Scribd
Upload
2K views

Evolutionary Algorithm For Decryption of Monoalphabetic Homophonic Substitution Ciphers Encoded As Constraint Satisfaction Problems

A homophonic substitution cipher maps each plaintext letter of a message to one or more ciphertext symbols [4]. Monoalphabetic homophonic ciphers do not allow ciphertext symbols to map to more than one plaintext letter. Homophonic ciphers conceal language statistics in the enciphered messages, making statistical-based attacks more difficult. We present a dictionary-based attack using a genetic algorithm that encodes solutions as plaintext word placements subjected to constraints imposed by the cipher symbols. We test the technique using a famous cipher (with a known solution) created by the Zodiac serial killer. We present several successful decryption attempts using dictionary sizes of up to 1,600 words.

Uploaded by

doranchak
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
2K views

Evolutionary Algorithm For Decryption of Monoalphabetic Homophonic Substitution Ciphers Encoded As Constraint Satisfaction Problems

A homophonic substitution cipher maps each plaintext letter of a message to one or more ciphertext symbols [4]. Monoalphabetic homophonic ciphers do not allow ciphertext symbols to map to more than one plaintext letter. Homophonic ciphers conceal language statistics in the enciphered messages, making statistical-based attacks more difficult. We present a dictionary-based attack using a genetic algorithm that encodes solutions as plaintext word placements subjected to constraints imposed by the cipher symbols. We test the technique using a famous cipher (with a known solution) created by the Zodiac serial killer. We present several successful decryption attempts using dictionary sizes of up to 1,600 words.

Uploaded by

doranchak
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Evolutionary Algorithm for Decryption of Monoalphabetic

Homophonic Substitution Ciphers Encoded as Constraint


Satisfaction Problems

David Oranchak
NTU School of Engineering and Applied Science
Roanoke, VA 24018
[email protected]

ABSTRACT
A homophonic substitution cipher maps each plaintext letter
of a message to one or more ciphertext symbols [4]. Monoal-
phabetic homophonic ciphers do not allow ciphertext sym-
bols to map to more than one plaintext letter. Homophonic
ciphers conceal language statistics in the enciphered mes-
sages, making statistical-based attacks more difficult. We
present a dictionary-based attack using a genetic algorithm
that encodes solutions as plaintext word placements sub-
jected to constraints imposed by the cipher symbols. We
test the technique using a famous cipher (with a known so-
lution) created by the Zodiac serial killer. We present several
successful decryption attempts using dictionary sizes of up
to 1,600 words.
Program Track: Real-World Applications
Categories and Subject Descriptors: E.3 Data Encryp- Figure 1: Left: Solved 408-character homophonic
tion: Code breaking substitution cipher sent by the Zodiac serial killer
General Terms: Algorithms, Experimentation to three San Francisco newspapers. Right: Unsolved
Keywords: Evolutionary computing, genetic algorithms, 340-character cipher sent by the killer.
Zodiac killer, Zodiac murder ciphers, codebreaking, cryp-
tography, constraint satisfaction, homophonic substitution
something gives me the most thrilling experence. It is even
better than getting your rocks off with a girl. The best part is
1. INTRODUCTION thae when I die I will be reborn in paradice and all the I have
Simple substitution ciphers encrypt plaintext messages killed will become my slaves. I will not give you my name be-
using symbols which map to individual plaintext letters. cause you will try to sloi down or stop my collecting of slaves
Monoalphabetic ciphers use the same mappings from plain- for my afterlife ebeorietemethhpiti.] The killer mailed a sec-
text to ciphertext throughout the encrypted message. Monoal- ond cipher to a San Francisco newspaper (Figure 1[2]). No
phabetic substitution ciphers are often easy to decipher with satisfactory solution to this cipher has yet been found. We
frequency analysis because the simple mappings preserve let- use the 408-symbol cipher as a test case for our technique,
ter frequencies of the plaintext message. Homophonic ci- which we hope can be used to attack the 340-symbol cipher.
phers hide letter frequencies of plaintext messages. Each Many effective decryption techniques for simple ciphers
letter of enciphered plaintext is mapped to one or more ci- have been studied, such as statistical analysis [3][1], evolu-
phertext units, called homophones, which flattens the distri- tionary computing[8][6], and dictionary-based attacks [5][7].
bution of ciphertext symbols. The Zodiac killer is a famous Our experiments combine the strengths of evolutionary search
serial killer who operated in California in the late 1960s [2]. and constraint-imposed dictionary-based attacks.
In 1969, the killer sent three letters to area newspapers. In
each letter, the killer took credit for recent shootings, and
included a part of the 408-symbol three-part cipher (Fig- 2. APPROACH
ure 1[2]). A high school teacher and his wife soon decoded The 408-character cipher has a keyspace size of 2654 . To
the cipher by hand.[2]: [I like killing people because it is so reduce the space, we attacked a 52-character region of ci-
much fun. It is more fun than killing wild game in the for- phertext that decodes over 90% of the entire message. The
rest because man is the most dangeroue animal of all to kill targeted section decodes to the following: killing wild game
in the forrest because man is the most danger. This
section is only 12.7% of the cipher, but decodes 369 char-
Copyright is held by the author/owner(s).
GECCO’08, July 12–16, 2008, Atlanta, Georgia, USA. acters (90.4%) of the plaintext. We also limit word place-
ACM 978-1-60558-131-6/08/07. ments to sets of unique words having a minimum length
Table 1: Results of experimental runs for different
word pool sizes. Fe is the evolved solution’s multi-
objective fitness, and Fs is the multiobjective fitness
of the known correct solution.
#Words Correctness Fe Fs Generations
500 1.0 58,880 58,880 874
859 1.0 82,1074 82,1074 668
1000 1.0 85,1099 85,1099 807
1395 1.0 104,1270 104,1270 1750
1600 0.9 110,1202 120,984 3222

