EC8553 Discrete Time Signal Processing

Unit I Discrete Fourier Transform

Discrete Fourier Transform and its inverse



 =  
 ⁄ ,  = 0, 1, 2, … … … ,  − 1



1
 = 
  ⁄ ,  = 0, 1, 2, … … … ,  − 1


Properties of DFT

If
 is an N point DFT of , then

Periodicity property

 +  =   !"" 

 +  =
  !"" 
Proof:



We know that

 =  
 ⁄




 +  =  
 #⁄




 +  =  
 ⁄
 ⁄




 +  =  
 ⁄
 

We know that
 
=1



Hence

 +  =  
 ⁄


 +  =


If
 and $ are the N point DFT of  and % respectively, and ! and & are arbitrary constants either
Linearity property

'()*! + &%+ = !
 + &$

real or complex valued, then



Proof:

'()*+ =  
 ⁄




'()*! + &%+ = *! + &%+

 ⁄


Discrete Time Signal Processing 2

 
 


'()*! + &%+ =  !

 ⁄
+  &%
 ⁄
 

 


'()*! + &%+ = !  

 ⁄ + &  %
 ⁄
 
'()*! + &%+ = !
 + &$

If
 and $ are the N point DFT of  and % respectively, then
Circular Convolution property

'()* ⊗ %+ =
$



where,

 ⊗ % =  -% − -, -. 

/



Proof:

'()*+ =  
 ⁄




'()* ⊗ %+ = * ⊗ %+

 ⁄


 


'()* ⊗ %+ =   -% − -, -. 

 ⁄
 /

 


'()* ⊗ %+ =   -% +  − -

 ⁄
 /
Put " =  +  − -

 ="+-−
If  = 0, " =  − -, -.  = −- = 0 (if - = 0)
If  =  − 1, " =  +  − 1 − -, -.  = 2 − 1 − -, -. 

" =  − 1 − - =  − 1 (if - = 0)

 


'()* ⊗ %+ =   -%"

 0#/
⁄
0 /

 


'()* ⊗ %+ =   -%"

 0⁄
 /⁄  ⁄
0 /

 


'()* ⊗ %+ =   -%"

 0⁄
 /⁄  
0 /
We know that  
=1

 


'()* ⊗ %+ =   -%"

 0⁄
 /⁄
0 /

Discrete Time Signal Processing 3

 
 


'()* ⊗ %+ =  -

 /⁄
 %"
 0⁄
0 /
'()* ⊗ %+ =
$

If
 is the N point DFT of the sequence , then

Time reversal property

'()*−, -. + =
−, -. 

'()* − + =
 − 



Proof:

'()*+ =  
 ⁄




'()* − + =   − 

 ⁄

Put - =  − 

 =−-
If  = 0, - = , -.  = 0
If  =  − 1, - =  −  + 1 = 1, -.  =  − 1



Hence,

'()* − + =  -

 
/⁄
/



'()* − + =  -

 ⁄  /⁄
/



'()* − + =  -

 
 /
⁄
/
We know that
  =
 / = 1



'()* − + =  -

 /
⁄
 /
/



'()* − + =  -

 /
⁄
 /⁄
/



'()* − + =  -

 /
⁄
/
'()* − + =
 − 

If
 is the DFT of the sequence , then

Circular time shift property

'()* − ", -. + =


 0⁄
where  − ", -.  =  +  − "
Proof:

Discrete Time Signal Processing 4

 


'()*+ =  
 ⁄




'()* − ", -. + =   − ", -. 

 ⁄




'()* − ", -. + =   +  − "

 ⁄

Put - =  +  − "

 ="−+-
If  = 0, - =  − ", -.  = −" = 0 (if " = 0)
If  =  − 1, - =  +  − 1 − ", -.  = 2 − 1 − ", -.  =  − 1 − "

- =  − 1 (if " = 0)



Hence

'()* − ", -. + =  -

 0
#/⁄
/



'()* − ", -. +  -

 0⁄  ⁄
 /⁄
/



'()* − ", -. + =  -

 0⁄  
 /⁄
/
We know that   = 1



'()* − ", -. + =  -

 0⁄
 /⁄
/



'()* − ", -. + =

 0⁄
 -
 /⁄
/
'()* − ", -. + =
 0⁄



If
 is the DFT of the sequence , then

Circular frequency shift property

'()1  0⁄ 2 =
 − ", -.

where,
 − ", -.  =
 +  − "



Proof:

 =  
 ⁄




 +  − " =  

 #
0⁄




 +  − " =  

 ⁄
 ⁄  0⁄


Discrete Time Signal Processing 5

 


 +  − " =  

 
 ⁄  0⁄

We know that,
  = 1



 +  − " = 1  0⁄ 2

 ⁄

'()1  0⁄
2 =
 +  − "

If '()*+ =
, then

Complex conjugate property

'()* ∗ + =
∗ −, -. =
∗  − 


 
 ∗
1 1
4'()*
∗ + = 
∗   ⁄ = 5 
  
⁄ 6
 
 



Proof:

 =  
 ⁄




 −  =  
 
⁄




 −  =  
 ⁄  ⁄




 −  =  
   ⁄

We know that
 
=1



 −  =    ⁄



 ∗

∗  −  = 5   ⁄ 6




∗  −  =   ∗ 
 ⁄

∗ 

−  = '()* ∗ +

If '()* + =
 and '()* + =
 , then

Circular correlation property

1
'()* 7 "+ = 87  =
 
∗ 




where,

7 " =   ∗  − ", -.



Discrete Time Signal Processing 6

 

Proof:

'()*+ =  
 ⁄




'()* 7 "+ = 7 "

 0⁄

0

 


'()* 7 "+ =    ∗  − ", -.

 0⁄
0 

 


'()* 7 "+ =    ∗  +  − "

 0⁄
0 
Put - =  +  − "

" =+−-
If " = 0, - =  + , -.  = − = 0 (if  = 0)
If " =  − 1, - =  +  −  + 1, -.  =  − 1 −  =  − 1 (if  = 0)


 

Hence,

'()* 7 "+ =    ∗ -

 #
/⁄
/ 

 


'()* 7 "+ =    ∗ -

 ⁄
 ⁄  /⁄
/ 

 


'()* 7 "+ =    ∗ -

 
 ⁄  /⁄
/ 
We know that
 
=1

 


'()* 7 "+ =    ∗ -

 ⁄  /⁄
/ 

 


'()* 7 "+ =   

 ⁄  ∗ -  /⁄
 /

 
 ∗

'()* 7 "+ =   

 ⁄ 5   -
 /⁄ 6
 /
'()* 7 "+ =
 
 
∗

If '()* + =
 and '()* + =
 , then

Multiplication of two sequences property

1
'()*+ = '()*  + = *
⨂
+
 



where,

 ⨂
  = 
 "
  − ", -. 

0
Proof:
Discrete Time Signal Processing 7
 

1 1
*
 ⨂
 + = 
 "
  − ", -. 
 
0


1 1
*
 ⨂
 + = 
 "
  +  − "
 
0
We know that the IDFT of the sequence
 is given by


1
4'()*
+ = 
  ⁄




1 1 1
4'() : *
 ⨂
 +; =  *
 ⨂
 +  ⁄
  


 

1 1 1
4'() : *
 ⨂
 +; =  
 "
  +  − "  ⁄
  
 0
Put - =  +  − "

 ="+-−
If  = 0, - =  − ", -.  = −" = 0 (if " = 0)
If  =  − 1, - = 2 − " − 1, -.  =  − 1 − " =  − 1 (if " = 0)

 

1 1 1
4'() : *
 ⨂
 +; =  
 "
 -  0#/
⁄
  
/ 0

 

1 1 1
4'() : *
 ⨂
 +; = 
 "  0⁄ 
 -  /⁄
 ⁄
  
0 /

 

1 1 1
4'() : *
 ⨂
 +; = 
 "  0⁄ 
 -  /⁄
 
  
0 /
We know that
 
=1

 

1 1 1
4'() : *
 ⨂
 +; = 
 "  0⁄ 
 -  /⁄
  
0 /
1
4'() : *
 ⨂
 +; =   

1
'()*  + = *
 ⨂
 +


For complex valued sequence and %, if '()*+ =
 and '()*%+ = $,
Parseval’s property


 

∗ 
1
 % = 
$ ∗ 

 
If  = %, then

 

1
|| = |
|

 
Proof:
Discrete Time Signal Processing 8
 


 =  
 ⁄




$ =  %
 ⁄




$ ∗  =  % ∗   ⁄



 

∗ 

$ =  
 ⁄
 % ∗   ⁄
 



$ ∗  =  % ∗ 



 
 



$ ∗  =  1  % ∗ 

  

 



$ ∗  =   % ∗ 

 

 

1

$ ∗  =  % ∗ 

 
If  = %,

 

1


∗  =   ∗ 

 

 

1
|
| = ||

 
Example 1: Find the 4 point DFT for the sequence => = ?@, A, B, CD.
Given: = ?1, 2, 3, 4D, = 4
To find:



Formula:

 =  
 ⁄ ,  = 0, 1, 2, … … … ,  − 1


H
Solution:

 =  
 ⁄G ,  = 0, 1, 2, 3

H

 =  
 ⁄ ,  = 0, 1, 2, 3


 = 0 + 1
 ⁄ + 2
  + 3
H ⁄ ,  = 0, 1, 2, 3

 = 1 + 2
 ⁄ + 3
  + 4
H ⁄ ,  = 0, 1, 2, 3
For  = 0,

Discrete Time Signal Processing 9


0 = 1 + 2 + 3 + 4 = 10
For  = 1,

1 = 1 + 2
 ⁄ + 3
 + 4
H ⁄

1 = 1 − 2I − 3 + 4I

1 = −2 + 2I
For  = 2,

2 = 1 + 2
 + 3
 + 4
H

2 = 1 − 2 + 3 − 4 = −2
For  = 3,

3 = 1 + 2
H ⁄ + 3
H + 4
J ⁄

3 = 1 + 2I − 3 − 4I

3 = −2 − 2I

 = ?10, −2 + 2I, −2, −2 − 2ID

Result:

Example 2: Find the 8 point DFT of the sequence => = KL, @, A, B, B, A, @, LM.

 = K0, 1, 2, 3, 3, 2, 1, 0M
Given:

=8
To find:




Formula:

 =  
 ⁄ ,  = 0, 1, 2, … … … ,  − 1


P
Solution:

 =  
 ⁄O ,  = 0, 1, 2, 3, 4, 5, 6, 7

P

 =  
 ⁄G ,  = 0, 1, 2, 3, 4, 5, 6, 7


 = 0 + 1 + 2

 ⁄G
 ⁄
+ 3
H ⁄G + 4
  + 5
T ⁄G + 6
H ⁄
+ 7
P ⁄G ,  = 0, 1, 2, 3, 4, 5, 6, 7

 =
 ⁄G + 2
 ⁄ + 3
H ⁄G + 3
  + 2
T ⁄G +
H ⁄ ,  = 0, 1, 2, 3, 4, 5, 6, 7
For  = 0,

0 = 1 + 2 + 3 + 3 + 2 + 1 = 12
For  = 1,

1 =
 ⁄G + 2
 ⁄ + 3
H ⁄G + 3
 + 2
T ⁄G +
H ⁄

1 = 0.707 − I0.707 − I2 − 2.121 − I2.121 − 3 − 1.414 + I1.414 + I

1 = −5.828 − I2.414
For  = 2,

2 =
 ⁄ + 2
 + 3
H ⁄ + 3
 + 2
T ⁄ +
H

2 = −I − 2 + I3 + 3 − I2 − 1

2 = 0
For  = 3,

Discrete Time Signal Processing 10


3 =
H ⁄G + 2
H ⁄ + 3
J ⁄G + 3
H + 2
T ⁄G +
J ⁄

3 = −0.707 − I0.707 + I2 + 2.121 − I2.121 − 3 + 1.414 + I1.414 − I

3 = −0.172 − I0.414
For  = 4,

4 =
 + 2
 + 3
H + 3
G + 2
T +
V

4 = −1 + 2 − 3 + 3 − 2 + 1 = 0

5 = −0.172 + I0.414;
6 = 0;
7 = −5.828 + I2.414

The remaining coefficients are

 = ?12, − 5.828 − I2.414, 0, − 0.172 − I0.414, 0, − 0.172 + I0.414, 0, − 5.828 + I2.414D
Result:

Example 3: Determine the 8 point DFT of the Sequence => = K@, @, @, @, @, L, LM.
Given:  = K1, 1, 1, 1, 1, 0, 0M, = 8
To find:




Formula:

 =  
 ⁄ ,  = 0, 1, 2, … … … ,  − 1


P
Solution:

 =  
 ⁄O ,  = 0, 1, 2, 3, 4, 5, 6, 7

P

 =  
 ⁄G ,  = 0, 1, 2, 3, 4, 5, 6, 7


 = 0 + 1
 ⁄G + 2
 ⁄ + 3
H ⁄G + 4
  + 5
T ⁄G + 6
H ⁄
+ 7
P ⁄G ,  = 0, 1, 2, 3, 4, 5, 6, 7

 = 1 +
 ⁄G +
 ⁄ +
H ⁄G +
  ,  = 0, 1, 2, 3, 4, 5, 6, 7
For  = 0,

0 = 1 + 1 + 1 + 1 + 1 = 5
For  = 1,

1 = 1 +
 ⁄G +
 ⁄ +
H ⁄G +


1 = 1 + 0.707 − I0.707 − I − 0.707 − I0.707 − 1

1 = −I2.414
For  = 2,

2 = 1 +
 ⁄ +
 +
H ⁄ +


2 = 1 − I − 1 + I + 1

2 = 1
For  = 3,

3 = 1 +
H ⁄G +
H ⁄ +
J ⁄G +
H

3 = 1 − 0.707 − I0.707 + I + 0.707 − I0.707 − 1

3 = −I0.414
For  = 4,

4 = 1 +
 +
 +
H +
G
Discrete Time Signal Processing 11

4 = 1 − 1 + 1 − 1 + 1 = 1

5 = I0.414;
6 = 1;
7 = I2.414

The remaining coefficients are

Result:
 = K5, − I2.414, 1, − I0.414, 1, I0.414, 1, I2.414M

Example 4: Find 4 point DFT for the sequence => = K@, − @, A, − AM.

0 1 1 1 1 0
Solution:

1 1 −I −1 I 1
X Y=X YX Y

2 1 −1 1 −1 2

3 1 I −1 −I 3

0 1 1 1 1 1

1 1 −I −1 I −1
X Y=X YX Y

2 1 −1 1 −1 2

3 1 I −1 −I −2

0 1−1+2−2

1 1 + I − 2 − I2
X Y=X Y

2 1+1+2+2

3 1 − I − 2 + I2

0 0

1 −1 − I
X Y=X Y

2 6

3 −1 + I
Result:
 = K0, − 1 − I, 6, − 1 + IM
Example 5: Find 8 point DFT for the sequence => = K@, @, @, @, A, A, A, AM

0 1 1 1 1 1 1 1 1 0
Solution:
\ _ \ _ \ _
[
1^ [1 0.707 − I0.707 −I −0.707 − I0.707 −1 −0.707 + I0.707 I 0.707 + I0.707 [1^
^
[
2^ [1 −I −1 I 1 −I −1 I
^[
2^
[
3^ [1 −0.707 − I0.707 I 0.707 − I0.707 −1 0.707 + I0.707 −I −0.707 + I0.707^ [3^
[
4^ = 1 −1 1 −1 1 −1 1 −1 [ ^
[ ^ [1 ^ [4^

5 −0.707 + I0.707 −I 0.707 + I0.707 −1 0.707 − I0.707 I −0.707 − I0.707^ 5
[ ^ [ [ ^
[
6^ [1 I −1 −I 1 I −1 −I ^ [6^
Z
7] Z1 0.707 + I0.707 I −0.707 + I0.707 −1 −0.707 − I0.707 −I 0.707 − I0.707 ] Z7]

0 1 1 1 1 1 1 1 1 1
\ _ \
[
1^ [1 0.707 − I0.707 −I −0.707 − I0.707 −1 −0.707 + I0.707 I 0.707 + I0.707 _ \1_
^[ ^
[
2^ [1 −I −1 I 1 −I −1 I
^ [1^
[
3^ [1 −0.707 − I0.707 I 0.707 − I0.707 −1 0.707 + I0.707 −I −0.707 + I0.707^ [1^
[
4^ = 1 −1 1 −1 1 −1 1 −1
[ ^ [1 ^ [2^

5 −0.707 + I0.707 −I 0.707 + I0.707 −1 0.707 − I0.707 I −0.707 − I0.707^ [2^
[ ^ [
[
6^ [1 I −1 −I 1 I −1 −I ^ [2^
Z
7] Z1 0.707 + I0.707 I −0.707 + I0.707 −1 −0.707 − I0.707 −I 0.707 − I0.707 ] Z2]

Discrete Time Signal Processing 12


0 1+1+1+1+2+2+2+2
\ _ \ _

1
[ ^ [1 + 0.707 − I0.707 − I − 0.707 − I0.707 − 2 − 1.414 + I1.414 + I2 + 1.414 + I1.414^
[
2^ [ 1 − I − 1 + I + 2 − I2 − 2 + I2
^
[
3^ [1 − 0.707 − I0.707 + I + 0.707 − I0.707 − 2 + 1.414 + I1.414 − I2 − 1.414 + I1.414^
[
4^ = 1−1+1−1+2−2+2−2
[ ^ [1 − 0.707 + I0.707 − I + 0.707 + I0.707 − 2 + 1.414 − I1.414 + I2 − 1.414 − I1.414^
[
5^ [ ^
[
6^ [ 1 + I − 1 − I + 2 + I2 − 2 − I2 ^
Z
7] Z1 + 0.707 + I0.707 + I − 0.707 + I0.707 − 2 − 1.414 − I1.414 − I2 + 1.414 − I1.414]

 = ?12, − 1 + I2.414, 0, − 1 + I0.414, 0, − 1 − I0.414, 0, − 1, − I2.414D
Example 6: Find the IDFT for the sequence `a = K@L, − A + Ab, − A, − A − AbM.



1
Solution:

 = 
  ⁄ ,  = 0, 1, 2, … … … ,  − 1



H
1
 = 
  ⁄G ,  = 0, 1, 2, 3
4

H
1
 = 
  ⁄ ,  = 0, 1, 2, 3
4

1
 = 1
0 +
1  ⁄ +
2   +
3 H ⁄ 2,  = 0, 1, 2, 3
4
1
 = 110 + −2 + I2  ⁄ − 2   + −2 − I2 H ⁄ 2,  = 0, 1, 2, 3
4
1 1
0 = *10 + −2 + I2 − 2 + −2 − I2+ = *4+ = 1
4 4
1
1 = 110 + −2 + I2  ⁄ − 2  + −2 − I2 H ⁄ 2
4
1
1 = *10 + −2 + I2I − 2−1 + −2 − I2−I+
4
1 1
1 = *10 − I2 − 2 + 2 + I2 − 2+ = *8+ = 2
4 4
1
2 = 110 + −2 + I2  − 2  + −2 − I2 H 2
4
1
2 = *10 + −2 + I2−1 − 2 + −2 − I2−1+
4
1 1
2 = *10 + 2 − I2 − 2 + 2 + I2+ = *12+ = 3
4 4
1
3 = 110 + −2 + I2 H ⁄ − 2 H + −2 − I2 J ⁄ 2
4
1
3 = *10 + −2 + I2−I − 2−1 + −2 − I2I+
4
1 1
3 = *10 + I2 + 2 + 2 − I2 + 2+ = *16+ = 4
4 4
Result:  = K1, 2, 3, 4M

Discrete Time Signal Processing 13

Example 7: Find the IDFT for the sequence `a = K@A, @ + b, L, + b, L, − b, L, @ − bM.



