Sie sind auf Seite 1von 131

Chapter 10

APPROXIMATIONS FOR ANALOG FILTERS


10.1 Introduction, 10.2 Realizability
10.3 to 10.7 Butterworth, Chebyshev, Inverse-Chebyshev,
Elliptic, and Bessel-Thomson Approximations
Copyright c 2005- by Andreas Antoniou
Victoria, BC, Canada
Email: aantoniou@ieee.org
October 28, 2010
Frame # 1 Slide # 1 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction
t
As mentioned in previous presentations, the solution of the
approximation problem for recursive lters can be accomplished by
using direct or indirect methods.
Frame # 2 Slide # 2 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction
t
As mentioned in previous presentations, the solution of the
approximation problem for recursive lters can be accomplished by
using direct or indirect methods.
t
In indirect approximation methods, digital lters are designed
indirectly through the use of corresponding analog-lter
approximations.
Frame # 2 Slide # 3 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction
t
As mentioned in previous presentations, the solution of the
approximation problem for recursive lters can be accomplished by
using direct or indirect methods.
t
In indirect approximation methods, digital lters are designed
indirectly through the use of corresponding analog-lter
approximations.
t
Several analog-lter approximations have been proposed in the past,
as follows:
Butterworth,
Chebyshev,
Inverse-Chebyshev,
elliptic, and
Bessel-Thomson approximations.
Frame # 2 Slide # 4 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction
t
As mentioned in previous presentations, the solution of the
approximation problem for recursive lters can be accomplished by
using direct or indirect methods.
t
In indirect approximation methods, digital lters are designed
indirectly through the use of corresponding analog-lter
approximations.
t
Several analog-lter approximations have been proposed in the past,
as follows:
Butterworth,
Chebyshev,
Inverse-Chebyshev,
elliptic, and
Bessel-Thomson approximations.
t
This presentation deals with the basics of these approximations.
Frame # 2 Slide # 5 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
An analog lter such as the one shown below can be represented by
the equation
V
o
(s)
V
i
(s)
= H(s) =
N(s)
D(s)
where
R
2

R
1

L
2
C
2
C
1
C
3

v
i
(t)
v
o
(t)
Frame # 3 Slide # 6 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
An analog lter such as the one shown below can be represented by
the equation
V
o
(s)
V
i
(s)
= H(s) =
N(s)
D(s)
where
V
i
(s) is the Laplace transform of the input voltage v
i
(t),
R
2

R
1

L
2
C
2
C
1
C
3

v
i
(t)
v
o
(t)
Frame # 3 Slide # 7 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
An analog lter such as the one shown below can be represented by
the equation
V
o
(s)
V
i
(s)
= H(s) =
N(s)
D(s)
where
V
i
(s) is the Laplace transform of the input voltage v
i
(t),
V
o
(s) is the Laplace transform of the output voltage v
o
(t),
R
2

R
1

L
2
C
2
C
1
C
3

v
i
(t)
v
o
(t)
Frame # 3 Slide # 8 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
An analog lter such as the one shown below can be represented by
the equation
V
o
(s)
V
i
(s)
= H(s) =
N(s)
D(s)
where
V
i
(s) is the Laplace transform of the input voltage v
i
(t),
V
o
(s) is the Laplace transform of the output voltage v
o
(t),
H(s) is the transfer function,
R
2

R
1

L
2
C
2
C
1
C
3

v
i
(t)
v
o
(t)
Frame # 3 Slide # 9 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
An analog lter such as the one shown below can be represented by
the equation
V
o
(s)
V
i
(s)
= H(s) =
N(s)
D(s)
where
V
i
(s) is the Laplace transform of the input voltage v
i
(t),
V
o
(s) is the Laplace transform of the output voltage v
o
(t),
H(s) is the transfer function,
N(s) and D(s) are polynomials in complex variable s.
R
2

R
1

L
2
C
2
C
1
C
3

v
i
(t)
v
o
(t)
Frame # 3 Slide # 10 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
The loss (or attenuation) is dened as
L(
2
) =
|V
i
(j )|
2
|V
o
(j )|
2
=

