CHAPTER 10

APERIODIC POLYALPHABETIC CIPHERS

Simple Manual Aperiodic Systems
10-1.
Chapter 9 showed that periodic polyalphabetic systems are generally more secure than
monoalphabetic systems. However, the regular, repeating nature of the keys in
periodic systems are a weakness that an analyst can exploit. Using factor analysis or
the phi test, the analyst can readily determine how many alphabets there are and
which letters are enciphered by which alphabets. Aperiodic polyalphabetic systems
eliminate the regular, repeating use of alphabets so the analyst cannot easily tell which
letters are enciphered by which alphabets. There area number of ways to use a limited
set of alphabets but suppress their regular repetition. The following subparagraphs
show the most common types of these, and briefly discuss their weaknesses and
approaches to their solution. They are presented to make you aware of the possibility
that such techniques can be used, but no detailed explanation of their solution is
given.
a. Word Length Aperiodic. The simplest type of aperiodic changes alphabets with
each word instead of each letter. The analyst cannot tell which letters are encrypted
by which alphabet until the text is recovered. However, the major weakness of this
system is that when repeats occur, they are likely to be word length, and plaintext
word patterns show through as clearly as with monoalphabetics. When alphabets
are known, the generatrix method makes the plaintext obvious.

b. Numerically Keyed Aperiodic. Another approach, similar to word-length
encipherment, is to change alphabets after a number of letters, determined by a
numerical key. The numerical key is often based on the repeating key. The key is
generated by the same process used with a numerically keyed transposition

10-0
sequence. The letters in the repeating keyword are numbered alphabetically. Then
the key determines how many letters are enciphered consecutively by each
alphabet. For example, here is a short message enciphered by a numerically keyed
aperiodic based on the keyword BLACK.

This system, while more complicated than a word-length aperiodic, allows many
repeats and patterns to appear. When the alphabets are known, use of the
generatrix method also quickly reveals the plaintext.
c. Interruptor Letter Aperiodic. Another approach to breaking up the cyclic nature
of periodic systems is through the use of an interruptor letter. In interruptor letter
systems, the alphabets are used in rotation like a periodic system, but whenever a
preselected plaintext (or alternatively, ciphertext) letter is encountered, the rota-
tion is interrupted and encipherment returns to the first alphabet. This is a more
secure method than the previous two, but it can have the effect of creating repeats
that would not otherwise occur. For example, if a plaintext R is used as an interrup-
tor letter, every time REINFORCEMENTS appears in the text, encipherment from
the second letter on will be identical every time. The letter after the initial R will be
enciphered by the first alphabet each time because of the interruption. The same
thing will happen with any word that begins with the interruptor letter. Use of a
ciphertext interruptor letter instead of a plaintext letter will avoid many of these
repeats, but the interruptions will generally occur much less often in such a case.

Long-Running Key Aperiodic
10-2.
Much more common than the simple manual aperiodic systems described in the
previous paragraph are those that use a long-running, ever changing key. These
systems may be enciphered manually, by cipher machine, or by computer, as first dis-
cussed in paragraph 8-1. Figure 8-1 gave an example of using a book key where the key

10-1
letters were a quotation. A quotation, particularly from a book, provides a ready source
of long-running keys, but it is relatively unsecure, because the key itself is so orderly.
More often, the keys will be random or pseudorandom. The keys are applied to the
plaintext using an alphabet chart like the Vigenere square in Figure 8-1. The keys may
be generated by a pseudorandom, repeatable process or by a random, nonrepeatable
process. Both the sending and receiving cryptographer must have a copy of the same
book or pad of keys. When these are intended for single usage of the keys, the system is
called a one-time pad system. Truly random one-time pad systems are absolutely
unbreakable when used properly. When keys are reused, however, whether by mistake
or by design, the messages with the reused keys are likely to be recoverable. Manual
one-time pad systems are slow systems to use and present logistics problems for any
large scale usage. The volume of keys must be at least equal to the volume of messages
to be sent, When more than one communications link shares the use of copies of the
same pad, careful procedures must be set up to prevent reuse of the same keys by dif-
ferent users.

