab.SBox(temp[iTemp]);
}
w[j+iTemp] = (byte)(w[j - 4*Nk + iTemp] ˆ temp[iTemp]);
j = j + 4;
}
}
// SubBytes: apply Sbox substitution to each byte of state

private void SubBytes(byte[][] state) {
state[row][col] = tab.SBox(state[row][col]);
}
284 Program VI.23.a
// ShiftRows: simple circular shift of rows 1, 2, 3 by 1, 2, 3

private void ShiftRows(byte[][] state) {
for (int r = 1; r < 4; r++) {
for (int c = 0; c < Nb; c++)
t[c] = state[r][(c + r)%Nb];
for (int c = 0; c < Nb; c++)
state[r][c] = t[c];
}
}
// MixColumns: complex and sophisticated mixing of columns

private void MixColumns(byte[][] s) {
int[] sp = new int[4];
byte b02 = (byte)0x02, b03 = (byte)0x03;
for (int c = 0; c < 4; c++) {
sp[0] = tab.FFMul(b02, s[0][c]) ˆ tab.FFMul(b03, s[1][c]) ˆ
s[2][c] ˆ s[3][c];
sp[1] = s[0][c] ˆ tab.FFMul(b02, s[1][c]) ˆ
tab.FFMul(b03, s[2][c]) ˆ s[3][c];
sp[2] = s[0][c] ˆ s[1][c] ˆ
tab.FFMul(b02, s[2][c]) ˆ tab.FFMul(b03, s[3][c]);
sp[3] = tab.FFMul(b03, s[0][c]) ˆ s[1][c] ˆ
s[2][c] ˆ tab.FFMul(b02, s[3][c]);
for (int i = 0; i < 4; i++) s[i][c] = (byte)(sp[i]);
}
}
// AddRoundKey: xor a portion of expanded key with state

private void AddRoundKey(byte[][] state) {
for (int c = 0; c < Nb; c++)
for (int r = 0; r < 4; r++)
state[r][c] = (byte)(state[r][c] ˆ w[wCount++]);
}
}
Java class: AEStables

// AEStables: construct various 256-byte tables needed for AES
public class AEStables {
public AEStables() {
loadE(); loadL(); loadInv();
loadS(); loadInvS(); loadPowX();
}
private byte[] E = new byte[256]; // "exp" table (base 0x03)

private byte[] L = new byte[256]; // "Log" table (base 0x03)
private byte[] S = new byte[256]; // SubBytes table
private byte[] invS = new byte[256]; // inverse of SubBytes table
private byte[] inv = new byte[256]; // multiplicative inverse table
private byte[] powX = new byte[15]; // powers of x = 0x02
// Routines to access table entries

public byte SBox(byte b) {
return S[b & 0xff];

}
public byte invSBox(byte b) {

return invS[b & 0xff];
}
public byte Rcon(int i) {

return powX[i-1];
}
// FFMulFast: fast multiply using table lookup

int t = 0;;
if (a == 0 || b == 0) return 0;
if (t > 255) t = t - 255;
}

while (aa != 0) {
if ((aa & 1) != 0)
t = (byte)(bb & 0x80);
bb = (byte)(bb << 1);
if (t != 0)
aa = (byte)((aa & 0xff) >> 1);
}
return r;
}

private void loadE() {
int index = 0;
for (int i = 0; i < 255; i++) {
E[index++] = y;
x = y;
}
}

private void loadL() { // careful: had 254 below several places
int index;
for (int i = 0; i < 255; i++) {
}
286 Program VI.23.a
// loadS: load in the table S

private void loadS() {
int index;
for (int i = 0; i < 256; i++)
S[i] = (byte)(subBytes((byte)(i & 0xff)) & 0xff);
}
// loadInv: load in the table inv

private void loadInv() {
int index;
for (int i = 0; i < 256; i++)
inv[i] = (byte)(FFInv((byte)(i & 0xff)) & 0xff);
}
// loadInvS: load the invS table using the S table

private void loadInvS() {
int index;
for (int i = 0; i < 256; i++) {
invS[S[i] & 0xff] = (byte)i;
}
}
// loadPowX: load the powX table using multiplication

private void loadPowX() {
int index;
byte xp = x;
powX[0] = 1; powX[1] = x;
for (int i = 2; i < 15; i++) {
xp = FFMul(xp, x);
powX[i] = xp;
}
}
// FFInv: the multiplicative inverse of a byte value

public byte FFInv(byte b) {
byte e = L[b & 0xff];
return E[0xff - (e & 0xff)];
}
// ithBIt: return the ith bit of a byte

public int ithBit(byte b, int i) {
int m[] = {0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80};
return (b & m[i]) >> i;
}
// subBytes: the subBytes function

public int subBytes(byte b) {
byte inB = b;
int res = 0;
if (b != 0) // if b == 0, leave it alone
b = (byte)(FFInv(b) & 0xff);

byte c = (byte)0x63;
for (int i = 0; i < 8; i++) {
int temp = 0;
temp = ithBit(b, i) ˆ ithBit(b, (i+4)%8) ˆ ithBit(b, (i+5)%8) ˆ
ithBit(b, (i+6)%8) ˆ ithBit(b, (i+7)%8) ˆ ithBit(c, i);
res = res | (temp << i);
}
return res;
}
}
Java class: GetBytes

// GetBytes: fetch array of bytes, represented in hex
import java.io.*;
public class GetBytes {
private String fileName; // input filename
private int arraySize; // number of bytes to read
private Reader in;
// GetBytes: constructor, opens input file

public GetBytes(String file, int n) {
fileName = file;
arraySize = n;
try {
in = new FileReader(fileName);
System.out.println("Exception opening " + fileName);
}
}
// getNextChar: fetches next char

try {
}
return ch;
}
// val: return int value of hex digit

private int val(char ch) {
if (ch >= ’0’ && ch <= ’9’) return ch - ’0’;
if (ch >= ’a’ && ch <= ’f’) return ch - ’a’ + 10;
if (ch >= ’A’ && ch <= ’F’) return ch - ’A’ + 10;
return -1000000;
}
// getBytes: fetch array of bytes in hex

public byte[] getBytes() {
byte[] ret = new byte[arraySize];
288 Program VI.23.a
for (int i = 0; i < arraySize; i++) {

char ch1 = getNextChar();
char ch2 = getNextChar();
ret[i] = (byte)(val(ch1)*16 + val(ch2));
}
return ret;
}
}
Java class: Copy

// Copy: copy arrays of bytes
public class Copy {
private static final int Nb = 4;
// copy: copy in to state
public static void copy(byte[][] state, byte[] in) {
int inLoc = 0;
for (int c = 0; c < Nb; c++)
for (int r = 0; r < 4; r++)
state[r][c] = in[inLoc++];
}
// copy: copy state to out

public static void copy(byte[] out, byte[][] state) {
int outLoc = 0;
for (int c = 0; c < Nb; c++)
for (int r = 0; r < 4; r++)
out[outLoc++] = state[r][c];
}
}
Java class: Print

// Print: print arrays of bytes
public class Print {
private static final int Nb = 4;
private static String[] dig = {"0","1","2","3","4","5","6","7",
"8","9","a","b","c","d","e","f"};

public static String hex(byte a) {
}
public static void printArray(String name, byte[] a) {

System.out.print(name + " ");
System.out.print(hex(a[i]) + " ");
}
public static void printArray(String name, byte[][] s) {

System.out.print(name + " ");
for (int c = 0; c < Nb; c++)
for (int r = 0; r < 4; r++)
System.out.print(hex(s[r][c]) + " ");
}
}
Java class: AEStest
// AEStest: test AES encryption
public class AEStest {

// for 128-bit key, use 16, 16, and 4 below
// for 192-bit key, use 16, 24 and 6 below
GetBytes getInput = new GetBytes("plaintext1.txt", 16);
byte[] in = getInput.getBytes();
GetBytes getKey = new GetBytes("key1.txt", 16);
byte[] key = getKey.getBytes();
AESencrypt aes = new AESencrypt(key, 4);
Print.printArray("Plaintext: ", in);
Print.printArray("Key: ", key);
byte[] out = new byte[16];
aes.Cipher(in, out);
Print.printArray("Ciphertext: ", out);
}
}
Program VI.24.a
AES Decryption
Classes Tables, GetBytes, Copy, and Print are the same as for encryption as presented
in the previous section.
The class AESdecrypt provides all the principle functions for the AES decryption algorithm,
while AESinvTest is a driver for testing decryption.
Java class: AESdecrypt
// AESdecrypt: AES decryption
public class AESdecrypt {
public final int Nb = 4; // words in a block, always 4 for now
public int Nk; // key length in words
public int Nr; // number of rounds, = Nk + 6
private int wCount; // position in w (= 4*Nb*(Nr+1) each encrypt)
AEStables tab; // all the tables needed for AES
byte[] w; // the expanded key
// AESdecrypt: constructor for class. Mainly expands key

public AESdecrypt(byte[] key, int NkIn) {
Nk = NkIn; // words in a key, = 4, or 6, or 8
Nr = Nk + 6; // corresponding number of rounds
tab = new AEStables(); // class to give values of various functions
w = new byte[4*Nb*(Nr+1)]; // room for expanded key
KeyExpansion(key, w); // length of w depends on Nr
}
// InvCipher: actual AES decryption

public void InvCipher(byte[] in, byte[] out) {
wCount = 4*Nb*(Nr+1); // count bytes during decryption
byte[][] state = new byte[4][Nb]; // the state array
Copy.copy(state, in); // actual component-wise copy
InvAddRoundKey(state); // xor with expanded key
for (int round = Nr-1; round >= 1; round--) {
Print.printArray("Start round " + (Nr - round) + ":", state);
InvShiftRows(state); // mix up rows
InvSubBytes(state); // inverse S-box substitution
InvMixColumns(state); // complicated mix of columns
}
Print.printArray("Start round " + Nr + ":", state);
InvShiftRows(state); // mix up rows
InvSubBytes(state); // inverse S-box substitution
Copy.copy(out, state);
}
// KeyExpansion: expand key, byte-oriented code, but tracks words

