# Exercise 4.3: Algebraic and Modulo Sum

A  "clocked"  random number generator returns a sequence  $\langle x_\nu \rangle$  of binary random numbers.

• It is assumed that the binary numbers  $0$  and  $1$  occur with equal probabilities and that the individual random numbers do not depend on each other.
• The random numbers  $x_\nu \in \{0, 1\}$  are entered into the first memory location of a shift register and shifted down one digit with each clock pulse.

Two new random sequences  $\langle a_\nu \rangle$  and  $\langle m_\nu \rangle$  are formed from the contents of the three-digit shift register. Here denotes:

• the  "algebraic sum"  $a_\nu$:
$$a_\nu=x_\nu+x_{\nu-1}+x_{\nu-2},$$
• the  "modulo–2 sum"  $m_\nu$:
$$m_\nu=x_\nu\oplus x_{\nu-1}\oplus x_{\nu-2}.$$

Hints:

### Questions

1

Calculate the probabilities of the random variable  $m_\nu$.  What is the probability that the modulo-2 sum is equal to  $0$ ?

 ${\rm Pr}(m_\nu = 0) \ = \$

2

Are there statistical dependencies within the sequence  $\langle m_\nu \rangle$?

 The sequence elements  $m_\nu$  are statistically independent. There are statistical bindings within the sequence  $\langle m_\nu \rangle$.

3

Determine the 2D–PDF  $f_{xm}(x_\nu, m_\nu)$.  Based on the result,  evaluate the following statements.

 The random variables  $x_\nu$  and  $m_\nu$  are statistically dependent. The random variables  $x_\nu$  and  $m_\nu$  are statistically independent. The random variables  $x_\nu$  and  $m_\nu$  are correlated. The random variables  $x_\nu$  and  $m_\nu$  are uncorrelated.

4

Do statistical dependencies exist within the sequence  $\langle a_\nu \rangle$ ?

 The sequence elements  $a_\nu$  are statistically independent. There are statistical bindings within the sequence  $\langle a_\nu \rangle$.

5

Determine the 2D–PDF  $f_{am}(a_\nu, m_\nu)$  and the correlation coefficient  $\rho_{am}$.  Which of the following statements are true?

 The random variables  $a_\nu$  and  $m_\nu$  are statistically dependent. The random variables  $a_\nu$  and  $m_\nu$  are statistically independent. The random variables  $a_\nu$  and  $m_\nu$  are correlated. The random variables  $a_\nu$  and  $m_\nu$  are uncorrelated.

### Solution

#### Solution

(1)  It can be seen from the table in the information section that for the modulo–2 sum,  the two values  $0$  and  $1$  have equal probability:

$${\rm Pr}(m_\nu = 0) = {\rm Pr}(m_\nu = 1)\hspace{0.15cm}\underline{=0.5}.$$

(2)  The table shows that for each preassignment   ⇒   $( x_{\nu-1}, x_{\nu-2}) = (0,0), (0,1), (1,0), (1,1)$,  the values  $m_\nu = 0$  and  $m_\nu = 1$  resp. are equally likely.

• Expressed differently:   ${\rm Pr}(m_{\nu}\hspace{0.05cm}|\hspace{0.05cm}m_{\nu-1}) = {\rm Pr}( m_{\nu}).$
• This exactly matches the definition of  "statistical independence"   ⇒   Answer 1.

(3)  Correct are  the second and the last suggested solutions.

• The 2D–PDF consists of four Dirac delta functions,  each with weight  $1/4$.
• One obtains this result,  for example,  by evaluating the table in the data section.
• Since  $f_{xm}(x_\nu, m_\nu)=f_{x}(x_\nu) \cdot f_{m}(m_\nu)$,  the quantities  $x_\nu$  and  $m_\nu$  are statistically independent.
• Statistically independent random variables,  however,  are also linearly statistically independent,  so they are certainly uncorrelated.

(4)  Within the sequence  $\langle a_\nu \rangle$  of algebraic sum there are statistical bindings   ⇒   Answer 2.

• You can see this because the unconditional probability is  ${\rm Pr}( a_{\nu} = 0) =1/8$,
• while,  for example,  ${\rm Pr}(a_{\nu} = 0\hspace{0.05cm}|\hspace{0.05cm}a_{\nu-1} = 3) =0$  holds.

(5)  Correct are  the first and the last suggested solutions:

• As in the subtask  (3)  there are again four Dirac delta functions,  but this time not with equal Dirac weights  $1/4$.
• The two-dimensional PDF thus cannot be written as a product of the two marginal probability densities.
• But this means that statistical bindings must exist between  $a_\nu$  and  $m_\nu$.
• For the joint expected value,  one obtains:
$${\rm E}\big[a\cdot m \big] = \rm \frac{1}{8}\cdot 0 \cdot 0 +\frac{3}{8}\cdot 2 \cdot 0 +\frac{3}{8}\cdot 1 \cdot 1 + \frac{1}{8}\cdot 3 \cdot 1 = \frac{3}{4}.$$
• With the linear means  ${\rm E}\big[a \big] = 1.5$  and  ${\rm E}[m] = 0.5$  it follows for the covariance:
$$\mu_{am}= {\rm E}\big[ a\cdot m \big] - {\rm E}\big[ a \big]\cdot {\rm E} \big[ m \big] = \rm 0.75-1.5\cdot 0.5 = \rm 0.$$
• Thus,  the correlation coefficient  $\rho_{am}= 0$.  That is:   The dependencies present are nonlinear.
• The quantities  $a_\nu$  and  $m_\nu$  are statistically dependent,  but still uncorrelated.