Difference between revisions of "Aufgaben:Exercise 3.2Z: (3, 1, 3) Convolutional Encoder"

From LNTwww
Line 70: Line 70:
 
===Solution===
 
===Solution===
 
{{ML-Kopf}}
 
{{ML-Kopf}}
'''(1)'''  For the index $l$ of the partial matrices, $0 ≤ l ≤ m$.  
+
'''(1)'''  For the index  $l$  of the partial matrices:  $0 ≤ l ≤ m$.  
*The coder under consideration has memory $m = 3$.  
+
*The coder under consideration has memory  $m = 3$.
*Thereby <u>four partial matrices</u> are to be considered.
+
 +
*Thereby&nbsp; <u>four partial matrices</u>&nbsp; are to be considered.
  
  
  
'''(2)'''&nbsp; Each partial matrix $\mathbf{G}_l$ consists of.
+
'''(2)'''&nbsp; Each partial matrix&nbsp; $\mathbf{G}_l$&nbsp; consists of  
*<u>one row</u> &nbsp;&#8658;&nbsp; $k = 1$, and  
+
*<u>one row</u> &nbsp; &#8658; &nbsp; $k = 1$,&nbsp; and
*<u>three columns</u> &nbsp;&#8658;&nbsp; $n = 3$.
+
 +
*<u>three columns</u> &nbsp; &#8658; &nbsp; $n = 3$.
  
  
  
 
'''(3)'''&nbsp; <u>All statements</u> are correct:  
 
'''(3)'''&nbsp; <u>All statements</u> are correct:  
*Since the current information bit $u_i$ affects all three outputs $x_i^{(1)}, \ x_i^{(2)}$ and $x_i^{(3)}$, $\mathbf{G}_0 = (1, 1, 1)$.  
+
*Since the current information bit&nbsp; $u_i$&nbsp; affects all three outputs&nbsp; $x_i^{(1)}, \ x_i^{(2)}$&nbsp; and&nbsp; $x_i^{(3)}$ &nbsp; &rArr; &nbsp;  $\mathbf{G}_0 = (1, 1, 1)$.
*In contrast, $\mathbf{G}_3 = (1, 1, 0)$ states that only the first two inputs are affected by $u_{i-3}$, but not $x_i^{(3)}$.
+
 +
*In contrast,&nbsp; $\mathbf{G}_3 = (1, 1, 0)$&nbsp; states that only the first two inputs are affected by&nbsp; $u_{i-3}$,&nbsp; but not $x_i^{(3)}$.
  
  
  
'''(4)'''&nbsp; Correct is the <u>proposed solution 2</u>:
+
'''(4)'''&nbsp; Correct is the&nbsp; <u>proposed solution 2</u>:
[[File:P_ID2626__KC_Z_3_2d.png|right|frame|Generator matrix $\mathbf{G}$]]
+
[[File:P_ID2626__KC_Z_3_2d.png|right|frame|Generator matrix&nbsp; $\mathbf{G}$]]
*The searched generator matrix $\mathbf{G}$ is shown on the right, where the four partial matrices $\mathbf{G}_0, \ ... , \mathbf{G}_3$ are distinguished by color.  
+
*The searched generator matrix&nbsp; $\mathbf{G}$&nbsp; is shown on the right,&nbsp; where the four partial matrices&nbsp; $\mathbf{G}_0, \ ... , \mathbf{G}_3$&nbsp; are distinguished by color.
 +
 
*The following vector equation gives the result corresponding to the second proposed solution 2:
 
*The following vector equation gives the result corresponding to the second proposed solution 2:
 
:$$\underline{x} = \underline{u} \cdot { \boldsymbol{\rm G}} = (1\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} 1) \cdot { \boldsymbol{\rm G}}. $$
 
:$$\underline{x} = \underline{u} \cdot { \boldsymbol{\rm G}} = (1\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} 1) \cdot { \boldsymbol{\rm G}}. $$
*The code sequence $\underline{x}$ is thereby equal to the modulo 2 sum of the matrix rows 1, 3 and 4.
+
*The code sequence &nbsp; $\underline{x}$ &nbsp; is thereby equal to the modulo-2 sum of the matrix rows 1, 3 and 4.
  