24. AES Decryption 291
// (the same as for encryption)

private void KeyExpansion(byte[] key, byte[] w) {
byte[] temp = new byte[4];
// first just copy key to w
int j = 0;
while (j < 4*Nk) {
w[j] = key[j++];
}
// here j == 4*Nk;
int i;
while(j < 4*Nb*(Nr+1)) {
i = j/4; // j is always multiple of 4 here
// handle everything word-at-a time, 4 bytes at a time
temp[iTemp] = w[j-4+iTemp];
if (i % Nk == 0) {
byte ttemp, tRcon;
byte oldtemp0 = temp[0];
for (int iTemp = 0; iTemp < 4; iTemp++) {
if (iTemp == 3) ttemp = oldtemp0;
else ttemp = temp[iTemp+1];
if (iTemp == 0) tRcon = tab.Rcon(i/Nk);
else tRcon = 0;
temp[iTemp] = (byte)(tab.SBox(ttemp) ˆ tRcon);
}
}
else if (Nk > 6 && (i%Nk) == 4) {
temp[iTemp] = tab.SBox(temp[iTemp]);
}
w[j+iTemp] = (byte)(w[j - 4*Nk + iTemp] ˆ temp[iTemp]);
j = j + 4;
}
}
// InvSubBytes: apply inverse Sbox substitution to each byte of state

private void InvSubBytes(byte[][] state) {
state[row][col] = tab.invSBox(state[row][col]);
}
// InvShiftRows: right circular shift of rows 1, 2, 3 by 1, 2, 3

private void InvShiftRows(byte[][] state) {
for (int r = 1; r < 4; r++) {
for (int c = 0; c < Nb; c++)
t[(c + r)%Nb] = state[r][c];
for (int c = 0; c < Nb; c++)
state[r][c] = t[c];
}
}
292 Program VI.24.a
// InvMixColumns: complex and sophisticated mixing of columns

private void InvMixColumns(byte[][] s) {
int[] sp = new int[4];
byte b0b = (byte)0x0b; byte b0d = (byte)0x0d;
byte b09 = (byte)0x09; byte b0e = (byte)0x0e;
for (int c = 0; c < 4; c++) {
sp[0] = tab.FFMul(b0e, s[0][c]) ˆ tab.FFMul(b0b, s[1][c]) ˆ
tab.FFMul(b0d, s[2][c]) ˆ tab.FFMul(b09, s[3][c]);
sp[1] = tab.FFMul(b09, s[0][c]) ˆ tab.FFMul(b0e, s[1][c]) ˆ
tab.FFMul(b0b, s[2][c]) ˆ tab.FFMul(b0d, s[3][c]);
sp[2] = tab.FFMul(b0d, s[0][c]) ˆ tab.FFMul(b09, s[1][c]) ˆ
tab.FFMul(b0e, s[2][c]) ˆ tab.FFMul(b0b, s[3][c]);
sp[3] = tab.FFMul(b0b, s[0][c]) ˆ tab.FFMul(b0d, s[1][c]) ˆ
tab.FFMul(b09, s[2][c]) ˆ tab.FFMul(b0e, s[3][c]);
for (int i = 0; i < 4; i++) s[i][c] = (byte)(sp[i]);
}
}
// InvAddRoundKey: same as AddRoundKey, but backwards

private void InvAddRoundKey(byte[][] state) {
for (int c = Nb - 1; c >= 0; c--)
for (int r = 3; r >= 0 ; r--)
state[r][c] = (byte)(state[r][c] ˆ w[--wCount]);
}
}
Java class: AESinvTest
// AESinvTest: test AES decryption
public class AESinvTest {

// for 128-bit key, use 16, 16, and 4 below
GetBytes getInput = new GetBytes("ciphertext1.txt", 16);
byte[] in = getInput.getBytes();
GetBytes getKey = new GetBytes("key1.txt", 16);
byte[] key = getKey.getBytes();
AESdecrypt aesDec = new AESdecrypt(key, 4);
Print.printArray("Ciphertext: ", in);
Print.printArray("Key: ", key);
byte[] out = new byte[16];
aesDec.InvCipher(in, out);
Print.printArray("Plaintext: ", out);
}
}
Program VI.24.b
Test Runs of the AES Algorithm
Here are results of test runs with all the sample test data supplied in the AES Specification
and in B. Gladman’s writeup about the AES. The values in the state variable are shown at
the start of each round. There are also test runs with plaintext and key all zeros and with a
single 1 inserted. The AES Specification and Gladman also show step-by-step results of the
key expansion for these cases, which was useful for my debugging, but I don’t show that data
here.
Gladman’s Test Data, 128-bit key
Encrypting ...
Plaintext: 32 43 f6 a8 88 5a 30 8d 31 31 98 a2 e0 37 07 34
Key: 2b 7e 15 16 28 ae d2 a6 ab f7 15 88 09 cf 4f 3c
Start round 1: 19 3d e3 be a0 f4 e2 2b 9a c6 8d 2a e9 f8 48 08
Start round 2: a4 9c 7f f2 68 9f 35 2b 6b 5b ea 43 02 6a 50 49
Start round 3: aa 8f 5f 03 61 dd e3 ef 82 d2 4a d2 68 32 46 9a
Start round 4: 48 6c 4e ee 67 1d 9d 0d 4d e3 b1 38 d6 5f 58 e7
Start round 5: e0 92 7f e8 c8 63 63 c0 d9 b1 35 50 85 b8 be 01
Start round 6: f1 00 6f 55 c1 92 4c ef 7c c8 8b 32 5d b5 d5 0c
Start round 7: 26 0e 2e 17 3d 41 b7 7d e8 64 72 a9 fd d2 8b 25
Start round 8: 5a 41 42 b1 19 49 dc 1f a3 e0 19 65 7a 8c 04 0c
Start round 9: ea 83 5c f0 04 45 33 2d 65 5d 98 ad 85 96 b0 c5
Start round 10: eb 40 f2 1e 59 2e 38 84 8b a1 13 e7 1b c3 42 d2
Ciphertext: 39 25 84 1d 02 dc 09 fb dc 11 85 97 19 6a 0b 32
Decrypting ...
Ciphertext: 39 25 84 1d 02 dc 09 fb dc 11 85 97 19 6a 0b 32
Start round 1: e9 31 7d b5 cb 32 2c 72 3d 2e 89 5f af 09 07 94
Start round 2: 87 6e 46 a6 f2 4c e7 8c 4d 90 4a d8 97 ec c3 95
Start round 3: be 3b d4 fe d4 e1 f2 c8 0a 64 2c c0 da 83 86 4d
Start round 4: f7 83 40 3f 27 43 3d f0 9b b5 31 ff 54 ab a9 d3
Start round 5: a1 4f 3d fe 78 e8 03 fc 10 d5 a8 df 4c 63 29 23
Start round 6: e1 fb 96 7c e8 c8 ae 9b 35 6c d2 ba 97 4f fb 53
Start round 7: 52 a4 c8 94 85 11 6a 28 e3 cf 2f d7 f6 50 5e 07
Start round 8: ac c1 d6 b8 ef b5 5a 7b 13 23 cf df 45 73 11 b5
Start round 9: 49 db 87 3b 45 39 53 89 7f 02 d2 f1 77 de 96 1a
Start round 10: d4 bf 5d 30 e0 b4 52 ae b8 41 11 f1 1e 27 98 e5

Encrypting ...
76 2e 71 60 f3 8b 4d a5
Start round 2: 72 48 f0 85 13 40 54 3f 5f 65 c0 61 17 35 e7 f1
Start round 3: 14 e2 0a 1f b3 dc 3a 62 36 27 2f d3 da 75 6f 70
294 Program VI.24.b
Start round 4: cb 42 fd 92 33 3f 28 43 21 11 fe 84 3c bc a8 1a
Start round 5: 94 99 c6 ee b9 78 94 12 bb 04 09 b7 a7 97 c0 25
Start round 6: 8a 6c 1e 3e db 78 a6 4e f5 db 78 62 ea d6 a4 01
Start round 7: 43 5c e2 58 97 7c 16 d8 71 7c 0f f7 79 19 e5 19
Start round 8: 70 b8 37 b9 ae fc 8b bc 5c d2 ab a5 cc 56 d7 4e
Start round 9: 94 a2 c3 31 ed 28 bf de d7 d6 c5 83 4b a9 ed 1e
Start round 10: 52 2d 88 c5 ed ab 19 4e 25 ec 73 1c 11 fa 6b 08
Start round 11: ab 82 54 06 da 72 4d 0c 2b cc f6 c2 39 32 12 01
Start round 12: 43 88 b3 26 6a f7 68 e8 4f cc a4 2a 3a 4d 45 5f
Ciphertext: f9 fb 29 ae fc 38 4a 25 03 40 d8 33 b8 7e bc 00
Decrypting ...
Ciphertext: f9 fb 29 ae fc 38 4a 25 03 40 d8 33 b8 7e bc 00
76 2e 71 60 f3 8b 4d a5
Start round 1: 1a 68 49 cf 02 4b 6e f7 84 e3 6d 9b 80 c4 45 e5
Start round 2: 62 40 42 7c 57 4b c9 6f f1 23 20 fe 12 13 e3 25
Start round 3: 00 62 8f 30 55 ce 7f a6 3f 2d c4 2f 82 d8 d4 9c
Start round 4: 22 34 a6 72 55 f6 55 c7 0e d3 2e 1d b3 3a 08 ec
Start round 5: 51 b0 62 2f e4 b5 0e 56 4a b1 9a 65 4b 6c 3d 06
Start round 6: 1a 10 76 d4 88 10 d9 6a a3 d4 98 61 b6 4a 47 68
Start round 7: 7e bc bc 7c b9 b9 49 b2 e6 f6 72 2f 87 50 24 aa
Start round 8: 22 bc 01 3f 56 f2 ba 28 ea 88 b4 c9 5c ee 22 a9
Start round 9: 1f 75 bb a2 c3 82 c2 4f fd 65 54 1a eb 2c 34 5f
Start round 10: fa 86 15 51 6d cc a8 c0 05 9d 67 aa 57 98 80 66
Start round 11: 40 09 ba a1 7d 4d 94 97 cf 96 8c 75 f0 52 20 ef

