Difference between revisions of "Aufgaben:Exercise 5.4:Is the BSC Model Renewing?"

From LNTwww
 
(4 intermediate revisions by 2 users not shown)
Line 1: Line 1:
  
{{quiz-Header|Buchseite=Digitalsignalübertragung/Binary Symmetric Channel (BSC)}}
+
{{quiz-Header|Buchseite=Digital_Signal_Transmission/Binary_Symmetric_Channel_(BSC)}}
  
[[File:P_ID1833__Dig_A_5_4.png|right|frame|Fehlerkorrelationsfunktion (FKF) <br>des BSC–Modells]]
+
[[File:P_ID1833__Dig_A_5_4.png|right|frame|Error correlation function&nbsp; $\rm (ECF)$&nbsp; of the BSC model]]
Zur Beschreibung von digitalen Kanalmodellen werden vorwiegend benutzt:
+
For the description of digital channel models are mainly used:
* die Fehlerabstandsverteilung (FAV)
+
* the error distance distribution&nbsp; $\rm (EDD)$
 
:$$V_a(k) =  {\rm Pr}(a \ge k) = 1 - \sum_{\kappa = 1}^{k}  {\rm Pr}(a = \kappa)\hspace{0.05cm},$$
 
:$$V_a(k) =  {\rm Pr}(a \ge k) = 1 - \sum_{\kappa = 1}^{k}  {\rm Pr}(a = \kappa)\hspace{0.05cm},$$
* die Fehlerkorrelationsfunktion (FKF)
+
* the error correlation function&nbsp; $\rm (ECF)$
 
:$$\varphi_{e}(k) = {\rm E}[e_{\nu} \cdot e_{\nu + k}]
 
:$$\varphi_{e}(k) = {\rm E}[e_{\nu} \cdot e_{\nu + k}]
 
\hspace{0.05cm}.$$
 
\hspace{0.05cm}.$$
  
Für eine große Klasse von Kanalmodellen besteht ein einfacher Zusammenhang zwischen diesen Beschreibungsgrößen, nämlich
+
For a large class of channel models,&nbsp;  there is a simple relationship between these descriptive quantities,&nbsp; viz.
 
:$$\varphi_{e}(k) =
 
:$$\varphi_{e}(k) =
 
  \left\{ \begin{array}{c} \varphi_{e}(0) \\
 
  \left\{ \begin{array}{c} \varphi_{e}(0) \\
 
  \sum_{\kappa = 1}^{k}  {\rm Pr}(a = \kappa) \cdot \varphi_{e}(k - \kappa)\end{array} \right.\quad
 
  \sum_{\kappa = 1}^{k}  {\rm Pr}(a = \kappa) \cdot \varphi_{e}(k - \kappa)\end{array} \right.\quad
\begin{array}{*{1}c} f{\rm \ddot{u}r }\hspace{0.15cm}k = 0  \hspace{0.05cm},
+
\begin{array}{*{1}c} f{\rm or }\hspace{0.15cm}k = 0  \hspace{0.05cm},
\\  f{\rm \ddot{u}r }\hspace{0.15cm} k > 0 \hspace{0.05cm}.\\ \end{array}$$
+
\\  f{\rm or }\hspace{0.15cm} k > 0 \hspace{0.05cm}.\\ \end{array}$$
  
Man nennt solche Kanalmodelle&nbsp; '''erneuernd'''. Sie zeichnen sich dadurch aus, dass bei ihnen die einzelnen Fehlerabstände statistisch voneinander unabhängig sind, so dass zur Generierung der Fehlerfolge der oft schnellere Weg über die Generierung der Fehlerabstände gegangen werden kann, wie in der&nbsp; [[Aufgaben:5.5_Fehlerfolge_und_Fehlerabstandsfolge| Aufgabe 5.5]]&nbsp; beschrieben wird.
+
Such channel models are called&nbsp; "'''renewing'''".  
  
