Introduction

• The Butterworth approximation of a LP filter is

maximally flat at =0 but becomes progressively
poor as  increases
• At =1 (normalised frequency), |T|=0.707 (3dB
down) of the maximum value
• In the stop band where  > 1 A() falls off with a
slope of 20ndB/decade (n is the order of the filter)
• The higher the value of n, the flatter is the filter
response and the more complex is the network
realisation

1
 Chebyshev response
• For better approximation to the ideal characteristic, the
deviation from the ideal can be distributed throughout the pass
band so that, |T| oscillates between 1 (0 dB) and 1/ (1+ ε2)
(minimum) for n times
• Chebyshev magnitude response for a LP filter is defined as
1
Tn ( j ) =
2

1 +  2C n2 ( ) ……………….(1)
•where Cn() is the Chebyshev polynomial of order n in 
•ε < 1 specifies the ripple in the pass band
•n is the order of the filter as well the Chebyshev polynomial
•n also gives the number of ripples in the pass band i.e. the
total number of maxima and minima in the range 0    1
•For  > 1, |T| decreases monotonically
2
• The nth order Chebyshev function Cn() is defined as
Cn ( ) = cos n ( cos −1  ) for   1 ……...(2)
Cn ( ) = cosh n ( cosh −1  ) for   1
Let  = cos for 0    1 ………..…….(3)
and  = cosh for   1
Then C n ( ) = cos n for 0    1 ………..…….(4)
and C n ( ) = cosh n for   1
jn
= cos n + j sin n = ( cos + j sin )
n
But e
cos n = Re ( cos + j sin )
n

•The real part has even powers of sin as factors (sin2,

sin4, … ) which can be expressed in terms of cos2 since
sin2 = 1- cos2 3
 Chebyshev polynomials
 It can be seen from (2) and (3) that for || ≤ 1
C0() = 1
C1() = cos = 
C2() = cos2 - sin2 = 22 - 1
C3() = cos3 - 3 cos sin2 = 43 - 3
 For 0 <   1, Cn()  1
 In general, from (4)
C n+1 ( ) + C n−1 ( ) = cos ( n + 1) + cos ( n − 1)
but cos( A + B ) + cos( A − B ) = 2cos A cos B 
Cn+1 ( ) + Cn−1 ( ) = 2cos n cos = 2Cn ………(5)
Equation (5) yields the recurrence relation
Cn+1 ( ) = 2Cn ( ) − C n−1 ( ) ………..……(6) 4
Fig. 1 shows plots of Chebyshev functions for -1 ≤  ≤ 1

5
Fig. 1
 As seen from Fig. 1, the Chebyshev polynomial Cn()
oscillates for the range -1 ≤  ≤ 1
 The ripples are always equal with the upper limit being 1
and the lower limit of -1
 This is also true for Cn2()
 Thus, the functions Cn() and Cn2() are known as equal-
ripple functions

6
 Properties of Chebyshev magnitude
response
1
Tn ( j ) =
2

1 +  2C n2 ( ) ……………….(1)
1. For 0 ≤  ≤ 1, the filter has equi-ripple behaviour between
the limits 1 and 1/ (1+ ε2)
(i) At  = 0
Cn(0) = 0 for n odd and |Cn(0)| = 1 for n even
|Tn(j0)| = 1 for n odd and |Tn(j0)| = 1/ (1+ ε2) for n even
(ii) At  = 1
Cn(1) = 1 for all n
|Tn(j1)| = 1/ (1+ ε2) for all n

7
3. The minima of Tn(j) occur when C2n() = 1 and
Tn(j)min = 1/ (1+ ε2) = -10log(1+ ε2) dB as shown in Fig.
2
4. The maxima occur when C2n() = 0 and Tn(j)max = 1= 0
dB as shown in Fig. 2
5. The maxima and minima occur when  = cos(k/2n)
where k = 0, 1,…., n
• For n = 6, T6(j) is maximum when  = 0.259, 0.707,
0.966 and minimum when  = 0, 0.500, 0.866, 1
6. The ripple N = 10log(1+ ε2) dB e.g. for ε= ½, the ripple  1
dB
7. There are n half-ripples in the range  = 0 → 1
8. For  > 1 Tn(j) falls monotonically to zer
8
Fig. 2(a)
9
Fig. 2(b)
10
 The attenuation is given by
 n = −20log Tn ( j ) dB
= 10log 1 +  2C n2 ( )  dB ……………….(2)
• The rippling nature of Chebyshev response is observed
for frequency in the range 0 – 1
• For larger values of 
( )
 n = 10log 1 +  cosh n cosh   dB ……….(3)
 2
2 −1
 