Encrypting ...
76 2e 71 60 f3 8b 4d a5 6a 78 4d 90 45 19 0c fe
Start round 2: 72 48 f0 85 13 40 54 3f 22 80 9e ea 6d 1f 2a b2
Start round 3: 59 f8 a8 d4 12 71 bf 44 e2 2b a6 5e 5d 69 9a 49
Start round 4: 88 a8 eb d5 66 49 40 5e 9f ad 55 e9 33 0d 7f 84
Start round 5: 6d 0c 80 51 d5 bc 1d b5 c5 1f 45 0f 18 46 7f 34
Start round 6: cf f5 04 43 27 6f 76 55 a5 5a fd 7b b6 99 f4 5f
Start round 7: 63 93 d6 68 2d da 2e 4f 42 88 77 37 12 57 8a 11
Start round 8: 29 26 ae 58 f4 32 23 4b f0 70 ff 6e 56 9e 44 23
Start round 9: 3f da e4 32 0e ce 55 ce c9 32 d6 55 e4 3a cd 2b
Start round 10: d6 40 90 18 38 64 11 35 61 ef 7c 37 99 00 31 fd
Start round 11: 6b 3c e6 72 ea e1 1d 52 e2 8f 1d 54 96 e0 c0 d0
Start round 12: 84 74 88 72 49 e5 0a 9f 17 c0 5a 37 a1 a6 9f 41
Start round 13: bf 8a 29 14 80 f8 06 21 44 3e 2b 81 aa 2f 4c 16
Start round 14: d3 20 3d d1 1a c7 2d 8b 5e c4 72 24 95 3d fe 5b
Ciphertext: 1a 6e 6c 2c 66 2e 7d a6 50 1f fb 62 bc 9e 93 f3
Decrypting ...
Ciphertext: 1a 6e 6c 2c 66 2e 7d a6 50 1f fb 62 bc 9e 93 f3
76 2e 71 60 f3 8b 4d a5 6a 78 4d 90 45 19 0c fe
Start round 1: 66 c6 40 39 a2 1c bb 3e 58 27 27 3d 2a b7 d8 36
24. AES Test Runs 295
Start round 2: 08 41 f1 47 cd b2 29 fa 1b 15 a5 fd ac 7e 6f 0c
Start round 3: 5f d9 be 83 3b ba db 40 f0 24 c4 db 32 92 67 9a
Start round 4: 7f f8 a4 70 87 73 ba 40 98 e1 8e 00 90 eb a4 20
Start round 5: f6 43 10 54 07 df c7 ad ef 63 60 96 ee 09 82 9a
Start round 6: 75 8b f6 f1 ab 23 bd 23 dd 80 69 8b 69 57 fc fc
Start round 7: a5 23 16 26 bf 51 1b 6a 8c 0b e4 b3 b1 f7 26 9f
Start round 8: fb 57 f5 82 d8 c4 7e 45 2c 5b f6 84 c9 dc 31 9a
Start round 9: 8a a8 54 cf cc be bf 1a 06 ee f2 fc 4e e6 38 21
Start round 10: 3c 65 6e 18 03 c0 d2 d1 a6 5a cd d5 ad fe a4 76
Start round 11: c4 3b fc 5f 33 95 d2 03 db d7 e9 58 c3 c2 09 1e
Start round 12: cb a3 24 3b c9 f1 b8 48 98 f9 c2 1b 4c 41 08 58
Start round 13: 40 09 0b 37 7d cd e5 97 93 c0 8c 75 3c 52 20 87
AES Specification Test Data, 128-bit key

Encrypting ...
Plaintext: 00 11 22 33 44 55 66 77 88 99 aa bb cc dd ee ff
Key: 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
Start round 1: 00 10 20 30 40 50 60 70 80 90 a0 b0 c0 d0 e0 f0
Start round 2: 89 d8 10 e8 85 5a ce 68 2d 18 43 d8 cb 12 8f e4
Start round 3: 49 15 59 8f 55 e5 d7 a0 da ca 94 fa 1f 0a 63 f7
Start round 4: fa 63 6a 28 25 b3 39 c9 40 66 8a 31 57 24 4d 17
Start round 5: 24 72 40 23 69 66 b3 fa 6e d2 75 32 88 42 5b 6c
Start round 6: c8 16 77 bc 9b 7a c9 3b 25 02 79 92 b0 26 19 96
Start round 7: c6 2f e1 09 f7 5e ed c3 cc 79 39 5d 84 f9 cf 5d
Start round 8: d1 87 6c 0f 79 c4 30 0a b4 55 94 ad d6 6f f4 1f
Start round 9: fd e3 ba d2 05 e5 d0 d7 35 47 96 4e f1 fe 37 f1
Start round 10: bd 6e 7c 3d f2 b5 77 9e 0b 61 21 6e 8b 10 b6 89
Ciphertext: 69 c4 e0 d8 6a 7b 04 30 d8 cd b7 80 70 b4 c5 5a
Decrypting ...
Ciphertext: 69 c4 e0 d8 6a 7b 04 30 d8 cd b7 80 70 b4 c5 5a
Key: 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
Start round 1: 7a d5 fd a7 89 ef 4e 27 2b ca 10 0b 3d 9f f5 9f
Start round 2: 54 d9 90 a1 6b a0 9a b5 96 bb f4 0e a1 11 70 2f
Start round 3: 3e 1c 22 c0 b6 fc bf 76 8d a8 50 67 f6 17 04 95
Start round 4: b4 58 12 4c 68 b6 8a 01 4b 99 f8 2e 5f 15 55 4c
Start round 5: e8 da b6 90 14 77 d4 65 3f f7 f5 e2 e7 47 dd 4f
Start round 6: 36 33 9d 50 f9 b5 39 26 9f 2c 09 2d c4 40 6d 23
Start round 7: 2d 6d 7e f0 3f 33 e3 34 09 36 02 dd 5b fb 12 c7
Start round 8: 3b d9 22 68 fc 74 fb 73 57 67 cb e0 c0 59 0e 2d
Start round 9: a7 be 1a 69 97 ad 73 9b d8 c9 ca 45 1f 61 8b 61
Start round 10: 63 53 e0 8c 09 60 e1 04 cd 70 b7 51 ba ca d0 e7

Encrypting ...
Key: 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
10 11 12 13 14 15 16 17
Start round 2: 4f 63 76 06 43 e0 aa 85 af f8 c9 d0 41 fa 0d e4
Start round 3: cb 02 81 8c 17 d2 af 9c 62 aa 64 42 8b b2 5f d7
296 Program VI.24.b
Start round 4: f7 5c 77 78 a3 27 c8 ed 8c fe bf c1 a6 c3 7f 53
Start round 5: 22 ff c9 16 a8 14 74 41 64 96 f1 9c 64 ae 25 32
Start round 6: 80 12 1e 07 76 fd 1d 8a 8d 8c 31 bc 96 5d 1f ee
Start round 7: 67 1e f1 fd 4e 2a 1e 03 df dc b1 ef 3d 78 9b 30
Start round 8: 0c 03 70 d0 0c 01 e6 22 16 6b 8a cc d6 db 3a 2c
Start round 9: 72 55 da d3 0f b8 03 10 e0 0d 6c 6b 40 d0 52 7c
Start round 10: a9 06 b2 54 96 8a f4 e9 b4 bd b2 d2 f0 c4 43 36
Start round 11: 88 ec 93 0e f5 e7 e4 b6 cc 32 f4 c9 06 d2 94 14
Start round 12: af b7 3e eb 1c d1 b8 51 62 28 0f 27 fb 20 d5 85
Ciphertext: dd a9 7c a4 86 4c df e0 6e af 70 a0 ec 0d 71 91
Decrypting ...
Ciphertext: dd a9 7c a4 86 4c df e0 6e af 70 a0 ec 0d 71 91
Key: 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
10 11 12 13 14 15 16 17
Start round 1: 79 3e 76 97 9c 34 03 e9 aa b7 b2 d1 0f a9 6c cc
Start round 2: c4 94 bf fa e6 23 22 ab 4b b5 dc 4e 6f ce 69 dd
Start round 3: d3 7e 37 05 90 7a 1a 20 8d 1c 37 1e 8c 6f bf b5
Start round 4: 40 6c 50 10 76 d7 00 66 e1 70 57 ca 09 fc 7b 7f
Start round 5: fe 7c 7e 71 fe 7f 80 70 47 b9 51 93 f6 7b 8e 4b
Start round 6: 85 e5 c8 04 2f 86 14 54 9e bc a1 7b 27 72 72 df
Start round 7: cd 54 c7 28 38 64 c0 c5 5d 4c 72 7e 90 c9 a4 65
Start round 8: 93 fa a1 23 c2 90 3f 47 43 e4 dd 83 43 16 92 de
Start round 9: 68 cc 08 ed 0a bb d2 bc 64 2e f5 55 24 4a e8 78
Start round 10: 1f b5 43 0e f0 ac cf 64 aa 37 0c de 3d 77 79 2c
Start round 11: 84 e1 dd 69 1a 41 d7 6f 79 2d 38 97 83 fb ac 70