1
Solution:

 = 
  ⁄ ,  = 0, 1, 2, … … … ,  − 1



P
1
 = 
  ⁄O ,  = 0, 1, 2, … … … , 7
8

P
1
 = 
  ⁄G ,  = 0, 1, 2, … … … , 7
8

1
 = 1
0 +
1  ⁄G +
2  ⁄ +
3 H ⁄G +
4   +
5 T ⁄G +
6 H ⁄
8
+
7 P ⁄G 2,  = 0, 1, 2, … … … , 7
1
 = 112 + 1 + I  ⁄G + I H ⁄G − I T ⁄G + 1 − I P ⁄G 2,  = 0, 1, 2, … … … , 7
8
1 14
0 = *12 + 1 + I + I − I + 1 − I+ = = 1.75
8 8
1
1 = 112 + 1 + I  ⁄G + I H ⁄G − I T ⁄G + 1 − I P ⁄G 2
8
1
1 = *12 + 1 + I0.707 + I0.707 + I−0.707 + I0.707 − I−0.707 − I0.707
8
+ 1 − I0.707 − I0.707+
1
1 = *12 + 0.707 + I0.707 + I0.707 − 0.707 − I0.707 − 0.707 + I0.707 − 0.707 + 0.707 − I0.707
8
− I0.707 − 0.707+
1 1
1 = *12 − 0.707 − 0.707+ = *10.586+ = 1.3233
8 8
1
2 = 112 + 1 + I  ⁄ + I H ⁄ − I T ⁄ + 1 − I P ⁄ 2
8
1
2 = *12 + 1 + II + I−I − II + 1 − I−I+
8
1 1
2 = *12 + I − 1 + 1 + 1 + −I − 1+ = *12+ = 1.5
8 8
1
3 = 112 + 1 + I H ⁄G + I J ⁄G − I T ⁄G + 1 − I  ⁄G 2
8
1
3 = *12 + 1 + I−0.707 + I0.707 + I0.707 + I0.707 − I0.707 − I0.707
8
+ 1 − I−0.707 − I0.707+
1
3 = *12 − 0.707 + I0.707 − I0.707 − 0.707 + I0.707 − 0.707 − I0.707 − 0.707 + −0.707 − I0.707
8
+ I0.707 − 0.707+
1 1
3 = *12 − 0.707 − 0.707 − 0.707 − 0.707 + −0.707 − 0.707+ = *7.758+ = 0.9698
8 8
1
4 = 112 + 1 + I  + I H − I T + 1 − I P 2
8
Discrete Time Signal Processing 14
 1
4 = *12 + 1 + I−1 + I−1 − I−1 + 1 − I−1+
8
1 1 1
4 = *12 − 1 − I − I + I − 1 + I+ = *12 − 1 − 1+ = *10+ = 1.25
8 8 8
1
5 = 112 + 1 + I T ⁄G + I T ⁄G − I T ⁄G + 1 − I HT ⁄G 2
8
1
5 = *12 + 1 + I−0.707 − I0.707 + I0.707 − I0.707 − I0.707 + I0.707
8
+ 1 − I−0.707 + I0.707+
1
5 = *12 − 0.707 − I0.707 − I0.707 + 0.707 + I0.707 + 0.707 − I0.707 + 0.707 − 0.707 + I0.707
8
+ I0.707 + 0.707+
1 1
5 = *12 + 0.707 + 0.707+ = *13.414+ = 1.6768
8 8
1
6 = 112 + 1 + I H ⁄ + I J ⁄ − I T ⁄ + 1 − I  ⁄ 2
8
1
6 = *12 + 1 + I−I + II − I−I + 1 − II+
8
1 1
6 = *12 − I + 1 − 1 − 1 + I + 1+ = *12+ = 1.5
8 8
1
7 = 112 + 1 + I P ⁄G + I  ⁄G − I HT ⁄G + 1 − I GJ ⁄G 2
8
1
7 = *12 + 1 + I0.707 − I0.707 + I−0.707 − I0.707 − I−0.707 + I0.707
8
+ 1 − I0.707 + I0.707+
1
7 = *12 + 0.707 − I0.707 + I0.707 + 0.707 − I0.707 + 0.707 + I0.707 + 0.707 + 0.707 + I0.707
8
− I0.707 + 0.707+
1 1
7 = *12 + 0.707 + 0.707 + 0.707 + 0.707 + 0.707 + 0.707+ = *16.242+ = 2.0303
8 8
Result:  = K1.75, 1.3233, 1.5, 0.9698, 1.25, 1.6768, 1.5, 2.0303M
Example 8: Find IDFT for the sequence `a = K@, L, L, b, L, − b, L, L M.

0 1 1 1 1 1 1 1 1
0
Solution:
\ _ \1 0.707 + I0.707 _ \ _
[1^ I −0.707 + I0.707 −1 −0.707 − I0.707 −I 0.707 − I0.707 [
1^
[ ^
[2^ 1 I −1 −I 1 I −1 −I
2^
[ ^[
[3^ 1 [1 −0.707 + I0.707 −I 0.707 + I0.707 −1 0.707 − I0.707 I −0.707 − I0.707^ [
3^
=
[4^ 8 1 [ ^
[ −1 1 −1 1 −1 1 −1 ^ [
4^
[ ^ 1 −0.707 − I0.707 I 0.707 − I0.707 −1 0.707 + I0.707 −I −0.707 + I0.707^
5
[5^ [ [ ^
[6^ [ 1 −I −1 I 1 −I −1 I ^ [
6^
Z7] Z1 0.707 − I0.707 −I −0.707 − I0.707 −1 −0.707 + I0.707 I 0.707 + I0.707 ] Z
7]

Discrete Time Signal Processing 15

 0 1 1 1 1 1 1 1 1 1
\ _ \1
[
1
^ 0.707 + I0.707 I −0.707 + I0.707 −1 −0.707 − I0.707 −I 0.707 − I0.707 _ \ 0 _
[ ^[ ^
[2^ 1 I −1 −I 1 I −1 −I
[ ^[ 0 ^
[3^ 1 [1 −0.707 + I0.707 −I 0.707 + I0.707 −1 0.707 − I0.707 I −0.707 − I0.707^ [ I ^
[4^ = 8 1 −1 1 −1 1 −1 1 −1
[ ^ [ ^[ 0 ^
[5^ [1 −0.707 − I0.707 I 0.707 − I0.707 −1 0.707 + I0.707 −I −0.707 + I0.707^ [−I^
[6^ [1 −I −1 I 1 −I −1 I ^[ 0 ^
Z7] Z1 0.707 − I0.707 −I −0.707 − I0.707 −1 −0.707 + I0.707 I 0.707 + I0.707 ] Z 0 ]
0 1+I−I
\ _ \ _
1 1 + I−0.707 + I0.707 − I−0.707 − I0.707
[ ^ [ ^
[2^ [ 1 + I−I − II ^
[3^ 1 [ 1 + I0.707 + I0.707 − I0.707 − I0.707 ^
[4^ = 8 [ 1 + I−1 − I−1 ^
[ ^ [ ^
[5^ [ 1 + I0.707 − I0.707 − I0.707 + I0.707 ^
[6^ [ 1 + II − I−I ^
Z7] Z1 + I−0.707 − I0.707 − I−0.707 + I0.707]
0 1 1 0.125
\ _ \1 − I0.707 − 0.707 + I0.707 − 0.707_ \−0.414_ \−0.0518_
1
[ ^ [ ^ [ ^ [ ^
[2^ [
1+1+1
^ [
3
^ [ 0.375 ^
[3^ 1 [1 + I0.707 − 0.707 − I0.707 − 0.707^ 1 [−0.414^ [−0.0518^
[4^ = 8 1−I+I
= =
8 [ 1 ^ [ 0.125 ^
[ ^ [ ^
[5^ [1 + I0.707 + 0.707 − I0.707 + 0.707^ [ 2.414 ^ [ 0.3018 ^
[6^ [ 1 − 1 − 1 ^ [ −1 ^ [ −0.125 ^
Z7] Z1 − I0.707 + 0.707 + I0.707 + 0.707] Z 2.414 ] Z 0.3018 ]

The N point DFT of the sequence  is given by

8 point DFT using DITFFT Algorithm




 =  
 ⁄ ,  = 0, 1, 2, … … … ,  − 1

Let us assume Twiddle factor d =
 ⁄ . Substitute twiddle factor in the sequence
,



 =  d ,  = 0, 1, 2, … … … ,  − 1

In DITFFT algorithm, the time domain sequence  is decimated into even and odd component.

 


 =  d +  d

,efe ,ghh
 



 

 =  2d +  2 + 1d

#

 
 



 

 =  2d +  2 + 1d d

 

Discrete Time Signal Processing 16

  



 

 =  2d + d  2 + 1d

 

d =
 ⁄
We know that

d = i
 ⁄ j


d = i
G ⁄ j


d = i
 ⁄⁄ j


d = id⁄ j


d = d
⁄

 
Hence



 

 =  2d
⁄ + d  2 + 1d⁄
 

 


 = k + d l, 0 ≤  ≤ −1
2
For the remaining values of , i.e.,  ≤  ≤  − 1,  is replaced by  + 
 

 
p# q  

 = k n + o + d  l n + o , ≤  ≤  − 1
2 2 2
 
 

 = k n + o + d d l n + o , ≤  ≤  − 1
2 2 2

d =
 ⁄
Again, we know that

 
d = i
 ⁄ j 

d =
 ⁄

d =


d = −1

  
Hence

 = k n + o − d l n + o , ≤  ≤  − 1
2 2 2


Thus,
k + d l, 0 ≤  ≤ −1

 = r 2
  
k n + o − d l n + o, ≤  ≤  − 1
2 2 2
Here k and l are the  point DFT and is given by


Discrete Time Signal Processing 17

 




k =  2d
⁄






l =  2 + 1d

⁄

Again repeat the same procedure for k and l and decomposes

G

point DFT. As a result we can get,

s + d⁄ t, 0 ≤  ≤ − 1

k = r 4
   
s n + o − d⁄ t n + o, ≤  ≤ − 1
4 4 4 2


and
u + d⁄ ', 0 ≤  ≤
−1
l = r 4
   
u n + o − d⁄ ' n + o, ≤  ≤ − 1
4 4 4 2
where s, t, u, ' are the G point DFT and are given by the following equations


 



G G

⁄G ; t =  4 + 2d⁄G

s =  4d 

 
 



G G

u =  4 + 1d

⁄G ; ' =  4 + 3d⁄G


 

Discrete Time Signal Processing 18

Example 1: Find the 8 point DFT of the sequence => = ?@, A, B, B, A, @, −@, −AD using DITFFT

Given:  = ?1, 2, 3, 3, 2, 1, −1, −2D

algorithm.

To find:
 using DITFFT Algorithm

v = vG = vO = 1; vG = vO = −I; vO = 0.707 − I0.707; vOH = −0.707 − I0.707
Formula:

Solution:

0 = 1 1 + 2 = 3 3 + 2 = 5 5+4=9
0

Input 2 point DFT 4 point DFT 8 point DFT Output

4 = 2 1 − 2 = −1 −1 + 4−I = −1 −1 − I4 + 1 − I50.707
1

− I4 − I0.707
2 = 3 3 + −1 = 2 3 − 2 = 1 1 + 2−I = 1 − I2
2
6 3 − −1 = 4 −1 − 4−I = −1 −1 + I4 + 1 + I5−0.707
3
= −1 + I4 − I0.707
1 = 2 2 + 1 = 3 3 + 1 = 4 5−4=1
4
5 = 1 2 − 1 = 1 1 + 5−I = 1 − I5 −1 − I4 − 1 − I50.707
5
− I0.707
3 = 3 3 + −2 = 1 3 − 1 = 2 1 − 2−I = 1 + I2
6
7 3− −2 = 5 1 − 5−I = 1 + I5 −1 1
+ I4 − + I5−0.707
7
= −2 − I0.707

 = ?9, −3.828 − I8.242, 1 − I2, 1.828 − I0.242, 1, 1.828 + I0.242, 1 + I2, −3.828 + I8.242D
Result:

The N point DFT of the sequence  is given by

8 point DFT using DIFFFT Algorithm




 =  
 ⁄ ,  = 0, 1, 2, … … … ,  − 1

Let us assume Twiddle factor d =
 ⁄ . Substitute twiddle factor in the sequence
,
Discrete Time Signal Processing 19
 


 =  d ,  = 0, 1, 2, … … … ,  − 1

In DIFFFT algorithm, the frequency domain sequence
 is decimated into even and odd component.


Hence the time domain sequence is decimated into two equal halves and is given as


 


 =  d +  d

 


 



 
 
p# q

 =  d +   n + o d 
2
 
 



 


 =  d +   n + o d d 
⁄
2
 
 



 


 =  d + d   n + o d
⁄
2
 

d =
 ⁄
We know that

⁄
d = i
 ⁄ j
⁄

d =
 ⁄
⁄

d =
 
⁄

d = −1
⁄

 
Hence,



 


 =  d + −1   n + o d
2
 






 =  w + −1  n + ox d
2

Now,
 is decimated into even and odd component.
For even component of
,  is replaced by 2,






2 =  w + −1  n + ox d
2


d =
 ⁄
We know that

d = i
 ⁄ j


d = i
G ⁄ j


Discrete Time Signal Processing 20

 d = i
 ⁄⁄ j


d = id⁄ j


d = d
⁄


Hence,





2 =  w +  n + ox d
2 ⁄






2 =  yd
⁄



where,
y =  +  n + o
2
For odd component of
,  is replaced by 2 + 1,






2 + 1 =  w + −1#  n + ox d
#
2







2 + 1 =  w + −1#  n + ox d d
2







2 + 1 =  w −  n + ox d d
2 ⁄






2 + 1 =  ℎd
⁄



where,
ℎ = w −  n + ox d
2
Again repeat the procedure for
2 and
2 + 1, we get as follow:



G

4 =  !d
⁄G




G

4 + 2 =  &d
⁄G




G

4 + 1 =  {d
⁄G


Discrete Time Signal Processing 21

 


G

4 + 3 =  .d
⁄G



where
! = y + y n + o
4

& = wy − y n + ox d⁄
4

{ = ℎ + ℎ n + o
4

. = wℎ − ℎ n + ox d⁄
4

Example 1: Find the 8 point DFT of the sequence => = ?@, A, B, C, C, B, A, @D using DIFFFT algorithm.
Given:  = ?1, 2, 3, 4, 4, 3, 2, 1D
To find:
 using DIFFFT Algorithm

v = vG = vO = 1; vG = vO = −I; vO = 0.707 − I0.707; vOH = −0.707 − I0.707
Formula:

Solution:

0 1+4=5 5 + 5 = 10 10 + 10 = 20
0

Input 8 point DFT 4 point DFT 2 point DFT Output

=1

Discrete Time Signal Processing 22

 1 2+3=5 5 + 5 = 10 10 − 10 = 0
4
=2
2 3+2=5 5 − 51 = 0 0+0= 0
2
=3
3 4+1=5 5 − 5−I = 0 0−0= 0
6
=4
4 1 − 41 = −3 −3 − I −3 − I − 2.828
1
=4 − I1.414
= −5.828 − I2.414
5 2 − 30.707 −0.707 + I0.707 −3 − I + 2.828
5
=3 − I0.707 +−2.121 − I2.121 + I1.414
= −0.707 + I0.707 = −2.828 − I1.414 = −0.172 + I0.414
6 3 − 2−I = −I −3 + I1 = −3 + I −3 + I + 2.828
3
=2 − I1.414
= −0.172 − I0.414
7 4 − 1−0.707 *−0.707 + I0.707 −3 + I − 2.828
7
=1 − I0.707 −−2.121 − I2.121+−I + I1.414
= −2.121 − I2.121 = 2.828 − I1.414 = −5.828 + I1.414

 = ?20, −5.828 − I2.414, 0, −0.172 − I0.414, 0, −0.172 + I0.414, 0, −5.828 + I2.414D
Result:

Step 1: Find the DFT of the first input sequence  to get
.
Circular Convolution using DFT Method

Step 2: Find the DFT of the second input sequence ℎ to get l.
Step 3: Multiply both
l to get $.
Step 4: Find the IDFT for the sequence $ to get %.
(Condition: Both sequence are equal in length and length must be 2/ , i.e., 4, 8, 16, 32, 64, … … …)
Example 1: Find the circular convolution of the two sequences => = ?@, A, B, CD and |> = ?A, B, @D
using DFT method.
Discrete Time Signal Processing 23
Length of ,  = 4
Solution:

Length of ℎ, } = 3
Condition for circular convolution is that both  and ℎ must be equal in length.

 = ?1, 2, 3, 4D
Hence,

ℎ = ?2, 3, 1, 0D
1 1 1 1 1 1+2+3+4 10
1 −I −1 I 2 1 − I2 − 3 + I4 −2 + I2

 = X YX Y = X Y=X Y
1 −1 1 −1 3 1−2+3−4 −2
1 I −1 −I 4 1 + I2 − 3 − I4 −2 − I2

 = ?10, −2 + I2, −2, −2 − I2D
1 1 1 1 2 2+3+1+0 6
1 −I −1 I 3 2 − I3 − 1 + I0 1 − I3
l = X YX Y = X Y=X Y
1 −1 1 −1 1 2−3+1−0 0
1 I −1 −I 0 2 + I3 − 1 − I0 1 + I3
l = ?6, 1 − I3, 0, 1 + I3D
$ =
l = ?60, 4 + I8, 0, 4 − I8D
1 1 1 1 60 60 + 4 + I8 + 0 + 4 − I8 68 17
1 1 I −1 −I 4 + I8 1 60 + I4 − 8 + 0 − I4 − 8 1 44 11
% = X YX Y= X Y= X Y=X Y
4 1 −1 1 −1 0 4 60 − 4 − I8 + 0 − 4 + I8 4 52 13
1 −I −1 I 4 − I8 60 − I4 + 8 + 0 + I4 + 8 76 19
% = ?17, 11, 13, 19D

Step 1: See the length of the input sequence  and impulse sequence %. Assume iits length as  and }
Linear Convolution using DFT Method

Step 2: Calculate the length of the output sequence ~ =  + } − 1.

respectively.

(Condition: Length must be in 2/ , i.e., 4, 8, 16, 32, 64, … … …). If the length is not in this form, then append
Step 3: Append zeros to the input and impulse sequence in accordance to the length of the output sequence

zero to the sequences to reach the minimum value greater than ~.

Step 4: Find the DFT of the first input sequence  to get
.
Step 5: Find the DFT of the second input sequence ℎ to get l.
Step 6: Multiply both
l to get $.
Step 7: Find the IDFT for the sequence $ to get %.
Example 6: Find the response of the system with impulse response |> = ?A, @D and input => =
?@, B, AD using DFT method.
Length of ,  = 3
Length of ℎ, } = 2
Length of the output sequence %, ~ =  + } − 1 = 3 + 2 − 1 = 4

ℎ = ?2, 1, 0, 0D
Hence,

 = 1, 3, 2, 0D
1 1 1 1 1 1+3+2+0 6
1 −I −1 I 3 1 − I3 − 2 + I0 −1 − I3

 = X YX Y = X Y=X Y
1 −1 1 −1 2 1−3+2−0 0
1 I −1 −I 0 1 + I3 − 2 − I0 −1 + I3

 = ?6, −1 − I3, 0, −1 + I3D

Discrete Time Signal Processing 24

 1 1 1 1 2 2+1+0+0 3
1 −I −1 I 1 2 − I1 − 0 + I0 2−I
l = X YX Y = X Y=X Y
1 −1 1 −1 0 2−1+0−0 1
1 I −1 −I 0 2 + I1 − 0 − I0 2+I
l = ?3, 2 − I, 1, 2 + ID
$ =
l = ?18, −5 − I5, 0, −5 + I5D
1 1 1 1 18 18 − 5 − I5 + 0 − 5 + I5 8 2
1 1 I −1 −I −5 − I5 1 18 − I5 + 5 + 0 + I5 + 5 1 28 7
% = X YX Y= X Y= X Y=X Y
4 1 −1 1 −1 0 4 18 + 5 + I5 + 0 + 5 − I5 4 28 7
1 −I −1 I −5 + I5 18 + I5 − 5 + 0 − I5 − 5 8 2
% = ?2, 7, 7, 2D

Step 1:  − 1 zeros are padded at the end of the impulse response sequence ℎ which is of length } and a
Overlap Add Method

sequence of length } +  − 1 = ~ is obtained. Then, this L – point FFT is performed and the output values are

Step 2: An L – point FFT on the selected data block is performed. Here each data block has  input data values
stored.

and } − 1 zeros.
Step 3: The stored frequency response of the filter, i.e., the FFT output sequence obtained in Step 1 is multiplied
by the FFT output sequence of the selected data block obtained in Step 2.

Step 5: The first } − 1 IFFT values obtained in Step 4 is overlapped with last } − 1 IFFT values for the
Step 4: An L point inverse FFT is performed on the product sequence obtained in Step 3.

previous block. Then addition is done to produce the final convolution output sequence %.

Example 7: Find the response of the system with impulse response |> = ?B, @D and input=> =
Step 6: For the next data block, go to step 2.

?@, −@, A, @, B, A, −@, B, A, @, −@D using overlap add method.

