Exercise 4.1Z: Log Likelihood Ratio at the BEC Model

From LNTwww

BEC channel model

We consider the so-called  $\text{BEC channel }$  ("binary erasure channel")  with

  • the input variable  $x ∈ \{+1, \, -1\}$,
  • the output variable  $y ∈ \{+1, \, -1, \, {\rm E}\}$,  and
  • the erasure probability  $\lambda$.


Here  $y = {\rm E}$  ("erasure")  means that the initial value  $y$  could neither be decided as  "$+1$"  nor as  "$-1$".

Also known are the input probabilities

$${\rm Pr}(x = +1) = 3/4\hspace{0.05cm}, \hspace{0.5cm}{\rm Pr}(x = -1) = 1/4\hspace{0.05cm}.$$

The log likelihood ratio of the binary random variable  $x$  is given by bipolar approach as follows:

$$L(x)={\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(x = +1)}{{\rm Pr}(x = -1)}\hspace{0.05cm}.$$

Correspondingly,  for the conditional log likelihood ratio in forward direction for all  $y ∈ \{+1, \, -1, \, {\rm E}\}$:

$$L(y\hspace{0.05cm}|\hspace{0.05cm}x) = {\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(y\hspace{0.05cm}|\hspace{0.05cm}x = +1)}{{\rm Pr}(y\hspace{0.05cm}|\hspace{0.05cm}x = -1)} \hspace{0.05cm}. $$



Hints:

  • Reference is made in particular to the sections 



Questions

1

What is the log likelihood ratio of the input variable  $x$?

$L(x) \ = \ $

2

What probability  ${\rm Pr}(x = \, -1)$  corresponds to  $L(x) = \, -2$?

${\rm Pr}(x = \, -1) \ = \ $

3

Calculate the conditional  L–value  $L(y = {\rm E}\hspace{0.05cm} |\hspace{0.05cm} x)$  in the forward direction.

$L(y = {\rm E} \hspace{0.05cm} |\hspace{0.05cm} x) \ = \ $

4

Which statements are true for the other two conditional log likelihood ratios?

$L(y = +1 \hspace{0.05cm} |\hspace{0.05cm} x)$  is positive and infinite in magnitude.
$L(y = \, -1 \hspace{0.05cm} |\hspace{0.05cm} x)$  is negative and infinite in magnitude.
It holds  $L(y = +1 \hspace{0.05cm} |\hspace{0.05cm} x) = L(y = \, -1 \hspace{0.05cm} |\hspace{0.05cm} x) = 0$.

5

Under what conditions do the results from subtasks  (3)  and  (4)  hold?

For  $0 ≤ \lambda ≤ 1$.
For  $0 < \lambda ≤ 1$.
For  $0 ≤ \lambda < 1$.
For  $0 < \lambda < 1$.


Solution

(1)  With the given symbol probabilities   ${\rm Pr}(x = +1) = 3/4$   and   ${\rm Pr}(x = -1) = 1/4$,  we obtain:

$$L(x)={\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(x = +1)}{{\rm Pr}(x = -1)} ={\rm ln} \hspace{0.15cm} \frac{3/4}{1/4}\hspace{0.15cm}\underline{= 1.099}\hspace{0.05cm}.$$


(2)  According to the definition

$$L(x)={\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(x = +1)}{{\rm Pr}(x = -1)}$$

yields the following equation for  $L(x) = \, -2$:

$$\hspace{0.15cm} \frac{{\rm Pr}(x = +1)}{1-{\rm Pr}(x = +1)} \stackrel{!}{=}{\rm e}^{-2} \approx 0.135 \hspace{0.25cm}\Rightarrow \hspace{0.25cm} 1.135 \cdot {\rm Pr}(x = +1)\stackrel{!}{=}0.135\hspace{0.3cm} \Rightarrow \hspace{0.3cm} {\rm Pr}(x = +1) = 0.119\hspace{0.05cm},\hspace{0.4cm}{\rm Pr}(x = -1) \hspace{0.15cm}\underline{= 0.881}\hspace{0.05cm}. $$


(3)  For the conditional log likelihood ratio   $L(y = {\rm E} \hspace{0.05cm} |\hspace{0.05cm} x)$   in the forward direction,  valid for the BEC model:

$$L(y = {\rm E}\hspace{0.05cm}|\hspace{0.05cm}x) = {\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(y= {\rm E}\hspace{0.05cm}|\hspace{0.05cm}x = +1)}{{\rm Pr}(y= {\rm E}\hspace{0.05cm}|\hspace{0.05cm}x = -1)} = {\rm ln} \hspace{0.15cm} \frac{\lambda}{\lambda}\hspace{0.15cm}\underline{= 0}\hspace{0.05cm}.$$


(4)  Analogous to the sample solution of subtask  (3),  we obtain for  $y = ±1$:

$$L(y = +1\hspace{0.05cm}|\hspace{0.05cm}x) \hspace{-0.15cm} \ = \ \hspace{-0.15cm} {\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(y= +1\hspace{0.05cm}|\hspace{0.05cm}x = +1)}{{\rm Pr}(y= +1\hspace{0.05cm}|\hspace{0.05cm}x = -1)} = {\rm ln} \hspace{0.15cm} \frac{1-\lambda}{0}\hspace{0.15cm}\underline{ \hspace{0.05cm}\Rightarrow \hspace{0.15cm}+\infty }\hspace{0.05cm},$$
$$L(y = -1\hspace{0.05cm}|\hspace{0.05cm}x) \hspace{-0.15cm} \ = \ \hspace{-0.15cm} {\rm ln} \hspace{0.15cm} \frac{{\rm Pr}(y= -1\hspace{0.05cm}|\hspace{0.05cm}x = +1)}{{\rm Pr}(y= -1\hspace{0.05cm}|\hspace{0.05cm}x = -1)} = {\rm ln} \hspace{0.15cm} \frac{0}{1-\lambda}\hspace{0.15cm}\underline{ \hspace{0.05cm}\Rightarrow \hspace{0.15cm}-\infty }\hspace{0.05cm}. $$
  • Accordingly,  the  proposed solutions 1 and 2  are correct.


(5)  Correct is the  last proposed solution:

  • For  $\lambda = 0$  $($"ideal channel"$)$   ⇒   $L(y = {\rm E} \hspace{0.05cm} |\hspace{0.05cm} x) = \ln {(0/0)}$   ⇒   indefinite result.
  • For  $\lambda = 1$  $($complete erasure,   $y ≡ {\rm E})$   ⇒   $L(y = +1 \hspace{0.05cm} |\hspace{0.05cm} x)$  and   $L(y = \, -1 \hspace{0.05cm} |\hspace{0.05cm} x)$   are undefined.