Butterworth Filters Lecture Notes by A. M.

Muhia

The selection of a realizable mathematical relationship which approximates to the given specifica-
tion usually involves algebraic manipulation.

Analog Filter Approximation Techniques

some of the frequently used analog filter approximation technologies are;

? Butterworth

? Chebychev

? Inverse chebychev

? Elliptic

? Bessel

1.4 Butterworth Filter Response

The design requires that the magnitude response is maximally flat

Figure 7: Butterworth filter response

In S-plane
2
H (jω) = H (s) H (−s) s=jω
(1)
The magnitude square function is even
2 2
H (jω) = H (−jω)
2
Representing the transfer function H (jω) as a quotient of polynomial, both the numerator and
denominator polynomials must be even.
Therefore

A ω2


2
H (jω) = (2)
B (ω 2 )
Let the numerator A ω 2 be a single function such that

8
Butterworth Filters Lecture Notes by A. M. Muhia

2 A0
H (jω) = (3)
B0 + B2 ω2 + B4 ω 4 + · · · · · · + B2n ω 2n
A special case in which all B coefficients except B0 and B2n have zero values , gain required A0 = B0
such that Hn (jω) = 1 and that
1
B2n = (4)
(ω0 )2n
Then expression (3) becomes

2 1
Hn (jω) =  2n (5)
ω
1+
ω0
Expression (5) is called the Butterworth response
NB: n is the Butterworth filter order or the number of poles in Hn (jω) . Coefficient A is normalized
to 1.
Normalizing the frequency such that ω0 = 1 makes expression (5) to be

2 1
Hn (jω) =
1 + ω 2n
=⇒
1
Hn (jω) = √ (6)
1 + ω 2n
From (6)

1. Hn (j0) = 1 for all n

i.e at ω = 0, Hn (jω) = 1 irrespective of n

1
2. Hn (j1) = √ for all n
2
3. A (ω) = 20 log H (jω)

" #1
1 2
= 20 log  2n
ω
1+
ω0
"  2n #
ω
= −10 log 1 +
ω0
For high frequencies
 2n
ω
A (ω) = −10 log
ω0

9
Butterworth Filters Lecture Notes by A. M. Muhia

 
ω
= −20n log
ω0
Therefore n increases the gradient; hence the higher the order the sleeper the gradient. This
reduces the transition

4.
  −1
2 2n
H (jω) = 1 + (ω) (7)

Assuming frequency scaling ω0 = 1

Equation (7) becomes
 − 1
2 2n
H (jω) = 1 + (ω)

−1
2n
 
= 1+ω 2 (8)

Expanding (8) using Taylor series

i.e
α (α − 1) x2 (α − 1) (α − 2) x3
(1 + x)α = 1 + αx + +α + ······
2! 2!
1 3
Hn (jω) = 1 − ω 2n + ω 4n + · · · · · ·
2 8
Taking the derivative of Hn (jω) w.r.t ω

dk
Hn (jω) =0
dω k ω=0
for k = 1, 2, 3, . . . . . . 2n − 1

for k = 2n

d2n −1
Hn (jω) =
dω 2n ω=0 2
This gives a flat response.All the derivatives except one are equal to zero near ω = 0
Hence the response is known as maximally flat captured in figure 8.

1.4.1 Butterworth Pole location

Assuming frequency scaling, ω0 = 1

2 1
Hn (jω) =
1 + ω 2n

For the poles 1 + ω 2n = 0

s
s = jω →ω =
j

10
Butterworth Filters Lecture Notes by A. M. Muhia

Figure 8: Maximally flat property of a butterworth filter

 2n
s
⇒ 1+ =0
j

For n = 1

1 − s2 = 0

s2 = ±1
1
for stability s = 1 ⇒ H (s) =
1+s
For n = 2

s4 = −1 = 1∠180
180 + k360
s = 1∠ k = 0, 1, 2, 3
4
pole location
−1 1
s1, s2 = √ ± j √
2 2
1
H2 (s) = √
s2 + 2s + 1
For n = 3
1 + (−1)3 s6 = 0

s6 = 1

= 1∠0
0 + k360
s = 1∠ k = 0, 1, 2, 3, 4, 5
6

11
Butterworth Filters Lecture Notes by A. M. Muhia

θ1 = 60 , θ2 = 120 , θ3 = 180 ,θ4 = 240 ,θ5 = 300 ,θ6 = 360

for stability
1
H3 (s) =
(s + 1) (s2 + s + 1)

90 (2k + n − 1)
In general θk = k = 1, 2, . . . . . . 2n
n
The poles on the LHS of the S-plane are always selected since they correspond to a stable circuit.