Length of ℎ, } = 2
~ = 2 = 2 = 4
~ =+}−1
 = ~−}+1=4−2+1= 3
  = ?1, −1, 2, 0D
  = ?1, 3, 2, 0D
H  = ?−1, 3, 2, 0D
G  = ?1, −1, 0, 0D
ℎ = ?3, 1, 0, 0D
1 0 2 −1 3 3
−1 1 0 2 1 −2
%  =   ⊛ ℎ = X YX Y = X Y
2 −1 1 0 0 5
0 2 −1 1 0 2
1 0 2 3 3 3
3 1 0 2 1 10
%  =   ⊛ ℎ = X YX Y = X Y
2 3 1 0 0 9
0 2 3 1 0 2
−1 0 2 3 3 −3
3 −1 0 2 1 8
%H  = H  ⊛ ℎ = X YX Y = X Y
2 3 −1 0 0 9
0 2 3 −1 0 2

Discrete Time Signal Processing 25

 1 0 0 −1 3 3
−1 1 0 0 1 −2
%G  = G  ⊛ ℎ = X YX Y = X Y
0 −1 1 0 0 −1
0 0 −1 1 0 0
> L @ A B C  ‚ ƒ „ … @L @@ @A
†@ > 3 −2 5 2
†A > 3 10 9 2
†B > −3 8 9 2
†C > 3 −2 −1 0
†> B −A   @L … −@ „ …  −A −@ L

% = ?3, −2, 5, 5, 10, 9, −1, 8, 9, 5, −2, −1, 0D

Result:

Step 1:  − 1 zeros are padded at the end of the impulse response ℎ which is of length } and a sequence
Overlap Save Method

of length } +  − 1 = ~ is obtained. Then this L – point FFT is performed and the output values are stored.

} − 1 values in the previous data block, except the first data block which begins with } − 1 zeros.
Step 2: An L – point FFT on the selected data block is performed. Here each data block begins with the last

Step 3: The stored frequency response of the filter, i.e., the FFT output sequence obtained in Step 1 is multiplied
by the FFT output sequence of the selected data block obtained in Step 2.

Step 5: The first } − 1 values from successive output of Step 4 are discarded and the last  values of the
Step 4: An L point inverse FFT is performed on the product sequence obtained in Step 3.

IFFT obtained in Step 4 is saved to produce the output %.

|> = ?B, A, @D and input => =

Step 6: For the next data block, go to step 2.

?−@, @, −B, A, −@, −A, B, B, −@, @, BD using overlap save method.

Example 8: Find the response of the system with impulse response

Length of ℎ, } = 3
~ = 2 = 2H = 8
~ =+}−1
 = ~−}+1=8−3+1= 6
  = ?0, 0, −1, 1, −3, 2, −1, −2D
  = ?−1, −2, 3, 3, −1, 1, 3, 0D
ℎ = ?3, 2, 1, 0, 0,0, 0, 0D
0 −2 −1 2 −3 1 −1 0 3 −5
\0 0 −2 −1 2 −3 1 −1_ \2_ \−2_
[−1 0 0 −2 −1 2 −3 1 ^ [1^ [−3^
[ ^[ ^ [ ^
1 −1 0 0 −2 −1 2 −3^ [0^ [ 1 ^
%  =   ⊛ ℎ = [ =
[−3 1 −1 0 0 −2 −1 2 ^ [0^ [−7^
[ 2 −3 1 −1 0 0 −2 −1^ [0^ [ 1 ^
[−1 2 −3 1 −1 0 0 −2^ [0^ [−2^
Z−2 −1 2 −3 1 −1 0 0 ] Z0] Z−6]
−1 0 3 1 −1 3 3 −2 3 0
\−2 −1 0 3 1 −1 3 3 _ \2_ \−8_
[ 3 −2 −1 0 3 1 −1 3 ^ [1 ^ [ 4 ^
[ ^[ ^ [ ^
3 3 −2 −1 0 3 1 −1^ [0^ [ 13 ^
%  =   ⊛ ℎ = [ =
[ −1 3 3 −2 −1 0 3 1 ^ [0 ^ [ 6 ^
[ 1 −1 3 3 −2 −1 0 3 ^ [0 ^ [ 4 ^
[3 1 −1 3 3 −2 −1 0 ^ [0^ [ 10 ^
Z0 3 1 −1 3 3 −2 −1] Z0] Z 7 ]
> −A −@ L @ A B C  ‚ ƒ „ … @L @@
Discrete Time Signal Processing 26
†@ > −5 −2 −3 1 −7 1 −2 6
†A > 0 −8 4 13 6 4 10 7
†> −B @ −ƒ @ −A ‚ C @B ‚ C @L ƒ

% = ?−3, 1, −7, 1, −2, 6, 4, 13, 6, 4, 10, 7D

Result:

Discrete Time Signal Processing 27

 Unit II IIR Filter Design
Impulse Invariant Transformation
Impulse Invariant Transformation is the many to one mapping from the s domain to z domain.

1
Consider an analog filter’s transfer function as
l‡ ˆ =
ˆ−!

ℎ‡ ‰ = ‡Š ‹‰
Taking inverse Laplace Transform on both side,

Sample the ℎ‡ ‰ by ‰ = ), then

ℎ = *ℎ‡ ‰+Š→
ℎ = ‡ ‹)
ℎ = ‡ ‹

1
Taking Z transform on both side, we get,
lŽ =
1 −
‡ Ž


1 1
Hence, from this we can get the mapping as
→
ˆ − ! 1 −
‡ Ž


1 −1/
 . /
 1
Similarly, we can derive remain three transform as

→ w x
ˆ + ! / - − 1! .ˆ /
 1 −
 Ž
 →‡
ˆ+! 1 −
‡ cos &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž

&
‡ sin &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


Here the analog filter with pole ˆ = ! is mapped to the digital filter with pole Ž = ‡ . Hence
Relation between analog and digital frequency

Ž = 

Ž = –
We know that

ˆ = — + IΩ

– = ™#ٍ
Hence

– = ™ ٍ

š = Ω)
From the above equation, it is clear that

This is the relationship between analog and digital frequency in impulse invariant transformation.
Disadvantage
This method is only applicable for low pass and band pass filter and is not applicable for high pass and band
reject filter. This method is easily affected by aliasing due to sampling of analog signal.
Summary:

Discrete Time Signal Processing 28

 1 1
→
ˆ − ! 1 − ‡ Ž

1 −1/
 . /
 1
→ w x
ˆ + ! / - − 1! .ˆ /
 1 −
 Ž
 →‡
ˆ+! 1 −
‡ cos &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž

&
‡ sin &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


@
Example 1: For the analog transfer function
›œ =
œ + @œ + A
Determine the digital transfer function using impulse invariant transformation. Assume  = @ œžŸ.

1
Given:
lˆ =
ˆ + 1ˆ + 2
) = 1 ˆ {
To find:lŽ

1 1
Formula:
→
ˆ − ! 1 − ‡ Ž

1 s t
Solution:
= +
ˆ + 1ˆ + 2 ˆ + 1 ˆ + 2
sˆ + 2 + tˆ + 1 = 1
Put ˆ = −1, then s = 1
Put ˆ = −2, then t = −1
1 1
lˆ = −
ˆ+1 ˆ+2
1 1
lŽ = −
1− Ž

 1 −
 Ž

1 1
lŽ = −
1 − 0.3679Ž
 1 − 0.1353Ž

1 − 0.1353Ž − 1 + 0.3679Ž



lŽ =
1 − 0.3679Ž
 1 − 0.1353Ž
 
0.2326Ž

lŽ =
1 − 0.5032Ž
 + 0.0498Ž

Example 2: Determine ›  using the impulse invariant transformation for the analog transfer function
@
›œ =
œ + L. œ + L. œ + A
A

1
Given:
lˆ =
ˆ + 0.5ˆ  + 0.5ˆ + 2
To Find: lŽ

1 1
Formula:
→
ˆ − ! 1 − ‡ Ž


Discrete Time Signal Processing 29

 ˆ+! 1 −
‡ cos &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž

&
‡ sin &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


1 s tˆ + u
Solution:
= + 
ˆ +
0.5ˆ  + 0.5ˆ + 2 ˆ + 0.5 ˆ + 0.5ˆ + 2
sˆ  + 0.5ˆ + 2 + tˆ + uˆ + 0.5 = 1
Put ˆ = −0.5, then s0.25 − 0.25 + 2 = 1 ⟹ 2s = 1 ⟹ s = 0.5
Put ˆ = 0, then 2s + 0.5u = 1 ⟹ 1 + 0.5u = 1 ⟹ 0.5u = 0 ⟹ u = 0
Put ˆ = 1, then 3.5s + 1.5t + 1.5u = 1 ⟹ 1.75 + 1.5t = 1 ⟹ 1.5t = −0.75 ⟹ t = −0.5
0.5 0.5ˆ
lˆ = − 
ˆ + 0.5 ˆ + 0.5ˆ + 2
0.5 0.5ˆ + 0.25 − 0.25
lˆ = −
ˆ + 0.5 ˆ + 0.25 + 1.3919
0.5 0.5ˆ + 0.25 − 0.125
lˆ = −
ˆ + 0.5 ˆ + 0.25 + 1.3919
0.5 0.5ˆ + 0.25 1 0.125 ∗ 1.3919
lˆ = − +
ˆ + 0.5 ˆ + 0.25 + 1.3919 1.3919 ˆ + 0.25 + 1.3919
0.5 0.5ˆ + 0.25 0.0898 ∗ 1.3919
lˆ = − +
ˆ + 0.5 ˆ + 0.25 + 1.3919
  ˆ + 0.25 + 1.3919

 
0.5 0.51 −
.T
cos 1.3919 Ž 0.0898 ∗
.T sin 1.3919 Ž

lˆ = − +
1 −
.T Ž
 1 − 2
.T cos 1.3919 Ž
 +
.T Ž
 1 − 2
.T cos 1.3919 Ž
 +
.T Ž

0.5 0.5 − 0.0693Ž
 0.0688Ž

lˆ = − +
1 − 0.6065Ž
 1 − 0.2772Ž
 + 0.6065Ž
 1 − 0.2772Ž
 + 0.6065Ž

0.5 0.5 − 0.1361Ž

lˆ = −
1 − 0.6065Ž
 1 − 0.2772Ž
 + 0.6065Ž

0.51 − 0.2772Ž
 + 0.6065Ž
  − 0.5 − 0.1361Ž
 1 − 0.6065Ž
 
lˆ =
1 − 0.6065Ž
 1 − 0.2772Ž
 + 0.6065Ž
 
0.5 − 0.1386Ž + 0.3033Ž
 − 0.5 + 0.3033Ž
 + 0.1361Ž
 − 0.0825Ž



lˆ =
1 − 0.2772Ž
 + 0.6065Ž
 − 0.6065Ž
 + 0.1681Ž
 − 0.3678Ž
H
0.3008Ž
 + 0.2208Ž

lˆ =
1 − 0.8837Ž
 + 0.7746Ž
 − 0.3678Ž
H
Bilinear Transformation
Bilinear Transformation is the one to one mapping from the s domain to z domain. It is a conformal
mapping that transforms jΩ axis in the s plane into the unit circle in the z domain only once, hence the output
signal are not affected by the aliasing due to sampling. It is used to design any type of filter such as low pass,
high pass, band pass and band reject filter.

&
Let the system transfer function of analog signal is

lˆ =
ˆ+!
$ˆ &
=

ˆ ˆ + !
$ˆˆ + ! = &
ˆ
ˆ$ˆ + !$ˆ = &
ˆ
Taking inverse Laplace Transform,
Discrete Time Signal Processing 30
 .%‰
+ !%‰ = &‰
.‰

  
Integrate the above equation by t over a interval of nT,
.%‰
¢ .‰ + ¢ !%‰.‰ = ¢ &‰.‰
.‰

 
 

  

¢ .%‰ + ¢ !%‰.‰ = ¢ &‰.‰


 
 



We know that by trapezoidal rule of numerical integration,
)
¢ !‰.‰ = *!) + !) − )+
2



!) &)
Hence,
*%‰+

 + *%) + %) − )+ = *) + ) − )+
2 2
!) !) &) &)
%) − %) − ) + %) + %) − ) = ) + ) − )
2 2 2 2
!) !) &) &)
n1 + o %) − n1 − o %) − ) = ) + ) − )
2 2 2 2

!) !) &) &)

Taking inverse Z transform,
n1 + o $Ž − n1 − o Ž
 $Ž =
Ž + Ž
Ž
2 2 2 2
!) !) &)
wn1 + o − n1 − o Ž
 x $Ž = 1 + Ž
 
Ž
2 2 2
&)
$Ž 1 + Ž
 
= 2

Ž p1 + !)q − p1 − !)q Ž

2 2
&)
1 + Ž
 
lŽ = 2
!) !)
1 + 2 − Ž
 + 2 Ž

&)
1 + Ž
 
lŽ = 2
!)
1 − Ž
  + 1 + Ž
 
2
&
lŽ =
2 1 − Ž

) n1 + Ž
 o + !
Now from the lˆ and lŽ equation, it is clear that
2 1 − Ž

ˆ→ £ ¤
) 1 + Ž

Relation between analog and digital frequency

Ž = –
We know that

ˆ = — + IΩ
Discrete Time Signal Processing 31
 2 1 − i – j


Thus

— + IΩ → ¥ ¦
) 1 +  – 

2 – − 1
— + IΩ → £ – ¤
) +1
2 cos š + I sin š − 1
— + IΩ → n o
) cos š + I sin š + 1
2 cos š − 1 + I sin š
— + IΩ → n o
) cos š + 1 + I sin š
2 cos š − 1 + I sin š cos š + 1 − I sin š
— + IΩ → n on o
) cos š + 1 + I sin š cos š + 1 − I sin š
2  − 1 + I2 sin š
— + IΩ → £  ¤
) + 2 cos š + 1

2 2 sin š
Equating imaginary part on both side,
Ω = w  x
) + 2 cos š + 1
For unity magnitude, i.e., = 1
2 2 sin š
Ω = w x
) 1 + 2 cos š + 1
2 2 sin š
Ω = w x
) 2 + 2 cos š
2 sin š
Ω = w x
) 1 + cos š
š š
2 2 cos p 2 q sin p 2 q
Ω = § š ¨
) 2 cos  p 2 q
š
2 sin p 2 q
Ω = § ¨
) cos pšq
2
2 š
Ω= tan p q
) 2
This is the relation between the analog and digital frequency in the bilinear transformation.
Here it is shown that the relation between analog and digital frequency is non linear and hence due to
this non linearity, warping effect will occur.

A
Example 1: Convert the analog transfer function
›œ =
œ + @œ + B
to digital transfer function using bilinear transformation. Assume  = L. @ œžŸ.

2
Given:
lˆ =
ˆ + 1ˆ + 3
) = 0.1 ˆ {
To Find:lŽ
Formula:
Discrete Time Signal Processing 32
 2 1 − Ž

ˆ→ £ ¤
) 1 + Ž


lŽ = *lˆ+
Solution:
 
« ¬­
→ n o
 #« ¬­
2
lŽ = w x
ˆ + 1ˆ + 3 →n
« ¬­ o
#« ¬­
2
lŽ =
1 − Ž
 1 − Ž

n20 n o + 1o n20 n o + 3o
1 + Ž
 1 + Ž

21 + Ž
 
lŽ =
20 − 20Ž
 + 1 + Ž
 20 − 20Ž
 + 3 + 3Ž
 
21 + Ž
 
lŽ =
21 − 19Ž
 23 − 17Ž
 
21 + Ž
 
lŽ =
483 − 794Ž
 + 323Ž

0.00421 + Ž
 
lŽ =
1 − 1.6439Ž
 + 0.6687Ž


œ + L. @
Example 2: Convert the analog transfer function
›œ =
œ + L. @A + …

of ®¯ = °⁄C.
into digital transfer function using bilinear transformation. The digital filter have a resonant frequency

ˆ + 0.1
Given:
lˆ =
ˆ + 0.1 + 9
š± = ²⁄4
To Find: lŽ

2 1 − Ž

Formula:
ˆ→ £ ¤
) 1 + Ž

2 š±
Ω³ = tan
) 2
Solution:

Ω³ = 9
From the analog transfer function,

Ω³ = 3

2 ²
Now,
3 = tan
) 8
0.8284
3=
)
0.8284
)=
3
) = 0.2761 ˆ {
lŽ = *lˆ+  
« ¬­
→ n o
 #« ¬­

Discrete Time Signal Processing 33

 ˆ + 0.1
lŽ = w x
ˆ + 0.1 + 9 →P.GHOn
« ¬­ o
#« ¬­
ˆ + 0.1
lŽ = w  x
ˆ + 0.2ˆ + 9.01 →P.GHOn
« ¬­
¬­ o #«
1 − Ž

7.2438 n o + 0.1
1 + Ž

lŽ = 
1 − Ž
 1 − Ž

£7.2438 n o¤ + 0.2 £7.2438 n o¤ + 9.01
1+Ž
 1 + Ž

7.3438 − 7.2438Ž
 1 + Ž
 
lŽ =
17.7026 − 15.9364Ž
 + 7.2438Ž

7.3438 − 7.2438Ž
 + 7.3438Ž
 − 7.2438Ž

lŽ =
17.7026 − 15.9364Ž
 + 7.2438Ž

7.3438 + 0.099Ž
 − 7.2438Ž

lŽ =
17.7026 − 15.9364Ž
 + 7.2438Ž

0.4148 + 0.0056Ž
 − 0.4092Ž

lŽ =
1 − 0.9002Ž
 + 0.4092Ž

Butterworth Low Pass Filter

´ ≤ µli – jµ ≤ 1, 0 ≤ š ≤ š¶
Specification:

µli – jµ ≤ ´ , š ≤ š ≤ ²

š¶
, 4-·‹"ˆ 4¸! ¹!‰ ) !ˆ -!‰¹
Step 1: Determination of Analog Edge Frequencies:

)
Ω¶ = r 2 š¶
tan , t¹"¹ ! ) !ˆ -!‰¹
) 2
š
, 4-·‹"ˆ 4¸! ¹!‰ ) !ˆ -!‰¹
)
ِ = r 2 š
tan , t¹"¹ ! ) !ˆ -!‰¹
) 2
1 1
log ½n  − 1o¾n  − 1o¿
Step 2: Determination of Order of the Filter:

1 ´ ´
≥
2 logiِ ⁄Ω¶ j

Ω¶
Step 3: Determination of Cut off Frequency:
Ω³ =
1 ⁄
n  − 1o
´
Step 4: Determination of Analog Transfer Function ›À œ:
For  even,
⁄
t Ω³
l‡ ˆ = Á
ˆ  + & Ω³ ˆ + { Ω³

For  odd,

⁄
t Ω³ t Ω³
l‡ ˆ = Á 
ˆ + { Ω³ ˆ + & Ω³ ˆ + { Ω³

Discrete Time Signal Processing 34
 2 − 1²
where,
& = 2 sin £ ¤
2
{ = 1
t can obtained from
For  even,
⁄

s = 1 = Á t

For  odd,

 ⁄

s=1= Á t

Step 5: Determination of Digital Transfer Function › :
lŽcan be obtained from l‡ ˆ using either impulse invariant transformation or bilinear transformation.

1 1
Using Impulse Invariant Transformation:
→
ˆ − ! 1 − ‡ Ž

1 −1 /

. /
 1
→ w x
ˆ + ! / - − 1! .ˆ /
 1 −
 Ž
 →‡
ˆ+! 1 −
‡ cos &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž

&
‡ sin &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


2 1 − Ž

Using Bilinear Transformation:
ˆ→
) 1 + Ž


√L.  ≤ µ›žb® µ ≤ @, L ≤ ® ≤ °⁄A

Example 1: Design a Digital Butterworth filter satisfying the following conditions

µ›žb® µ ≤ L. A, B°⁄C ≤ ® ≤ °
with  = @ œžŸ. Apply Impulse Invariant Transformation.