*The three code sequences of the individual branches are distinguished by color. For example, the following applies to the lower output:
+
*The three code sequences of the individual branches are distinguished by color.&nbsp; For example,&nbsp; the following applies to the lower output:
 
:$$\underline{x}^{(3)} =  (1\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} ... \hspace{0.05cm}) \hspace{0.05cm}.$$
 
:$$\underline{x}^{(3)} =  (1\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} ... \hspace{0.05cm}) \hspace{0.05cm}.$$
  
Using the equations given above, this result can be verified:
+
*Using the equations given above,&nbsp; this result can be verified:
 
:$${x}_1^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm}  u_1 + u_{-1} = 1+ (0) = 1 \hspace{0.05cm},$$
 
:$${x}_1^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm}  u_1 + u_{-1} = 1+ (0) = 1 \hspace{0.05cm},$$
 
:$${x}_2^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm}  u_2 + u_{0} = 0+ (0) = 0 \hspace{0.05cm},$$
 
:$${x}_2^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm}  u_2 + u_{0} = 0+ (0) = 0 \hspace{0.05cm},$$
Line 105: Line 109:
 
:$${x}_5^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm}  u_5 + u_{3} = 0+ 1 = 1 \hspace{0.05cm}.$$
 
:$${x}_5^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm}  u_5 + u_{3} = 0+ 1 = 1 \hspace{0.05cm}.$$
  
Notes:  
+
<u>Notes:</u>
*The memory preallocation with zeros is taken into account here: $u_0 = u_{&ndash;1} = 0$.
+
*The memory preallocation with zeros is taken into account here:&nbsp; $u_0 = u_{&ndash;1} = 0$.
*If, as assumed here, the information sequence is limited to four bits, then ones can occur in the code sequence up to the position $(4 + m) \cdot n = 21$.
+
 
 +
*If,&nbsp; as assumed here,&nbsp; the information sequence is limited to four bits,&nbsp; then&nbsp; "ones"&nbsp; can occur in the code sequence up to the position&nbsp; $(4 + m) \cdot n = 21$.
 
{{ML-Fuß}}
 
{{ML-Fuß}}
  

Revision as of 17:14, 9 November 2022

Convolutional encoder with parameters  $k = 1, \ n = 3, \ m = 3$.

The presented convolutional encoder is defined by the parameters 

  • $k = 1$  $($only one information sequence  $\underline{u})$,  and 
  • $n = 3$  $($three code sequences  $\underline{x}^{(1)}, \ \underline{x}^{(2)}, \ \underline{x}^{(3)}).$ 


From the number of memory cells,  the memory  $m = 3$.

With the information bit  $u_i$  to the coding step  $i$,  the following code bits are obtained:

$$x_i^{(1)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_{i} + u_{i-1} + u_{i-3}\hspace{0.05cm},$$
$$x_i^{(2)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_{i} + u_{i-1} + u_{i-2} + u_{i-3} \hspace{0.05cm},$$
$$x_i^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_{i} + u_{i-2} \hspace{0.05cm}.$$


From this,  partial matrices   $\mathbf{G}_l$   can be derived,  as described on the page  "Division of the generator matrix into partial matrices" .

  • For the generator matrix can thus be written:
$$ { \boldsymbol{\rm G}}=\begin{pmatrix} { \boldsymbol{\rm G}}_0 & { \boldsymbol{\rm G}}_1 & { \boldsymbol{\rm G}}_2 & \cdots & { \boldsymbol{\rm G}}_m & & & \\ & { \boldsymbol{\rm G}}_0 & { \boldsymbol{\rm G}}_1 & { \boldsymbol{\rm G}}_2 & \cdots & { \boldsymbol{\rm G}}_m & &\\ & & { \boldsymbol{\rm G}}_0 & { \boldsymbol{\rm G}}_1 & { \boldsymbol{\rm G}}_2 & \cdots & { \boldsymbol{\rm G}}_m &\\ & & & \ddots & \ddots & & & \ddots \end{pmatrix}\hspace{0.05cm}.$$
  • For the code sequence   $\underline{x} = (x_1^{(1)}, \ x_1^{(2)}, \ x_1^{(3)}, \ x_2^{(1)}, \ x_2^{(2)}, \ x_2^{(3)}, \ \text{...})$   holds:
$$\underline{x} = \underline{u} \cdot { \boldsymbol{\rm G}} \hspace{0.05cm}.$$



Hints:



Questions

1

From how many partial matrices  $\mathbf{G}_l$  is the matrix  $\mathbf{G}$  composed?