Solution of Long-Running Key Aperiodic
10-3.
The solution of messages enciphered in long-running key systems may be possible in
three situations. First, the key generation process may be known in advance from prior
recoveries or other sources. Second, the keys may be so orderly that they are
recognizable when partially recovered, as can occur when plaintext is used as the
source of keys. Third, the same sequence of keys is reused. We are primarily concerned
with the third case, where keys are reused.

a. Depth Recognition. A reuse of long-running keys is called a depth. Messages using
the same keys are called messages in depth. If the keys begin at the same point in
two or more messages, the messages are in flush depth. If the keys begin at different
points in two or more messages, but include reused keys for at least part of the
messages, they are in offset depth. The solution of messages in depth first requires
you to recognize that the depth exists.

(1) One way to recognize depth is through exploitation of indicator systems. In one-
time pad systems and in many types of cipher machine or computer systems,
the starting point or settings for the keys must be known by the enciphering and
deciphering cryptographers. This information on the keys is often passed from
cryptographer to cryptographer through the use of an indicator system. The
first way to recognize a depth is to find two messages or transmissions with
identical indicators. Identical indicators will often tip-off that a flush depth is
occurring.

(2) The second way to recognize depth is to find repeated text between two or more
messages. Except for short accidental repeats, repeated ciphertext will only
occur when the same plaintext is enciphered with the same keys. In periodic

10-2
systems and simple manual aperiodic, this will often occur within a single
message as the same keys are reused. With long-running key aperiodic, this
will only occur between messages when keys are reused. If all depths are expec-
ted to be flush depths, the search for repeats is a matter of superimposing
messages and looking for repeats in the same position in each message. If depths
are offset, they are more difficult to find by inspection alone.

(3) The third way to recognize depth is to use a type of coincidence test known as
the kappa test. Whether whole words and phrases are repeated using the same
keys or not, individual characters using the same keys will occur frequently
when depths are present. When two messages are matched together, letter by
letter, and do not use the same keys, 1 out of 26 letters (or 3.85 percent) will ran-
domly match. Of course, if a different alphabet is used, or if characters other
than letters are also used, the expected number of matches by chance alone will
be 1 out of the total number of different characters used. On the other hand, if
the messages are correctly placed in depth, a letter by letter comparison (the
kappa test) will produce matches about 6.67 percent of the time. Also, the
results can be expressed as a kappa index of coincidence showing the ratio of
observed coincidences to random expectation. As with searching for repeats, it
is much easier to find flush depths than it is to find offset depths, but with com-
puter support, messages can be matched in every possible alignment to search
for depths.

(4) As an example of depth recognition, consider the three messages that follow.
Each has similar indicator groups that suggest the messages may be in depth
with each other. Messages 1 and 2 have identical indicators. Message 3 differs
only in the last digit of the second group.

10-3
(5) There are no repeats longer than three letters between any of the three
messages. Because of the identical indicators, we first try to match messages 1
and 2 at a flush depth using the kappa test. The number of matches multiplied
by 26 and divided by the number of comparisons equals the kappa IC. Do not
count the indicator groups in the comparisons.

(6) As shown by the kappa test, the number of matches is well above random expec-
tation. The two messages appear to be in flush depth with each other. Next we
try message 3 matched with the first two at a flush depth.

10-4
(7) The flush match of message 3 is clearly not a correct match, because of the low
kappa index of coincidence. We next try offsets of 1, 2, 3, 4, and 6 letters to the
right.

10-5
(8) The offset of five is clearly the best match of those tried, and the kappa index of
coincidence is a good value for a correct match. The three messages are now
correctly placed in depth.
b. Depth Reading. When the messages are superimposed properly, they can be solved
by a process known as depth reading. With only a few messages, the process of
applying the key must be known. With manual systems, standard alphabets are
commonly used. With cipher machine or computer based systems, the process of
baud addition is usually known or can be figured out easily. The three messages in
our example use the standard alphabet Vigenere square of Figure 10-1.