NB: If n is odd, there is always a pole at the real axis and if n is even there are poles at ψ−angle
from negative real axis
±90
ψ=
n
The poles are separated by an angle of

180°
n
Example 1.3
Tabulate the information of a fifth order Butterworth response and obtain the transfer function

solution
n is odd → a pole exists on the negative axis
180
other poles separated by = 36°
5
s1 = 1∠0 (s + 1)

s2 = 1∠108 (s + 0.31 − j0.95)

s3 = 1∠144 (s + 0.81 − j0.59)

s4 = 1∠216 (s + 0.81 + j0.59)

s5 = 1∠252 (s + 0.31 + j0.95)

1
H (s) =
(s + 1) (s + 0.31 − j0.95) (s + 0.81 − j0.59) (s + 0.81 + j0.59) (s + 0.31 + j0.95)

1.4.2 Low Pass Filter specification

The pass band extends from ω = 0 to ω = ωp and the attenuation should not exceed αmax .
Transition band is from ωp to ωs . For frequencies above ωs , the attenuation should not be less than
αmin .
The filter specification is always given as ωp , αmax , ωs and αmin . For Butterworth response we
need to find n and ω0

12
Butterworth Filters Lecture Notes by A. M. Muhia

Figure 9: Low pass filter specification in terms of attenuation

2 1
Hn (jω) =  2n
ω
1+
ω0
1
α = −10 log  2n (1)
ω
1+
ω0
"  2n #
ω
α = 10 log 1 +
ω0
"  2n #
α ω
= log 1 +
10 ω0
α  2n
ω
10 10 = 1 +
ω0
 2n α
ω
10
= 10 − 1
ω0
1
α !

ω
 2n
= 10 10 − 1
ω0

ω ω (2)
0=
1
α !
2n
10 10 − 1

Using the values of the low pass filter above; ωp and αmax

13
Butterworth Filters Lecture Notes by A. M. Muhia

ωp
ω0 =
1
" αmax #
2n
10 10 − 1

 2n αmax
" #
ωp
=⇒ = 10 10 − 1 (3)
ω0
Similarly using ωs and αmin
 2n "αmin #
ωs
= 10 10 − 1 (4)
ω0
Dividing (4) by (3) and making n the subject
αmin 
2n
 10 10 − 1 

ωs
=  αmax
ωp

10 10 − 1
 αmin 
 10 10 − 1 
log  αmax 
10 10 − 1
n= 
ωs
2 log
ωp
 αmin !
 10 10 − 1 
 
log 
 
 αmax 

10 10 − 1
 

n=   (5)
ωs
2 log
ωp
The design is carried out in two steps
1. using expression (5), find n
Normally this value will be a non integer and therefore rounding up is done to the nearest
integer value.
2. using n obtained above, find ω0
Example
Design a low pass filter with the following specifications αmax = 0.5dB , αmin = 20dB, ωp =
1000rad/s , ωs = 2000rad/s
solution

(10αmin − 1)
 
log
(10αmax − 1)
n=  
ωs
2 log
ωp

14
Butterworth Filters Lecture Notes by A. M. Muhia

"
#
102 − 1
log
(100.05 − 1)
2 log 2

= 4.83

n=5

ωp 1000
ω0 = =
1 1
" αmax #
2n [100.05 − 1] 10
10 10 − 1

= 1234.12rad/s

ωs 2000
ω0 = =
1 1
" αmin #
10 [102 − 1] 10
10 10 − 1

= 1263.18rad/s

15
Chebychev Filters Lecture Notes by A. M. Muhia

1.5 Chebychev Type I Response

Unlike Butterworth filter which has a maximally flat pass band amplitude and monotonic fall outside
the pass band, Chebychev type (I) filter has a magnitude response that ripples in the pass band
and monotonically falls outside the pass band
Chebychev polynomials are used for this purpose. The magnitude response of the Chebychev
response is given by

2 1
Hn (jω) = (1)
1+ 2 Cn2 (ω)
where
Cn - chebychev polynomial
Cn (ω) = cos n cos−1 ω
for ω ≤ 1.............(i)

