[Spooks] going form numbers to characters ?

[JMM Moi-Meme Maitre du Monde]

On Mon, Feb 21, 2005 at 11:50:25AM -0800, Nicholas Gessler wrote:
> Visit http://mailman.qth.net/mailman/listinfo/spooks to unsubscribe from 
> this list
> 
---8<--- SNIP ---
> 
> Irrespective of whether numbers stations use code books, one-time-keys or 
> quasi-random-generated-keys, I'm wondering what the algorithm is for 
> translating the numbers (0-9) into characters (a-z).  The most obvious 
> solutions that come to mind are:
> 
> 1)  Two digits select the entry coordinates to a 10x10 table of common 
> letters and letter digraphs and trigraphs (e.g. a, b, th, ion...)
> 2)  Two digits (modulus 25) select the alphabetic character 0-24 in a 
> shortened alphabet (a-y)
> 3)  Two digits select an ASCII value from a shortened set of 100 characters 
> (32-127 plus 5 control characters).
> 

There is a 1997 documentary ((c) FILMROOS Inc.) produced for the Discovery
Channel and entitled 'Top Secret NSA' which explained one of the way to
do it.

It actually refered to 'Venona' Soviet cipher. The explanation given
by William Cromwell, NSA Deputy Director, is that you, as a code clerk,

1 - Use a code book (eg. 'atomic' = 3856 ) to substitute common words
    or expression in a number

2 - Then, on top of that, add the the number given by the OTP (e.g 1349)
    using the so-called 'Chinese arithmetic method' (i.e. each digit
    is added to its counterpart, and the 10 modulo of the sum is written 
    down) 3856 + 1349 = 5205, but in fact you will write down 4195

This part is approximately 25 minutes from the beginning.

Of course, as it is a general public documentary, it does not go deep into
the details (how do you spell a name or a location that is not in the code book)

> If digits are grouped by two, then why send them in groups of 
> five?  Custom?  Clarity?  It seems to me that numbers are clearer than 

Groups of five is due to historical reasons. When the first numbers messages
were transmitted by wire (diplomatic traffic for instance), telegraph companies 
had to find a way to calculate how much they will charge as they used to do it
on a per word basis. There came a general consensus that a group of five
will be counted as a 'word'.

> letters and female voices more clear than male voices.  Would you 
> agree?  Are there any studies on this?  That is to say, are there any 
> publications investigating how much noise it takes to render a female 

hmm ! that is a matter personal appreciation.

Most of Anglo-Saxon numberists who listened to E17 found it sounded female
(some of them called it 'hermaphrodite', though). Personnaly, it is a 'young
lady' I would not have kissed, being afraid of a possible beard.

(http://www.cvni.net/mp3/2001-01-17-0215z-4783U-E17.mp3)

(Same for G04 voice, but for other reasons. The way the voice used to sound
made think of an utterly dominative female with very bad temper *lol*)

(http://www.cvni.net/mp3/2000-08-10-2035-5320U-G04.mp3)

The audio spectrum in SW is limited. Depending on the language used and 
the speaker native language and/or origins (foreign or regional accent)
the range of sound frequencies varies.

For instance, Portuguese Portuguese is not the same as Brazilian Portuguese.
Nevertheless, it is a language with a lot of 'jee' and 'sh' sounds which can
easily rise to 7000 kHz, while French very scarcely use sounds above 800 Hz. 

There are also some radio habits, like the use in English of 'niner' instead
of 'nine' to avoid confusion with 'one'.

> versus a male voice intelligible?  Or studies on the relative 
> intelligibility of large symbol sets (the alphabet) versus mall symbol sets 

I do not know of any station using raw alphabet. They do use some alphabetics
('Adam', 'Baker', 'Charlie',... or 'Alpha', 'Bravo', 'Charlie' ...)

> (the decimal digits)?  Clearly, binary digits are the most robust, so maybe 
> a study would be redundant.

Somehow, 'trinary' digits are used. You can consider Morse code as some
kind of 0 (no signal/blank), 1 (dot), and 2 (dash) set of values.


> 
> I'm teaching a course in computer simulation and I and a few students are 
> interested in building some small computer applications demonstrating how 
> it "might" work.

(The use of common digraphs and trigraphs you mentionned above makes your
cipher cleartext language dependent.)

> 
> (I put a simple OTP simulation on the Web.  I probably should have coded 
> "a" as "0" instead of "1," but it works.  It will run on any PC.)
> http://www.sscnet.ucla.edu/geog/gessler/borland/crypto-alpha.exe
> 
> Cheers,
> Nick