´ = √0.5 = 0.707; ´ = 0.2; š¶ = ²⁄2 ; š = 3²⁄4 ; ) = 1 ˆ {

Given:

To find: lŽ
Step 1: Determination of Analog Edge Frequencies:
š¶
Ω¶ =
Formula:

)
š
ِ =
)
²
Ω¶ = = 1.5708 ˆ.⁄ˆ {
Solution:

2
3²
ِ = = 2.3562 !.⁄ˆ {
4
Step 2: Determination of Order of the Filter:
Formula:

Discrete Time Signal Processing 35

 1 1
log ½n  − 1o¾n  − 1o¿
1 ´ ´
≥
2 logiِ ⁄Ω¶ j

1 1
1 log Ãp0.04 − 1qÄp0.5 − 1qÅ
Solution:

≥
2 log2.3562⁄1.5708
1 log?24⁄1D
≥
2 log2.3562⁄1.5708
1 1.3802
≥
2 0.1761
 ≥ 3.9188
=4
Step 3: Determination of Cut off Frequency:

Ω¶
Formula:
Ω³ =
1 ⁄
n  − 1o
´

1.5708
Solution:
Ω³ =
1⁄O
Ω³ = 1.5708 !.⁄ˆ {

Step 4: Determination of Analog Transfer Function ›À œ:

Since  even,
Formula:

⁄
t Ω³
l‡ ˆ = Á
ˆ  + & Ω³ ˆ + { Ω³


2 − 1²
where,
& = 2 sin £ ¤
2
{ = 1
t can obtained from
For  even,
⁄

s = 1 = Á t



t Ω³
Solution:

l‡ ˆ = Á 
ˆ + & Ω³ ˆ + { Ω³

t Ω³ t Ω³
l‡ ˆ = £  ¤ £ ¤
ˆ + & Ω³ ˆ + { Ω³ ˆ  + & Ω³ ˆ + { Ω³
t t ΩG³
l‡ ˆ = 
ˆ + & Ω³ ˆ + { Ω³ ˆ  + & Ω³ ˆ + { Ω³ 

Discrete Time Signal Processing 36

 2 − 1²
& = 2 sin £ ¤
8
²
& = 2 sin p q = 0.7653
8
3²
& = 2 sin n o = 1.8478
8
{ = { = 1


1 = Á t

t t = 1
6.0881
l‡ ˆ = 
ˆ + 1.2021ˆ + 2.4674ˆ  + 2.9025ˆ + 2.4674
Step 5: Determination of Digital Transfer Function › :
6.0881 sˆ + t uˆ + '
=  + 
ˆ + 1.2021ˆ + 2.4674ˆ + 2.9025ˆ + 2.4674 ˆ + 1.2021ˆ + 2.4674 ˆ + 2.9025ˆ + 2.4674
 
sˆ + tˆ  + 2.9025ˆ + 2.4674 + uˆ + 'ˆ  + 1.2021ˆ + 2.4674 = 6.0881
Put ˆ = 0,
2.4674t + 2.4674' = 6.0881
t + ' = 2.4674 −→ 1
Comparing ˆ coefficients,
H

s + u = 0 −→ 2
Put ˆ = 1,
s + t6.3699 + u + '4.6695 = 6.0881
6.3699s + 6.3699t + 4.6695u + 4.6695' = 6.0881 −→ 3
Put ˆ = −1,
−s + t0.5649 + −u + '2.2653 = 6.0881
−0.5649s + 0.5649t − 2.2653u + 2.2653' = 6.0881 −→ 4

0 1 0 1 s 2.4674
Equation (1), (2), (3) and (4) can be written in the matrix form as

1 0 1 0 t 0
X YX Y = X Y
6.3699 6.3699 4.6695 4.6695 u 6.0881
−0.5649 0.5649 −2.2653 2.2653 ' 6.0881
0 1 0 1
1 0 1 0
∆= Ç Ç = −21.1556 − 17.0675 − −17.0675 + 7.1967 = 5.7827
6.3699 6.3699 4.6695 4.6695
−0.5649 0.5649 −2.2653 2.2653
2.4674 1 0 1
0 0 1 0
∆È = Ç Ç = 2.4674−11.7919 − 14.637 − −35.3414 = −8.3909
6.0881 6.3699 4.6695 4.6695
6.0881 0.5649 −2.2653 2.2653
0 2.4674 0 1
1 0 1 0
∆É = Ç Ç = −2.467421.1556 − 17.0675 − −42.2198 + 42.2198
6.3699 6.0881 4.6695 4.6695
−0.5649 6.0881 −2.2653 2.2653
= −10.0868
0 1 2.4674 1
1 0 0 0
∆³ = Ç Ç = −−14.637 + 2.467411.7919 − 35.3414 = 8.3909
6.3699 6.3699 6.0881 4.6695
−0.5649 0.5649 6.0881 2.2653
Discrete Time Signal Processing 37
 0 1 0 2.4674
1 0 1 0
∆Ê = Ç Ç = −42.2198 − 42.2198 − 2.4674−17.0675 + 7.1967
6.3699 6.3699 4.6695 6.0881
−0.5649 0.5649 −2.2653 6.0881
= 24.3552
∆È 8.3909
s= =− = −1.451
∆ 5.7827
ƃ 10.0868
t= =− = −1.7443
∆ 5.7827
∆³ 8.3909
u= = = 1.451
∆ 5.7827
∆Ê 24.3552
'= = = 4.2117
∆ 5.7827
1.451ˆ + 1.7443 1.451ˆ + 4.2117
l‡ ˆ = −  + 
ˆ + 1.2021ˆ + 2.4674 ˆ + 2.9025ˆ + 2.4674
1.451ˆ + 1.2021 1.451ˆ + 2.9026
l‡ ˆ = − +
ˆ + 0.6011 + 1.4513 ˆ + 1.4513 + 0.6011
1.451ˆ + 0.6011 + 0.601 1.451ˆ + 1.4513 + 1.4513
l‡ ˆ = − +
ˆ + 0.6011 + 1.2047 ˆ + 1.4513 + 0.7753
1.451ˆ + 0.6011 0.7239 ∗ 1.2047 1.451ˆ + 1.4513
l‡ ˆ = − − +
ˆ + 0.6011 + 1.2047 ˆ + 0.6011 + 1.2047 ˆ + 1.4513 + 0.7753
2.7162 ∗ 0.7753
+
ˆ + 1.4513 + 0.7753
1.45111 −
.V cos1.2047 Ž
  0.7239
.V sin1.2047 Ž

lŽ = − −
1 − 2
.V cos1.2047 Ž
 +
. Ž
  1 − 2
.V cos1.2047 Ž
 +
. Ž
 
1.4511 −
.GTH cos0.7753 Ž
 
+
1 − 2
.GTH cos0.7753 Ž
 +
.JV Ž
 
2.7162
.GTH sin0.7753 Ž

+
1 − 2
.GTH cos0.7753 Ž
 +
.JV Ž
 
−1.4511 + 0.2848Ž
 −0.3705Ž
 1.451 − 0.2428Ž

lŽ = + +
1 − 0.3925Ž
 + 0.3005Ž
 1 − 0.3925Ž
 + 0.3005Ž
 1 − 0.3346 + 0.0549Ž

0.4454Ž

+
1 − 0.3346 + 0.0549Ž

−1.4511 − 0.0857Ž
 1.451 + 0.2026Ž

lŽ = +
1 − 0.3925Ž
 + 0.3005Ž
 1 − 0.3346 + 0.0549Ž

Example 2: Design a digital Butterworth filter to satisfy the following constraint using Impulse Invariant
Transformation.

L. ƒ ≤ µ›žb® µ ≤ @, L ≤ ® ≤ L. B°

µ›žb® µ ≤ L. B, L. ° ≤ ® ≤ °

Assume  = L. B œžŸ.

s¶ = 0.7
Specification:

s = 0.35
š¶ = 0.3²
š = 0.5²
Discrete Time Signal Processing 38
 ) = 0.3 ˆ {
Type of transformation: Impulse Invariant Transformation
Type of filter: Butterworth Low Pass filter

š¶ 0.3²
Step 1: To find analog edge frequencies
Ω¶ = = = 3.1416 !.⁄ˆ {
) 0.3
š 0.5²
ِ = = = 5.236 !.⁄ˆ {
) 0.3
1 1
log ½n  − 1o¾n  − 1o¿
Step 2: To find the order of the filter

1 s s¶
≥
2 logiِ ⁄Ω¶ j
1 1
1 log Ãp0.35 − 1qÄp0.7 − 1qÅ
≥
2 log5.236⁄3.1416
1 log?7.1633⁄1.0408D
≥
2 log5.236⁄3.1416
1
 ≥ 3.7762
2
 ≥ 1.8881
=2

Ω¶
Step 3: To find cut off frequency
Ω³ =
1 ⁄
n  − 1o
s¶
3.1416
Ω³ =
1.0408⁄G
Ω³ = 3.1104 !.⁄ˆ {

The transfer function for normalized Butterworth Low Pass Filter with order  = 2 is
Step 4: To find the analog transfer function

1
l ˆ = 
ˆ + 1.414ˆ + 1
The transfer function for Butterworth Low Pass Filter with order  = 2 and cut off frequency Ω³ =
3.1104 !.⁄ˆ { is
l‡ ˆ = *l ˆ+→⁄ΩË
1
l‡ ˆ = w  x
ˆ + 1.414ˆ + 1 →⁄H.G
1
l‡ ˆ =
ˆ  1.414ˆ
p3.1104q + 3.1104 + 1
3.1104
l‡ ˆ = 
ˆ + 1.414 ∗ 3.1104ˆ + 3.1104
9.6746
l‡ ˆ = 
ˆ + 4.3988ˆ + 9.6746
9.6746
Step 5: To find the digital transfer function
l‡ ˆ =
4.3988 4.3988  4.3988 
ˆ  + 2 p 2 q ˆ + p 2 q + 9.6746 − p 2 q

Discrete Time Signal Processing 39

 9.6746
l‡ ˆ =
ˆ + 2.1994 + 4.8372
9.6746
l‡ ˆ =
ˆ + 2.1994 + i√4.8372j


1
9.6746 ∗ 2.1994 ∗
l‡ ˆ = 2.1994
ˆ + 2.1994 + 2.1994
4.3988 ∗ 2.1994
l‡ ˆ =
ˆ + 2.1994 + 2.1994

&
‡ sin &) Ž

We know that
→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


4.3988
.JJG∗.H sin2.1994 ∗ 0.3 Ž

Hence
lŽ =
1 − 2
.JJG∗.H cos2.1994 ∗ 0.3 Ž
 +
∗.JJG∗.H Ž

1.3939Ž

lŽ =
1 − 0.8169Ž
 + 0.2672Ž

Since ) = 0.3 sec < 1 ˆ {,
0.3 ∗ 1.3939Ž

lŽ =
1 − 0.8169Ž
 + 0.2672Ž

0.4182Ž

lŽ =
1 − 0.8169Ž
 + 0.2672Ž

Example 3: Design a digital Butterworth Filter to satisfy the following constraint using impulse invariant

2.5 dB ripple in the pass band L ≤ ® ≤ L. C°

transformation.

12 dB attenuation in the stop band L. ƒ° ≤ ® ≤ °

ζ = −20 log s¶ = 2.5 .t

Specification:

log s¶ = −0.125
s¶ = 10
.T = 0.7499
ΐ = −20 log s = 12 .t
log s = −0.6
s = 10
.V = 0.2512
š¶ = 0.4²
š = 0.7²
Assume ) = 1 ˆ {
Type of transformation: Impulse Invariant Transformation
Type of filter: Butterworth Low Pass Filter

š¶ 0.4²
Step 1: To find analog edge frequencies
Ω¶ = = = 1.2566 !.⁄ˆ {
) 1
š 0.7²
ِ = = = 2.1991 !.⁄ˆ {
) 1
1 1
log ½n  − 1o¾n  − 1o¿
Step 2: To find the order of the filter

1 s s¶
≥
2 logiِ ⁄Ω¶ j

Discrete Time Signal Processing 40

 1 1
1 log Ãp0.2512 − 1qÄp0.7499 − 1qÅ
≥
2 log2.1991⁄1.2566
1 log?14.8475⁄0.7783D
≥
2 log2.1991⁄1.2566
1
 ≥ 5.2685
2
 ≥ 2.6343
=3

Ω¶
Step 3: To find cut off frequency
Ω³ =
1 ⁄
n  − 1o
s¶
1.2566
Ω³ =
0.7783⁄V
Ω³ = 1.3102 !.⁄ˆ {

The transfer function for normalized Butterworth Low Pass Filter with order  = 3 is
Step 4: To find the analog transfer function

1
l ˆ =
ˆ + 1ˆ  + ˆ + 1
The transfer function for Butterworth Low Pass Filter with order  = 3 and cut off frequency Ω³ =
1.3102 !.⁄ˆ { is
l‡ ˆ = *l ˆ+→⁄ΩË
1
l‡ ˆ = w x
ˆ + 1ˆ  + ˆ + 1 →⁄.H
1
l‡ ˆ =
ˆ ˆ  ˆ
p1.3102 + 1q np1.3102q + 1.3102 + 1o
1.3102H
l‡ ˆ =
ˆ + 1.3102ˆ  + 1.3102ˆ + 1.3102 
2.2491
l‡ ˆ =
ˆ + 1.3102ˆ + 1.3102ˆ + 1.7166


2.2491 s tˆ + u
Step 5: To find digital transfer function
= + 
ˆ + 1.3102ˆ + 1.3102ˆ + 1.7166 ˆ + 1.3102 ˆ + 1.3102ˆ + 1.7166


sˆ  + 1.3102ˆ + 1.7166 + tˆ + uˆ + 1.3102 = 2.2491

When ˆ = −1.3102,
s−1.3102 + 1.3102−1.3102 + 1.7166 = 2.2491
1.7166s = 2.2491
2.2491
s=
1.7166
s = 1.3102

1.3102ˆ  + 1.3102ˆ + 1.7166 + tˆ + uˆ + 1.3102 = 2.2491

Therefore,

Comparing ˆ  coefficient,
1.3102 + t = 0
t = −1.3102
Comparing constant coefficient,
Discrete Time Signal Processing 41
 2.2491 + 1.3102u = 2.2491
1.3102u = 0
u=0

1.3102 −1.3102ˆ
Substitute the value of A, B and C in the analog transfer function,
l‡ ˆ = + 
ˆ + 1.3102 ˆ + 1.3102ˆ + 1.7166
1.3102 1.3102ˆ
l‡ ˆ = −
ˆ + 1.3102 1.3102 1.3102  1.3102 
ˆ + 2 p qˆ + p q + 1.7166 − p q
2 2 2
1.3102 1.3102ˆ
l‡ ˆ = −
ˆ + 1.3102 ˆ + 0.6551 + 1.2875
1.3102 1.3102ˆ
l‡ ˆ = −
ˆ + 1.3102 ˆ + 0.6551 + i√1.2875j
1.3102 1.3102ˆ
l‡ ˆ = −
ˆ + 1.3102 ˆ + 0.6551 + 1.1347
1.3102 1.3102ˆ + 0.6551 − 0.6551
l‡ ˆ = −
ˆ + 1.3102 ˆ + 0.6551 + 1.1347
1.3102 1.3102ˆ + 0.6551 1.31020.6551
l‡ ˆ = − +
ˆ + 1.3102 ˆ + 0.6551 + 1.1347 ˆ + 0.6551 + 1.1347
1
1.3102 1.3102ˆ + 0.6551 0.8583 ∗ 1.1347 ∗ 1.1347
l‡ ˆ = − +
ˆ + 1.3102 ˆ + 0.6551 + 1.1347 ˆ + 0.6551 + 1.1347
1.3102 1.3102ˆ + 0.6551 0.7564 ∗ 1.1347
l‡ ˆ = − +
ˆ + 1.3102 ˆ + 0.6551 + 1.1347 ˆ + 0.6551 + 1.1347

1 1
We know that
→
ˆ − ! 1 − ‡ Ž

ˆ+! 1 −
‡ cos &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž

&
‡ sin &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


1.3102 1.31021 −
.VTT cos1.1347 Ž
 
Hence
lŽ = −
1 −
.H Ž
 1 − 2
.VTT cos1.1347 Ž
 +
∗.VTT Ž

0.7564
.VTT sin1.1347 Ž

+
1 − 2
.VTT cos1.1347 Ž
 +
∗.VTT
1.3102 1.31021 − 0.2194Ž
  0.3561Ž

lŽ = − +
1 − 0.2698Ž
 1 − 0.4388Ž
 + 0.2698Ž
 1 − 0.4388Ž
 + 0.2698Ž

1.3102 1.31021 − 0.2194Ž
  − 0.3561Ž

lŽ = −
1 − 0.2698Ž
 1 − 0.4388Ž
 + 0.2698Ž

1.3102 1.3102 − 0.6436Ž

lŽ = −
1 − 0.2698Ž
 1 − 0.4388Ž
 + 0.2698Ž

0.4221Ž
 + 0.1798Ž

lŽ =
1 − 0.7086Ž
 + 0.3882Ž
 − 0.0728Ž
H

´ ≤ µli – jµ ≤ 1, 0 ≤ š ≤ š¶
Chebyshev Low Pass Filter

µli – jµ ≤ ´ , š ≤ š ≤ ²
Discrete Time Signal Processing 42
 š¶
, 4-·‹"ˆ 4¸! ¹!‰ ) !ˆ -!‰¹
Step 1: Determination of Analog Edge Frequencies:

)
Ω¶ = r 2 š¶
tan , t¹"¹ ! ) !ˆ -!‰¹
) 2
š
, 4-·‹"ˆ 4¸! ¹!‰ ) !ˆ -!‰¹
)
ِ = r 2 š
tan , t¹"¹ ! ) !ˆ -!‰¹
) 2

1 1 ⁄
Step 2: Determination of Order of the Filter:
cosh
 Ð Ñ n  − 1o Ò
´
 ≥
cosh
 iِ ⁄Ω¶ j

1
⁄
where

Ñ = £  − 1¤
´

Ω³ = Ω¶
Step 3: Determination of Cut off Frequency:

Step 4: Determination of Analog Transfer Function ›À œ:

For  even,
⁄
t Ω³
l‡ ˆ = Á
ˆ  + & Ω³ ˆ + { Ω³

For  odd,

⁄
t Ω³ t Ω³
l‡ ˆ = Á 
ˆ + { Ω³ ˆ + & Ω³ ˆ + { Ω³


2 − 1²
where,
& = 2% sin £ ¤
2
2 − 1²
{ = % + {ˆ  £ ¤
2
{ = %
⁄
⁄
1 1 ⁄
1 1 ⁄
1
% = Ó:n  + 1o + ; − :n  + 1o + ; Ô
2 Ñ Ñ Ñ Ñ
t can obtained from
For  even,
⁄
s 1 t
= =Á
1 + Ñ 
 ⁄ 1 + Ñ 
 ⁄ {

For  odd,

 ⁄
t
s=1= Á
{

Alternate Way to find out ›À œ:
The poles of the Chebyshev LPF is found by
Discrete Time Signal Processing 43
 ˆ = − sin  sinh % + I cos  cosh % ,  = 1, 2, 3, … … … , 

2 − 1²
where
 = ,  = 1, 2, 3, … … … , 
2
1 1
% = sinh
 n o
 Ñ

1
where

Ñ =Õ  −1
s¶
The denominator polynomial of l‡ ˆ is given by
' ˆ = ˆ − ˆ ˆ − ˆ  … … … ˆ − ˆ 
The numerator polynomial of l‡ ˆ is given by
' 0
, vℎ   ¸ 
 ˆ = Ó √1 + Ñ 
' 0, vℎ   ..

 ˆ
The normalized transfer function of the Butterworth LPF is given by
l ˆ =
' ˆ
Step 5: Determination of Digital Transfer Function › :
lŽcan be obtained from l‡ ˆ using either impulse invariant transformation or bilinear transformation.