Encrypting ...
Key: 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f
Start round 2: 4f 63 76 06 43 e0 aa 85 ef a7 21 32 01 a4 e7 05
Start round 3: 18 59 fb c2 8a 1c 00 a0 78 ed 8a ad c4 2f 61 09
Start round 4: 97 5c 66 c1 cb 9f 3f a8 a9 3a 28 df 8e e1 0f 63
Start round 5: 1c 05 f2 71 a4 17 e0 4f f9 21 c5 c1 04 70 15 54
Start round 6: c3 57 aa e1 1b 45 b7 b0 a2 c7 bd 28 a8 dc 99 fa
Start round 7: 7f 07 41 43 cb 4e 24 3e c1 0c 81 5d 83 75 d5 4c
Start round 8: d6 53 a4 69 6c a0 bc 0f 5a ca ab 5d b9 6c 5e 7d
Start round 9: 5a a8 58 39 5f d2 8d 7d 05 e1 a3 88 68 f3 b9 c5
Start round 10: 4a 82 48 51 c5 7e 7e 47 64 3d e5 0c 2a f3 e8 c9
Start round 11: c1 49 07 f6 ca 3b 3a a0 70 e9 aa 31 3b 52 b5 ec
Start round 12: 5f 9c 6a bf ba c6 34 aa 50 40 9f a7 66 67 76 53
Start round 13: 51 66 04 95 43 53 95 03 14 fb 86 e4 01 92 25 21
Start round 14: 62 7b ce b9 99 9d 5a aa c9 45 ec f4 23 f5 6d a5
Ciphertext: 8e a2 b7 ca 51 67 45 bf ea fc 49 90 4b 49 60 89
Decrypting ...
Ciphertext: 8e a2 b7 ca 51 67 45 bf ea fc 49 90 4b 49 60 89
Key: 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 0d 0e 0f
10 11 12 13 14 15 16 17 18 19 1a 1b 1c 1d 1e 1f
Start round 1: aa 5e ce 06 ee 6e 3c 56 dd e6 8b ac 26 21 be bf
24. AES Test Runs 297
Start round 2: d1 ed 44 fd 1a 0f 3f 2a fa 4f f2 7b 7c 33 2a 69
Start round 3: cf b4 db ed f4 09 38 08 53 85 02 ac 33 de 18 5c
Start round 4: 78 e2 ac ce 74 1e d5 42 51 00 c5 e0 e2 3b 80 c7
Start round 5: d6 f3 d9 dd a6 27 9b d1 43 0d 52 a0 e5 13 f3 fe
Start round 6: be b5 0a a6 cf f8 56 12 6b 0d 6a ff 45 c2 5d c4
Start round 7: f6 e0 62 ff 50 74 58 f9 be 50 49 76 56 ed 65 4c
Start round 8: d2 2f 0c 29 1f fe 03 1a 78 9d 83 b2 ec c5 36 4c
Start round 9: 2e 6e 7a 2d af c6 ee f8 3a 86 ac e7 c2 5b a9 34
Start round 10: 9c f0 a6 20 49 fd 59 a3 99 51 89 84 f2 6b e1 78
Start round 11: 88 db 34 fb 1f 80 76 78 d3 f8 33 c2 19 4a 75 9e
Start round 12: ad 9c 7e 01 7e 55 ef 25 bc 15 0f e0 1c cb 63 95
Start round 13: 84 e1 fd 6b 1a 5c 94 6f df 49 38 97 7c fb ac 23
Plaintest and Key all zeros, 128-bit key

Encrypting ...
Plaintext: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Key: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Start round 1: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Start round 2: 01 00 00 00 01 00 00 00 01 00 00 00 01 00 00 00
Start round 3: c6 e4 e4 8b a4 87 87 e8 c6 e4 e4 8b a4 87 87 e8
Start round 4: 28 2d f3 c4 6a f3 86 25 4a 4e 90 a7 08 90 e5 46
Start round 5: ab d2 cd fe 37 5a b5 49 50 a0 af c0 75 9a 6a 5f
Start round 6: d4 6f 4f 6c 55 b8 96 33 7e 05 bb 3d 79 79 de 23
Start round 7: 04 f2 ca 97 07 78 28 45 e2 2f 01 96 49 c5 d7 10
Start round 8: b7 aa e4 c5 1d 25 2d 4f 6c 92 0f 81 94 e5 81 50
Start round 9: 23 e7 8c 3c 13 21 63 db aa c0 c6 57 2e 03 cb 95
Start round 10: 7f fe 0e 95 51 a5 66 35 0e 34 7c 47 29 29 ec cb
Ciphertext: 66 e9 4b d4 ef 8a 2c 3b 88 4c fa 59 ca 34 2b 2e
Decrypting ...
Ciphertext: 66 e9 4b d4 ef 8a 2c 3b 88 4c fa 59 ca 34 2b 2e
Key: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Start round 1: d2 06 10 1f d1 18 ce 2a ab a5 ab 96 a5 bb 33 a0
Start round 2: 26 fd b4 2a 7d ba 1f eb ac 7b 64 b9 31 94 fb 5b
Start round 3: a9 3f 76 53 a4 4f 0c a6 50 d9 69 84 22 ac d8 0c
Start round 4: f2 bc 7c ca c5 15 0e 88 98 a6 74 6e 3b 89 34 90
Start round 5: 48 6c ea 26 fc 6b 1d 50 f3 b6 84 c3 b6 a8 90 27
Start round 6: 62 be 79 cf 9a e0 02 bb 53 b8 bd 3b 9d b5 d5 ba
Start round 7: 34 0d 60 5a 02 2f d9 1c d6 60 0d 3f 30 d8 44 5c
Start round 8: b4 17 69 9b 49 69 17 3d b4 17 69 9b 49 69 17 3d
Start round 9: 7c 63 63 63 7c 63 63 63 7c 63 63 63 7c 63 63 63
Start round 10: 63 63 63 63 63 63 63 63 63 63 63 63 63 63 63 63
Plaintext: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Plaintest all zeros, key a single 1, 128-bit key

Encrypting ...
Plaintext: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Key: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01
Start round 1: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01
Start round 2: 1e 1f 3e 3e 01 00 1f 00 01 00 1f 00 01 00 1f 01
Start round 3: 66 e0 d8 04 d6 43 e3 f2 17 67 13 5a f9 cf 28 e9
Start round 4: 7e 41 45 5e 09 86 08 b5 a1 69 f0 da 70 61 e1 bf
298 Program X.27.b
Start round 5: 79 89 5e 0d d2 19 17 93 a2 96 41 74 c2 e1 0d 64
Start round 6: a0 f7 36 25 ca ae 92 22 a4 76 49 4b 04 d0 d6 8b
Start round 7: a8 1b dd b6 3b f7 72 10 81 6b 51 a9 70 27 04 e1
Start round 8: 89 51 a6 38 37 41 9e 27 9c 5d fe a7 d0 20 3c 26
Start round 9: 4c 39 fa 23 09 9f 8d 0b 53 f7 13 4f 0a 53 02 53
Start round 10: dd 9d a7 03 9c a1 f1 58 42 43 46 94 5d c8 68 7a
Ciphertext: 05 45 aa d5 6d a2 a9 7c 36 63 d1 43 2a 3d 1c 84
Decrypting ...
Ciphertext: 05 45 aa d5 6d a2 a9 7c 36 63 d1 43 2a 3d 1c 84
Key: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01
Start round 1: c1 32 5a da de 1a 45 7b 2c e8 5c 6a 4c 5e a1 22
Start round 2: 29 db 7d ed 01 68 77 26 ed ed 2d 2b 67 12 5d 84
Start round 3: a7 83 bb f7 9a 4c eb 07 de b7 24 cc 70 d1 0b 5c
Start round 4: c2 68 d1 f8 e2 7f f2 4e 0c cc c1 ca 51 af 40 d3
Start round 5: e0 e4 3b 3d 74 38 f6 3f 49 70 05 93 f2 68 4f b3
Start round 6: b6 d4 83 43 b5 90 d7 d7 3a f8 58 dc 25 a7 f0 92
Start round 7: f3 44 8c 08 01 f9 f8 58 32 ef 6e d5 51 83 30 57
Start round 8: 33 1a 7d 1e f6 85 34 f2 f0 8a 61 89 99 e1 11 be
Start round 9: 72 63 c0 7c 7c 63 c0 b2 7c 63 b2 63 7c c0 c0 63
Start round 10: 63 63 63 7c 63 63 63 63 63 63 63 63 63 63 63 63
Plaintext: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Plaintest a single 1, key all zeros, 128-bit key