In dieser Aufgabe soll überprüft werden, ob das BSC&ndash;Modell gemäß der oberen Grafik erneuernd ist.  
+
*They are characterized by the fact that in them the individual error distances are statistically independent of each other,&nbsp;
*Die Fehlerkorrelationsfunktion&nbsp; $\varphi_e(k)$&nbsp; ist in der unteren Grafik dargestellt.  
+
*so that to generate the error sequence the often faster way can be followed via the generation of the error distances,&nbsp; as described in&nbsp; [[Aufgaben:Exercise_5.5:_Error_Sequence_and_Error_Distance_Sequence| "Exercise 5.5"]].&nbsp;
*Die Wahrscheinlichkeiten der einzelnen Fehlerabstände sind beim BSC&ndash;Modell wie folgt gegeben:
+
 
 +
 
 +
In this exercise,&nbsp; we want to check whether the BSC model is renewing according to the upper graph.
 +
*The error correlation function&nbsp; $\varphi_e(k)$&nbsp; is shown in the bottom graph.
 +
 +
*The probabilities of the individual error distances are given by the BSC model as follows:
 
:$${\rm Pr}(a = k) = (1-p)^{k-1}\cdot p \hspace{0.05cm}.$$
 
:$${\rm Pr}(a = k) = (1-p)^{k-1}\cdot p \hspace{0.05cm}.$$
  
Line 28: Line 33:
  
  
''Hinweise:''
+
<u>Notes:</u>
* Die Aufgabe gehört zum Kapitel&nbsp; [[Digitalsignal%C3%BCbertragung/Binary_Symmetric_Channel_(BSC)| Binary Symmetric Channel (BSC)]].  
+
* The exercise belongs to the chapter&nbsp; [[Digital_Signal_Transmission/Binary_Symmetric_Channel_(BSC)|"Binary Symmetric Channel"]].
* Verwenden Sie für numerische Berechnungen den BSC&ndash;Parameter&nbsp; $p = 0.01$.
+
* Die mittlere Fehlerwahrscheinlichkeit&nbsp; $p_{\rm M}$ hat dann den gleichen Wert.
+
* Use the BSC parameter&nbsp; $p = 0.01$&nbsp; for numerical calculations.&nbsp; The mean error probability&nbsp; $p_{\rm M}$ then has the same value.
 
   
 
   
  
  
  
===Fragebogen===
+
===Questions===
 
<quiz display=simple>
 
<quiz display=simple>
{Berechnen Sie den FKF&ndash;Wert&nbsp; $\varphi_e(k = 0)$.
+
{Calculate the ECF value&nbsp; $\varphi_e(k = 0)$.
 
|type="{}"}
 
|type="{}"}
 
$\varphi_e(k = 0) \ = \ ${ 1 3% } $\ \cdot 10^{\rm -2} $
 
$\varphi_e(k = 0) \ = \ ${ 1 3% } $\ \cdot 10^{\rm -2} $
  
{Berechnen Sie den FKF&ndash;Wert&nbsp; $\varphi_e(k = 1)$.
+
{Calculate the ECF value&nbsp; $\varphi_e(k = 1)$.
 
|type="{}"}
 
|type="{}"}
 
$\varphi_e(k = 1) \ = \ ${ 0.01 3% } $\ \cdot 10^{\rm -2} $
 
$\varphi_e(k = 1) \ = \ ${ 0.01 3% } $\ \cdot 10^{\rm -2} $
  
{Berechnen Sie den FKF&ndash;Wert&nbsp; $\varphi_e(k = 2)$.
+
{Calculate the ECF value&nbsp; $\varphi_e(k = 2)$.
 
|type="{}"}
 
|type="{}"}
 
$\varphi_e(k = 2) \ = \ ${ 0.01 3% } $\ \cdot 10^{\rm -2} $
 
$\varphi_e(k = 2) \ = \ ${ 0.01 3% } $\ \cdot 10^{\rm -2} $
  
{Liefern Sie ein begründetes Resumé dieser Aufgabe.
+
{Provide a reasoned summary of this exercise.
 
|type="()"}
 
|type="()"}
+ Das BSC&ndash;Modell ist erneuernd.
+
+ The BSC model is renewing.
- Das BSC&ndash;Modell ist nicht erneuernd.
+
- The BSC model is non-renewing.
 