${\rm Number \ of \ partial \ matrices} \ = \ $

2

What dimension do the partial matrices  $\mathbf{G}_l$ have?

${\rm Rows \ of \ the \ partial \ matrices} \hspace{0.425cm} = \ $

${\rm Columns \ of \ the \ partial \ matrices} \ = \ $

3

Which statements are correct?

It holds  $\mathbf{G}_0 = (1, 1, 1)$.
It holds  $\mathbf{G}_ 1 = (1, 1, 0)$.
It holds  $\mathbf{G}_2 = (0, 1, 1)$.
It holds  $\mathbf{G}_3 = (1, 1, 0)$.

4

Create the generator matrix  $\mathbf{G}$  with five rows and fifteen columns.  What code sequence results for  $\underline{u} = (1, 0, 1, 1, 0)$?

It holds  $\underline{x} = (1, 0, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 0, \text{...}).$
It holds  $\underline{x} = (1, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 1, 0, 1, \text{...}).$
It holds  $\underline{x} = (0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, \text{...}).$


Solution

(1)  For the index  $l$  of the partial matrices:  $0 ≤ l ≤ m$.

  • The coder under consideration has memory  $m = 3$.
  • Thereby  four partial matrices  are to be considered.


(2)  Each partial matrix  $\mathbf{G}_l$  consists of

  • one row   ⇒   $k = 1$,  and
  • three columns   ⇒   $n = 3$.


(3)  All statements are correct:

  • Since the current information bit  $u_i$  affects all three outputs  $x_i^{(1)}, \ x_i^{(2)}$  and  $x_i^{(3)}$   ⇒   $\mathbf{G}_0 = (1, 1, 1)$.
  • In contrast,  $\mathbf{G}_3 = (1, 1, 0)$  states that only the first two inputs are affected by  $u_{i-3}$,  but not $x_i^{(3)}$.


(4)  Correct is the  proposed solution 2:

Generator matrix  $\mathbf{G}$
  • The searched generator matrix  $\mathbf{G}$  is shown on the right,  where the four partial matrices  $\mathbf{G}_0, \ ... , \mathbf{G}_3$  are distinguished by color.
  • The following vector equation gives the result corresponding to the second proposed solution 2:
$$\underline{x} = \underline{u} \cdot { \boldsymbol{\rm G}} = (1\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} 1) \cdot { \boldsymbol{\rm G}}. $$
  • The code sequence   $\underline{x}$   is thereby equal to the modulo-2 sum of the matrix rows 1, 3 and 4.
  • The three code sequences of the individual branches are distinguished by color.  For example,  the following applies to the lower output:
$$\underline{x}^{(3)} = (1\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 0\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} 1\hspace{0.05cm},\hspace{0.05cm} ... \hspace{0.05cm}) \hspace{0.05cm}.$$
  • Using the equations given above,  this result can be verified:
$${x}_1^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_1 + u_{-1} = 1+ (0) = 1 \hspace{0.05cm},$$
$${x}_2^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_2 + u_{0} = 0+ (0) = 0 \hspace{0.05cm},$$
$${x}_3^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_3 + u_{1} = 1+1 = 0 \hspace{0.05cm},$$
$${x}_4^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_4 + u_{2} = 1+0 = 1 \hspace{0.05cm},$$
$${x}_5^{(3)} \hspace{-0.15cm} \ = \ \hspace{-0.15cm} u_5 + u_{3} = 0+ 1 = 1 \hspace{0.05cm}.$$

Notes:

  • The memory preallocation with zeros is taken into account here:  $u_0 = u_{–1} = 0$.
  • If,  as assumed here,  the information sequence is limited to four bits,  then  "ones"  can occur in the code sequence up to the position  $(4 + m) \cdot n = 21$.