Cn ω = cosh n cosh−1 ω for ω > 1..............(ii)

Using equation (i)

Cn+1 (ω) + Cn−1 (ω) = cos (n + 1) cos−1 ω + cos (n − 1) cos−1 ω

= 2ωCn (ω)

Cn+1 (ω) = 2ωCn (ω) − Cn−1 (ω)

C0 (ω) = 1

C1 (ω) = ω

C2 (ω) = 2ω 2 − 1

C3 (ω) = 2ωC2 ω − C1 ω

= 2ω 2ω 2 − 1 − ω

= 4ω 3 − 3ω

C4 (ω) = 2ωC3 (ω) − C2 (ω)

= 2ω 4ω 3 − 3ω − 2ω 2 − 1

= 8ω 4 − 8ω 2 + 1
Plotting this on an ω scale for −1 ≤ ω ≤ 1 gives functions captured in Figure 10

16
Chebychev Filters Lecture Notes by A. M. Muhia

(a) C1 (b) C2 (c) C3

(d) C4

Figure 10: Plots of chebychev functions for n = 1 to 4

As seen from the above figures, the Chebychev polynomial has an equiripple characteristic between
the limits ω = −1 and ω = 1

From the magnitude response

2 1
Hn (jω) =
1 + 2 Cn2 (ω)
1
Hn (jω) = p
1 + 2 Cn2 (ω)
The magnitude response of the filter has the following properties
1
1. For 0 ≤ ω ≤ 1 , the filter has an equiripple behavior between the limits 1 and √
1 + 2
at ω = 0

Cn (0) = ± forn even

= 0 for n odd

1
T (j0) = √ for n even
1 + 2

17
Chebychev Filters Lecture Notes by A. M. Muhia

T (j0) = 1 for n odd

at ω = 1

Cn (1) = cos n cos−1 1

= 1 for all n

1
T j (1) = √
1 + 2

2. For ω > 1 , Tn (jω) falls monotonically to 0

Cn (ω) → increases w.r.t ω. Denominator polynomial increases.
The minimum of Tn (jω) occurs when Cn2 (ω) = 1 and the maxima when Cn2 (ω) = 0
There are n half ripples in the frequency range of1rad/s e.g for n = 6, the response is as
shown in Figure 11.

Figure 11: Properties of chebychev response

Attenuation Characteristics
By definition

α = −20 log Hn (jω) dB

Thus for chebychev response
−1
2 2
 
αn = −20 log 1 +  Cn (ω) 2 dB

αn
= log 1 + 2 Cn2 (ω)
 
(2)
10
For large values of ω then

18
Chebychev Filters Lecture Notes by A. M. Muhia

Figure 12: Chebychev response for n = 7 in terms of attenuation

h

2 i
αn = 10 log 1 + 2 cosh n cosh−1 ω (3)

This gives the attenuation in the stop band

Normally, two attenuation values are important in the design of low pass filter; the maximum value
of attenuation αmax which occurs at Cn2 (ω) = 1

Therefore
αmax
= log 1 + 2


10
αmax
r

=⇒  = 10 10 − 1

when 2 Cn2 (ω) = 1 then

α = 10 log 2 = 3.01dB

This is the attenuation that defines the half power frequency ωhp .

ωhp is obtained as

2 Cn2 (ωhp ) = 1

1
Cn (ωhp ) =


= cosh n cosh−1 ωhp

 
−1 −1 1
n cosh (ωhp ) = cosh


19
Chebychev Filters Lecture Notes by A. M. Muhia

  
1 1
⇒ ωhp = cosh cosh−1 (4)
n 
but
r
αmax
= 10 10 − 1
  
1 −1 
 1 
⇒ ωhp = cosh 
 n cosh  r αmax


10 −1
10
−1
 
" αmax #
1 2 
= cosh  cosh−1 10 10 − 1 (5)
 
n



From the plots of the low pass filter, the pass band extends from ω = 0 to ω = 1. At some higher
frequency ωs the stop band begins and the attenuation is required to have a minimum value of αmin

Hence the specifications of the chebychev response are the values of αmax , αmin , and ωs .