1 1
Impulse Invariant Transformation:
→
ˆ − ! 1 − ‡ Ž

1 −1/
 . /
 1
→ w x
ˆ + !/ - − 1! .ˆ /
 1 −
 Ž
 →‡
ˆ+! 1 −
‡ cos &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž

&
‡ sin &) Ž

→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


2 1 − Ž

Bilinear Transformation:
ˆ→
) 1 + Ž

Example 1 Design a digital Chebyshev low pass filter to satisfy the following constraint using impulse
invariant transformation

L. ‚ ≤ µ›žb® µ ≤ @, L ≤ ® ≤ L. B°

µ›žb® µ ≤ L. B, L. ° ≤ ® ≤ °

Assume  = L. B œžŸÖ>×.

s¶ = 0.65
Specification

s = 0.35
š¶ = 0.3²
š = 0.55²
) = 0.3 ˆ {
Type of transformation: Impulse Invariant Transformation
Discrete Time Signal Processing 44
Type of filter: Chebyshev Low Pass filter

š¶ 0.3²
Step 1: To find analog edge frequencies
Ω¶ = = = 3.1416 !.⁄ˆ {
) 0.3
š 0.55²
ِ = = = 5.7596 !.⁄ˆ {
) 0.3
Step 2: To find the order of the filter
1 1
cosh
 ÓØ − 1ÙÕ  − 1Ô
s
 s¶
≥
cosh
 iِ ⁄Ω¶ j
1 1
cosh
 ÐØ − 1¾Ø − 1Ò
0.35 0.65
≥
cosh
 5.7596⁄3.1416
cosh
?2.6764⁄1.1691D
≥
cosh
 5.7596⁄3.1416
1.4699
≥
1.2149
 ≥ 1.2041
=2

Ω³ = Ω¶
Step 3: To find cut off frequency

Ω³ = 3.1416 !.⁄ˆ {
Step 4: To find the analog transfer function

ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh % ,  = 1, 2, 3, … … … , 

The poles of the Chebyshev LPF is found by

ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh % ,  = 1, 2

2 − 1²
where
 = ,  = 1, 2, 3, … … … , 
2
2 − 1²
 = ,  = 1, 2
4
²
 =
4
3²
 =
4
1 1
% = sinh
 n o
 Ñ

1
where

Ñ = Õ  − 1 = 1.1691
s¶

1 1
Hence
% = sinh
 n o
2 1.1691
% = 0.3877

ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh %

Now the poles are

Discrete Time Signal Processing 45

 ² ²
ˆ = −3.1416 sin p q sinh0.3877 + IΩ³ cos p q cosh0.3877
4 4
ˆ = −0.883 + I2.3905
ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh %
3² 3²
ˆ = −3.1416 sin n o sinh0.3877 + IΩ³ cos n o cosh0.3877
4 4
ˆ = −0.883 − I2.3905
The denominator polynomial of l‡ ˆ is given by
' ˆ = ˆ − ˆ ˆ − ˆ 
' ˆ = ˆ + 0.883 − I2.3905ˆ + 0.883 + I2.3905
' ˆ = ˆ + 0.883 + 2.3905
Since  = 2, which is even, the numerator polynomial of l‡ ˆ is given by
' 0
 ˆ =
√1 + Ñ 
0.883 + 2.3905
 ˆ =
√1 + 1.1691
 ˆ = 4.2213

 ˆ
The transfer function of the Butterworth LPF is given by
l‡ ˆ =
' ˆ
4.2213
l‡ ˆ =
ˆ + 0.883 + 2.3905
4.2213
l‡ ˆ = 
ˆ + 1.766ˆ + 6.4942
1
4.2213 ∗ 2.3905 ∗
Step 5: To find the digital transfer function

l‡ ˆ = 2.3905
ˆ + 0.883 + 2.3905
1.7659 ∗ 2.3905
l‡ ˆ =
ˆ + 0.883 + 2.3905

&
‡ sin &) Ž

We know that
→
ˆ + ! + &  1 − 2
‡ cos &) Ž
 +
‡ Ž


1.7659
.OOH∗.H sin2.8905 ∗ 0.3 Ž

Hence
lŽ =
1 − 2
.OOH∗.H cos2.8905 ∗ 0.3 Ž
 +
∗.OOH∗.H Ž

1.0331Ž

lŽ =
1 − 0.9929Ž
 + 0.5887Ž

Since ) = 0.3 sec < 1 ˆ {,
0.3 ∗ 1.0331Ž

lŽ =
1 − 0.9929Ž
 + 0.5887Ž

0.3099Ž

lŽ =
1 − 0.9929Ž
 + 0.5887Ž

Example 2: Design a digital Chebyshev filter using Bilinear transformation to satisfy the following

2 dB ripples in the pass band L ≤ ® ≤ L. CA°

constraint

15 dB attenuation in the stop band L. ° ≤ ® ≤ °

Solution:
Discrete Time Signal Processing 46
 ζ = −20 log s¶ = 2 .t
ΐ = −20 log s = 15 .t
š¶ = 0.42²
š = 0.55²
Assume ) = 1 ˆ {
Type of Transformation: Bilinear Transformation

−20 log s¶ = 2 .t
Type of Filter: Chebyshev LPF

log s¶ = −0.1
s¶ = 10
. = 0.7943
−20 log s = 15 .t
log s = −0.75
s = 10
.PT = 0.1778

2 š¶ 0.42²
o = 1.5514 !.⁄ˆ {
Step 1: To find the analog edge frequencies
Ω¶ = tan p q = 2 tan n
) 2 2
2 š 0.55²
ِ = tan p q = 2 tan n o = 3.3417 !.⁄ˆ {
) 2 2
Step 2: To find the order of the filter
1 1
cosh
 ÚØ − 1ÙÕ  − 1Û
s s¶
≥
cosh
 iِ ⁄Ω¶ j
1 1
cosh
 £Ø − 1¾Ø − 1¤
0.1778  0.7943
≥
cosh
 3.3417⁄1.5514
cosh
 5.5347⁄0.7649
≥
cosh
 3.3417⁄1.5514
 ≥ 2.7476
=3

Ω³ = Ω¶ = 1.5514 !.⁄ˆ {
Step 3: To find the cut off frequency

Step 4: To find the analog transfer function

ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh % ,  = 1, 2, 3, … … … , 

The poles of the Chebyshev LPF is found by

ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh % ,  = 1, 2, 3

2 − 1²
where
 = ,  = 1, 2, 3, … … … , 
2
2 − 1²
 = ,  = 1, 2, 3
6
²
 =
6
3² ²
 = =
6 2
5²
H =
6

Discrete Time Signal Processing 47

 1
Ñ = Õ  − 1 = 0.7649
s¶
1 1
%= sinh
 n o = 0.361
 Ñ
ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh %
² ²
ˆ = −1.5514 sin p q sinh0.361 + I1.5514 cos p q cosh0.361
6 6
ˆ = −0.2862 + I1.4321
ˆ = −Ω³ sin  sinh % + IΩ³ cos  cosh %
² ²
ˆ = −1.5514 sin p q sinh0.361 + I1.5514 cos p q cosh0.361
2 2
ˆ = −0.5723
ˆH = −Ω³ sin H sinh % + IΩ³ cos H cosh %
5² 5²
ˆH = −1.5514 sin n o sinh0.361 + I1.5514 cos n o cosh0.361
6 6
ˆ = −0.2862 − I1.4321
The denominator polynomial for l‡ ˆ is
' ˆ = ˆ − ˆ ˆ − ˆ ˆ − ˆH 
' ˆ = ˆ + 0.2862 − I1.4321ˆ + 0.5723ˆ + 0.2862 + I1.4321
' ˆ = ˆ + 0.5723ˆ + 0.2862 + 1.4321 
' ˆ = ˆ + 0.5723ˆ  + 0.5724ˆ + 2.1328
' ˆ = ˆ H + 1.447ˆ  + 2.4604ˆ + 1.2206
The numerator polynomial of l‡ ˆ when  odd is
 ˆ = ' 0
 ˆ = 1.2206

 ˆ
Hence,
l‡ ˆ =
' ˆ
1.2206
l‡ ˆ = H
ˆ + 1.447ˆ  + 2.4604ˆ + 1.2206

lŽ = *l‡ ˆ+  

« ¬­
Step 5: To find the digital transfer function
→ n o
 #« ¬­
1.2206
lŽ = w x
ˆH + 1.447ˆ  + 2.4604ˆ + 1.2206 →n
« ¬­
¬­ o
 #«
1.2206
lŽ = w H x
ˆ + 1.447ˆ + 2.4604ˆ + 1.2206 →n
« ¬­

¬­ o #«
1.2206
lŽ = H 
1 − Ž
 1 − Ž
 1 − Ž

£2 n o¤ + 1.447 £2 n o¤ + 2.4604 £2 n o¤ + 1.2206
1 + Ž
 1 + Ž
 1 + Ž

1.22061 + Ž
 H
lŽ =
81 − Ž
 H + 4.57881 − Ž
  1 + Ž
  + 4.93281 − Ž
 1 + Ž
  + 1.22061 + Ž
 H
1.2206 + 3.6618Ž
 + 3.6618Ž
 + 1.2206Ž
H
lŽ =
18.7322 − 19.8942Ž
 + 18.1502Ž
 − 7.1334Ž
H
0.0652 + 0.1955Ž
 + 0.1955Ž
 + 0.0652Ž
H
lŽ =
1 − 1.0668Ž
 + 0.9689Ž
 − 0.3808Ž
H

Discrete Time Signal Processing 48

Frequency Transformation:

Low pass filter with cut off frequency ΩŸ to Low pass filter with cut off frequency Ω∗Ÿ :
Analog Frequency Transformation:

Ω³
ˆ→ ∗ˆ
Ω³
Low pass filter with cut off frequency ΩŸ to High pass filter with cut off frequency Ω∗Ÿ :
Ω³ Ω∗³
ˆ→
ˆ
Low pass filter with cut off frequency ΩŸ to Band pass filter with cut off frequencyΩ@ and ΩA :
ˆ  + Ω Ω
ˆ → Ω³
ˆΩ − Ω 
Low pass filter with cut off frequency ΩŸ to Bandstop filter with cut off frequencyΩ@ and ΩA :
ˆΩ − Ω 
ˆ → Ω³ 
ˆ + Ω Ω

Low pass filter with cut off frequency ®Ÿ to Low pass filter with cut off frequency ®∗Ÿ :
Digital Frequency Transformation:

Ž
 − !
Ž →


1 − !Ž


sin*š³ − š³∗ ⁄2+

where
!=
sin*š³ + š³∗ ⁄2+
Low pass filter with cut off frequency ®Ÿ to High pass filter with cut off frequency ®∗Ÿ :
Ž
 + !
Ž
 → − : ;
1 + !Ž


cos*š³ − š³∗ ⁄2+

where
!=
cos*š³ + š³∗ ⁄2+
Low pass filter with cut off frequency ®Ÿ to Band pass filter with cut off frequency®@ and ®A :
Ž
 − ! Ž
 + !
Ž
 → − : ;
! Ž
 − ! Ž
 + 1

2Î
where
! = −
+1
−1
! =
+1
cos*š + š ⁄2+
Î=
cos*š − š ⁄2+
š − š š³
 = cot p q tan p q
2 2
Low pass filter with cut off frequency ®Ÿ to Band stop filter with cut off frequency®@ and ®A :
Ž
 − ! Ž
 + !
Ž → −:


;
! Ž
 − ! Ž
 + 1

2Î
where
! = −
+1
1−
! =
1+
Discrete Time Signal Processing 49
 cos*š + š ⁄2+
Î=
cos*š − š ⁄2+
š − š š³
 = tan p q tan p q
2 2
Structure of IIR Filter
• Direct form I realization
• Direct form II realization
• Cascade realization
• Parallel realization

Direct Form – I

! + ! Ž
 + ! Ž
 + … … … + ! Ž

The general form of the IIR filter is given by
lŽ =
1 + & Ž
 + & Ž
 + … … … + & Ž

$Ž ! + ! Ž
 + ! Ž
 + … … … + ! Ž

=

Ž 1 + & Ž
 + & Ž
 + … … … + & Ž

$Ž*1 + & Ž
 + & Ž
 + … … … + & Ž
 + =
Ž*! + ! Ž
 + ! Ž
 + … … … + ! Ž
 +
$Ž =
Ž*! + ! Ž
 + ! Ž
 + … … … + ! Ž
 + − $Ž*& Ž
 + & Ž
 + … … … + & Ž
 +

Direct Form – II

! + ! Ž
 + ! Ž
 + … … … + ! Ž

The general form of the IIR filter is given by
lŽ =
1 + & Ž
 + & Ž
 + … … … + & Ž


Discrete Time Signal Processing 50

 $Ž ! + ! Ž
 + ! Ž
 + … … … + ! Ž

=

Ž 1 + & Ž
 + & Ž
 + … … … + & Ž

$Ž )Ž 1
∗ = *! + ! Ž
 + ! Ž
 + … … + ! Ž
 + ∗ w x
)Ž
Ž 1 + & Ž + & Ž + … … … + & Ž





The system lŽ can be divided into two system connected in cascade where one system having input of 
which produces a temporary output ‰ and the other system accept the intermediate input ‰ and produces
an output %.

lŽ.
Consider the transfer function of these two systems as follow such that the product of two transfer function is

$Ž
= ! + ! Ž
 + ! Ž
 + … … + ! Ž

)Ž

)Ž 1
and
=

Ž 1 + & Ž + & Ž + … … … + & Ž





Cascade realization

! + ! Ž
 + ! Ž
 + … … … + ! Ž

The general form of the IIR filter is given by
lŽ =
1 + & Ž
 + & Ž
 + … … … + & Ž


! + ! Ž
 + ! Ž
 + … … … + ! Ž

Now factorize the numerator and denominator polynomial as follow
lŽ =
1 + & Ž
 + & Ž
 + … … … + & Ž

iv + v Ž + … … … + v⁄ Ž
⁄ ji% + % Ž
 + … … … + %⁄ Ž
⁄ j


lŽ =
1 +  Ž
 + … … … +  Ž
⁄ 1 + Ž Ž
 + … … … + Ž Ž
⁄ 

Discrete Time Signal Processing 51

 v + v Ž
 + … … … + v⁄ Ž
⁄ % + % Ž
 + … … … + %⁄ Ž
⁄
lŽ = £ ¤£ ¤
1 +  Ž
 + … … … + ⁄ Ž
⁄ 1 + Ž Ž
 + … … … + Ž⁄ Ž
⁄
lŽ = l Žl Ž

v + v Ž
 + … … … + v⁄ Ž
⁄
where

l Ž =
1 +  Ž
 + … … … + ⁄ Ž
⁄
% + % Ž
 + … … … + %⁄ Ž
⁄
l Ž =
1 + Ž Ž
 + … … … + Ž⁄ Ž
⁄

Parallel Realization
Parallel realization of the IIR filter is realized due to the fast computation by giving the input to the various part

! + ! Ž
 + ! Ž
 + … … … + ! Ž

of the system. The general form of the IIR filter is given by
lŽ =
1 + & Ž
 + & Ž
 + … … … + & Ž

Here the denominator polynomial of the system transfer function can be factorized into two polynomials

! + ! Ž
 + ! Ž
 + … … … + ! Ž

as follow:
lŽ =
1 +  Ž
 + … … … +  Ž
⁄ 1 + Ž Ž
 + … … … + Ž Ž
⁄ 

v + v Ž
 + … … … + v
⁄ Ž

⁄ % + % Ž
 + … … … + %
⁄ Ž

⁄
Now taking partial fraction for the above expression,

lŽ = +
1 +  Ž
 + … … … + ⁄ Ž
⁄ 1 + Ž Ž
 + … … … + Ž⁄ Ž
⁄
lŽ = l Ž + l Ž

v + v Ž
 + … … … + v
⁄ Ž

⁄
where

l Ž =
1 +  Ž
 + … … … + ⁄ Ž
⁄
% + % Ž
 + … … … + %
⁄ Ž

⁄
l Ž =
1 + Ž Ž
 + … … … + Ž⁄ Ž
⁄
Here the inputs are given to the two systems having transfer function l Ž and l Ž respectively.
Discrete Time Signal Processing 52
The output of the two systems are added to get the output of the overall system with transfer function lŽ.
Each system in the above transfer function lŽ can be realized in the direct form II because the delay element
needed is very less compared to direct form I realization.

@ + L. C 
@ + L. LC 
A
Example 1: Realize the following IIR filter with transfer function
›  =
@ − L.  
@ + L. LB 
A
using direct form I, direct form II, and cascade realization.
Solution:
Direct Form – I

Discrete Time Signal Processing 53

Direct Form II

1 + 0.15Ž
 1 + 0.3Ž
  1 + 0.15Ž
 1 + 0.3Ž

Cascade Realization
lŽ = = £ ¤£ ¤
1 − 0.2Ž
 1 − 0.3Ž
  1 − 0.2Ž
 1 − 0.3Ž


@ + L. ‚ 
@
Example 2: Realize the following IIR system with transfer function
›  =
@ + L. B 
@ + L. LA 
A
using parallel realization.
Solution:

Discrete Time Signal Processing 54

 1 + 0.6Ž

lŽ =
1 + 0.3Ž
 + 0.02Ž

1 + 0.6Ž

lŽ =
1 + 0.1Ž
 1 + 0.2Ž
 
1 + 0.6Ž

s t
= +
1 + 0.1Ž
 1 + 0.2Ž
  1 + 0.1Ž
 1 + 0.2Ž

s1 + 0.2Ž
  + t1 + 0.1Ž
  = 1 + 0.6Ž

Put Ž
 = −5, then
t1 − 0.5 = 1 − 3
0.5t = −2
t = −4
Put Ž
 = −10, then
s1 − 2 = 1 − 6
s=5
5 −4
lŽ = +
1 + 0.1Ž
 1 + 0.2Ž


Discrete Time Signal Processing 55

 Unit III FIR Filter Design
Linear Phase FIR Filters
The FIR filter has the finite impulse response and has only zeros. FIR filter have an exact linear phase and

The Discrete Time Fourier Transform of a finite sequence impulse response ℎ is given by
stable compared to IIR filter.




li – j =  ℎ
– = µl – µ ∅–

Any frequency response contains two components namely magnitude and phase responses which are

}š = µli – jµ = Ý?8 *l – +D + ?4-*l – +D 

given by

4- pli – jq
∅š = tan

Ú Û
8 il – j
Filters are classified into linear and non – linear phase filter based on the delay function namely the
phase delay and group delay.

∅š
The phase and group delays of the filter are given by

Þ¶ = −
š

.∅š
and

Þß = −
.š
The group delay is defined as the delayed response of the filter as a function of š to a signal.
Linear phase FIR filters is defined as the filter in which both the phase delay and group delay are
independent of frequency. Hence Linear phase filters are also called constant time delay filters.

phase delay of the FIR filter Þ¶ must be a constant (let it be Þ)

Let us obtain the conditions for FIR filters to have the linear phase. For the phase response to be linear,

∅š
− = Þ, − ² ≤ š ≤ ²
š

∅š = −šÞ
Therefore,

4- pli – jq
We know that

∅š = − tan

Ú Û
8 il – j

4- pli – jq
−šÞ = − tan

Ú Û
8 il – j
∑

 ℎ sin š
šÞ = tan
 £ 
 ¤
∑ ℎ cos š
∑

 ℎ sin š
tan šÞ = 

∑ ℎ cos š
Discrete Time Signal Processing 56
 sin šÞ ∑

 ℎ sin š
= 

cos šÞ ∑ ℎ cos š

 


 ℎ sin š cos šÞ =  ℎ cos š sin šÞ

 

 


 ℎ cos š sin šÞ −  ℎ sin š cos šÞ = 0

 



 ℎcos š sin šÞ − sin š cos šÞ = 0





 ℎ sinšÞ − š = 0


The above equation is the condition for the FIR filter to be linear phase and the solutions of above

}−1
equation are
Þ=
2

ℎ = ℎ} − 1 −   0 <  < } − 1

and

If the above two solutions are satisfied, then the FIR filter will have constant phase and group delays and
thus the phase of the filter will be linear. The phase and group delay of the linear phase FIR filter are equal and
constant over the frequency band. Whenever a constant group delay alone is preferred, the impulse response

ℎ = −ℎ} − 1 − 
will be of the form

and is antisymmetric about the center of the impulse response sequence.

Frequency response of Linear Phase FIR filters

The Discrete Time Fourier Transform of a finite sequence impulse response ℎ is given by
Case 1: Symmetric, M odd




li –
j =  ℎ
– = µl – µ ∅–



H
If the filter length M is odd, then the above equation becomes
 

} − 1
–p
q
li –
j =  ℎ
–
+ℎn o  +  ℎ
–
2
 #



ℎ = ℎ} − 1 − 
If the frequency response is symmetric,


H
Hence the frequency response can be written as

} − 1
–p
q
li – j =  ℎ1
– +
–

 2 + ℎ n o 
2


Factorizing
ᬭ

–p q
â in the above equation,

Discrete Time Signal Processing 57

 
H
\  _


q [

 
 }−1
li j =  ℎ w + x+ℎn o^
–
–p –p
q
–p
q
  
[  2 ^
Z ]
Put  = − ,





\  _


q [ }−1 }−1
li j =  ℎn − o 1 – +
– 2 + ℎ n o^
–
–p

[  2 2 ^
Z ]


\  _


q [ }−1
li j =
–
–p
  ! cos š) + !0ℎ n o ^
[  2 ^
Z ]





li – j =
–p  ! cos š)
q


li –
j=
∅–
}š

}−1 }−1
where,
! = 2ℎ n − o , !0 = ℎ n o
2 2
The above equation defines the frequency response of symmetric linear phase FIR filter with order of
the filter be odd.