</quiz>
 
</quiz>
  
===Musterlösung===
+
===Solution===
 
{{ML-Kopf}}
 
{{ML-Kopf}}
'''(1)'''&nbsp; Nach der allgemeinen Definition ist $\varphi_e(k = 0) = {\rm E}[e_{\nu}^2]$.  
+
'''(1)'''&nbsp; According to the general definition,&nbsp; $\varphi_e(k = 0) = {\rm E}[e_{\nu}^2]$.  
*Wegen $e_{\nu} &#8712; \{0, 1\}$ gilt aber gleichzeitig $\varphi_e(k = 0) = {\rm E}[e_\nu]$, was der mittleren Fehlerwahrscheinlichkeit $p_{\rm M} = p$&nbsp; entspricht &nbsp; &#8658; &nbsp; $\varphi_e(k = 0) \ \underline { = 0.01}$.
+
*However,&nbsp; because of&nbsp; $e_{\nu} &#8712; \{0, 1\}$ &nbsp; &rArr; &nbsp;  $\varphi_e(k = 0) = {\rm E}[e_\nu]$,&nbsp; holds simultaneously,&nbsp; which corresponds to the mean error probability&nbsp; $p_{\rm M} = p$&nbsp; &nbsp; &#8658; &nbsp; $\varphi_e(k = 0) \ \underline { = 0.01}$.
  
  
  
'''(2)'''&nbsp; Nach der allgemeinen FKF&ndash;Definition gilt unter Berücksichtigung des BSC&ndash;Modells:
+
'''(2)'''&nbsp; According to the general ECF definition,&nbsp; considering the BSC model:
 
:$$\varphi_{e}(k = 1) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} {\rm E}[e_{\nu} \cdot e_{\nu + 1}]={\rm Pr}[e_{\nu} = 1 \cap
 
:$$\varphi_{e}(k = 1) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} {\rm E}[e_{\nu} \cdot e_{\nu + 1}]={\rm Pr}[e_{\nu} = 1 \cap
 
e_{\nu + 1} = 1] = p \cdot p = p^2 \hspace{0.15cm}\underline {= 10^{-4}}\hspace{0.05cm}.$$
 
e_{\nu + 1} = 1] = p \cdot p = p^2 \hspace{0.15cm}\underline {= 10^{-4}}\hspace{0.05cm}.$$
  
Zum gleichen Ergebnis kommt man über die nur für erneuernde Kanalmodelle gültige Gleichung:
+
*The same result is obtained via the equation valid only for renewing channel models:
 
:$$\varphi_{e}(k = 1) = {\rm Pr}(a=1) \cdot \varphi_{e}(0) = p \cdot
 
:$$\varphi_{e}(k = 1) = {\rm Pr}(a=1) \cdot \varphi_{e}(0) = p \cdot
 
p = p^2 = 10^{-4}\hspace{0.05cm}.$$
 
p = p^2 = 10^{-4}\hspace{0.05cm}.$$
  
Das bedeutet: Der FKF&ndash;Wert $\varphi_e(1)$ spricht nicht dagegen, dass das BSC&ndash;Modell erneuernd ist.
+
*That means:&nbsp; The ECF value&nbsp; $\varphi_e(1)$&nbsp; does not argue against the BSC model being renewing.
  
  
  
'''(3)'''&nbsp; Aus der Grafik erkennt man bereits, dass $\varphi_e(k = 2) = \varphi_e(k = 1) = 10^{&ndash;4}$ gelten wird. Die explizite Berechnung bestätigt dieses Ergebnis:
+
'''(3)'''&nbsp; From the graph,&nbsp; we can already see that&nbsp; $\varphi_e(k = 2) = \varphi_e(k = 1) = 10^{&ndash;4}$ will hold.&nbsp; The explicit calculation confirms this result:
 
