Chapter 4

Random Numbers

The Security of many cryptographic systems depends upon the generation of unpre-
dictable quantities. As we have seen earlier, the one time pad requires a random key
in order to operate effectively. As well as that, the DES key needs to be random and
unpredictable. It must be remembered that a cryptographic algorithm’s security rests
with its key (this is known as Kerckhoffs’ principle which was stated by Auguste Ker-
ckhoffs in the 19th Century). As a result, any predictability in the value of the key
leads to a weakness in the entire system.
This chapter will look at the role of random numbers in cryptography and a couple of
methods for producing them. It will be seen that true random numbers are very hard
to produce accurately and require some physical phenomena which is not always prac-
tical. Failing the use of a true random source, a number of so called pseudorandom
sources have been developed. These work much better in practical environments.
Some other examples of random numbers include the following:

• Reciprocal authentication schemes. The use of random numbers for nonce’s

(described in the last chapter) frustrate replay attacks. Figure 3.6 is an example.

• Session key generation for conventional encryption, and generation of keys for
the RSA public key algorithm.

4.1 Generating Random numbers

It is not that easy to generate random numbers and the numbers produced must be
random in some well-defined statistical sense. Two criteria used to validate the ran-
domness of a sequence of numbers are:

1. Uniform distribution - the frequency of occurrence of each of the numbers

should be approximately the same.

2. Independence - no one value in the sequence can be inferred from the others.

Although there are many tests to determine whether a sequence has uniform distribu-
tion there is no test to prove independence. However, a number of tests can be applied
to demonstrate that a sequence doesn’t exhibit independence. The approach therefore

32
Chapter 4 Random Numbers

is to apply a sufficient number of tests to achieve a certain level of confidence that

independence exists.
Sources of true random numbers are hard to come by. Physical noise generators such
as ionising radiation events, leaky capacitors etc. are examples but are not too useful in
network security applications. Deterministic algorithmic techniques are used instead
and if the algorithm is good, the resulting sequence will pass many tests of randomness.
These are called PseudoRandom Number Generators (PRNG).

4.1.1 Linear Congruential Generator

The most widely used technique is the linear congruential method which was first
introduced by D.H. Lehmer in 1948. The algorithm is parameterised as follows:

m the modulus m>0

a the multiplier 0<a<m
c the increment 0≤c<m
X0 the starting value, or seed 0 ≤ X0 < m

The sequence of random numbers Xn is obtained via the following iterative equation:

Xn+1 = (aXn + c) mod m (4.1)

If m,a,c and X0 are integers then the technique will produce a sequence of integers
with each integer in the range 0 ≤ Xn < m. The selection of a,c and m is critical
in developing a good generator. For example, if a = c = 1 the sequence produced
is of no use as we simply obtain a sequence of integers incremented by 1 each time.
However, consider a = 7, c = 0, m = 32 and X0 = 1. The resultant sequence is
also clearly unsatisfactory {7, 17, 23, 1, 7, . . .}. Of the 32 possible values, only 4 are
used (the sequence is said to have a period of 4). Changing a = 5 then the sequence
is {5, 25, 29, 17, 21, 9, 13, 1, etc.} which gives a period of 8. This is better but still far
from what is desired.
The value m should be large so that there is potential for long sequences, usually m is
nearly equal to the maximum nonnegative integer for a given computer e.g. 231 − 1.
Three criteria that have been proposed to assess random number generators are as
follows:

T1 : The function should be a full-period generating function (i.e. should gener-

ate all numbers between 0 and m before repeating).

T2 : The generated sequence should appear random. The sequence is not random
at all but there is a wide variety of tests to assess the degree of randomness
exhibited.

T3 : The function should implement efficiently with 32-bit arithmetic.