10-6
10-7
(1) With three messages in depth, almost any correct assumption of plaintext will
lead to a quick solution. For example, trying the word REPLACEMENT as the
first word of message 3 produces the results shown below.

(2) Recovering the key from the assumption of REPLACEMENT and using it to
decipher the other two messages produces good segments of plaintext in each
message. It is easy to build on these assumptions to recover additional plain-
text. For example, assuming that the second message begins PROTECTIVE
GEAR and that the word after TEAM in the first message is ARRIVING leads

10-8
(3) This process of assuming text can be continued to a complete solution. Correct
assumptions are easily verified. Incorrect assumptions are quickly disproved.

(4) The most difficult step is making the first correct assumption. Message begin-
nings are the most likely area to yield results, because they are likely to be very
stereotyped. Sometimes, just trying the letters RE at the beginning of a message
will be enough to suggest the text of the messages in depth with it. When
message beginnings do not yield results, more powerful techniques are
available.

c. Crib Dragging. When you cannot assume the beginning of a message, you can still
often correctly assume a particular word that will be in a message. The assumptions
can come from familiarity with previous messages, results of traffic analysis and
direction finding, or other intelligence sources. Once you suspect a word is in one of
two or more messages in depth, you can systematically try the word at every posi-
tion, recover the keys each position would produce, and try the keys in the other
message or messages to see if the keys produce more plaintext. This is a laborious
process performed manually, but a sure one. Fortunately, there are some short cuts
that can be used to simplify the process.
(1) Two messages in depth can generally be combined in such a way that you can
skip the step of key recovery and proceed directly to checking for plaintext.
With the Vigenere square of Figure 10-1, this can be accomplished by treating
one message as if it were plaintext, the other as ciphertext, and producing the
resulting key stream, which is actually a combination of the two ciphertexts. To
demonstrate this process, consider the beginnings of messages 1 and 2 from the
previous example. If we combine message 1 and message 2 as if they were plain-
text and ciphertext respectively, it produces a combination text for the first
groups of YNWPE, Message 1 letters are used as keys in the Vigenere square.
Message 2 letters represent the internals of the Vigenere square. For example,
key H matched against internal F produces plaintext Y.

(2) If we now apply the correct plaintext of message 1 to the combination text using
the Vigenere square, it will directly produce the plaintext of message 2. The

10-9
combination text is again found in the key letter position in the square, and the
plaintext is found in the same position for each message as the original cipher-
texts.

(3) The combination text can be systematically used to try out a plaintext assump-
tion in every position by a process known as crib dragging. Crib is a common
synonym for assumption in cryptanalysts. Consider the following two messages
in depth. The first message was sent by a unit undergoing an artillery barrage. It
is likely that the word ARTILLERY will be found in the message.

(4) The first step to trying out ARTILLERY in message 1 is to create the combina-
tion text. Message 1 is treated as plaintext and message 2 as ciphertext.

(5) The results of trying ARTILLERY in each of the first three positions are shown
below.

10-10
(6) Obviously, not one of the first three tries is the correct placement of
ARTILLERY. The process can be speeded up, however, by plotting the crib ver-
tically and the resulting text for message 2 on a descending diagonal.

(7) The plot above is identical in results to the three separate plots that preceded.
Once this format is adopted, it is easier to write in a whole row at a time.

(8) The plaintext for message 2 appears on the sixth diagonal, as highlighted above.
Once the text is spotted and the crib confirmed, it becomes a matter of depth
reading, as before. The worksheet can now be set up and the rest of the text
recovered.

10-11
(9) With cipher machine and computer based systems that use baud addition,
adding two messages in depth together by baud addition eliminates the key.
The baud addition of the two ciphertexts is identical to the baud addition of the
two original plaintexts.

(10) Whatever type of alphabet square or system of combining bauds is used, there
is usually a way to combine texts in depth to eliminate the effects of the key. If
you are unsure how to approach a particular type of system, test samples you
create for yourself in the system to see how ciphertext can be combined to
eliminate the effect of the key.

10-12