•The attenuation characteristic is as shown in Fig. 3

11
  maxoccurs when C n2 ( ) = 1

  max = 10log(1 +  ) 2

max 10
  = 10 −1

(n = 7)
Fig. 3
1
When  C ( ) = 1  C n (  ) =
2 2
and   3dB
n

This is the attenuation that defines half-power frequency hp
Its value may be determined as follows
12
 C n ( hp ) =
1

(
= cosh n cosh −1  hp )
1
n cosh ( hp ) = cosh  
−1 −1

 
1 −1  1 
cosh  hp = cosh  
−1

n  
1 −1  1  
  hp = cosh  cosh   
n   
hp > 1 since 0 < ε < 1
Substituting for ε gives
1 1 
 hp = cosh  cosh −1


n 10 max 10
−1  13
• For a Chebyshev filter, the end of the pass band is
always  = 1
• Hence the specifications for Chebyshev response are
values of max, min and s
 min =  ( s ) = 10log ( 1 +  2C n2 ( s ) )
Since s > 1
(
= 10log 1 +  cosh n cosh  s 
 )
2
 min 2 −1
 
( cosh n cosh s )
 min 10 2
10 = 1+  2 −1

 min 10
−1
But  = 10  max 10
(
− 1  cosh n cosh  s −1
) =
10
10max 10
−1
14
  min 10
10 −1
n cosh  s = cosh
−1 −1

10max 10
−1
 −1 10  min 10
−1 
 cosh 
 10  max 10
− 1 
n=  
cosh −1  s

15
 Example
• Determine the order of a Chebyshev LP filter given the
following specifications: p = 1, s = 2.33, max = 0.5dB
and min = 22dB.
Solution
 −1 10 22 10
−1 
 cosh 
 10 0.5 10
− 1 
n=   = 2.87 (round up to 3)
cosh −1 2.33

1 1 
 hp = cosh  cosh −1

 = 1.167
n 10 max − 1 
10

16
• Same specifications for Butterworth response

n=
(
log  10min 10
−1 )( 10max 10
)
− 1 
= 4.234 (round up to 5)
 s 
2log  
  p 

17
 Location of the Chebyshev poles
• As in the Butterworth response, j = s   = s/j

Tn ( j )  = s = T ( s )T ( − s )
2

1
T ( s )T ( − s ) =
( j)
1 +  2C n2 s
• The loci of the poles are determined as
1 +  2C n2 s ( j) = 0
Cn ( j) = 0 j
s 1
……………………….(1)

18
 ( )
For   1, C n s
j

( )
= cos  n cos −1 s 


j 

Let cos ( s ) = x = u + jv ……………..…(2)

−1
j

 C ( s ) = cos nx = cos( nu + jnv )

n j
= cos nu cosh nv − j sin nu sinh nv …….…(3)
Equating real and imaginary parts in (1) and (3)
cos nu cosh nv = 0 and sin nu sinh nv = 1

cosh nv  0  cos nu = 0
The values of u for which this is true are

uk = ( 2k + 1) for k = 0,1, ,2n − 1 …….…(4)
2n 19
• For these values of uk, sin nuk =  1 so that
1 −1  1 
vk =  sinh   =  a
n   …………..(5)
 For each value of xk = uk + jvk, the corresponding
value of sk can be obtained from (2) as
  …………..(6)
sk = j cos xk = j cos  ( 2k + 1)  ja 
 2n 
If sk =  k + jk and as in (3)
 2k + 1 
 k =  sinh a sin  
 2n  ……………..…..(7)
k  2k + 1 
 =  sin  
sinh a  2n  20
 2k + 1
 From (6)  k = cosh a cos 
2n ……………..…..(8)
k  2k + 1 
 = cos  
cosh a  2n 
 From (7) and (8)
   k   k 
2 2
 2k + 1  2  2k + 1
2
sin    + cos  =  +  =1
 2n   2n   sinh a   cosh a 