33
Chapter 4 Random Numbers

With appropriate values of a, c and m, these criteria may be achieved. It can be shown
that if m is prime and c = 0, then for certain values of a the period of the generated
sequence is m − 1, with only the value 0 missing. For 32 bit arithmetic, a convenient
large prime of m is 231 − 1 and the function is as follows:

Xn+1 = (aXn ) mod (231 − 1)

Of the more that 2 billion choices for a only a handful of multipliers pass all three
tests. One of these is a = 75 = 16, 807, which was originally designed for use in the
IBM360 family of computers and has been extensively tested.
The strength of the linear congruential algorithm is that if the multiplier and modulus
are chosen properly, then the resultant sequence will be indistinguishable from a se-
quence drawn randomly (but without replacement) from the set 1, 2, 3, 4, . . . , m − 1.
But there is nothing random about the algorithm, apart from the choice of the initial
value X0 . Once that value has been chosen, the remaining numbers in the sequence
follow deterministically. If an opponent knows the algorithm is being used, he/she can
easily discover the sequence.
For example, assume the attacker knows X0 , X1 , X2 , X3 then it is possible to set up
three simultaneous equations as follows:

X1 = (a · X0 + c) mod m
X2 = (a · X1 + c) mod m
X3 = (a · X2 + c) mod m

As there are three equations in three unknowns, it is possible to solve for the three
unknowns.
To make the sequence less reproducible, the sequence could be restarted after every
N numbers. This can be achieved using the current clock value (mod(m)) as a new
seed after N numbers (starting the new sequence). Another method could be to add
the current value of the clock (mod(m)) to the random numbers produced.

4.1.2 Cyclic Encryption

It makes sense to take advantage of the encryption logic available to produce the ran-
dom numbers. An example of Cyclic Encryption is shown in fig 4.1. This is a method
of generating a session key from a master key. A counter with period N provides input
to the encryption logic. If 56 bit DES keys are to be produced, then a counter with a
period of 256 may be used. After each key is produced the counter is incremented, thus
the pseudorandom numbers generated by this scheme cycle through a full period. To
further strengthen the algorithm, the counter could be replaced by a full period PRNG.

34
Chapter 4 Random Numbers

Figure 4.1: Pseudorandom number generation from a counter.

4.1.3 ANSI X9.17: Pseudorandom Number Generator

This is considered one of the strongest (cryptographically speaking) pseudorandom

number generators. It is employed by a number of applications including financial
security schemes and PGP (to be discussed later in the course). It is a U.S. Federal
Information Processing Standard (FIPS) approved method. It makes use of Triple DES
to produce random numbers. Although it uses triple DES three times, it only uses two
keys as the same two keys are used three times. This is effectively the same as using a
112 bit key.
Figure 4.2 shows the scheme. The algorithm makes use of triple DES as mentioned
and has the following inputs and output:

• Input: Two pseudorandom inputs drive the algorithm. One is a 64 bit representa-
tion of the current date and time updated on each number generation. The other
is a 64 bit seed value initialized to some arbitrary value and updated during the
generation process.

• Keys: The generator makes use of three triple DES encryption modules. All
three make use of the same pair of 56 bit keys, which must be kept secret.

• Output: These are a 64 bit pseudorandom number (Ri ) and a 64 bit seed value
(Vi+1 ).

The strength of the PRNG of figure 4.2 can be attributed to the fact that there are a total
of nine DES encryptions and a 112 bit key (effectively). As well as this, the scheme
takes two pseudorandom inputs to begin with, both of which are not revealed (although
it might be possible to work out DTi ). As a result the amount of information needed

35
Chapter 4 Random Numbers

Figure 4.2: ANSI X9.17 Scheme.

by and attacker is formidable. Even if the attacker learns Ri , it would be impossible to

deduce vi+1 due to extra EDE encryption.

4.2 Reducing Possibilities for Traffic Analysis

With the use of link encryption it is possible to mitigate traffic analysis due to the
header being encrypted. However it is still possible to see the amount of traffic being
sent. This may allow the attacker to deduce certain information about the communica-
tion. It is possible to use random number generation as a countermeasure to this attack.
Figure 4.3 shows how.
The output of the model produces ciphertext continuously when no input (plaintext) is
available as the random number generator takes over producing a series of encrypted
random numbers. When an input becomes available the PRNG is switched away from
the output and the input is encrypted.

Figure 4.3: Traffic padding encryption device.

36