With these specifications given, n is determined as follows

αmin = α (ωs ) = 10 log 1 + 2 Cn2 (ωs )

 

αmin
= log 1 + 2 Cn2 (ωs )
 
10
αmin
10 10 = 1 + 2 Cn2 (ωs )

αmin
2
 Cn2 (ωs ) = 10 10 − 1
but

Cn2 (ωs ) = cosh2 n cosh−1 ωs

 

αmin
2 2 −1


=⇒  cosh n cosh ωs = 10 10 − 1
but
r
αmax
= 10 10 − 1

" αmax # αmin

2 −1


=⇒ 10 10 − 1 cosh n cosh (ωs ) = 10 10 − 1

20
Chebychev Filters Lecture Notes by A. M. Muhia

 αmin 
10 −1
cosh2 n cosh−1 ωs =

10
αmax !
10 10 − 1

1
αmin !

2
 10 10 − 1 

 
=⇒ cosh n cosh−1 ωs = 
 
 αmax ! 

10 10 − 1
 

1
αmin
 !
2
 10 10 − 1 
 
n cosh−1 ωs = cosh−1 
 
 αmax ! 

10 10 − 1
 

1
 αmin !
2
 10 10 − 1 
 
cosh−1 
 
 αmax ! 

10 10 − 1
 

n= (6)
cosh−1 ωs
Pole location
From
2 1
H (jω) =   (7)
s=jω −js
1+ 2 Cn2
ωp
The poles of the two expressions are obtained from
 
2 2 −js
1 +  Cn =0
ωp
 
2 2 −js
 Cn =1
ωp
 
−js 1
Cn = ±j (8)
ωp 
j
Assuming  < 1 then ± >  and the hyperbolic function of Cn is used.

Therefore
    
−1 −js −js
cosh n cosh = Cn
ωp ωp

21
Chebychev Filters Lecture Notes by A. M. Muhia

  
−1 −js j
cosh n cosh =±
ωp 
−js
since is complex then
ωp
−js
Let = cosh (ω)where ω = u + jv with u and v real
ωp
Therefore
 
−js
Cn = cosh nω
ωp

= cosh (nu + jnv)

= cosh (nu) cosh (jnv) + sinh (nu) sinh (jnv) (9)

both
cosh (jnv) = cos nv (10)
and
sinh (jnv) = j sin nv (11)
then  
−js j
Cn = cosh nu cosh nv + j sin nv sinh nu = ±
ωp 
Equating the real and imaginary parts

cosh nu cos nv = 0 (12)

1
sinh nu sin nv = ± (13)

since cosh nu cannot be zero (its minimum value is 1), then

cos nv = 0

Π
⇒ vk = (2k + 1) (14)
2n
where
k = 0, 1, 2, 3, . . . . . . 2n + 1
1
for the value of v given by sin nu sin nv = ± , sin nv becomes ±1

sin nv = ±1

1
⇒ sin (nu) = ±

 
1 1
u = ± sinh−1 (15)
n 

22
Chebychev Filters Lecture Notes by A. M. Muhia

 
−jsk
= cosh ω (16)
ωp

− sk = ωp cosh (ω) (17)

 
1 1 Π
sk = jωp cosh sinh−1 + j (2k + 1) (18)
n  2n

          
1 1 Π 1 −1 1 Π
= ωp − sinh sinh−1 sin (2k + 1) + j cosh sinh cos (2k + 1)
n  2n n  2n

for the left half of S-plane

sk = σk + jωk

where
    
1 1 Π
σk = ±ωp sinh sinh−1 sin (2k + 1) (19)
n  2n
    
1 −1 1 Π
ωk = ωp cosh sinh cos (2k + 1) (20)
n  2

Guillenim’s algorithms
This algorithm relates the Butterworth and Chebychev angles.

Π
Given that vk = (2k + 1)................(a)
2n
To avoid redundancy; expression (18) and (19) simplifies as

− σk = ωp [sinh a sin vk ] (21)

± ωk = ωp [cosh a cos vk ] (22)
This algorithm has shown that vk can be replaced by an angle θk of Butterworth such that expression
(a) becomes

(n − 2k − 1) Π
θk = (23)
n k
Hence

sin vk = cos θk (24)

cos vk = sin θk (25)

Substituting (23) and (24) into (20) and (21)

23
Chebychev Filters Lecture Notes by A. M. Muhia

− σk = ωp [sinh a cos θk ] (26)

± ωk = ωp [cosh a sin θk ] (27)

