Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!ames!ll-xn!mit-eddie!uw-beaver!cornell!rochester!PT.CS.CMU.EDU!NL.CS.CMU.EDU!mlm
From: mlm@NL.CS.CMU.EDU (Michael Mauldin)
Newsgroups: sci.crypt
Subject: Re: turning a plaintext source into a good "random" key
Message-ID: <1009@PT.CS.CMU.EDU>
Date: 1 Mar 88 23:41:02 GMT
References: <8803011848.AA05787@decwrl.dec.com>
Sender: netnews@PT.CS.CMU.EDU
Organization: Carnegie-Mellon University, CS/RI
Lines: 70

In article <8803011848.AA05787@decwrl.dec.com>, kruger@16bits.dec.com writes:
> The trick is to correctly extract the random element, while somehow
> 'chopping up' the regularities.

For something more demonstrably "random" why not use your favorite
redundancy reducer (read: file compressor) to destroy the pattern?
Huffman code based on bigrams, or use Lempel-Ziv compress.  The better
the compress routine, the less redundancy there will be in key stream.
You may still have to deal with some regularities in the resulting bits.

Your suggestion of ignoring vowels and using the low order bits doesn't
seem to work well at all (did you test it?).  Here are histograms for the
text of the US Constitution (50,090 characters):

Character counts (ignoring punctuation and digits < 100 occurences):

SP   10397   ****************************************************************
,      642   ***
.      377   **
a     2779   *****************
b      659   ****
c     1276   *******
d     1307   ********
e     5385   *********************************
f     1077   ******
g      454   **
h     2041   ************
i     2607   ****************
j      100   
k       46   
l     1496   *********
m      792   ****
n     2708   ****************
o     2856   *****************
p      834   *****
q       52   
r     2325   **************
s     2831   *****************
t     3913   ************************
u      883   *****
v      481   **
w      385   **
x      129   
y      553   ***
z       31   


Low 4 bits of all letters except vowels:

0000 11248   ****************************************************************
0001  2943   ****************
0010  3037   *****************
0011  4149   ***********************
0100  5241   *****************************
0101  6287   ***********************************
0110  1584   *********
0111   862   ****
1000  2198   ************
1001  3200   ******************
1010   160   
1011   128   
1100  2166   ************
1101   918   *****
1110  3113   *****************
1111  2856   ****************

So you see there is still quite a lot of structure left.

Michael L. Mauldin (Fuzzy)		Department of Computer Science
ARPA: Michael.Mauldin@NL.CS.CMU.EDU	Carnegie Mellon University
Phone: (412) 268-3065			Pittsburgh, PA  15213-3890