Encrypting ...
Plaintext: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01
Key: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Start round 1: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01
Start round 2: 1e 1f 21 3e 01 00 00 00 01 00 00 00 01 00 00 00
Start round 3: da ea ea 99 75 56 ef 51 58 5d c3 15 5a da 24 4b
Start round 4: 6b 78 40 4b 44 dd 5b fb 89 73 17 e4 78 87 db aa
Start round 5: 14 47 70 88 31 17 26 e7 1f 03 f1 40 92 6c 27 a3
Start round 6: 30 dd 66 d8 e8 fe 56 7f 23 dd e9 d2 68 54 93 04
Start round 7: de d8 c5 d3 dc 39 2f e5 54 1e 04 f1 bc 15 a2 69
Start round 8: 26 b8 f7 a9 e8 1a 48 1e 78 15 35 0a 1b 22 26 19
Start round 9: 44 24 06 b6 d9 51 38 9d 40 4d 63 c4 95 fb 34 85
Start round 10: 83 f7 5d d0 88 f3 8d e2 2f 83 89 19 59 35 7f 36
Ciphertext: 58 e2 fc ce fa 7e 30 61 36 7f 1d 57 a4 e7 45 5a
Decrypting ...
Ciphertext: 58 e2 fc ce fa 7e 30 61 36 7f 1d 57 a4 e7 45 5a
Key: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Start round 1: ec 0d a7 05 c4 ec d2 70 15 96 4c 98 cb 68 5d d4
Start round 2: 1b d1 fb 97 35 e3 18 4e 09 0f 6f 5e 2a 36 07 1c
Start round 3: f7 a2 96 d4 9b 59 f7 d3 bc 93 68 72 af 6c 52 67
Start round 4: 1d 12 f2 f9 86 72 3a 66 20 59 a6 d9 65 61 15 a1
Start round 5: 04 bb 1e f2 9b c1 dc 61 26 20 33 d2 45 c1 b1 b5
Start round 6: fa f0 a1 0a c7 7b cc c4 c0 50 51 94 4f a0 f7 09
Start round 7: 7f c1 f0 ac 1b 8f b9 b3 a7 17 09 0f bc bc 39 69
Start round 8: 57 b1 2e b3 9d 4c 36 ee 6a 57 87 d1 be 87 df 59
Start round 9: 72 63 63 63 7c 63 63 b2 7c 63 fd 63 7c c0 63 63
Start round 10: 63 63 63 7c 63 63 63 63 63 63 63 63 63 63 63 63
Plaintext: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01
Program X.28.a
Shamir’s Threshold Scheme
The code below implements Shamir’s threshold scheme using 6 Java classes:
CreateThreshold: This class uses parameters fed into the constructor to create a
new threshold scheme. Inputs are the secret s, the threshold value t, the number of
users n, and the prime p. It first creates a random polynominal of degree t-1 by
choosing the coefficients at random. (The random number generator used is just Java’s
Math.random(), so an actual production system would need a better generator.) Then
the class evaluates the polynomial at successive positive integers to create the shares.
Secret: This class also uses parameters fed into the constructor to recover the secret
value. The input parameters are the threshold value t, the prime p, and x and y coor-
dinates for t shares. There is also a boolean parameter debug that will produce debug
output if it is true. After calculating the secret, the method getSecret() will return
it.
NewThreshold: The main of this class uses command line parameters to create a new
threshold scheme. On the command line are: the secret s, the threshold value t, the
number of users n, and the prime p. The class supplies arrays into which the n shares are
placed by the createThreshold class. The class finally writes the parameters (on the
first line) and the shares (one to a line) to the standard output as 4 + 2*n integers.
RecoverSecret: The main of this class reads integers from the standard input, first
the threshold value t and the prime p, and then t shares (x and y coordinates). The class
finally writes the secret to the standard output.
ThresholdTest: This class has a main that creates instances of CreateThresh-
old and Secret in debug mode. Using the same 4 command line parameters as
NewThreshold, it creates a threshold scheme instance, providing debug output show-
ing the n shares. Then t of these shares are chosen at random for input to the Secret
class, which also provides debug output.
GetNext: A class to read ints from the standard input. The integers are delimited by
any non-digit characters, which are ignored.
Java class: CreateThreshold

// CreateThreshold: Shamir’s threshold scheme: return shares
public class CreateThreshold {
long s; // "secret"
int t; // threshold value, t <= n
int n; // number of users, n >= 2
300 Program X.28.a
long p; // prime for system, p > S, p > t

long[] X; // array of n shares, x coordinates
long[] Y; // array of n shares, y coordinates
long[] A; // array of t random coefficients for polynomial
boolean debug = false; // output debug information
// CreateThreshold: constructor, does most of the work

public CreateThreshold(long s1, int t1, int n1, long p1,
long[] X1, long[] Y1, boolean d) {
s = s1; t = t1; n = n1; p = p1; debug = d;
if ( n < 2 || t > n || p <= s || p <= t) {
System.out.println("Parameter out of range");
System.exit(1);
}
X = X1;
Y = Y1;
A = new long[t];
createF(); // puts random coefficients into A, to create poly
if (debug) { // printout for debugging and demonstrations
System.out.print("New (" + t + "," + n + ") threshold scheme,");
System.out.println(" with p = " + p + " and s = " + s);
System.out.print("Function f(x) = ");
for (int i = 0; i < t; i++) {
System.out.print(A[i] + "*xˆ" + i + " ");
if (i != t-1) System.out.print("+ ");
}
}
createShares(); // use poly to create shares
if (debug) { // more debug printout
System.out.print("All " + n + " Output Shares: ");
for (int i = 0; i < n; i++)
System.out.print("(" + X[i] + "," + Y[i] + ") ");
System.out.println("\n");
}
}
// evalF: evaluate the function f

private long evalF(long x) {
long y = 0;
for (int i = t - 1; i >= 0; i--) {
y = y*x % p;
y = (y + A[i]) % p;
}
return y;
}
// createF: create F with random coefficients

// that is, load A with random coeffs
private void createF() {
A[0] = s; // the secret
for (int i = 1; i < t; i++)
A[i] = randlong(0, p-1);
// randlong: return a random long x, a <= x <= b

private long randlong(long a, long b) {
return (long)( Math.random()*(b - a + 1) + a);
}
// createShares: load X and Y with (x,y) coords of each share

public void createShares() {
for(int i = 0; i < n; i++) {
X[i] = i+1;
Y[i] = evalF(i+1);
}
}
}
Java class: Secret

// Secret: Shamir’s threshold scheme: return secret
public class Secret {
long s; // "secret"
int t; // threshold value, t <= n
long p; // prime for system, p > S, p > t
long[] X; // array of t shares, x coordinates
long[] Y; // array of t shares, y coordinates
long[] C; // array of t coefficents
boolean debug = false; // output debug information
// Secret: constructor, does most of the work

public Secret(int t1, long p1, long[] X1, long[] Y1, boolean d) {
t = t1; p = p1; debug = d; X = X1; Y = Y1;
if (p <= t) {
System.out.println("Parameter out of range");
System.exit(1);
}
if (debug) {
System.out.print("Recover secret from t = " + t);
System.out.println(" shares, with p = " + p);
System.out.print("All " + t + " Input Shares: ");
for (int i = 0; i < t; i++)
System.out.print("(" + X[i] + "," + Y[i] + ") ");
}
C = new long[t];
createC();
}
// getSecret: return the secret value

public long getSecret() {
return s;
}
// createC: do the calculation of the secret

// Note: interspresed debug output
302 Program X.28.a
private void createC() {

for (int i = 0; i < t; i++) { // calculate C[i]
if (debug) System.out.print("C[" + i + "] = ");
C[i] = 1;
for (int j = 0; j < t; j++)
if (i != j) {
if (debug)
System.out.print(X[j] + "/(" + X[j] + "-" +
X[i] + ") ");
long term = modP(X[j]*invModP(modP(X[j] - X[i])));
if (debug)
System.out.print(" ( or " + term + ") ");
C[i] = modP(C[i]*term );
}
if (debug) System.out.println("= " + C[i]);
}
s = 0;
if (debug) System.out.print("Secret = ");
for (int i = 0; i < t; i++) {
if (debug)
System.out.print(C[i] + "*" + Y[i] + " ");
if (i != t-1) if (debug) System.out.print("+ ");
s = modP(s + C[i]*Y[i]);
}
if (debug) System.out.println("= " + s);
}
// modP: does actual x mod p, even if x < 0

private long modP(long x) {
long y = x%p;
if (y < 0) y = y + p;
return y;
}
// invModP: calculate an inverse value mod p

private long invModP(long x) {
long[] res = new long[3];
res = GCD(x, p);
return modP(res[0]);
}
// GCD: extended GCD algorithm

private static long[] GCD(long x, long y) {
while (v[2] != 0) {
long q = u[2]/v[2];
for (int i = 0; i < 3; i++) {
t[i] = u[i] -v[i]*q; u[i] = v[i]; v[i] = t[i];
}
}
return u;
}
}
Java class: NewThreshold

// NewThreshold: create instance of Shamir’s threshold scheme
// Input (on command line):
// s (the secret), t (threshold value), n (# users), p (prime)
// Output (in System.out):
// s t n p
// n (x,y) threshold pairs, for the n users
public class NewThreshold {

long s = Long.parseLong(args[0]); // "secret"
int t = Integer.parseInt (args[1]); // threshold, t <= n
int n = Integer.parseInt (args[2]); // number of users, n >= 2
long p = Long.parseLong(args[3]); // prime , p > S, p > t
long[] X = new long[n];
long[] Y = new long[n];
CreateThreshold cT = new CreateThreshold(s, t, n, p, X, Y, false);
System.out.println(s + " " + t + " " + n + " " + p);
for (int i = 0; i < n; i++)
System.out.println(X[i] + " " + Y[i]);
}
}
Java class: RecoverSecret