:$$\varphi_{e}(k \hspace{-0.1cm} \ = \ \hspace{-0.1cm} 2) = {\rm
 
:$$\varphi_{e}(k \hspace{-0.1cm} \ = \ \hspace{-0.1cm} 2) = {\rm
 
Pr}[e_{\nu} = 1 \cap e_{\nu + 2} = 1] =  
 
Pr}[e_{\nu} = 1 \cap e_{\nu + 2} = 1] =  
Line 82: Line 87:
 
e_{\nu + 1} = 0 \cap e_{\nu + 2} = 1] \hspace{0.05cm}.$$
 
e_{\nu + 1} = 0 \cap e_{\nu + 2} = 1] \hspace{0.05cm}.$$
  
*Der erste Term lautet beim BSC-Modell mit den bedingten Wahrscheinlichkeiten (nur erste Ordnung erforderlich):
+
*The first term in the BSC model with the conditional probabilities&nbsp; $($only first order required$)$&nbsp; is:
 
:$${\rm Pr}[1\hspace{0.1cm}1 \hspace{0.1cm} 1] \hspace{-0.1cm} \ = \
 
:$${\rm Pr}[1\hspace{0.1cm}1 \hspace{0.1cm} 1] \hspace{-0.1cm} \ = \
 
\hspace{-0.1cm}{\rm Pr}[e_{\nu +2} = 1
 
\hspace{-0.1cm}{\rm Pr}[e_{\nu +2} = 1
Line 90: Line 95:
 
p^3\hspace{0.05cm}.$$
 
p^3\hspace{0.05cm}.$$
  
*Entsprechend gilt für den zweiten Term:
+
*Correspondingly,&nbsp; for the second term:
 
:$${\rm Pr}[1\hspace{0.1cm}0 \hspace{0.1cm} 1] \hspace{-0.1cm} \ = \
 
:$${\rm Pr}[1\hspace{0.1cm}0 \hspace{0.1cm} 1] \hspace{-0.1cm} \ = \
 
\hspace{-0.1cm}{\rm Pr}[e_{\nu +2} = 1
 
\hspace{-0.1cm}{\rm Pr}[e_{\nu +2} = 1
Line 102: Line 107:
 
10^{-4}}\hspace{0.05cm}.$$
 
10^{-4}}\hspace{0.05cm}.$$
  
*Mit der nur für erneuernde Kanalmodelle gültigen Gleichung erhält man:
+
*Using the equation valid only for renewing channel models,&nbsp; we obtain:
 
:$$\varphi_{e}(k = 2) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} {\rm
 
:$$\varphi_{e}(k = 2) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} {\rm
 
Pr}(a=1) \cdot \varphi_{e}(k= 1) + {\rm Pr}(a=2) \cdot
 
Pr}(a=1) \cdot \varphi_{e}(k= 1) + {\rm Pr}(a=2) \cdot
Line 108: Line 113:
 
\cdot p = p^2 \hspace{0.05cm}.$$
 
\cdot p = p^2 \hspace{0.05cm}.$$
  
*Auch dieses Ergebnis spricht also nicht dagegen, dass das BSC&ndash;Modell erneuernd ist.
+
*Thus,&nbsp; this result also does not argue against the BSC model being renewing.
  
  
  
  
'''(4)'''&nbsp; Die bisherigen Ergebnisse lassen schon darauf schließen, dass das BSC&ndash;Modell erneuernd ist. Und auch die Tatsache, dass hier die einzelnen Fehlerabstände statistisch unabhängig voneinander sind, spricht für diese These.  
+
'''(4)'''&nbsp; The previous results already suggest that the BSC model is renewing.  
  