The Discrete Time Fourier Transform of a finite sequence impulse response ℎ is given by
Case 2: Symmetric, M even




li –
j =  ℎ
– = µl – µ ∅–



H
If the filter length M is even, then the above equation becomes
 


li –
j =  ℎ
–
+  ℎ
–
 



H 
H
 

li – j =  ℎ
– +  ℎ} − 1 − 
–


 

ℎ = ℎ} − 1 − 
If the frequency response is symmetric,


H 
H
Hence the frequency response can be written as
 

li – j =  ℎ
– +  ℎ
–


 

Discrete Time Signal Processing 58

 
H 
H
\  




_

–
⁄ [
ä ^
li j =  ℎ +  ℎ
– –ã
ä
–ã
 
[  ^
Z ]

 
\   _

–
⁄ [
}  } 

–ã
ä ^
li j =  ℎ n − o +  ℎ n − o
– –ã
ä
 
[ 2 2 ^
Z ]



} 1
li – j =
–
⁄  2ℎ n − o cos ½š w − x )¿
2 2



1
li – j =
–
⁄  & cos ½š w − x )¿
2


where & = 2ℎ p − q


li – j =
–
⁄ l å i – j
li – j = l å i – j æ–
å i – j 
ç–
li – j = l


where

1
å i – j =  & cos ½š w − x )¿
l
2

èš = −Κ

}−1
and
Î=
2
The above equation defines the frequency response of symmetric linear phase FIR filter with order of
the filter be even.
Case 3: Antisymmetric, M odd

}−1
For this type of sequence,
ℎn o=0
2
The Discrete Time Fourier Transform of a finite sequence impulse response ℎ is given by



li –
j =  ℎ
– = µl – µ ∅–



H
If the filter length M is odd, then the above equation becomes
 

} − 1
–p
q
li –
j =  ℎ
–
+ℎn o  +  ℎ
–
2 #
 


Discrete Time Signal Processing 59

 
H
 


li –
j =  ℎ
–
+  ℎ
–
 #



ℎ = −ℎ} − 1 − 
If the frequency response is antisymmetric,


H
Hence the frequency response can be written as


li – j =  ℎ1
– −
–

 2


Factorizing
ᬭ

–p q
â

H
in the above equation,


 
 

li – j =  ℎ w − x

–p q –p
q
–p
q
  

Put  = − ,





\  _

 }−1
li j =
–
–p

q [  ℎn − o 1 –
+
–
2^
[  2 ^
Z ]




\  _
 ⁄ [
li j =
–
–p

q
 { sin š)^
[  ^
Z ]
li – j = ∅– l å š

}−1
where,
{ = 2ℎ n − o
2

² ² }−1
and
∅š =
− Κ = − n oš
2 2 2
The above equation defines the frequency response of antisymmetric linear phase FIR filter with order
of the filter be odd.

The Discrete Time Fourier Transform of a finite sequence impulse response ℎ is given by
Case 4: Antisymmetric, M even




li –
j =  ℎ
– = µl – µ ∅–



H
If the filter length M is even, then the above equation becomes
 


li –
j =  ℎ
–
+  ℎ
–
 



Discrete Time Signal Processing 60

 
H 
H
 

li – j =  ℎ
– +  ℎ} − 1 − 
–


 

ℎ = −ℎ} − 1 − 
If the frequency response is antisymmetric,


H 
H
Hence the frequency response can be written as
 

li – j =  ℎ
– −  ℎ
–


 

H 
H
\  




_
li –
j=
–
⁄ [  ℎ –ã 

ä
−  ℎ

–ã


ä ^
[  ^
Z ]

 
\   _

–
⁄  ⁄ [
}  } 

–ã
ä ^
li j =
–
 ℎ n − o
–ã


ä
−  ℎ n − o 
[ 2 2 ^
Z ]



} 1
li – j =
–
⁄  ⁄  2ℎ n − o sin ½š w − x )¿
2 2



1
li – j =
–
⁄  ⁄  . sin ½š w − x )¿
2


where . = 2ℎ p  − q


å i – j
li – j =
–
⁄  ⁄ l
li – j = lå i – j æ–

å i – j
li – j = l
p
ç–q



where

1
å i – j =  . sin ½š w − x )¿
l
2

²
èš = − Κ
2

}−1
and
Î=
2
The above equation defines the frequency response of antisymmetric linear phase FIR filter with order of the
filter be even.
Design of FIR filter by Fourier Series Method

Desired frequency response, lh i – j

Given specification:

Cut off frequency š³ for Low pass and High pass, and š³ and š³ for Band pass and Band stop filter
Discrete Time Signal Processing 61
The number of samples, }
Note: If the frequency is given in Hz, then š³ (rad/sec) can be obtained by
2²³
š³ =

³ is the cut off frequency in Hz and  is the sampling frequency in Hz.
Filter Desired frequency response:

1, − š³ ≤ š ≤ š³
lh i – j = ½
Low Pass Filter:

0, š³ < |š| ≤ ²

0, − š³ ≤ š ≤ š³
lh i – j = ½
High Pass Filter:

1, š³ < |š| ≤ ²

1, š³ ≤ |š| ≤ š³
Band Pass Filter:

lh i j = Ó 0, − š³ ≤ š ≤ š³
–

0, š³ ≤ |š| ≤ ²

0, š³ ≤ |š| ≤ š³
Band Reject Filter:

lh i j = Ó 1, − š³ ≤ š ≤ š³
–

1, š³ ≤ |š| ≤ ²
For non ideal (practical) filter, replace 1 by .
–ç

Step 1: Choose the desired frequency response lh i – j

Step 2: Determination of desired impulse response ℎh 
1
ℎh  = ¢ l i – j – .š
2²
h
Step 3: Calculate M samples of ℎh  for  = 0 ‰ } − 1
ℎ = ℎh / Šg 

Step 4: Determination of lŽ


For causal (non ideal or practical) filter,

lŽ =  ℎŽ





For non causal (ideal) filter,


lŽ =  ℎŽ





li – j = lŽ/«→e êë
Step 5: Determination of Frequency Response

Example 1: Design a FIR HPF with cut off frequency of 0.45π radian/samples using Fourier Series.
Assume M=7.

}=7
Specification:

š³ = 0.45² !.⁄ˆ!-·" ˆ
Type of filter: FIR HPF
Type of design method: Hanning Window
Step 1: Choose the desired frequency response

Discrete Time Signal Processing 62

 0, |š| ≤ š³
lh i – j = ½
–ç
, š³ ≤ |š| ≤ ²
}−1 7−1
Î= = =3
2 2
0, |š| ≤ 0.45²
lh i – j = ½
H–
, 0.45² ≤ |š| ≤ ²
Step 2: To find |× >

1
ℎh  = ¢ lh i – j – .š
2²



.GT
1
ℎh  = § ¢
H– – .š + ¢
H– – .š¨
2²
.GT


.GT
1
ℎh  = § ¢
H
– .š + ¢
H
– .š¨
2²
.GT

1
H
–

H
–
.GT
ℎh  = 5: ; +: ; 6
2² −I3 −  .GT −I3 − 

−1
ℎh  = 1
H
 −
H
.GT + H
.GT − H
 2
2I3 − ²
1
ℎh  = 1sini3 − ²j − sini3 − 0.45²j2,  ≠ 3
3 − ²
For  = 3, apply L’ hospitals rule,
−² cosi3 − ²j + 0.45² cosi3 − 0.45²j
ℎh 3 = : ;
−² →H
−² + 0.45²
ℎh 3 = = 0.55
−²
For remaining values in the limit 0 ≤  ≤ } − 1, i.e., 0 ≤  ≤ 6
sin3² − sin3 ∗ 0.45²
ℎh 0 = ℎh 6 = = 0.0945
3²
sin2² − sin2 ∗ 0.45²
ℎh 1 = ℎh 5 = = −0.0492
2²
sin1² − sin1 ∗ 0.45²
ℎh 2 = ℎh 4 = = −0.3144
1²

ℎ = ℎh  = ?0.0945, − 0.0492, − 0.3144, 0.55, − 0.3144, − 0.0492, 0.0945D

Hence,



Step 5: To find the transfer function

lŽ =  ℎŽ


V

lŽ =  ℎŽ


lŽ = Ž
H *0.55 − 0.3144Ž + Ž
  − 0.0492Ž  + Ž
  + 0.0945Ž H + Ž
H +

li – j = lŽ/«→e êë
Step 6: To find the frequency response

li – j =
H– *0.55 − 0.6288 cos š − 0.0984 cos 2š + 0.189 cos 3š+
Discrete Time Signal Processing 63
 }i – j = 0.55 − 0.6288 cos š − 0.0984 cos 2š + 0.189 cos 3š
Magnitude Response

∅i – j = −3š
Phase Response

Fourier Series method. Assume í = ƒ.

Example 2: Design an ideal LPF with cut off frequency of 7 kHz and sampling frequency of 18 kHz using

}=7
Specification:

³ = 7 lŽ
 = 18 lŽ
Type of filter: Ideal FIR LPF

2²³ 2² ∗ 7 ∗ 10H
Type of design method: Fourier Series
š³ = = = 0.7778² !.⁄ˆ {
 18 ∗ 10H

1, |š| ≤ š³
Step 1: Choose the desired frequency response
lh i – j = ½
0, š³ ≤ |š| ≤ ²
1, |š| ≤ 0.7778²
lh i – j = ½
0, 0.7778² ≤ |š| ≤ ²
Step 2: To find |× >

1
ℎh  = ¢ lh i – j – .š
2²


.PPPO
1
ℎh  = ¢ – .š
2²

.PPPO
1 – .PPPO
ℎh  = : ;
2² I
.PPPO
1
ℎh  = 1
.PPPO − .PPPO 2
2I²
1
ℎh  = sin0.7778² ,  ≠ 0
²
For  = 0, apply L’ hospitals rule,
0.7778² cos0.7778²
ℎh 0 = : ;
² →
ℎh 0 = 0.7778
sin0.7778 ∗ 1²
ℎh 1 = ℎh −1 = = 0.2046
1²
sin0.7778 ∗ 2²
ℎh 2 = ℎh −2 = = −0.1567
2²
sin0.7778 ∗ 3²
ℎh 3 = ℎh −3 = = 0.0919
3²
> −B −A −@ L @ A B
î× > 0.0919 −0.1567 0.2046 0.7778 0.2046 −0.1567 0.0919
Step 3: To find the transfer function

Discrete Time Signal Processing 64

 



lŽ =  ℎŽ





lŽ =  ℎŽ


H
lŽ = 0.7778 + 0.2046Ž + Ž
  − 0.1567Ž  + Ž
  + 0.0919Ž H + Ž
H 

li – j = lŽ/«→e êë
Step 4: To find the frequency response

li – j = 0.7778 + 0.4092 cos š − 0.3134 cos 2š + 0.1838 cos 3š

Design of FIR Filter using Windowing technique

Desired frequency response, lh i – j

Given specification:

Cut off frequency š³ (rad/sec) for Low pass and High pass, and š³ (rad/sec) and š³ (rad/sec) for Band pass

The number of samples, }

and Band stop filter

Note: If the frequency is given in Hz, then š³ (rad/sec) can be obtained by

2²³
š³ =

³ is the cut off frequency in Hz and  is the sampling frequency in Hz.
Step 1: Choose the desired frequency response lh i – j
Step 2: Determination of ℎh :
1
ℎh  = ¢ l i – j – .š
2²
h
Step 3: Determination of ℎ:
ℎ = ℎh v
where v is the windowing function
Step 4: Determination of lŽ


For causal (non ideal or practical) filter,

lŽ =  ℎŽ





For non causal (ideal) filter,


lŽ =  ℎŽ





li – j = lŽ/«→e êë
Step 5: Determination of Frequency Response

Window Function
Rectangular Window Function:

1 , 0 ≤  < } − 1
vï  = Ã
Causal Rectangular Window Function:

0 , ‰ℎ v¹ˆ
Non Causal Rectangular Window Function:

Discrete Time Signal Processing 65

 }−1
vï  = ð 1, || ≤
2
0, ‰ℎ v¹ˆ
Hamming Window Function:

2²
Causal Hamming Window Function:

vñ  = ð 0.54 − 0.46 cos , 0 ≤  < } − 1

}−1
0 , ‰ℎ v¹ˆ

2² }−1
Non Causal Hamming Window Function:

vñ  = ð 0.54 + 0.46 cos } − 1 , || ≤ 2

0 , ‰ℎ v¹ˆ
Hanning Window Function:

2²
Causal Hanning Window Function:

vñ‡  = ð0.5 − 0.5 cos } − 1 , 0 ≤  < } − 1

0 , ‰ℎ v¹ˆ

2² }−1
Non Causal Hanning Window Function:

vñ  = ð 0.5 + 0.5 cos , || ≤

}−1 2
0 , ‰ℎ v¹ˆ
Blackmann Window Function:

2² 4²
Causal Blackmann Window Function:

vÉ  = ð 0.42 − 0.5 cos + 0.08 cos , 0 ≤  < } − 1

}−1 }−1
0 , ‰ℎ v¹ˆ

2² 4² }−1

Non Causal Blackmann Window Function:

vÉ  = ð 0.42 + 0.5 cos + 0.08 cos , || ≤

}−1 }−1 2
0 , ‰ℎ v¹ˆ
Example 1: Design an ideal FIR LPF with cut off frequency of 990 kHz and sampling frequency of 2
MHz using Hamming window. Assume M=9.

}=9
Specification:

³ = 990 lŽ
 = 2 }lŽ
Type of filter: Ideal FIR LPF

2²³ 2² ∗ 990 ∗ 10H

Type of design method: Hamming Window
š³ = = = 0.99² !.⁄ˆ {
 2 ∗ 10V

1, |š| ≤ š³
Step 1: Choose the desired frequency response
lh i – j = ½
0, š³ ≤ |š| ≤ ²
1, |š| ≤ 0.99²
lh i – j = ½
0, 0.99² ≤ |š| ≤ ²
Step 2: To find |× >

1
ℎh  = ¢ lh i – j – .š
2²

Discrete Time Signal Processing 66

 .JJ
1
ℎh  = ¢ – .š
2²

.JJ
1 – .JJ
ℎh  = : ;
2² I
.JJ
1
ℎh  = 1
.JJ − .JJ 2
2I²
1
ℎh  = sin0.99² ,  ≠ 0
²
For  = 0, apply L’ hospitals rule,
0.99² cos0.99²
ℎh 0 = : ;
² →
ℎh 0 = 0.99
sin0.99 ∗ 1²
ℎh 1 = ℎh −1 = = 0.01
1²
sin0.99 ∗ 2²
ℎh 2 = ℎh −2 = = −0.01
2²
sin0.99 ∗ 3²
ℎh 3 = ℎh −3 = = 0.01
3²
sin0.99 ∗ 4²
ℎh 4 = ℎh −4 = = −0.01
4²
2² }−1 }−1
dñ  = 0.54 + 0.46 cos n
Step 3: To find the window sequence
o , − ≤≤
}−1 2 2
2²
dñ  = 0.54 + 0.46 cos n o , − 4 ≤  ≤ 4
8
²
dñ  = 0.54 + 0.46 cos p q , − 4 ≤  ≤ 4
4
0²
dñ 0 = 0.54 + 0.46 cos n o = 1
4
1²
dñ 1 = dñ −1 = 0.54 + 0.46 cos n o = 0.8653
4
2²
dñ 2 = dñ −2 = 0.54 + 0.46 cos n o = 0.54
4
3²
dñ 3 = dñ −3 = 0.54 + 0.46 cos n o = 0.2147
4
4²
dñ 4 = dñ −4 = 0.54 + 0.46 cos n o = 0.08
4
Step 4: To find |>
ℎ = ℎh dñ 
> î× > ò› > î>
−C −0.01 0.08 −0.0008
−B 0.01 0.2147 0.0022
−A −0.01 0.54 −0.0054
−@ 0.01 0.8653 0.0087
L 0.99 1 0.99
@ 0.01 0.8653 0.0087
A −0.01 0.54 −0.0054
Discrete Time Signal Processing 67
 B 0.01 0.2147 0.0022
C −0.01 0.08 −0.0008


Step 5: To find the transfer function


lŽ =  ℎŽ





lŽ =  ℎŽ


G
lŽ = 0.99 + 0.0087Ž + Ž
  − 0.0054Ž  + Ž
  + 0.0022Ž H + Ž
H  − 0.0008Ž G + Ž
G 

li – j = lŽ/«→e êë
Step 6: To find the frequency response

li – j = 0.99 + 0.0174 cos š − 0.0108 cos 2š + 0.0044 cos 3š − 0.0016 cos 4š

Example 2: Design a FIR HPF with cut off frequency of 0.45π radian/samples using Hanning window.
Assume M=7.

}=7
Specification:

š³ = 0.45² !.⁄ˆ!-·" ˆ
Type of filter: FIR HPF
Type of design method: Hanning Window

0, |š| ≤ š³
Step 1: Choose the desired frequency response
lh i – j = ½
–ç
, š³ ≤ |š| ≤ ²
}−1 7−1
Î= = =3
2 2
0, |š| ≤ 0.45²
lh i – j = ½
H–
, 0.45² ≤ |š| ≤ ²
Step 2: To find |× >

1
ℎh  = ¢ lh i – j – .š
2²



.GT
1
ℎh  = § ¢
H– – .š + ¢
H– – .š¨
2²
.GT


.GT
1
ℎh  = § ¢
H
– .š + ¢
H
– .š¨
2²
.GT

1
H
–

H
–
.GT
ℎh  = 5: ; +: ; 6
2² −I3 −  .GT −I3 − 

−1
ℎh  = 1
H
 −
H
.GT + H
.GT − H
 2
2I3 − ²
1
ℎh  = 1sini3 − ²j − sini3 − 0.45²j2,  ≠ 3
3 − ²
For  = 3, apply L’ hospitals rule,

Discrete Time Signal Processing 68

 −² cosi3 − ²j + 0.45² cosi3 − 0.45²j
ℎh 3 = : ;
−² →H
−² + 0.45²
ℎh 3 = = 0.55
−²
sin3² − sin3 ∗ 0.45²
ℎh 0 = ℎh 6 = = 0.0945
3²
sin2² − sin2 ∗ 0.45²
ℎh 1 = ℎh 5 = = −0.0492
2²
sin1² − sin1 ∗ 0.45²
ℎh 2 = ℎh 4 = = −0.3144
1²
2²
dñ  = 0.5 − 0.5 cos n
Step 3: To find the window sequence
o , 0 ≤  ≤ } − 1
}−1
2²
dñ  = 0.5 − 0.5 cos n o , 0 ≤  ≤ 6
6
²
dñ  = 0.5 − 0.5 cos p q , 0 ≤  ≤ 6
3
0²
dñ 0 = dñ 6 = 0.5 − 0.5 cos n o = 0
3
1²
dñ 1 = dñ 5 = 0.5 − 0.5 cos n o = 0.25
3
2²
dñ 2 = dñ 4 = 0.5 − 0.5 cos n o = 0.75
3
3²
dñ 3 = 0.5 − 0.5 cos n o = 1
3
Step 4: To find |>
ℎ = ℎh dñ 
> î× > ò› > î>
L 0.0945 0 0
@ −0.0492 0.25 −0.0123
A −0.3144 0.75 −0.2358
B 0.55 1 0.55
C −0.3144 0.75 −0.2358
 −0.0492 0.25 −0.0123
‚ 0.0945 0 0



Step 5: To find the transfer function

lŽ =  ℎŽ


V

lŽ =  ℎŽ


lŽ = Ž
H *0.55 − 0.2358Ž + Ž
  − 0.0123Ž  + Ž
 +

li – j = lŽ/«→e êë
Step 6: To find the frequency response

li – j =
H– *0.55 − 0.4716 cos š − 0.0246 cos 2š+

}i – j = 0.55 − 0.4716 cos š − 0.0246 cos 2š

Magnitude Response

Phase Response
Discrete Time Signal Processing 69
 ∅i – j = −3š
Example 3: Design a Band Pass filter with a cut off frequencies of 0.2π and 0.5π radian/ samples using
Hamming window. Assume M=9.

