5.0 KiB
TWENTY4/160 - a fun stream cipher
THIS IS JUST FOR LEARINING CRYPTO, DO NOT EVER USE THIS FOR ANYTHING
This is the implementation of the fun stream cipher TWENTY4/160 by T.v. Dein, 09/2015. Published under the public domain, Creative Commons Zero License. It works bytewise, uses a 160 bit key bits in 8 rounds, applies an S-Box. From the key various PRNGs are seeded, all those PRNGs are recombined into an output key stream, which is being xored with the input (after applying of the sbox).
The name TWENTY4 is a reference to article 20 paragraph 4 of the german constitution which reads:
All Germans shall have the right to resist any person seeking to abolish this constitutional order, if no other remedy is available.
Also, the cipher uses the contents of the german constitution as the source for its S-Box.
S-Box generation
TWENTY4 uses the german constitution (called "basic law" in germany) as the source for S-Boxes. The EPUB version (included in sbox/ subdir) is encrypted using AES-256-CBC with the passphrase "grundgesetz\n". The resulting byte stream is used as the source for S-Boxes.
The following version of the german constitution is used:
Basic Law for the Federal Republic of Germany in the revised version published in the Federal Law Gazette Part III, classification number 100-1, as last amended by the Act of 23 December 2014 (Federal Law Gazette I p. 2438).
Linux Shell commands to generate the S-Boxes:
curl -o BJNR000010949.epub http://www.gesetze-im-internet.de/gg/BJNR000010949.epub
echo grundgesetz > BJNR000010949.pass
cat BJNR000010949.epub | openssl aes-256-cbc -kfile BJNR000010949.pass | ./gen-static-sbox
'gen-static-sbox' compiled from gen-static-sbox.c in this directory, which has SHA256 checksum: 29bfd8bd6dbca696d4d8b7ca997497e091875d6bf939e9702b1edf669d0742b0.
However, it just prints out bytes which it reads from STDIN, collecting them into an 256 byte array, ignoring possible duplicates, and prints it out as hex.
Both S-Boxes are bijective and have the following properties (calculated using analyze.c):
Char distribution: 100.000000%
Char redundancy: 0.000000%
Char entropy: 8.000000 bits/char
Compression rate: 0.000000%
TWENTY4 uses two S-Box arrays, one for key expansion and one for encryption.
Key expansion
FIXME.
Encryption
FIXME.
Analysis so far
While this stuff only exists for me to play around with crypto, I tried to test the cipher a little bit. Here are my results using a couple of statistical measurements:
I encrypted the GPLv3 contents using the key "1". To compare, I encrypted the same file with AES-256-CBC using the same passphrase.
Check using analyze.c
My own measurement, see analyze.c:
File size: 35147 bytes
Char distribution: 100.000000% Char redundancy: 0.000000% Char entropy: 7.994904 bits/char Compression rate: 0.000000% (35147 => 35168 bytes)
For comparision, AES result:
File size: 35168 bytes
Char distribution: 100.000000%
Char redundancy: 0.000000%
Char entropy: 7.994892 bits/char
Compression rate: 0.000000% (35168 => 35189 bytes)
Check using ent
(ent from http://www.fourmilab.ch/random/):
Entropy = 7.995333 bits per byte.
Optimum compression would reduce the size
of this 35147 byte file by 0 percent.
Chi square distribution for 35147 samples is 229.98, and randomly
would exceed this value 86.79 percent of the times.
Arithmetic mean value of data bytes is 127.6631 (127.5 = random).
Monte Carlo value for Pi is 3.172955438 (error 1.00 percent).
Serial correlation coefficient is -0.004405 (totally uncorrelated = 0.0).
Entropy = 7.994904 bits per byte.
Optimum compression would reduce the size
of this 35147 byte file by 0 percent.
Chi square distribution for 35147 samples is 248.29, and randomly
would exceed this value 60.64 percent of the times.
Arithmetic mean value of data bytes is 127.9724 (127.5 = random).
Monte Carlo value for Pi is 3.101929315 (error 1.26 percent).
Serial correlation coefficient is -0.000624 (totally uncorrelated = 0.0).
For comparision, AES result:
Entropy = 7.994892 bits per byte.
Optimum compression would reduce the size
of this 35168 byte file by 0 percent.
Chi square distribution for 35168 samples is 250.98, and randomly
would exceed this value 55.94 percent of the times.
Arithmetic mean value of data bytes is 127.8717 (127.5 = random).
Monte Carlo value for Pi is 3.151680601 (error 0.32 percent).
Serial correlation coefficient is 0.002014 (totally uncorrelated = 0.0).
Check using dieharder
I ran the cipher against the dieharder test suite this way:
dd if=/dev/zero of=/dev/stdout | ./twenty4 00000000000000000001 | dieharder -g 200 -a
Find the results in analyze/dieharder160.log
So, all those checks don't look that bad, but of course this doesn't say much about TWENTY4/160's security. However, not THAT bad for the first cipher :)