*Als letzten Beweis zeigen wird, dass die Gleichung
+
*And also the fact that here the individual error distances are statistically independent of each other speaks in favor of this thesis.
 +
 
 +
*As a final proof,&nbsp; we show that the equation
 
:$$\varphi_{e}(k) = \sum_{\kappa = 1}^{k}  {\rm Pr}(a = \kappa) \cdot
 
:$$\varphi_{e}(k) = \sum_{\kappa = 1}^{k}  {\rm Pr}(a = \kappa) \cdot
 
\varphi_{e}(k - \kappa)= \varphi_{e}(0) \cdot  {\rm Pr}(a = k)+
 
\varphi_{e}(k - \kappa)= \varphi_{e}(0) \cdot  {\rm Pr}(a = k)+
Line 121: Line 128:
 
\kappa)$$
 
\kappa)$$
  
:das richtige Ergebnis liefert, wenn $\varphi_e(0) = p$ und $\varphi_e(1) = \ \text{...} \ = \varphi_e(k&ndash;1) = p^2$ eingesetzt wird.  
+
:yields the correct result when&nbsp; $\varphi_e(0) = p$&nbsp; and&nbsp; $\varphi_e(1) = \ \text{...} \ = \varphi_e(k&ndash;1) = p^2$&nbsp; are used.
*Man erhält
+
 
 +
*One obtains
 
:$$\varphi_{e}(k) \hspace{-0.1cm} \ = \ \hspace{-0.1cm}  p \cdot
 
:$$\varphi_{e}(k) \hspace{-0.1cm} \ = \ \hspace{-0.1cm}  p \cdot
 
(1-p)^{k-1} \cdot p + p^2 \cdot \sum_{\kappa = 1}^{k-1}
 
(1-p)^{k-1} \cdot p + p^2 \cdot \sum_{\kappa = 1}^{k-1}
Line 128: Line 136:
 
\cdot \sum_{\kappa = 0}^{k-2} (1-p)^{\kappa}\hspace{0.05cm}.$$
 
\cdot \sum_{\kappa = 0}^{k-2} (1-p)^{\kappa}\hspace{0.05cm}.$$
  
*Mit der Summenformel einer geometrischen Reihe
+
*Using the summation formula of a geometric series
 
:$$\sum_{\kappa = 0}^{n} x^{\kappa} = \frac{1 - x^{n+1}}{1 -
 
:$$\sum_{\kappa = 0}^{n} x^{\kappa} = \frac{1 - x^{n+1}}{1 -
 
x}\hspace{0.05cm},$$
 
x}\hspace{0.05cm},$$
  
:lässt sich dieser Ausdruck wie folgt darstellen:
+
:this expression can be represented as follows:
 
:$$\varphi_{e}(k) \hspace{-0.1cm} \ = \ \hspace{-0.1cm}  p^2 \cdot
 
:$$\varphi_{e}(k) \hspace{-0.1cm} \ = \ \hspace{-0.1cm}  p^2 \cdot
 
(1-p)^{k-1} + p^3 \cdot \frac{1 - (1-p)^{k-1}}{1 -
 
(1-p)^{k-1} + p^3 \cdot \frac{1 - (1-p)^{k-1}}{1 -
Line 138: Line 146:
 
(1-p)^{k-1} + 1 - (1-p)^{k-1} \right ] = p^2\hspace{0.05cm}.$$
 
(1-p)^{k-1} + 1 - (1-p)^{k-1} \right ] = p^2\hspace{0.05cm}.$$
  
*Das bedeutet:&nbsp; Das BSC&ndash;Modell ist tatsächlich erneuernd &nbsp; &#8658; &nbsp; <u>Lösungsvorschlag 1</u>.
+
*This means:&nbsp; The BSC model is in fact renewing &nbsp; &#8658; &nbsp; <u>solution 1</u>.
 
{{ML-Fuß}}
 
{{ML-Fuß}}
  
  
[[Category:Digital Signal Transmission: Exercises|^5.2 Binary Symmetric Channel (BSC)^]]
+
[[Category:Digital Signal Transmission: Exercises|^5.2 Binary Symmetric Channel^]]

Latest revision as of 15:29, 5 September 2022

Error correlation function  $\rm (ECF)$  of the BSC model

For the description of digital channel models are mainly used:

  • the error distance distribution  $\rm (EDD)$
$$V_a(k) = {\rm Pr}(a \ge k) = 1 - \sum_{\kappa = 1}^{k} {\rm Pr}(a = \kappa)\hspace{0.05cm},$$
  • the error correlation function  $\rm (ECF)$
$$\varphi_{e}(k) = {\rm E}[e_{\nu} \cdot e_{\nu + k}] \hspace{0.05cm}.$$