of four. Our attack uses a variation of the techniques de-


scribed by Olson[7] and Lucks[5]. We encode an attack as
a non-conflicting set T of word and position selection tu-
ples [(wi , pi ); 0 <= wi < D, 0 <= pi < 52; D is dictionary
size]. The GA performs generative placement of words into Figure 2: Performance plot for each dictionary size.
the 52-character cipher region. Words are selected from pre-
made dictionaries composed of common words in the Zodiac
corpus. The GA uses tournament selection of size two but the ciphertext. By concentrating the search on a small 52-
sometimes selects individuals with lesser fitness. Crossover character section of the Zodiac killer’s 408-character cipher,
randomly merges feasible tuples from both parents. Sim- we reduced the search space and evolved correct decodings
ple mutation is applied to each offspring. All operators are of the 52-character section that aid in the decryption of the
restricted from producing infeasible tuple sets. Diversity is entire ciphertext. Removal of approximate minimum ver-
preserved using fitness sharing and elitism via Pareto sam- tex cover helps prevent exploration of word-dense plaintext
pling. decodings that in fact contain many conflicts.
The first fitness measure is computed by counting par-
tial or complete dictionary words that form beyond the 52-
character region. The GA computes the second fitness mea-
5. ACKNOWLEDGEMENTS
sure by forming a graph G of the found words. Each word We thank the following people for their contributions and
is a node, and any conflict between two words is an edge. assistance: Chris McCubbin, Brax Sisco, and Edwin Olson.
We want to remove a minimal subset of nodes in G so no
remaining edges (conflicts) remain, which is a minimum ver- 6. REFERENCES
tex cover problem. We obtain vertex cover within a factor [1] J. M. Carrol and S. Martin. The automated
of 2 of optimal by removing the maximal matching set of cryptanalysis of substitution ciphers. Cryptologia,
edges from the graph G to form G′ . From G′ the GA com- X(4):193–209, 1986.
putes the second objective, Equation 1. Words from G′ are [2] R. Graysmith. Zodiac. St. Martin’s, New York NY,
counted for the entire cipher C except for the 52-character 1986.
section (between k0 and k1 ). For each position, the score pi
[3] T. Jakobsen. A fast method for cryptanalysis of
is l (word length) if at least one word is found that covers
substitution ciphers. Cryptologia, 19(3):265–274, 1995.
position i, or 0 otherwise.
[4] J. C. King and D. R. Bahler. A framework for the
X study of homophonic ciphers in classical encryption and
f itness1 = max l × pi (1) genetic systems. Cryptologia, XVII(1):45–54, 1993.
l∈{4,5,6,7,8,9,10}
i∈C,i<k0 or i>k1 [5] M. Lucks. A constraint satisfaction algorithm for the
automated decryption of simple substitution ciphers. In
3. RESULTS CRYPTO, pages 132–144, 1988.
We ran experiments using the following parameters: pop- [6] R. A. J. Matthews. The use of genetic algorithms in
ulation size 10, 000; mutation probability: 0.1; genome size: cryptanalysis. 17(2):187–201, Apr. 1993. cryptanalysis;
25 tuples; dictionary sizes: {500, 859, 1000, 1395, 1600}. genetic algorithms; cryptographic systems; keyspaces;
Performance measure is determined by comparing the plain- GENALYST.
text of the best evolved solutions to the plaintext of the [7] E. Olson. Robust dictionary attack of short simple
known solution (using the given dictionary): killing wild substitution ciphers. 2007.
game inthe forrest because ??? isthe most danger [8] R. Spillman, M. Janssen, B. Nelson, and M. Kepner.
(”man” is missing because it is too short.) Results are shown Use of a genetic algorithm in the cryptanalysis of
in Table 3. Figure 2 plots performance versus number of gen- simple substitution ciphers. 17(1):31–44, Jan. 1993.
erations for each dictionary size. Each experiment found a
correct or very close to correct solution.

4. CONCLUSIONS
We encoded a homophonic substitution cipher attack as
an evolutionary search of a combinatorial space of dictio-
nary word placements subjected to constraints imposed by

You might also like