// RecoverSecret: read t shares and recover the secret
// Input (in System.in):
// s t
// t (x,y) threshold pairs
// Output (in System.out):
// s (the secret)
public class RecoverSecret {

long s; // "secret"
int t; // threshold, t <= n
long p; // prime , p > S, p > t
GetNext getNext = new GetNext();
t = getNext.getNextInt();
p = (long)getNext.getNextInt();
long[] X = new long[t];
long[] Y = new long[t];
for (int i = 0; i < t; i++) {
X[i] = (long)getNext.getNextInt();
Y[i] = (long)getNext.getNextInt();
}
Secret secret = new Secret(t, p, X, Y, false);
s = secret.getSecret();
}
}
Java class: ThresholdTest

// ThresholdTest: Test Shamir’s threshold scheme, produce debug output
public class ThresholdTest {
304 Program X.28.a

long s = Long.parseLong(args[0]); // "secret"
int t = Integer.parseInt (args[1]); // threshold, t <= n
int n = Integer.parseInt (args[2]); // number of users, n >= 2
long p = Long.parseLong (args[3]); // prime, p > S, p > t
long[] X = new long[n];
long[] Y = new long[n];
CreateThreshold cT = new CreateThreshold(s, t, n, p, X, Y, true);
long[] Xs = new long[t];
long[] Ys = new long[t];
// choose t of the n shares at random
int[] select = new int[n];
for (int i = 0; i < n; i++) select[i] = i; // indexes of shares
for (int i = 0; i < t; i++) { // interchange first t at random
int j = (int)(Math.random()*((n-1) - i + 1) + i);
int temp = select[i];
select[i] = select[j];
select[j] = temp;
}
for (int i = 0; i < t; i++) {
Xs[i] = X[select[i]];
Ys[i] = Y[select[i]];
}
Secret secret = new Secret(t, p, Xs, Ys, true);
s = secret.getSecret();
System.out.println("Secret: " + s);
}
}
Java class: GetNext

// GetNext: fetch next char or unsigned integer from System.in
import java.io.*;
public class GetNext {
private Reader in; // internal file name for input stream
// GetNext: constructor
public GetNext () {
}
// getNextChar: fetches next char

try {
}
return ch;
}
// getNextInt: fetch unsigned int

public int getNextInt() {
String s ;
char ch;
while (!Character.isDigit(ch = getNextChar()))
;
s = "" + ch;
while (Character.isDigit(ch = getNextChar()))
s += ch;
return Integer.parseInt(s);
}
}
Here are two runs testing the threshold scheme implementation:

Function f(x) = 1111*xˆ0 + 1981*xˆ1 + 196*xˆ2 + 961*xˆ3 + 288*xˆ4 +
1696*xˆ5
All 9 Output Shares: (1,236) (2,461) (3,456) (4,1049) (5,850) (6,1870)
(7,1147) (8,363) (9,1468)

All 6 Input Shares: (6,1870) (8,363) (4,1049) (7,1147) (5,850) (2,461)
C[0] = 8/(8-6) ( or 4) 4/(4-6) ( or 1997) 7/(7-6) ( or 7) 5/(5-6)
( or 1994) 2/(2-6) ( or 999) = 1859
C[1] = 6/(6-8) ( or 1996) 4/(4-8) ( or 1998) 7/(7-8) ( or 1992)
5/(5-8) ( or 1331) 2/(2-8) ( or 666) = 1321
C[2] = 6/(6-4) ( or 3) 8/(8-4) ( or 2) 7/(7-4) ( or 1335) 5/(5-4)
( or 5) 2/(2-4) ( or 1998) = 1929
C[3] = 6/(6-7) ( or 1993) 8/(8-7) ( or 8) 4/(4-7) ( or 665) 5/(5-7)
( or 997) 2/(2-7) ( or 1199) = 64
C[4] = 6/(6-5) ( or 6) 8/(8-5) ( or 669) 4/(4-5) ( or 1995) 7/(7-5)
( or 1003) 2/(2-5) ( or 1332) = 1482
C[5] = 6/(6-2) ( or 1001) 8/(8-2) ( or 1334) 4/(4-2) ( or 2) 7/(7-2)
( or 801) 5/(5-2) ( or 668) = 1342
Secret = 1859*1870 + 1321*363 + 1929*1049 + 64*1147 + 1482*850 +
1342*461 = 1111
Secret: 1111
Function f(x) = 444444444*xˆ0 + 321956576*xˆ1 + 166564884*xˆ2 +
237875836*xˆ3
All 6 Output Shares: (1,97099922) (2,436398366) (3,205240247)
(4,294009672) (5,519348930) (6,161029401)

All 4 Input Shares: (3,205240247) (4,294009672) (6,161029401)
(5,519348930)
C[0] = 4/(4-3) ( or 4) 6/(6-3) ( or 2) 5/(5-3) ( or 268435457) = 20
C[1] = 3/(3-4) ( or 536870906) 6/(6-4) ( or 3) 5/(5-4) ( or 5) =
536870864
C[2] = 3/(3-6) ( or 536870908) 4/(4-6) ( or 536870907) 5/(5-6)
( or 536870904) = 536870899
306 Program X.28.a
C[3] = 3/(3-5) ( or 268435453) 4/(4-5) ( or 536870905) 6/(6-5)

( or 6) = 36
Secret = 20*205240247 + 536870864*294009672 + 536870899*161029401 +
36*519348930 = 444444444
Secret: 444444444
The following code illustrates the acutal threshold schemes, without the debug information.
First using standard input and output:
% java NewThreshold 2222222 5 8 10316017
2222222 5 8 10316017
1 9512402
2 8010272
3 7372056
4 8834487
5 5214807
6 1542801
7 4744780
8 3011547
% java RecoverSecret
5 10316017
3 7372056
4 8834487
7 4744780
8 3011547
2 8010272
2222222
Next using redirected files in Unix:

% java NewThreshold 2222222 5 8 10316017 > thresh.txt
% cat thresh.txt
2222222 5 8 10316017
1 1769097
2 766836
3 4599213
4 4208181
5 7041923
6 6106801
7 599390
8 6222495
% cat secret.txt
5 10316017
3 4599213
6 6106801
2 766836
7 599390
1 1769097
% java RecoverSecret < secret.txt
2222222
Appendices
308
A
Using Printed Log Tables
This is a lesson from prehistoric times. It brings back nostalgic memories. Before calculators,
one used printed tables to carry out calculations. The example in the main section was to
calculate 23.427 * 23.427 * 3.1416. To do this, one first needed the logarithms (base
10) of the two numbers. In colored bold italic below are the actual table entries (using a book
of tables dating from 1957) — everything else you had to do mentally or on paper:
Number Log Explanation
2342 36959 using interpolation entry:
23427 369716 7th entry under 18 is 12.6
2343 36977 take 369590 + 126 to get 369716
This means that log(2.3427) = 0.369716 approximately. Then log(23.427) =

log(2.3427 * 10) = log(2.3427) + log(10) = 0.369716 + 1 = 1.369716
Similarly, look up 3.1416:
31416 497154 6th entry under 14 is 8.4
3142 49721 take 497070 + 84 to get 497154
This means that log(3.1416) = 0.497154 approximately.

Form the sum: 1.369716 + 1.369716 + 0.497154 = 3.236586 (this must be done
by hand, with pencil and paper).
Now finally, one has to look up the “anti-log” in the same table:
17242 23659 2nd entry under 25 is 5.0
1725 23679 take 17240 + 2 to get 17242
This means that log(1.7242) = 0.23659 approximately, so 3.23659 = 3 +

0.23659 = log(1000) + log(1.7242) = log(1000 * 1.7242) = log(1724.2),
or (finally), the answer is 1724.2 approximately. So the area of a circle of radius 23.427 is
approximately 1724.2.
All this pain just to multiply 3 numbers together, to get 4 or 5 digits of accuracy in the answer.
The next two lessons from our primitive ancestors: how to use tables of the logarithms of trig
functions (to save one lookup), and how to use a slide rule. (Just kidding.)
B
Unsigned bytes in Java
The Advanced Encryption Standard (AES) makes extensive use of the unsigned byte type.
This creates awkward code in Java, because Java supports only the signed byte type. In
addition, some of the Java operators do not work as in the documentation, creating further
problems. For example:
public class TestRightShift0 {
byte b = (byte)0x80;
int c = b >>> 4; // <----------------------------------
System.out.println("b: " + b + ", c: " + c);
c = 0x0ffffff8;
System.out.println("c: " + c);
}
}
/* Output:
b: -128, c: 268435448
c: 268435448
*/
The output shows that b has the value 0x80 or 1000 0000 in binary, as one would expect.
According to The Java Programming Language, Third Edition, page 164, the >>> operator
should fill new high-order bits with zeros. In fact, though, Java is converting b to int type
with sign-extended value 0xffffff80, right shifting this and putting just four zeros at the
right, to give 0x0ffffff8. To get the desired value, one can use either of the following:
int c = (b & 0xf0) >> 4;} // <----------------------------------
c = (b >> 4) & 0xf;} // <----------------------------------
}
}
/* Output:
b: -128, c: 8
b: -128, c: 8
*/
Similarly, a right shift of 3 could use either of the lines:
int c = (b & 0xf1) >> 3; // <----------------------------------
311