For a large class of channel models,  there is a simple relationship between these descriptive quantities,  viz.

$$\varphi_{e}(k) = \left\{ \begin{array}{c} \varphi_{e}(0) \\ \sum_{\kappa = 1}^{k} {\rm Pr}(a = \kappa) \cdot \varphi_{e}(k - \kappa)\end{array} \right.\quad \begin{array}{*{1}c} f{\rm or }\hspace{0.15cm}k = 0 \hspace{0.05cm}, \\ f{\rm or }\hspace{0.15cm} k > 0 \hspace{0.05cm}.\\ \end{array}$$

Such channel models are called  "renewing".

  • They are characterized by the fact that in them the individual error distances are statistically independent of each other, 
  • so that to generate the error sequence the often faster way can be followed via the generation of the error distances,  as described in  "Exercise 5.5"


In this exercise,  we want to check whether the BSC model is renewing according to the upper graph.

  • The error correlation function  $\varphi_e(k)$  is shown in the bottom graph.
  • The probabilities of the individual error distances are given by the BSC model as follows:
$${\rm Pr}(a = k) = (1-p)^{k-1}\cdot p \hspace{0.05cm}.$$



Notes:

  • Use the BSC parameter  $p = 0.01$  for numerical calculations.  The mean error probability  $p_{\rm M}$ then has the same value.



Questions

1

Calculate the ECF value  $\varphi_e(k = 0)$.

$\varphi_e(k = 0) \ = \ $

$\ \cdot 10^{\rm -2} $

2

Calculate the ECF value  $\varphi_e(k = 1)$.

$\varphi_e(k = 1) \ = \ $

$\ \cdot 10^{\rm -2} $

3

Calculate the ECF value  $\varphi_e(k = 2)$.

$\varphi_e(k = 2) \ = \ $

$\ \cdot 10^{\rm -2} $

4

Provide a reasoned summary of this exercise.

The BSC model is renewing.
The BSC model is non-renewing.


Solution

(1)  According to the general definition,  $\varphi_e(k = 0) = {\rm E}[e_{\nu}^2]$.

  • However,  because of  $e_{\nu} ∈ \{0, 1\}$   ⇒   $\varphi_e(k = 0) = {\rm E}[e_\nu]$,  holds simultaneously,  which corresponds to the mean error probability  $p_{\rm M} = p$    ⇒   $\varphi_e(k = 0) \ \underline { = 0.01}$.


(2)  According to the general ECF definition,  considering the BSC model:

$$\varphi_{e}(k = 1) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} {\rm E}[e_{\nu} \cdot e_{\nu + 1}]={\rm Pr}[e_{\nu} = 1 \cap e_{\nu + 1} = 1] = p \cdot p = p^2 \hspace{0.15cm}\underline {= 10^{-4}}\hspace{0.05cm}.$$
  • The same result is obtained via the equation valid only for renewing channel models:
$$\varphi_{e}(k = 1) = {\rm Pr}(a=1) \cdot \varphi_{e}(0) = p \cdot p = p^2 = 10^{-4}\hspace{0.05cm}.$$
  • That means:  The ECF value  $\varphi_e(1)$  does not argue against the BSC model being renewing.


(3)  From the graph,  we can already see that  $\varphi_e(k = 2) = \varphi_e(k = 1) = 10^{–4}$ will hold.  The explicit calculation confirms this result:

$$\varphi_{e}(k \hspace{-0.1cm} \ = \ \hspace{-0.1cm} 2) = {\rm Pr}[e_{\nu} = 1 \cap e_{\nu + 2} = 1] = {\rm Pr}[e_{\nu} = 1 \cap e_{\nu + 1} = 1 \cap e_{\nu + 2} = 1] + {\rm Pr}[e_{\nu} = 1 \cap e_{\nu + 1} = 0 \cap e_{\nu + 2} = 1] \hspace{0.05cm}.$$
  • The first term in the BSC model with the conditional probabilities  $($only first order required$)$  is:
$${\rm Pr}[1\hspace{0.1cm}1 \hspace{0.1cm} 1] \hspace{-0.1cm} \ = \ \hspace{-0.1cm}{\rm Pr}[e_{\nu +2} = 1 \hspace{0.1cm}|\hspace{0.1cm} e_{\nu + 1} = 1] \cdot {\rm Pr}[e_{\nu +1} = 1 \hspace{0.1cm}|\hspace{0.1cm} e_{\nu } = 1] \cdot {\rm Pr}[ e_{\nu } = 1]=p \cdot p \cdot p = p^3\hspace{0.05cm}.$$
  • Correspondingly,  for the second term:
$${\rm Pr}[1\hspace{0.1cm}0 \hspace{0.1cm} 1] \hspace{-0.1cm} \ = \ \hspace{-0.1cm}{\rm Pr}[e_{\nu +2} = 1 \hspace{0.1cm}|\hspace{0.1cm} e_{\nu + 1} = 0] \cdot {\rm Pr}[e_{\nu +1} = 0 \hspace{0.1cm}|\hspace{0.1cm} e_{\nu } = 1] \cdot {\rm Pr}[ e_{\nu } = 1]=p \cdot (1-p) \cdot p = p^2 -p^3\hspace{0.05cm}.$$
$$\Rightarrow \hspace{0.3cm} \varphi_{e}(k = 2) = {\rm Pr}[1\hspace{0.1cm}1 \hspace{0.1cm} 1] + {\rm Pr}[1\hspace{0.1cm}0 \hspace{0.1cm} 1] = (p^3) + (p^2 -p^3)= p^2\hspace{0.15cm}\underline { = 10^{-4}}\hspace{0.05cm}.$$
  • Using the equation valid only for renewing channel models,  we obtain:
$$\varphi_{e}(k = 2) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} {\rm Pr}(a=1) \cdot \varphi_{e}(k= 1) + {\rm Pr}(a=2) \cdot \varphi_{e}(k= 0)= p \cdot p^2 + (1-p) \cdot p \cdot p = p^2 \hspace{0.05cm}.$$
  • Thus,  this result also does not argue against the BSC model being renewing.



(4)  The previous results already suggest that the BSC model is renewing.

  • And also the fact that here the individual error distances are statistically independent of each other speaks in favor of this thesis.
  • As a final proof,  we show that the equation
$$\varphi_{e}(k) = \sum_{\kappa = 1}^{k} {\rm Pr}(a = \kappa) \cdot \varphi_{e}(k - \kappa)= \varphi_{e}(0) \cdot {\rm Pr}(a = k)+ \sum_{\kappa = 1}^{k-1} \varphi_{e}(k - \kappa) \cdot {\rm Pr}(a = \kappa)$$
yields the correct result when  $\varphi_e(0) = p$  and  $\varphi_e(1) = \ \text{...} \ = \varphi_e(k–1) = p^2$  are used.
  • One obtains
$$\varphi_{e}(k) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} p \cdot (1-p)^{k-1} \cdot p + p^2 \cdot \sum_{\kappa = 1}^{k-1} (1-p)^{\kappa-1} \cdot p =p^2 \cdot (1-p)^{k-1} + p^3 \cdot \sum_{\kappa = 0}^{k-2} (1-p)^{\kappa}\hspace{0.05cm}.$$
  • Using the summation formula of a geometric series
$$\sum_{\kappa = 0}^{n} x^{\kappa} = \frac{1 - x^{n+1}}{1 - x}\hspace{0.05cm},$$
this expression can be represented as follows:
$$\varphi_{e}(k) \hspace{-0.1cm} \ = \ \hspace{-0.1cm} p^2 \cdot (1-p)^{k-1} + p^3 \cdot \frac{1 - (1-p)^{k-1}}{1 - (1-p)}= p^2 \cdot \left [ (1-p)^{k-1} + 1 - (1-p)^{k-1} \right ] = p^2\hspace{0.05cm}.$$
  • This means:  The BSC model is in fact renewing   ⇒   solution 1.