Example 1.4
Determine the pole location and the polynomial for chebychev filter with specifications n = 5 and
αmax = 0.5dB. Assume a normalized case

Solution    
1 1 n − 2k − 1 Π
a= sinh−1 θk =
n  n 2
αmax
r

 = 10 10 − 1
√
= 100.5 − 1
= 0.35
 
1 −1 1
a = sinh
5 0.35
= 0.3545
cosh a = 1.063
sinh a = 0.362
4 180
k=0 θ0 = ∗ = 72°
5 2
2 180
k=1 θ1 = ∗ = 36°
5 2
2 180
k=2 θ2 = ∗ = 0°
5 2
−2 180
k=3 θ3 = ∗ = −36°
5 2
−4 180
k=4 θ4 = ∗ = −72°
5 2
θk = ±72 , ±36 ,0
k = 0 ⇒ σk = − sinh 0.3545 cos 72 = −0.112
ωk = cosh 0.3545 sin 72 = 1.011
k = 1 ⇒ σk = −0.293
ωk = 0.62
k = 2 ⇒ σk = −0.362
ωk = 0
sk = σk + jωk

1
H (s) =
(s + 0.362) (s + 0.112 ± j1.011) (s + 0.293 ± j0.625)

24
Chebychev Filters Lecture Notes by A. M. Muhia

1
=
(s + 0.362) (s2 + 0.224s + 1.035) (s2 + 0.586s + 0.476)
Comparison of Butterworth and Chebychev responses

1. In Butterworth response, ω = 1 identifies the half power frequency while in chebychev response
ω = 1 identifies the end of the ripple band or the pass band. These two frequencies are different
unless  = 1 (αmax = 3dB)

Figure 13: Butterworth and Chebychev responses at ω = 1rad/sec

To overcome the problem,the Butterworth function can be modified by introducing a scaling

factor ε2

2 1
⇒ Hn (jω) = (28)
1 + 2 ω 2n
Then when ω = 1 expression (28) becomes

2 1
Hn (jω) =
1 + 2
Both the modified Butterworth function and chebychev function hold at ω = 1

2. Attenuation corresponding to expression (28)

α (ω) = 10 log 1 + 2 ω 2n
 
(29)

For large values of ω wrt to 1 the function (29) simplifies to

α (ω) = 20 log (ω n )

= 20 log  + 20n log ω (30)

The attenuation for the chebychev function is

25
Chebychev Filters Lecture Notes by A. M. Muhia

α (ω) = 10 log 1 + 2 Cn2 (ω)

 
(31)

for ω  1 , Cn (ω) has the approximate form

Cn (ω) = 2n−1 ω n (32)

h
2 i
⇒ α (ω) = 10 log 21 2n−1 ω n

= 20 log 2n−1 ω n
 

= 20 log 2n−1 + 20 log  + 20 log ω n

= 20 (n − 1) log 2 + 20 log  + 20 log ω

= 6 (n − 1) + 20 log  + 20n log ω (33)

Subtracting (30) from (33)

αCh (ω) − αBw (ω) = 6 (n − 1) dB (34)

Expression (34) shows that chebychev response has a greater attenuation than Butterworth
response at ω >> 1

26
Inverse Chebychev Filters Lecture Notes by A. M. Muhia

1.6 Inverse Chebychev response

Its response is maximally flat in the pass band and ripples in the stop band as shown in Figure 14.

Figure 14: Inverse chebychev responses in terms of attenuation

The transfer function can be obtained in three steps

1. From the Chebychev type 1 response

2 1
H (jω) = (1)
1+ ε2 Cn2 (ω)

Figure 15: Step 1

2. Subtracting expression (1) from 1

2 ε2 Cn2 (ω)
1 − H (jω) = (2)
1 + ε2 Cn2 (ω)

1
3. Replacing ωwith we obtain the inverse chebychev transfer function as
ω
 
2 2 1
ε Cn
2 ω
H (jω) =   (3)
2 2
1
1 + ε Cn
ω

27
Inverse Chebychev Filters Lecture Notes by A. M. Muhia

Figure 16: Step 2

Figure 17: Step 3

In terms of attenuation expression (3) can be written as

2
α = −10 log H (jω)

  
1

ε2 Cn2
 ω 
= −10 log   
 1 
1 + ε2 Cn2
ω
 
 
 