c = (b >> 3) & 0x1f; // <----------------------------------
}
}
/* Output:
b: -128, c: 16
b: -128, c: 16
*/
This seems to require different constants for different shifts, but actually, the first method works
with a fixed constant:
for (int i = 0; i < 9; i++) {
int c = (b & 0xff) >> i; // <----------------------------------
System.out.println("b: " + b + ", shift by: " + i + ", c: " + c);
}
}
}
/* Output:
b: -128, shift by: 0, c: 128
b: -128, shift by: 1, c: 64
b: -128, shift by: 2, c: 32
b: -128, shift by: 3, c: 16
b: -128, shift by: 4, c: 8
b: -128, shift by: 5, c: 4
b: -128, shift by: 6, c: 2
b: -128, shift by: 7, c: 1
b: -128, shift by: 8, c: 0
*/
Law JAVA-BYTES-1:
In the Java language, to right shift an integer amount
shiftAmount, use the code
int shiftedValue = (byteValue & 0xff) shiftAmount;**
where byteValue is of type byte and shiftAmount is an int in the
range from 0 to 8. A 0 for shiftAmount is the same as not do-
ing the shift, but just to store an unsigned byte into an int type
requires
int shiftedValue = byteValue & 0xff;
Left shifts work as they ought to, but the result is an int, so it needs to be cast to a byte if
that is needed.
public class TestLeftShift {
312 B. Unsigned bytes in Java
for (int i = 0; i < 9; i++) {
int c = (b << i);} // <----------------------------------
System.out.println("b: " + b + ", shift by: " + i + ", c: " + c);
byte bb = (byte)(b << i);} // <----------------------------------
System.out.println("b: " + b + ", shift by: " + i + ", bb: " + bb);
}
}
}
/* Output:
b: 1, shift by: 0, c: 1
b: 1, shift by: 0, bb: 1
b: 1, shift by: 1, c: 2
b: 1, shift by: 2, c: 4
b: 1, shift by: 3, c: 8
b: 1, shift by: 4, c: 16
b: 1, shift by: 5, c: 32
b: 1, shift by: 6, c: 64
b: 1, shift by: 7, c: 128
b: 1, shift by: 7, bb: -128
b: 1, shift by: 8, c: 256
*/
Law JAVA-BYTES-2:
In the Java language, logical and shifting operators work
as follows:
All operators return an int, so they must be cast to a byte if
a byte is needed. This includes: &, , *5* *5* *
, and .
Hex constants such as 0xff actually define an integer, so this
is the same as 0x000000ff. For values bigger than Ox7f a
cast to byte is needed.
Arithmetic (except for / and %) with Java’s signed bytes
works just as if the bytes were unsigned, since there is no
overflow.
C
Projects
Project 1: Designing a Stream Cipher.

For this project you are to start with a very simple cryptosystem based on the exclusive-or of a
pseudo-random number sequence with plain text to produce ciphertext, followed by the same
exclusive-or to transform ciphertext back into plaintext. This is the simplest possible stream
cipher.
As a stream cipher, an unending stream of bits is generated from a key (or seed to the random
number generator) and is xored with successive message bits to form the ciphertext. Everything
can be thought of occurring one bit at a time.
The code below uses the Java Random class for the pseudo-random number generator. For
convenience of the Java input function, the encryption is done byte at a time, using 8 bits from
the generator at each step. Despite the byte orientation, it is still essentially a stream cipher,
with encryption occurring one bit at a time.
Since the generator is of the multiplicative linear congruence type, one needs to extract the high
order 8 bits, rather than low order bits. (Knuth notes that with these generators, the high-order
bits are much more random than low-order ones.)
Here is the program. I deliberately wrote this using as few lines of Java as possible (well, except
for a few extra variables).
Java class: Crypto
// Cipher.java: simple cipher
import java.io.*;
class Cipher {
public static void main(String[] args) throws IOException {
long seed = Long.parseLong(args[0]);
Random rand = new Random(seed);
int ch;
while ((ch = System.in.read()) != -1) {
double x = rand.nextDouble();
int m = (int)(256.0*x);
ch = (ch ˆ m);
System.out.write(ch);
}
System.out.close();
}
314 C. Projects
Here is a sample run, starting with a binary PDF file utsa.pdf. Using Unix redirection, the
encrypted output goes to a binary file utsa.binary of the same size. Finally, another run
using the same key recovers the original file, this time named utsa2.pdf. Notice that the key
is the long with value 98765432123456789, nearly a full 64-bit integer.
% pandora% javac Cipher.java
% java Cipher 98765432123456789 < utsa.pdf > utsa.binary
% java Cipher 98765432123456789 < utsa.binary > utsa2.pdf
% ls -l
total 18
-rw-r--r-- 1 wagner faculty 769 May 26 22:33 Cipher.class
-rw-r--r-- 1 wagner faculty 462 May 26 20:34 Cipher.java
-rw-r--r-- 1 wagner faculty 3116 May 26 22:34 utsa2.pdf
-rw-r--r-- 1 wagner faculty 3116 May 26 22:33 utsa.binary
-rw-r--r-- 1 wagner faculty 3116 May 26 16:32 utsa.pdf
Analysis of the initial system: The above system uses the Random class from Java as the
random number generator (RNG). This class is provided for simulation and for other similar
uses and is not intended to be a cryptographically secure RNG, where this means a RNG whose
future outputs cannot be efficiently calculated from earlier outputs. In the case of the class
Random, the multiplier and modulus are known, so any output immediately allows calculation
of later outputs. Even if the multiplier and modulus are not known, given a sequence of integers
produced by such a RNG, there are efficient (though difficult) algorithms to calculate them.
In the case above, one is far from knowing any of the integer outputs, however. One knows
only a succession of 8-bit inital values from the floating point output of the generator. A known
plaintext or chosen plaintext attack will produce such a sequence of these values immediately.
I have no idea if there are efficient algorithms to deduce the portion of the RNG in use just
from a succession of these 8-bit values. It most likely would be harder to break if one only had
a succession of 1-bit values output by the RNG, the leading bits of the floating point outputs.
(This step doesn’t help against the brute-force attack mentioned in the next paragraph below.)
Project 1.1 Modify the code above so that at each stage 8 calls to the RNG produce
8 most significant bits that are assembled into an 8-bit byte for use in xor as in the
original code.
The RNG Random is described as a “48-bit random number linear congruence generator.” In
= < =?D $(!BE %

this case, by investigating the details of the generator, one should be able to mount a brute-
force attack requiring
or about 300 trillion steps. One starts with a
known sequence of outputs, either the first 8 bits or the first bit in each case. At each step, the
generator needs to be exercised until that initial value is eliminated or until the search succeeds.
This calculation easily allows parallelization. If one could complete a step each giga-second, it
would still take on the average 150 000 seconds or just over 40 hours. In parallel 10 000 such
processors would take 15 seconds. So it all depends on how much hardware you have!
Project 2: Designing a Block Ciphcer 315
Systems with a modified generator: One could try modifying the generator. For example,
one could use two seeds and two separate copies of the Random RNG. The input seeds would
each be 64 bit long, but the RNGs are actually 48 bits, for 96 bits total in two generators. How
can one use two generators? Consider some possibilities.
✦ Alternate the use of the two generators, back and forth, or according to some scheme
publicly known. (Argue that this is essentially no help at all.)
✦ In order to get 8 bits for each byte of source, take 4 bits from one generator and 4
bits from the other. (Argue that this is essentially no help at all.)
✦ How about just averaging two floating point outputs from the two generators? (What
is the matter with this? Can it be made to work?)
Choose a value
for the address
"(
✦ Here is a scheme of Knuth’s to use two generators:
size of a buffer, say,
for a buffer of size . Fill the buffer with 8-bit
values from the first generator. Then get 10-bit values from the second generator,
using this number to fetch the 8-bit value from the first generator at that address in
the buffer, and then replacing the value in the buffer using the first generator again.
One could also use one of the two double-seeded RNGs described at the end of Section 16.2.
= =D((!B
!%
In each case these are two 32-bit seeds for a total of 64 bits. Thus the brute-force attack takes
or 20 billion billion steps. Finally one could use a more sophisticated
RNG such as the one based on chaos theory in Chapter 17, or the perfect RNGs of Chapter
(?19? in Section V). I have no proof of the difficulty of breaking any scheme of the type shown
here, but even the very first piece of Java code seems like it would be quite hard to break,
if it could be broken at all except by a brute-force search (an attack which always succeeds,
though perhaps taking an unacceptable amount of time). Each refinement may make the system
stronger; it’s hard to imagine any way to break a system based on the RNG in Chapter 17, since
no one has ever solved such 2-dimensional chaotic equations analytically.
Project 1.2 Modify the code above so it uses a better RNG with a longer total
number of seed bits, or uses two or more RNGs.
The key is somewhat awkwardly handled up to now. It would be better to have an arbitrary
character string as the input key. This string should be hashed into whatever actual inputs are
needed for the random number generator. The final project could look somewhat like the Unix
crypt utility.
Project 1.3 Look up the man page on crypt and modify the code above so it
handles an arbitrary character string as the input key. It should be indistinguishable
from crypt.
316 C. Projects
Project 2: Designing a Block Cipher.