}=9
Specification:

š³ = 0.2² !.⁄ˆ!-·" ˆ

š³ = 0.5² !.⁄ˆ!-·" ˆ
Type of filter: FIR BPF
Type of design method: Hamming Window

0, |š| ≤ š³
Step 1: Choose the desired frequency response

lh i j = Ó 0. š³ ≤ |š| ≤ ²
–

–ç , š³ ≤ |š| ≤ š³
}−1 9−1
Î= = =4
2 2
0, |š| ≤ 0.2²
lh i j = Ó 0. 0.5² ≤ |š| ≤ ²
–

–ç , 0.2² ≤ |š| ≤ 0.5²
Step 2: To find |× >

1
ℎh  = ¢ lh i – j – .š
2²


.T
.
1
ℎh  = § ¢
G– – .š + ¢
G– – .š¨
2²
.
.T
.T
.
1
ℎh  = § ¢
G
– .š + ¢
G
– .š¨
2²
.
.T

1

G
– .T
G
–
.
ℎh  = 5: ; +: ; 6
2² I4 −  . I4 − 
.T
1
ℎh  = 1
G
.T −
G
. + G
. − G
.T 2
2I4 − ²
1
ℎh  = 1sini4 − 0.5²j − sini4 − 0.2²j2,  ≠ 4
4 − ²
For  = 4, apply L’ hospitals rule,
−0.5² cosi4 − 0.5²j + 0.2² cosi4 − 0.2²j
ℎh 4 = : ;
−² →G
−0.5² + 0.2²
ℎh 4 = = 0.3
−²
sin4 ∗ 0.5² − sin4 ∗ 0.2²
ℎh 0 = ℎh 8 = = −0.0468
4²
sin3 ∗ 0.5² − sin3 ∗ 0.2²
ℎh 1 = ℎh 7 = = −0.2070
3²
sin2 ∗ 0.5² − sin2 ∗ 0.2²
ℎh 2 = ℎh 6 = = −0.1514
2²

Discrete Time Signal Processing 70

 sin1 ∗ 0.5² − sin1 ∗ 0.2²
ℎh 3 = ℎh 5 = = 0.1312
1²
2²
dñ  = 0.54 − 0.46 cos n
Step 3: To find the window sequence
o , 0 ≤  ≤ } − 1
}−1
2²
dñ  = 0.54 − 0.46 cos n o , 0 ≤  ≤ 8
8
²
dñ  = 0.54 − 0.46 cos p q , 0 ≤  ≤ 8
4
0²
dñ 0 = dñ 8 = 0.54 − 0.46 cos n o = 0.08
4
1²
dñ 1 = dñ 7 = 0.54 − 0.46 cos n o = 0.2147
4
2²
dñ 2 = dñ 6 = 0.54 − 0.46 cos n o = 0.54
4
3²
dñ 3 = dñ 5 = 0.54 − 0.46 cos n o = 0.8653
4
4²
dñ 4 = 0.54 − 0.46 cos n o = 1
4
Step 4: To find |>
ℎ = ℎh dñ 
> î× > ò› > î>
L −0.0468 0.08 −0.0038
@ −0.2070 0.2147 −0.0445
A −0.1514 0.54 −0.0818
B 0.1312 0.8653 0.1135
C 0.3 1 0.3
 0.1312 0.8653 0.1135
‚ −0.1514 0.54 −0.0818
ƒ −0.2070 0.2147 −0.0445
„ −0.0468 0.08 −0.0038



Step 5: To find the transfer function

lŽ =  ℎŽ


V

lŽ =  ℎŽ


lŽ = Ž
G *0.3 + 0.1135Ž + Ž
  − 0.0818Ž  + Ž
  − 0.0445Ž H + Ž
H  − 0.0038Ž G + Ž
G +

li – j = lŽ/«→e êë
Step 6: To find the frequency response

li – j =
G– *0.3 + 0.227 cos š − 0.1636 cos 2š − 0.089 cos 3š − 0.0076 cos 4š+

}i – j = 0.3 + 0.227 cos š − 0.1636 cos 2š − 0.089 cos 3š − 0.0076 cos 4š

Magnitude Response

∅i – j = −4š
Phase Response

Example 4: Design an ideal FIR BRF with cut off frequencies of 12 kHz and 20 kHz and sampling
frequency of 60 kHz using Hanning window. Assume M=5.
Specification:
Discrete Time Signal Processing 71
 }=5
³ = 12 lŽ
³ = 20 lŽ
 = 60 lŽ
Type of Filter: FIR BRF

2²³ 2² ∗ 12 ∗ 10H
Type of Design Method: Hanning Window
š³ = = = 0.4²
 60 ∗ 10H
2²³ 2² ∗ 20 ∗ 10H
š³ = = = 0.6667²
 60 ∗ 10H

1, |š| ≤ š³
Step 1: Choose the desired frequency response

lh i j = Ó 1. š³ ≤ |š| ≤ ²
–

0, š³ ≤ |š| ≤ š³
1, |š| ≤ 0.4²
lh i – j = ð1. 0.6667² ≤ |š| ≤ ²
0, 0.4² ≤ |š| ≤ 0.6667²
Step 2: To find |× >

1
ℎh  = ¢ lh i – j – .š
2²


.G
.VVVP
1
ℎh  = § ¢ – .š + ¢ –
.š + ¢ – .š¨
2²

.G .VVVP

1
– .G –
–
.VVVP
ℎh  = 5: ; +: ; +: ; 6
2² I
.G I .VVVP I

1
ℎh  = 1 .G −
.G +  − .VVVP +
.VVVP −
 2
2I²
1
ℎh  = *sin0.4² + sin² − sin0.6667²+,  ≠ 0
²
For  = 0, apply L’ hospitals rule,
0.4² cos0.4² + ² cos² − 0.6667² cos0.6667²
ℎh 0 = : ;
² →
0.4² + ² − 0.6667²
ℎh 0 = = 0.7333
²
sin0.4 ∗ 1² + sin1² − sin0.6667 ∗ 1²
ℎh 1 = ℎh −1 = = 0.0271
1²
sin0.4 ∗ 2² + sin2² − sin0.6667 ∗ 2²
ℎh 2 = ℎh −2 = = 0.2314
2²
2² }−1 }−1
dñ  = 0.5 + 0.5 cos n
Step 3: To find the window sequence
o , − ≤≤
}−1 2 2
2²
dñ  = 0.5 + 0.5 cos n o , − 2 ≤  ≤ 2
4
²
dñ  = 0.5 + 0.5 cos p q , − 2 ≤  ≤ 2
2

Discrete Time Signal Processing 72

 0²
dñ 0 = 0.5 + 0.5 cos n o = 1
2
1²
dñ 1 = dñ −1 = 0.5 + 0.5 cos n o = 0.5
2
2²
dñ 2 = dñ −2 = 0.5 + 0.5 cos n o = 0
2
Step 4: To find |>
ℎ = ℎh dñ 
> î× > ò› > î>
−A 0.2314 0 0
−@ 0.0271 0.5 0.0136
L 0.7333 1 0.7333
@ 0.0271 0.5 0.0136
A 0.2314 0 0


Step 5: To find the transfer function


lŽ =  ℎŽ





lŽ =  ℎŽ



lŽ = 0.7333 + 0.0136Ž + Ž
 

li – j = lŽ/«→e êë
Step 6: To find the frequency response

li – j = 0.7333 + 0.0272 cos š

• Choose the desired frequency response lh i – j

Design of FIR filter by Type 1 frequency sampling method

Sample lh i – j at M – points by taking š =  where  = 0, 1, 2, … … … , } − 1, to generate the

 
•

l = lh i – j/     = 0, 1, 2, … … … , } − 1
sequence
–

Compute the M samples of impulse response ℎ using following equation


\ _
•

1
ℎ = [ l0 + 2  8 1l   ⁄ ^
2 , } ..
}[ ^
Z ]


\ _



1
ℎ = [l0 + 2  8 1l   ⁄
2^ , } ¸ 
}[ ^
Z ]


Take Z transform of the impulse response ℎ to get the filter transfer function lŽ


•

lŽ =  ℎŽ


Example 1: Design a Low pass filter with cut off frequency of 3 kHz and sampling frequency of 9 kHz
using Type 1 frequency sampling method. Assume M=11.
Specification:
Discrete Time Signal Processing 73
 ³ = 3 lŽ
 = 9 lŽ
} = 11
Type of filter: Low Pass Filter

2²³ 2² ∗ 3 ∗ 10H
Type of Design Method: Frequency Sampling Method
š³ = = = 0.6667²
 9 ∗ 10H

ç– , 0 ≤ š ≤ š³
Step 1: Choose the desired frequency response

lh i – j = Ó
ç– , 2² − š³ ≤ š ≤ 2²
0, š³ ≤ š ≤ 2² − š³
}−1
Î= =5
2

T– , 0 ≤ š ≤ 0.6667²
lh i j = Ó
T– , 2² − 0.6667² ≤ š ≤ 2²
–

0, 0.6667² ≤ š ≤ 2² − 0.6667²

T– , 0 ≤ š ≤ 0.6667²
lh i – j = Ó
T– , 1.3333² ≤ š ≤ 2²
0, 0.6667² ≤ š ≤ 1.3333²

2²
Step 2: Sampling the Frequency Response
š = ,  = 0, 1, 2, … … … , } − 1
}
2²
š = ,  = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
11
š = 0
2²
š = = 0.1818²
11
4²
š = = 0.3636²
11
6²
šH = = 0.5454²
11
8²
šG = = 0.7272²
11
10²
šT = = 0.9090²
11
10²
šT = = 0.9090²
11
12²
šV = = 1.0909²
11
14²
šP = = 1.2727²
11
16²
šO = = 1.4545²
11
18²
šJ = = 1.6363²
11
20²
š = = 1.8181²
11
l = 1lh i – j2  
–→–ó 


Discrete Time Signal Processing 74

 l = 1lh i – j2  
–→–ó 


l = ½ ,  = 0, 1, 2, 3, 8, 9, 10

 ⁄

0,  = 4, 5, 6, 7
Step 3: To find |>


\  _
1
ℎ = [l0 + 2  8 1l   ⁄
2^ ,  = 0, 1, 2, … … … , } − 1
}[ ^
Z ]

H
1
ℎ = 51 + 2  8 1
 ⁄  ⁄ 26 ,  = 0, 1, 2, … … … , 10
11

H
1
ℎ = 51 + 2  8 1
 T
⁄ 26 ,  = 0, 1, 2, … … … , 10
11

H
1 2²5 − 
ℎ = 51 + 2  cos £ ¤6 ,  = 0, 1, 2, … … … , 10
11 11

1 2²5 −  4²5 −  6²5 − 
ℎ = :1 + 2 cos £ ¤ + 2 cos £ ¤ + 2 cos £ ¤; , 
11 11 11 11
= 0, 1, 2, … … … , 10
1 10² 20² 30²
ℎ0 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = −0.0497
11 11 11 11
1 8² 16² 24²
ℎ1 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = 0.0989
11 11 11 11
1 6² 12² 18²
ℎ2 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = −0.0339
11 11 11 11
1 4² 8² 12²
ℎ3 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = −0.1271
11 11 11 11
1 2² 4² 6²
ℎ4 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = 0.2935
11 11 11 11
1 0² 0² 0²
ℎ5 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = 0.6364
11 11 11 11
1 −2² −4² −6²
ℎ6 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = 0.2935
11 11 11 11
1 −4² −8² −12²
ℎ7 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = −0.1271
11 11 11 11
1 −6² −12² −18²
ℎ8 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = −0.0339
11 11 11 11
1 −8² −16² −24²
ℎ9 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = 0.0989
11 11 11 11
1 −10² −20² −30²
ℎ10 = w1 + 2 cos n o + 2 cos n o + 2 cos n ox = −0.0497
11 11 11 11


Step 4: To find Transfer Function

lŽ =  ℎŽ




lŽ =  ℎŽ



Discrete Time Signal Processing 75

 lŽ = Ž
T *0.6364 + 0.2935Ž + Ž
  − 0.1271Ž  + Ž
  − 0.0339Ž H + Ž
H  + 0.0989Ž G + Ž
G 
− 0.0497Ž T + Ž
T +

li – j = *lŽ+«→e êë
Step 5: To find Frequency Response

li – j =
T– *0.6364 + 0.587 cos š − 0.2542 cos 2š − 0.0678 cos 3š + 0.1978 cos 4š
− 0.0994 cos 5š+
Example 2: Determine the coefficient |> of a linear phase FIR filter of length í = @ which has

@, a = L, @, A, B
symmetric unit sample response and frequency response

›a = ð L. C, a = C
L, a = , ‚, ƒ
Solution:

1
–óç ,  = 0,1, 2, 3
Write this same equation for non ideal filter

l = Ó 0.4
–óç ,  = 4
0,  = 5, 6,7
}−1
Î= =7
2

P–ó ,  = 0,1, 2, 3
l = Ó 0.4
P–ó ,  = 4
0,  = 5, 6,7

2² 2²
For type 1 frequency sampling technique,
š = =
} 15

G ⁄T
,  = 0,1, 2, 3
l = Ó 0.4
G ⁄T ,  = 4
0,  = 5, 6,7


\  _
1
ℎ = [l0 + 2  8 1l  ⁄ 2^
[ ^
Z ]

P
1
ℎ = 5l0 + 2  8 1l  ⁄T 26
15

H
1
ℎ = §1 + 2  8 1
G ⁄T  ⁄T 2 + 28 10.4
TV ⁄T O ⁄T 2¨
15

H
1
ℎ = §1 + 2  8 1
 P
⁄T 2 + 28 10.4
O P
⁄T 2¨
15

H
1 2²7 −  8²7 − 
ℎ = 51 + 2  cos + 0.8 cos 6
15 15 15

1 2²7 −  4²7 −  6²7 −  8²7 − 
ℎ = :1 + 2 cos + 2 cos + 2 cos + 0.8 cos ;
15 15 15 15 15
ℎ0 = ℎ14 = −0.0141
ℎ1 = ℎ13 = −0.0018
Discrete Time Signal Processing 76
 ℎ2 = ℎ12 = 0.0401
ℎ3 = ℎ11 = 0.0121
ℎ4 = ℎ10 = −0.0915
ℎ5 = ℎ9 = −0.0179
ℎ6 = ℎ8 = 0.3135
ℎ7 = 0.52
ℎ
= ?−0.0141, − 0.0018, 0.0401, 0.0121, − 0.0915, − 0.0179, 0.3135, 0.52, 0.3135, − 0.0179, − 0.0915,

• Choose the desired frequency response lh i – j

Design of FIR filter by Type 2 frequency sampling method

Sample lh i – j at M – points by taking š =  where  = 0, 1, 2, … … … , } − 1, to generate

 #
•

l = lh i – j/  #   = 0, 1, 2, … … … , } − 1
the sequence
–

Compute the M samples of impulse response ℎ using following equation

H
\  _
•

2
ℎ = [  8 1l  #⁄ ^
2 , } ..
[ ^
Z ]


\  _


2
ℎ = [2  8 1l  # ⁄
2^ , } ¸ 
[ ^
Z ]


Take Z transform of the impulse response ℎ to get the filter transfer function lŽ


•

lŽ =  ℎŽ


Realization of FIR Filter
The FIR filter can be realized using various structures namely
• Direct form realization
• Cascade realization
• Linear phase realization

Direct Form Realization



The FIR filter can be described by the difference equation as follow:

% =  !  − 

% = !  + !  − 1 + !  − 2 + … … … + !
  −  + 1
$Ž = !
Ž + ! Ž

Ž + ! Ž

Ž + … … … + !
 Ž


Ž
$Ž
= ! + ! Ž
 + ! Ž
 + … … … + !
 Ž



Ž
lŽ = ! + ! Ž
 + ! Ž
 + … … … + !
 Ž



Discrete Time Signal Processing 77

Cascade Realization

To realize the FIR filter in cascade realization, the transfer function lŽ is factorized into two or more
The cascade realization of the FIR filter is formed by the series connection of two or more FIR filter section.

polynomial.

lŽ = ! + ! Ž
 + ! Ž
 + … … … + !
 Ž


Given the transfer function of the FIR filter be

Here we clear that the transfer function is looks like an Nth order polynomial. Therefore the above function can
be factorized as two systems (polynomial) with (N–1)th order as

 

lŽ = n& + & Ž
 + … … … + &
 Ž
 o n{ + { Ž
 + … … … + {
 Ž
 o

Linear Phase Realization

The FIR filter has a linear phase, that is, the impulse response sequence of the FIR filter has follow a
symmetrical property of the sequence. The length of the impulse response may be even or odd.
Case 1: Length of impulse response sequence is even

lŽ = ! + ! Ž
 + ! Ž
 + … … … + ! Ž

H + ! Ž

 + ! Ž


The impulse response of the linear phase FIR filter with the even number of sequence is given by

lŽ = ! i1 + Ž

 j + ! iŽ
 + Ž

 j + ! iŽ
 + Ž

H j + … … …
This type of filter can be rewritten as follow

Discrete Time Signal Processing 78

Case 2: Length of impulse response sequence is odd


lŽ = ! + ! Ž
 + ! Ž
 + … + !
 Ž
+ … + ! Ž

H + ! Ž

 + ! Ž


The impulse response of the linear phase FIR filter with the odd number of sequence is given by





lŽ = ! i1 + Ž

 j + ! iŽ
 + Ž

 j + ! iŽ
 + Ž

H j + … … … + !
 Ž

This type of filter can be rewritten as follow




Example 1: Obtain the direct form and cascade form realization of the transfer function of the FIR filter

@ B @ @
given by
›  = n@ −  
@ +  
A o n@ −  
@ −  
A o
C „ „ A
Direct Form:

Discrete Time Signal Processing 79

 1 3 1 1
lŽ = n1 − Ž
 + Ž
 o n1 − Ž
 − Ž
 o
4 8 8 2
3
 3
 5
H 3
G
lŽ = 1 − Ž − Ž + Ž − Ž
8 32 64 16

Cascade Realization:

@ @ @ A A
Example 2: Realize the system with transfer function
›  = n@ +  
@ +  
A +  
B +  
C o n@ +  
@ +  
A +  
B o
A B A B B
in direct form, cascade and linear phase realization.
Solution:

1 1 1 2 2
Direct Form
lŽ = n1 + Ž
 + Ž
 + Ž
H + Ž
G o n1 + Ž
 + Ž
 + Ž
H o
2 3 2 3 3
1
 1
 1
H 2 1 2 1 2 2 1 2
lŽ = 1 + Ž + Ž + Ž + Ž
G + Ž
 + Ž
 + Ž
H + Ž
G + Ž
T + Ž
 + Ž
H + Ž
G
2 3 2 3 3 9 3 3 3 3 9
1
T 2
V 1
G 1
T 1
V
+ Ž + Ž +Ž + Ž + Ž + Ž +Ž

H
P
3 3 2 3 2
7
 4
 37
H 37
G 4
T 7
V
lŽ = 1 + Ž + Ž + Ž + Ž + Ž + Ž + Ž
P
3 3 18 18 3 3

Cascade Form

Discrete Time Signal Processing 80

 1 1 1 2 2
lŽ = n1 + Ž
 + Ž
 + Ž
H + Ž
G o n1 + Ž
 + Ž
 + Ž
H o = l Žl Ž
2 3 2 3 3

1 1 1
where
l Ž = 1 + Ž
 + Ž
 + Ž
H + Ž
G
2 3 2
2
 2

l Ž = 1 + Ž + Ž + Ž
H
3 3

Linear Phase Realization

Cascade with Linear Phase Realization

Discrete Time Signal Processing 81

 @ @ @ @
Example 3: Realize the following FIR system in linear phase realization.
†> = => − => − @ + => − A + => − B − => − C + => − 
B C C B
1 1 1 1
Solution:
% =  −  − 1 +  − 2 +  − 3 −  − 4 +  − 5
3 4 4 3
1 1 1 1
Taking Z transform on both side,
$Ž =
Ž − Ž

Ž + Ž

Ž + Ž
H
Ž − Ž
G
Ž + Ž
T
Ž
3 4 4 3
1
 1
 1
H 1
G
$Ž =
Ž w1 − Ž + Ž + Ž − Ž + Ž
T x
3 4 4 3
$Ž 1
 1
 1
H 1
G
= 1 − Ž + Ž + Ž − Ž + Ž
T

Ž 3 4 4 3
1
 1
 1
H 1
G
lŽ = 1 − Ž + Ž + Ž − Ž + Ž
T
3 4 4 3
1 1
lŽ = 1 + Ž
T  − Ž
 + Ž
G  + Ž
 + Ž
H 
3 4

Discrete Time Signal Processing 82

 Unit – IV Finite Word Length Effect
Types of Number Representation
• Fixed point representation
• Floating point representation
Fixed Point Representation
• It is a generalization of the familiar decimal representation of a number as a string of digits with a
decimal points.
• In this notation, the digits to the left of the decimal point represent the integer part of the number and the
digit to the right of the decimal point represent the fractional part of the number.
• There are three ways to represent the negative number.

Example: −2 = 1010

• Sign Magnitude Format: In this, the MSB is set to 1 to represent the negative sign.

• One’s Complement Format: In this, the MSB is set to 1 and all the other digits are represented by its

Example:−2 = 1101
complement.

• Two’s Complement Format: In this, a negative number is represented by forming the two’s
complement of the corresponding positive number. In other words, the negative number is obtained by

Example:−2 = 1ô ˆ{-·" - ‰2 + 1 = 1101 + 1 = 1110

subtracting the positive number from 2.

