This machine is described in an article in a recent issue of Cryptologia magazine, which also provides accompanying information of historical and human interest.
It is a telecipher device built around a five-rotor machine.
The original version supplied thirteen live inputs to the input rotor. Five outputs from the other end were XORed to the five bits of a teletypewriter character. The rotors advanced conventionally for each character enciphered, and had only one carry position. Like the SIGABA, the letter O, and not A, indicated the zero position of the rotors. The five rotors included one that moved for every letter enciphered, and one that moved only when that first rotor completed a full rotation and so on, like an odometer, but with one difference: which rotor moved with which frequency could be chosen arbitrarily. The original Hebern machine had three classes of rotor movement, and it too allowed the selection of how the individual rotors moved, but in a less straightforwards manner.
Here is a diagram of the way the original SIGCUM was wired:
Since 13 inputs out of 26 were live, each bit of the five outputs had a 50-50 chance of being live, but not all 32 combinations of all five outputs were equally likely: five ones or five zeroes were the least likely, and 2 or 3 of each was the most likely.
This simple design was found to be insecure, after being put into use, for a particular case of operator error. A considerably revised version served for some time, and its description is as follows:
In the revised version, there were only five live inputs to the input rotor. Fifteen outputs from the other end were wired together in five groups of three to provide the five output bits. This gave each bit a probability of about 48.846%, or exactly 381/780, of being a one.
In the original design, all five rotors moved in the conventional manner of wheels on an odometer, except that the position of the rotor that moved with every character enciphered, and the position of each rotor that moved when another rotor completed a full circle, could be set arbitrarily. In the revised design, three rotors moved in that way, and two other rotors had movements called fast bump and slow bump that depended on the bits the rotors generated.
If bit 1, bit 3, and bit 5 of the generated bits (to be XORed with the plaintext to create ciphertext) were all ones, the slow bump rotor would advance for the next character. If bit 2 and bit 4 were both 1, the fast bump rotor would advance. In addition, the slow bump rotor skipped an extra position when it reached O, as well as stepping the slow rotor (if the slow rotor was also at O; this may have been a misprint for 'slow bump', since that won't happen very often).
Here is a diagram of the modified version of the SIGCUM:
The bump rotors meant that extra parts did have to be added to the machine to effect the conversion; it was not a simple case of rewiring. Even the early version of the machine appeared to be highly secure at first glance, which makes Frank Rowlett's feat of establishing its weaknesses impressive.
An issue of Cryptologia some years previous to the issue in which SIGCUM or the M-228 was described, quoted a government memo which referred to the M-294. This machine was very similar to the SIGABA, except that the five "cipher rotors" supplied a bitstream to encipher a teletypewriter message; presumably using an arrangement similar to that seen here. Such a device would certainly provide a method of encipherment for teletypewriter messages of impressive security. SIGNIN was an alternative designation for the M-294. The M-138 and M-138-A, indirect predecessors of the SIGABA, were rotor machines, of the Hebern type, acting upon the 26 letters of the alphabet, but with the motions of their five rotors controlled by a paper tape; in a way, these machines were the exact inverse or complement of SIGCUM.
Frank B. Rowlett passed away on June 29, 1998, at the age of 90.
Table of Contents