This project asks you to work on a simple, symmetric, conventional, block-oriented cryptosys-
tem. I am giving you a skeleton Java program below that implements the Caesar cipher on each
byte, using a rotation from 0 to 255.
One could regard this project as a generalization of Project 1, but I mean for you to handle it as
a block cryptosystem.
✤ You are to produce a cipher that uses the same short secret key for encryption and
decryption. It would be fairly easy to modify the code below so that it implements
the Beale cipher, but you are not supposed to do that, since the Beale cipher uses a
key the same size as the plaintext.
✤ Your system should be a block cipher, so each ciphertext block should depend only
on the key and the plaintext block. Two separate but identical plaintext blocks should
encrypt to the same ciphertext block. If you wish to give your cipher some of the
characteristics of a stream cipher, with state and without the property of the previous
sentence, then you should just use the cipher in the cipher block chaining (CBC)
mode.
✤ The idea is to try to implement something that might not be easy to break. For the
strongest possible system, you could assume that opponents do not know your secret
key, but
✥ have access to your actual Java code, so that they know exactly what your
algorithms are and how they work, and
✥ have access to a “black” encryption/decryption box (without access to the
secret key), so that they can carry out a chosen plaintext attack, that is,
can choose arbitrary plaintexts and get the corresponding ciphertexts.
✤ The code below is strictly to get students started who might otherwise not know
where to start. You are also encouraged to write your own Java skeleton, perhaps
entirely different from the one below. In particular you do not have to get file names
as command line parameters, but you could even have the file names hard-wired into
your code.
✤ Specifically, the minimum you should do is to modify the methods encode and
decode of the class Code so that they will reverse one another, and so they will
make a cryptosystem harder to break than the Caesar cipher (hopefully much harder
to break). These methods appear in boldface in the listing below.
A few hints:
You are welcome to ignore these hints if you want.
✤ Remember that a system can’t possibly be strong unless there are at least or
more keys. You can use anything you like for the key, but it must not be arbitrarily
long (as it is with the Beale cipher). However, making the key long is something
you can also leave as a feature you might have included if you had had more time.
✤ One standard trick is to mess around with the key, and exclusive-or the result with
a plaintext block. Then to reverse this, mess around with the key the same way and
exclusive-or again. Notice that in this case you do not need to be able to reverse
what you do to the key.
✤ Block ciphers often use combinations of three basic elements, repeated in rounds
(to decrypt, the rounds are carried out in the opposite order):
✥ Use the trick above: xor with something to encrypt and xor with the same
thing during decryption.
✥ Use a permutation of elements of the plaintext or of the key. Here the
word permutation means a rearrangement of elements of the plaintext or
of the key. Of course one uses the reverse of the rearrangement to decrypt.
✥ Use a substitution of new items for elements of the plaintext or key. Here
something completely different is inserted in place of each element. This
operation also needs to be reversed for decryption.
✤ See another appendix at the end of this book for material about bit operations in
Java. In particular, the result of b1ˆb2 (both bytes) in Java is an int with the
proper bits in the 8 least significant bits. You then need to cast to a byte (using
(byte) to get a byte value for assignment or other uses.
✤ If your block size in only 8 (as in this case), this is also a weakness (why?). How
might you attack any block cipher with block size of 8? Obviously you could in-
crease the block size, but what else might you do to eliminate this weakness?
✤ You could increase the block size using the skeleton below by just handling more
than one byte at a time, say 4, 8, or 16 bytes at a time. In this case you may want to
pad the file at the end so that its length is a multiple of the block size.
Java Code for a Skeleton System:

The first class below (Crypto) just accesses command line arguments (which say whether to
encrypt or decrypt, give the key, give the name of the input file, and give the name of the output
file), reads the key, opens the files, creates an instance of the other class (Code) that does the
work, and invokes a method in that class (transform).
Java class: Crypto
// Crypto: simple encode and decode of a file, byte-at-a-time
import java.io.*;
public class Crypto {
static InputStream in; // input file args[2]
static OutputStream out; // output file args[3]
public static void openFiles(String infile, String outfile) {

try {
318 C. Projects
in = new FileInputStream(infile);
out = new FileOutputStream(outfile);
System.err.println("Error opening files");
System.exit(-1);
}
}

boolean encode = true; // encode or decode
int key = 0;
String usage =
"Usage: java Crypto (-encode | -decode) key infile outfile";
System.err.println(usage);
System.exit(-1);
}
try {
key = Integer.parseInt(args[1]);
} catch (NumberFormatException e) {
System.err.println("Error converting key \"" +
args[1] + "\" to int");
System.exit(-1);
}
if (args[0].equals("-encode")) encode = true;
else if (args[0].equals("-decode")) encode = false;
else {
System.err.println(usage);
System.exit(-1);
}
Code code = new Code(encode, key, in, out);
code.transform();
}
}
The second class below (Code) actually reads and writes the files, byte-at-a-time. Between
reading and writing a byte, it either encodes or decodes the byte, depending on the value of
a boolean switch encode. As mentioned before, your main work (in a simple version of the
assignment) could be to find more complicated functions to use for the methods encodeByte
and decodeByte, but remember that they must be inverses of one another.
Java class: Code
// Code: encode or decode a file
import java.io.*;
public class Code {
int key; // input key
boolean encode = false; // encode or decode
// code: constructor, pass parameters

public Code(boolean mode, int keyP,

InputStream inP, OutputStream outP) {
encode = mode;
key = keyP;
in = inP;
out = outP;
}
// transform: read bytes, encode or decode each byte, write

public void transform() {
try {
int inB; // input byte
int outB; // output byte
// read input file, byte-at-a-time
while ((inB = in.read()) != -1) { // till end-of-file
// make a simple change
if (encode) outB = encodeByte(key, inB);
else outB = decodeByte(key, inB);
writeByte(outB);
} // end of while
System.err.println("Error reading file");
System.exit(-1);
} // end try
}
// encodeByte: encode a byte

private int encodeByte(int key, int inB) {
return (inB + key)%256; // encode
}
// decodeByte: decode a byte

private int decodeByte(int key, int inB) {
return (inB - key)%256; // decode
}
// writeByte: then write byte

private void writeByte(int outB) {
try {
out.write(outB);
System.exit(-1);
}
}
}
Here are the results of a simple run on a Unix box, using the JDK directly. User input is in
boldface. Notice that after encoding and decoding, the original file is recovered.
% cat mess.text
Now is the time for all good men to come to the aid of their party
% java Crypto -encode 13 mess.text cipher2.text
320 C. Projects
% java Crypto -decode 13 cipher2.text mess2.text

% vi cipher2.text
[|x84-vx80-x81ur-x81vzr-s|ˆ?-nyy-t||q-zr-x81|-p|zr-x81|
-x81ur-nvq-|s-x8
1urvˆ?-nˆ?x81x86W (binary garbage)
% cat mess2.text
Now is the time for all good men to come to the aid of their party

The Laws of Cryptography With Java Code

Uploaded by

Copyright:

Available Formats

The Laws of Cryptography With Java Code

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Laws of Cryptography With Java Code

Uploaded by

Copyright:

Available Formats

The Laws of

II. Coding and Information Theory . . . . . . . . . . . . 22

III. Introduction to Cryptography . . . . . . . . . . . . . 60

IV. Public Key Cryptography . . . . . . . . . . . . . . . 78

V. Random Number Generation . . . . . . . . . . . . . 98

VI. The Advanced Encryption Standard (AES) . . . . . . . . 114

19. Introduction to the AES . . . . . . . . . . . . . . . . . . 115

VII. Hash Functions and Digital Signatures

Java Programs . . . . . . . . . . . . . . . . . . . . . . 159

IV. Public Key Cryptography

12. Public Key Distribution Systems

V. Random Number Generation

16. Traditional Random Number Generators

VI. The Advanced Encryption Standard (AES)

19. Introduction to the AES

VII. Hash Functions and Digital Signatures

II. Coding and Information Theory

3. Coding and Information Theory

III. Introduction to Cryptography

9. Cryptograms and Terminology

IV. Public Key Cryptography

12. Public Key Distribution Systems

V. Random Number Generation

16. Traditional Random Number Generators

VI. The Advanced Encryption Standard (AES)

19. Introduction to the AES

a. Generate AES Tables . . . . . . . . . . . . . . 127 277

VII. Hash Functions and Digital Signatures

A one-semester undergraduate course in cryptography might cover the following material:

San Antonio, Texas

✤ Confidentiality: keep the data secret.

Figure 1.1 Cryptosystem Using XOR.

Here are several other formulas involving logarithms:

  .  0/  1 for all 1 * 

 ;< 2 ;  , (mathematics)

8=?A    @@@E)>>A@D  $( .)

Table 1.2 Logarithms base 2.

A Java program that demonstrates these formulas is found on page 162.

takes   bits to represent 

integers mod n form a group for any integer

Table 1.3 Addition in the integers mod 10,

1.5 Fermat’s Theorem.

Table 1.4 Fermat’s Theorem for   .

  mod $A  %  mod A

 is the Euler phi function:

itself in case it is a prime.

Table 1.5 Euler’s theorem for  and

62   -  4   K4  . Table 1.5 illustrates

   , and

2.1 The Extended Euclidean Algorithm.

D @ (>  D @ A A B $ A

D @ (>    / AEB

public static long gcd2(long x, long y) {

for (int i = 0; i < 3; i++) {

A complete Java program using the above function is on page 166.

public static void check(long x, long y, long[] w) {

Here is the result of a run with the data shown above:

( @   "( $(  :

2.2 Fast Integer Exponentiation (Raise to a Power).

for (int i = k; i >= 0; i--) {

. 0/ 1 for all 1 *

;< 2 ; , (mathematics)

8=?A @@@E)>>A@D $( .)

takes bits to represent

Table 1.4 Fermat’s Theorem for .

mod $A % mod A

is the Euler phi function:

Table 1.5 Euler’s theorem for and

62 - 4 K4 . Table 1.5 illustrates

, and

D @ (> D @ A A B $ A

D @ (> / AEB

( @ "( $( :

) 62 /.= = ="/ ) 62 =

62 / 2

62 " 2 62 / = = =6/ 62 ) 2 62

where consists of two messages with probabilities and 04

/ 6 2)( 62)( / A2)( A2)(

<= DDB =$=$=

Repetition Codes, Error Probability is 8= E

2 6E2 8= $DDB B@<=?(DA

E E E E =$=$=

6E >)( , D , and (

<= 6E> E

= > BD = E

= B A E

= > BD

A ( E >

of each >bytes or >8= E

>2 6E = >>> >8= E

theory, the binary number system is regarded as the finite field

& 4 % / / =$=$=6/ 7 % mod

/ A /.==$=6/ @ / %'& mod

62 8= BE

D( =?A 6E<=B