• The Chebyshev poles lie on an ellipse as shown in Fig. 4

• Only poles located on the LHP are considered

21
 j

j
cosh a

sinh a

-j

Fig. 4

22
 Example
Determine the pole location for a 3rd order LP
Chebyshev filter whose max = 1dB
Solution
 
1 −1  1  1 −1  1  = 0.476
a = sinh   = sinh
  n  max 10 1 
n
 10( −1 ) 2

sinh0.476 = 0.494 and cosh0.476 = 1.115
From (7) and (8),
2k + 1
 k =  sinh 0.476sin 
6
2k + 1
 k = cosh 0.476cos  for k = 0,1,.....,5
6 23
  0 = 0.247 and 0 = 0.966
 1 = 0.494 and 1 = 0
 2 = 0.247 and 2 = −0.966
 3 = −0.247 and 3 = −0.966
 4 = −0.494 and 4 = 0
 5 = −0.247 and 5 = 0.966
Neglecting poles on the RHP
s1 = −0.494; s2 , s3 = −0.247  j 0.966
d ( s ) = ( s + 0.494 )( s + 0.247 − j 0.966 )( s + 0.247 + j 0.966 )
(
= ( s + 0.494 ) s 2 + 0.494 s + 0.994 )
• For the 2nd order factor 02 = 0.994  0 = 0.997 and
0/Q = 0.494  Q = 2.02 24
 Guillemin’s Algorithm
 In equations (7) and (8) for calculating k and k
respectively, let k be
2k + 1 
Let k = k = 0,1,2,..........2n − 1
n 2
measured with respect to the positive real axis.
• The loci of the Chebyshev poles are given by
sk = − k  j k = sinh a sin  k  j cosh a cos k
❑ Let k = (90 - k), which is the Butterworth angle
measured with respect to the negative real axis.
 2k + 1    n − 2k − 1 
k = − . =   k = 0,1, , n - 1
2 n 2 2 n 
sin k = cos k and cos k = sin k
25
• Therefore the poles loci are
sk = − k  j k = sinh a cos k  j cosh a sin k

Example
Determine the pole locations and TF for a LP Chebyshev
filter for the specifications; n = 5 and max= 0.5 dB.

26
 Solution
1 −1  1 
a = sinh   = 0.3548
n  
sinh0.3548 = 0.3623 and cosh0.3548 = 1.0636
 k = 72 ,  36 ,0
s1,2 = −0.1120  j1.0116
s 3 ,4 = −0.2931  j 0.6252
s 5 = −0.3623
1
T ( s) =
(s 2
+ 0.224 s + 1.0358 )( )
s 2 + 0.5863 s + 0.4768 ( s + 0.3623 )

Q1,2,3 = 4.54, 1.18,0.5

27
 Q of Chebyshev Poles
1
Q=  +
2 2

2
1 sinh 2 a cos 2  k + cosh 2 a sin 2  k
=
2cos k sinh 2 a

n even n odd
2 4 6 3 5 7
0.86 0.71 0.68 1.71 1.18 1.09
2.94 1.81 4.54 2.58
6.51 8.84
For n odd, there is a real pole for which Q = 0.5
28
 Comparison of Butterworth and Chebyshev responses
• In Butterworth response  = 1 identifies half-power
frequency whereas for Chebyshev response, it identifies
the end of the ripple band. For Chebyshev response
1 −1  1  
 hp = cosh  cosh   
n   
 Chebyshev response has a steeper slope at cut-off and
provides more attenuation in the stop band.
 However, Chebyshev approximation departs from the
linear curve more than the Butterworth does (delay
distortion)
 Chebyshev approximations require higher pole-Q
realizations thus demanding tighter component
tolerances. 29