Floating Point Representation

• A floating point representation can be employed as a means for covering a large dynamic range.

the fractional part of the number and falls in the range  ≤ } < 1, multiplied by the exponential factor

• The binary floating point representation commonly used in practice, consists of a mantissa M, which is

2õ , where the exponent E is either a positive or negative integer.

= }. 2õ
• Hence a number X is represented by

• Example 1: The number
= 5 is represented as

5
5 = 2H = 0.625 ∗ 2H
8
Mantissa:} = 0.625 = 0.101000
Exponent:ö = 3 = 011
5 = 0.101000 ∗ 2
• Example 2: The number
= O is represented as
H

3 6

= 2 = 0.75 ∗ 2

8 8
Mantissa:} = 0.755 = 0.110000
Exponent:ö = −1 = 101
3
= 0.110000 ∗ 2
8
Types of Quantization Error
• Rounding or truncation introduces an error whose magnitude depends on the number of bits truncated or
rounded off.

• Consider a number , whose original length is ‘~’ bits.

• Also, the characteristic of the error depends on the form of binary number representation.

• Let this number be quantized (truncated or rounded) to ‘t’ bits.

• This quantized number is represented as ù.
Discrete Time Signal Processing 83
 Both  and ù are shown above.
Note that t < ~.
•

A truncation error, э , is introduced in the input signal and thus quantized signal is
•

ù =  + Ñ 
•

• The range of values of the error due to truncation of the signal is analyzed here for both sign magnitude
and two’s complement.
• Quantization error follows the uniform distribution
Truncation Error
Truncation is defined as the removal of excessive bits. This leads to the reduction in the magnitude of the
number.
Truncation error depends on type of number representation.
Truncation Error for Fixed Point Number Representation
Range of Error Range of Error
Representation Error

ú 
−2 − 2

É
≤ э
(Finite Precision) (Infinite Precision)

−2
É ≤ э ≤ 0
≤0
Positive number Negative

0 ≤ э ≤ 2
É − 2
ú  0 ≤ э ≤ 2
É
0 ≤ э ≤ 2
É − 2
ú  0 ≤ э ≤ 2
É
Sign Magnitude Negative Number Positive

−2
É − 2
ú  ≤ э
One’s Complement Negative Number Positive

−2
É ≤ э ≤ 0
≤0
Two’s Complement Negative Number Negative

Probability Density Function for Fixed Point Number Representation

Overall range: −2
É ≤ э ≤ 2
É
Sign Magnitude and one’s complement:

2É
·э  = , −2
É ≤ э ≤ 2
É
2

Overall range: −2
É ≤ э ≤ 0
Two’s Complement:

·э  = 2É , −2
É ≤ э ≤ 0
Truncation Error for Floating Point Number Representation
Range of Error Range of Error
Representation Error

ú 
−22 − 2

É
≤ э
(Finite Precision) (Infinite Precision)

−2 ∗ 2
É ≤ э ≤ 0
≤0
Positive Mantissa Negative

−22
É − 2
ú  ≤ э
−2 ∗ 2
É ≤ э ≤ 0
≤0
Sign Magnitude Negative Mantissa Negative

−22
É − 2
ú  ≤ э
−2 ∗ 2
É ≤ э ≤ 0
≤0
One’s Complement Negative Mantissa Negative

0 ≤ э ≤ 22
É
0 ≤ э ≤ 2 ∗ 2
É
− 2
ú 
Two’s Complement Negative Mantissa Positive

Probability Density Function for Floating Point Number Representation

Overall range: −2 ∗ 2
É ≤ э ≤ 0
Sign Magnitude and one’s complement:

Discrete Time Signal Processing 84

 2É
·э  = , −2
É ≤ э ≤ 0
2

Overall range: −2 ∗ 2
É ≤ э ≤ 2 ∗ 2
É
Two’s Complement:

2É
·э  = , −2 ∗ 2
É ≤ э ≤ 2 ∗ 2
É
4
Rounding Error
Rounding is defined as changing a fractional value to the nearest integer. This leads to the decrease or increase
in the magnitude of the number.
Rounding error does not depends on types of number representation
For the positive number, the rounding error is positive and for the negative number, the rounding error is
negative

2
É − 2
ú 2
É − 2
ú
Therefore, the range of rounding error is
− ≤ Ñï ≤
2 2

2
É 2
É
For infinite precision,
− ≤ Ñï ≤
2 2

2
É 2
É
The probability density function of the rounding error is
·ѱ  = 2É , − ≤ ѱ ≤
2 2
Quantization effects in Analog to Digital Conversion of signals
• The process of analog to digital conversion involves
 Sampling the continuous time signal at a rate much greater than the Nyquist rate
 Quantizing the amplitude of the sampled signal into a set of discrete amplitude levels

In ADC, when B bits binary code is selected, we can generate 2É different binary numbers.
If the range of analog signal to be quantized be 8, then the quantization step size is given by
•

8
•
û= É
2

The difference between the quantised signal amplitude ü  and the actual signal amplitude  is
• This quantiser rounds the sampled signal to the nearest quantised output level.

called quantization error . That is

•

 = ü  − 

• Since rounding is involved in the process of quantization, the range of values for the quantization error

û û
− ≤  ≤
is

2 2

Discrete Time Signal Processing 85

 82
É 82
É
− ≤  ≤
2 2

• The power of the quantization noise, which is nothing but the variance (—e ) is given by
Input Quantization Noise Power from ADC:

ï¬þ


ýe = —e = ¢  ·Ñï  .
ï¬þ


ï¬þ

1
ýe = ¢  .
82
É
ï¬þ


ï¬þ
1 H 
ýe = : ;
82
É 3
ï¬þ

1 8 H 2
HÉ 8 H 2
HÉ
ýe = : + ;
82
É 24 24
1 8 H 2
HÉ
ýe =
É ∗ 2 ∗
82 24
8  2
É
ýe =
12
ý7
Signal to Noise Ratio:
ܵ8 = 10 log
ýe
ܵ8 = 10 log ý7 − 10 log ýe
8  2
É
ܵ8 = 10 log ý7 − 10 log £ ¤
12
ܵ8 = 10 log ý7 − 10 log 8  + 10 log 2
É − 10 log 12
ܵ8 = 10 log ý7 − 10 log 8  + 10 log 2É + 10 log 12
ܵ8 = 10 log ý7 − 20 log 8 + 10 log 2É + 10 log 12
ܵ8 = 10 log ý7 − 20 log 8 + 20t log 2 + 10 log 12
ܵ8 = 10 log ý7 − 20 log 8 + 6t + 10.8−→ 4
Consider the range 8 = 1ܸ,
ܵ8 = 10 log ý7 + 6t + 10.8
Dynamic Range:

'8 = −10 log ýe = −20 log 8 + 6t + 10.8

• The dynamic range (DR) is given by

• Consider the range 8 = 1ܸ,

'8 = −10 log ýe = 6t + 10.8
Output Quantization Power from digital System:

is applied as an input to a digital system with a transfer function lŽ.

• After converting the continuous time signal into a digital signal, let us assume that this quantised signal

 The unquantised input signal 

• This quantised input to the digital systems consists of two components

 The quantization error signal 

Discrete Time Signal Processing 86
  The output %ü  due to the quantised input signal
• The output of the digital system, therefore, consists of two components

The error output   due to the quantization error signal at the input of the digital system
From the figure above, it can be seen that the output $ of the digital system is given by
•

$ = $ü  +  −→ 1

•

The error output   is a random process and it is the response of the digital system to the input error
signal .
•

The error output   is obtained by convolving the system impulse response ℎ with input error
• The digital system is assumed to be causal.

signal .
•

• Thus,
ஶ

  =  ℎ  −  −→ 2


• Let us relate the statistical characteristics of the output error signal to the statistical characteristics of the

The autocorrelation sequence for the output error signal   is

input error signal and the characteristics of the system.

ߛeబeబ - = ö* ∗    + -+−→ 3

•

where E represents the statistical expectation

ஶ ஶ

ߛeబ eబ - = ö 5 ℎ ∗  −   ℎ  + - − 6

 
ஶ

ߛeబ eబ - =  ℎ ö* ∗  −   + - − +


ஶ

ߛeబ eబ - =  ℎ ߛee -



ߛeబ eబ - = —eబ
• It has been assumed that the noise resulting from the quantization process is a white noise. For this case,

ߛee - = —e
and

where, —eబ is the output noise power and —e is the input noise power.
• Thus,
ஶ

—eబ = —e  ℎ 


1
• Using Parseval’s theorem,
ஶ

 ℎ  = ර lŽlŽ
 Ž
 .Ž
2²I


—e
• Thus,
—eబ
= ර lŽlŽ
 Ž
 .Ž
2²I
where the closed contour of integration is around the unit circle |Ž| = 1.
Coefficient Quantization Effect
• The realization of the digital filters in hardware or software has some limitation due to the finite word
length of the registers that are available to store these filter coefficients.

Discrete Time Signal Processing 87

 • Since the coefficients stored in these registers are either truncated or rounded off, the system that is
realized using these coefficients is not accurate.
• The location of poles and zeros of the resulting system will be different from the original location and
consequently the system may have a different frequency response than the one desired.
• We know that the stability and system performance of a digital filter depends on the poles and zeros
location.
• Thus if the poles and zeros location are changed, then the system performance can vary.
• For example, if we want to design a Low Pass Filter and the filter coefficients are quantized, then it will
change the system as a High Pass Filter or Band Pass Filter or Band Reject Filer.

Product Quantization
• The error due to the quantization of the output of multiplier is referred to as product quantization error.
• When two B – bit numbers are multiplied, the product must be rounded to B – bits in all digital
processing applications.

ù*Î௜ + = Î௜  + 

• The output of a finite word length multiplier can be expressed as

whereÎ௜  is the product which is 2B – bit long and  is the error resulting from rounding the
product to B – bits.
• The fixed point, finite word length multiplier can be modeled as given below:

• In digital system, the product quantization is performed by rounding due to the following characteristics
of rounding.
 In rounding the error signal is independent of the type of arithmetic employed.
 The mean value of error signal due to rounding is zero.
 The variance of the error signal due to the rounding is least.
• The analysis of product quantization error is similar to the analysis of quantization error due to ADC.
• But, in product quantization error analysis, it is necessary to define the noise transfer function, which
depends on the structure of the digital network.
• The noise transfer function (NTF) is defined as transfer function from the noise source to the filter
output.
• NTF is the transfer function obtained by treating the noise source as actual input.
• The product quantization error signal is treated as a random process with uniform probability
distribution function.
• In general the following assumptions are made regarding the statistical independence of the various
noise sources of the digital filter.
 Any two different samples from the same noise source are uncorrelated.
 Any two different noise sources, when considered as random processes are uncorrelated.
 Each noise source is uncorrelated with the input sequence.
Product Quantization Noise Models for IIR filter
First Order Direct Form I

Discrete Time Signal Processing 88

Second Order Direct Form I

First Order Direct Form II

Second Order Direct Form II

Discrete Time Signal Processing 89

Cascading of two second order section

Cascading of two first order section

Analysis

2
É
The average power (variance) is given by
—e =
12

Let ℎ be the system response and   be the response of the system to the input error .
The effects of rounding due to multiplication in cascaded IIR sections are discussed now.

Then the output noise power is given by

ஶ

—eబ = —e  ℎ 

Using Parseval’s relation

Discrete Time Signal Processing 90

 —e
—eబ
= ර lŽlŽ
 Ž
 .Ž
2²I
Assume that the M cascaded sections, then the output noise power at the kth product in the ith section is given by
ஶ

—e

= —e  ℎ௜ 

Then the overall output noise power is given by
ஶ

—e±±

=  —e௜


௜
Limit Cycle Oscillation
• In recursive systems, when the input is zero or some nonzero constant value, the nonlinearities due to

• During periodic oscillations, the output % of a system will oscillate between a finite positive and
finite precision arithmetic operators may cause periodic oscillations in the output.

negative value for increasing  or the output will become constant for increasing .
• Such oscillations are called limit cycle oscillation.
• These oscillations are due to round off errors in multiplications and overflow in additions.
Types of Limit cycle oscillation
• Zero input limit cycle oscillation
• Overflow limit cycle oscillation
Zero input limit cycle oscillation
In recursive systems, if a system output enters a limit cycle, it will continue to remain in limit cycle even when
the input is made zero.
Hence these limit cycles are called zero input limit cycle.

% = !% − 1 + 

Consider the difference equation of first order system with only pole as

The system has one product !% − 1.

If the product is quantised to finite word length then the response % will deviate from actual value.
Let % ô  be the response of the system when the product is quantised in each recursive realization. Now,
% ô  = ù*!% − 1+ + 
whereù stands for quantization operation.
In the first order system with only pole, the coefficient “a” will be the pole of the systems.
Let us examine the nature of response of first order system for an impulse input and various values of poles.
For simplicity, let us choose sign magnitude representation for binary product and response.
Let the product be quantised to five bit binary.

% ô  = 0,   < 0
Let

15
and

 = ð 16 ,  = 0
0,  ≠ 0
and! = 


 t  ‹.¹y s‰ ‹.¹y5 &¹‰ˆ

' {¹-!" t¹! % t¹! % ' {¹-!"
−1 0 0.00000 0.0000 0
0 0.9375 0.11110 0.1111 0.9375
1 0.46875 0.01111 0.1000 0.5
2 0.25 0.01000 0.0100 0.25
Discrete Time Signal Processing 91
 3 0.125 0.00100 0.0010 0.125
4 0.0625 0.00010 0.0001 0.0625
5 0.3125 0.00001 0.0001 0.0625
Dead Band
In limit cycle, the amplitudes of the output are confined to the range of values, which is called dead band of the

For a first order system described by the equation % = !% − 1 + , the dead band is given by
filter.

2
É
' !. &!. = ±
1 − |!|
For a second order system described by the equation % = ! % − 1 + ! % − 2 + , the dead band

2
É
is given by
' !. &!. = ±
1 − |! |
Overflow Limit Cycle
The oscillation occurs due to the truncation of output of the adder or multiplier is called overflow limit cycle
oscillations.
Methods used to prevent overflow:
• Saturation arithmetic
• Scaling
Saturation arithmetic:
In saturation arithmetic, if the output exceeds the maximum value then the output is set to maximum value and
if the output goes below the minimum value then the output is set to minimum value.
Scaling:

Consider % be the output of the system ℎ for the input .
Scaling can be done by scale the input at certain points in the digital filter to prevent overflow.

Then the output % can be represented as

ஶ

% =  ℎ − 

ஶ
Apply magnitude on both side,
ஶ

|%| = อ  ℎ − อ


ஶ
Using Schwarz’s inequality,
ஶ

|%| ≤  |ℎ|| − |


ஶ
If the maximum value of input  is set as
, then
ஶ

|%| ≤
 |ℎ|  !"" 


ஶ
If the maximum range of the data handled by DSP processor is −1,1, then the output must be lesser than 1
ஶ

 |ℎ| < 1

ஶ
1

<
∑ஶ

ஶ|ℎ|
This is necessary condition for preventing overflow in a digital IIR system.
For an FIR Filter, the condition will be

Discrete Time Signal Processing 92

 1

<
∑

 |ℎ|

L. C 
Example 1: The output of an ADC is applied to a digital filter with the system function
›  =
  − L. ƒA
Find the output noise power of the digital filter when the input signal is quantized to 7 bits.

0.45Ž
Given:
lŽ =
Ž − 0.72
t = 7 &¹‰ˆ
To find: Output Noise Power —e 

—e
Formula:
—e

= ර lŽlŽ
 Ž
 .Ž
2²I
where, —e = 8  ⁄12 ∗ 2É 

Let consider the range of the input as −1ܸto +1ܸ.

Solution:

8 = 1+1= 2
4
—e = = 2.0345 ∗ 10
T d
12 ∗ 2 G
— 
0.45Ž 0.45Ž


—e = Ž .Ž
 e
ර
2²I Ž − 0.72 Ž
 − 0.72
—e 0.45Ž 0.45
—e

= ර Ž
 .Ž
2²I Ž − 0.72 1 − 0.72Ž
—e 0.2025
—e

= .Ž
2²I Ž − 0.721 − 0.72Ž
ර
—e

= —e ∗ ˆ‹-  8 ˆ¹.‹  ·" ˆ  lŽlŽ
 Ž
 vℎ¹{ℎ ¹ˆ ¹ˆ¹. ‰ℎ {¹ {"
Here the poles of lŽlŽ
 Ž
 are Ž = 0.72 (inside the unit circle) and Ž = 1.3889 (outside the unit circle)
8 = 18 ˆ *lŽlŽ
 Ž
 +2«→.P
0.2025
8 = w x
1 − 0.72Ž «→.P
8 = 0.4205
—e = 0.4205—e


—e = 0.4205 ∗ 2.0345 ∗ 10

T


—e

= 8.5551 ∗ 10
V d
Example 2: Consider the transfer function ›  = ›@  ›A   where ›@   = @⁄@ − L.  
@ and
›A   = @⁄@ − L. ‚ 
@ . Find the output round off noise power.

—e
We know that
—e

= ර lŽlŽ
 Ž
 .Ž
2²I

l Ž and noise power due to input noise at l Ž.

For the cascade connection, total output noise power is equal to the sum of noise power due to input noise at

) Ž = l Žl Ž

Hence we define two system functions as

) Ž = l Ž
Noise power due to input noise at ›@  :

Discrete Time Signal Processing 93

 —e
—e

= ර lŽlŽ
 Ž
 .Ž
2²I
—e Ž

—e

= .Ž
2²I 1 − 0.5Ž
 1 − 0.6Ž
 1 − 0.5Ž1 − 0.6Ž
ර
—e Ž
—e

= .Ž
2²I Ž − 0.5Ž − 0.61 − 0.5Ž1 − 0.6Ž
ර
—e

= —e ∗ ܵ‹-!""8 ˆ¹.‹ ˆ!‰·" ˆ"¹ ˆ¹ˆ¹. ‰ℎ {¹ {" 
Poles inside the unit circle: Ž = 0.5 !. 0.6
Poles outside the unit circle: Ž = 2 !. 1.6667
8 = 8 ˆ*lŽlŽ
 Ž
 +«→.T
Ž
8 = w x
Ž − 0.61 − 0.5Ž1 − 0.6Ž «→.T
0.5
8 = = −9.5238
−0.1 ∗ 0.75 ∗ 0.7
8 = 8 ˆ*lŽlŽ
 Ž
 +«→.V
Ž
8 = w x
Ž − 0.51 − 0.5Ž1 − 0.6Ž «→.V
0.5
8 = = 13.3929
0.1 ∗ 0.7 ∗ 0.64

—e

= —e ∗ 8 + 8 
—e

= 3.8691—e
Noise power due to input noise at ›A  :
—e
—e

= ර ) Ž) Ž
 Ž
 .Ž
2²I
—e Ž

—e

= .Ž
2²I 1 − 0.6Ž
 1 − 0.6Ž
ර
—e 1
—e =

.Ž
2²I Ž − 0.61 − 0.6Ž
ර
—e

= —e ∗ ܵ‹-!""8 ˆ¹.‹ ˆ!‰·" ˆ"¹ ˆ¹ˆ¹. ‰ℎ {¹ {" 
Poles inside the unit circle: Ž = 0.6
Poles outside the unit circle: Ž = 1.6667
8 = 8 ˆ*) Ž) Ž
 Ž
 +«→.V
1
8 = w x
1 − 0.6Ž «→.V
1
8 = = 1.5625
0.64
—e

= —e ∗ 8 
—e

= 1.5625—e

—e

= —e

+ —e
Total Power:

—e = 5.4316—e


the difference equation †> = L. …†> − @ + =>. Determine the dead band of the filter.
Example 3: Explain the characteristics of a limit cycle oscillation with respect to the system described by

Given that % = 0.95% − 1 + 

Here ! = 0.95, Assume t = 4 &¹‰ˆ

Discrete Time Signal Processing 94

 0.5,  = 0
Assume  = ½
0,  ≠ 0
 t  8‹.¹y 4 &¹‰ˆ s‰ 8‹.¹y 4 &¹‰ˆ
' {¹-!" t¹! % t¹! % ' {¹-!"
−1 0 0.0000 0.000 0
0 0.75 0.1100 0.110 0.75
1 0.7125 0.1011 0.110 0.75

2
É
Dead Band:
' !.t!. = ±
1 − |!|
2
G
' !.t!. = ±
1 − 0.95
' !.t!. = ±1.25

Discrete Time Signal Processing 95