 1 
= −10 log  

1 + 1 
 
 1 
ε2 Cn2
ω
 
 1 
= 10 log 
1 +   (4)
1 
ε2 Cn2
ω

at ω = 1 , Cn (1) = 1 , for all n

28
Inverse Chebychev Filters Lecture Notes by A. M. Muhia

Since at stop band the minimum attenuation is αmax equation (4) can be used to obtain the expres-
sion of 
 
1
αmin = 10 log 1 + 2

αmin  
1
10 10 = 1 + 2

" αmin #
1
= 10 10 − 1
2
1
= r (5)
αmin
10 10 − 1
To obtain the zeros
 
1
ε2 Cn2
2 ω
H (jω) =  
2 2
1
1 + ε Cn
ω
 
1
⇒ ε2 Cn2 =0
ω
 
2 1
⇒ Cn =0 (6)
ω
  
−1 1
cos n cos =0
ω
 
−1 1 Π
n cos = (2k + 1)
ω 2
where k = 0, 1, 2, 3, . . . . . . n − 1
 
−1 1 Π
cos = (2k + 1)
ω 2n
 
s
ω=
j
 
−1 j Π
⇒ cos = (2k + 1)
s 2n
 
j Π
= cos (2k + 1)
s 2n
j
s=  
Π
cos (2k + 1)
2n

29
Inverse Chebychev Filters Lecture Notes by A. M. Muhia

 
Π
sk = j sec (2k + 1)
2n
The expression for n , order of the filter is the same as that of chebychev type 1 response. There-
fore the poles of an inverter chebychev can be found by obtaining the chebychev type 1 poles and
inverting them.

Example 1.5
Given that the specification of an inverse chebychev filter are n = 5 , αmax = 0.3dB and assuming
normalized case, obtain the transfer function of the filter and put the poles and zeros on an s-plane

Solution
2
H (jω) =

αmax
r

ε= 10 10 − 1
r
0.03
= 10 10 − 1

= 0.2674
 
1 −1 1
a = sinh
n 
 
1 −1 1
= sinh
5 0.2674

= 0.406

Poles for chebychev type 1

cosh a = 1.0835

sinh a = 0.417
(n − 2k − 1) Π
θk =
n k
θ0 = 72 θ2 = 0 θ4 = −72

θ1 = 36 θ3 = −36

sk = −σk ± jωk

σk = − sinh a cos θk ± j cosh a sin θk

s0 = − sinh a cos 0 ± j cosh a sin 0

= −0.417

30
Bessel Response Lecture Notes by A. M. Muhia

s12 = − sinh a cos 36 ± j cosh a sin 36

= −0.337 ± j0.637

s34 = − sinh a cos 72 ± j cosh a sin 72

= −0.129 ± j1.0305

for inverse chebychev poles

1
s0 = = −2.398
−0.417
1
s12 = = −0.6489 ± j1.227
−0.337 ± j0.637
1
s34 = = −0.1186 ± j0.9514
−0.129 ± j1.0305

Zeros  
Π
sk = j sec (2k + 1)
2n

Π
s0 = j sec = j1.0515
10
Π
s1 = j sec ∗ 3 = j1.703
10
Π
s2 = j sec ∗5=∞
10
Π
s3 = j sec ∗ 7 = −j1.703
10
Π
s4 = j sec ∗ 9 = −j1.0515
10
(s − j1.0515) (s − j1.703) (s + j1.703) (s + 1.0515)
H (s) =   
(s + 2.398) (s + 0.6489)2 − (j1.227)2 (s + 0.1186)2 − (j0.9514)2

1.7 Bessel Thompson response

Consider and input V1 (t) which is passed through a filter network such that the output is V1 (t) as
shown in Figure 18 With T being the delay in seconds
Mathematically

V2 (t) = V1 (t − T ) (1)
Taking Laplace transform on both sides we obtain

V2 (s) = V1 (s) e−sT (2)

The transfer function is obtained as

31
Bessel Response Lecture Notes by A. M. Muhia

Figure 18: Bessel Filter

V2 (s)
= e−sT (3)
V1 (s)

H (jω) = 1 (4)

arg H (jω) = −ωT (5)

The delay is defined as the negative derivative of arg H (jω)

From expression (3)

let y = sT
1
H (y) = e−y =
ey
1
= (6)
cosh y + sinh y

y2 y4
cosh y = 1 + + + ··· (7)
2! 4!

y3 y5
sinh y = y + + + ··· (8)
3! 5!
1
1 sinh y
H (y) = =
cosh y + sinh y 1 + coth y
cosh y
now coth y =
sinh y
y2 y4 y6
1+ + + ···
= 2! 4! 6!
y 3 y 5 y 7
y+ + + ···
3! 5! 7!
A polynomial of coth y is obtained by cosh y and sinh y using continued function expansion as shown
in page 2 appendix A as.

32
Bessel Response Lecture Notes by A. M. Muhia

1 1
coth y = +
y 3 1
+
y 5 1
+
y 7
+ ......
y

(2n − 1)
For a given continued fraction function is truncated at
y
For example if n = 4 the continued fraction function becomes

1 1
coth y = +
y 3 1
+
y 5 1
+
y 7
y

5 1 35 + y 2
+ =
y 7 7y
y
3 1 10y 2 + 105
+ =
y 35 + y 2 y 3 + 35y
7y
1 1 y 4 + 45y 2 + 105
+ 2
=
y 10y + 105 10y 3 + 105y
y 3 + 35y
y 4 + 45y 2 + 105 cosh y
coth y = =
10y 3 + 105y sinh y

Adding the denominator and the numerator of coth y the transfer function becomes
1
H (y) =
y 4 + 10y 3 + 45y 2 + 105y + 105
Normalizing such that
H (0) = 1 we get

105
H (y) =
y4 + 10y 3+ 45y 2 + 105y + 105
Generally after truncating the infinite continued fraction expansion let the numerator polynomial
be M and the denominator polynomial be N
cosh y M (y)
coth y = =
sinh y N (y)

33
Bessel Response Lecture Notes by A. M. Muhia

C0
Then H (y) = with C0 being introduced such that H (0) = 1
M (y) + N (y)

By truncating coth y at various values of n , denominator polynomial is generalized as

Bn (y) = M (y) + N (y)

= Cn y n + Cn−1 y n−1 + · · · · · · + C1 y + C0

i.e
B0 (y) = 1

B1 (y) = y + 1

B2 (y) = y 2 + 3y + 3

B3 (y) = y 3 + 16y 2 + 15y + 15

B4 (y) = y 4 + 10y 3 + 45y 2 + 105y + 105

B5 (y) = y 5 + 15y 4 + 105y 3 + 420y 2 + 945y + 945

The coefficients of the polynomial Bn (y) can be obtained from

(2n − k)!
Ck = where k = 0, 1, 2 . . . . . . , n
2(n−k) k! (n
− k)!

Hence Bn (y) is called the Bessel polynomial with order n

For each choice of n , the transfer function in the Bessel Thompson assumes the form

Bn (0)
H (y) =
Bn (y)

and the delay of which can be calculated as a function of ωt

Example 1.6
for n = 3 ,the transfer function becomes
15
H (y) = ; y = sT = jωT
y3 + 16y 2+ 15y + 15
The phase function is obtained as
15
H (jωT ) =
(jωT ) + 16 (jωT )2 + 15jωT + 15
3

15
=
−j (ωT ) − 16 (ωT )2 + 15jωT + 15
3

15ωT − (ωT )3
 
−1
arg H (jωT ) = − tan
15 − 16 (ωT )2

34
Bessel Response Lecture Notes by A. M. Muhia

Figure 19: Bessel Response

Differentiating the above function w.r.t. ωT gives the normalized delay function

6 (ωT )4 + 45 (ωT )2 + 225

D (ωT ) =
36 (ωT )4 − 180 (ωT )2 + 225

A plot of D (ωT )as a function of the normalized frequency ωT is shown in Figure 19 below
This shows that for small values of ωT the delay function is approximately 1 for all n . The larger
the value of n the longer the band of frequencies for flat delay response

When the specifications are given in terms of delay T and maximum deviation for a given frequency,
the maximum order n of the Bessel polynomial required to satisfy the specifications can be deter-
mined

This n together with the value of T is sufficient to specify the desired transfer function H (y) from
which a network can be synthesized

Example 1.7
Determine the minimum order of the Bessel Thompson response that realizes the following specifi-
cations

1. at zero frequency the delay is 10ms

2. for ω ≤ 250rad/s the delay error should be less than 1%

3. for ω ≤ 250rad/s the max deviation in magnitude cannot exceed 2dB

35