V
i
(j )
V
o
(j )

2
=
1
|H(j )|
2
= 10 log
1
H(j )H(j )
Hence the loss in dB is given by
A() = 10 log L(
2
) = 10 log
1
|H(j )|
2
= 20 log|H(j )|
In eect, the loss in dB is the negative of the gain in dB.
Frame # 4 Slide # 11 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
The loss (or attenuation) is dened as
L(
2
) =
|V
i
(j )|
2
|V
o
(j )|
2
=

V
i
(j )
V
o
(j )

2
=
1
|H(j )|
2
= 10 log
1
H(j )H(j )
Hence the loss in dB is given by
A() = 10 log L(
2
) = 10 log
1
|H(j )|
2
= 20 log|H(j )|
In eect, the loss in dB is the negative of the gain in dB.
t
As a function of , A() is said to be the loss characteristic.
Frame # 4 Slide # 12 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
The phase shift and group delay of analog lters are dened
just as in digital lters, namely, the phase shift is the phase
angle of the frequency response and the group delay is the
negative of the derivative of the phase angle with respect to
, i.e.,
() = arg H(j ) and () =
d()
d
Frame # 5 Slide # 13 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
The phase shift and group delay of analog lters are dened
just as in digital lters, namely, the phase shift is the phase
angle of the frequency response and the group delay is the
negative of the derivative of the phase angle with respect to
, i.e.,
() = arg H(j ) and () =
d()
d
t
As functions of , () and () are the phase response and
delay characteristic, respectively.
Frame # 5 Slide # 14 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
As was shown earlier, the loss can be expressed as
L(
2
) =
1
H(j )H(j )
Frame # 6 Slide # 15 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
As was shown earlier, the loss can be expressed as
L(
2
) =
1
H(j )H(j )
t
If we replace by s/j in L(
2
), we get the so-called loss
function
L(s
2
) =
D(s)D(s)
N(s)N(s)
Frame # 6 Slide # 16 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
As was shown earlier, the loss can be expressed as
L(
2
) =
1
H(j )H(j )
t
If we replace by s/j in L(
2
), we get the so-called loss
function
L(s
2
) =
D(s)D(s)
N(s)N(s)
t
Thus if the transfer function of an analog lter is known, its
loss function can be readily deduced.
Frame # 6 Slide # 17 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
If
H(s) =
N(s)
D(s)
=

M
i =1
(s z
i
)

N
i =1
(s p
i
)
then
L(s
2
) =
D(s)D(s)
N(s)N(s)
=

N
i =1
(s p
i
)

N
i =1
(s p
i
)

M
i =1
(s z
i
)

M
i =1
(s z
i
)
= (1)
NM

N
i =1
(s p
i
)

N
i =1
[s (p
i
)]

M
i =1
(s z
i
)

M
i =1
[s (z
i
)]
Frame # 7 Slide # 18 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
If
H(s) =
N(s)
D(s)
=

M
i =1
(s z
i
)

N
i =1
(s p
i
)
then
L(s
2
) =
D(s)D(s)
N(s)N(s)
=

N
i =1
(s p
i
)

N
i =1
(s p
i
)

M
i =1
(s z
i
)

M
i =1
(s z
i
)
= (1)
NM

N
i =1
(s p
i
)

N
i =1
[s (p
i
)]

M
i =1
(s z
i
)

M
i =1
[s (z
i
)]
t
Therefore,
the zeros of the loss function are the poles of of the transfer
function and their negatives, and
the poles of the loss function are the zeros of the transfer
function and their negatives.
Frame # 7 Slide # 19 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
Zero-pole plots for transfer function and loss function:
s plane s plane
H(s)
2
2
2
2


j
j
L(s
2
)
Frame # 8 Slide # 20 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
An ideal lowpass lter is one that will pass only low-frequency
components. Its loss characteristic assumes the form shown in the
gure.

c
A()
(a)

Frame # 9 Slide # 21 A. Antoniou Digital Signal Processing Secs. 10.1-10.7


Introduction Contd
t
An ideal lowpass lter is one that will pass only low-frequency
components. Its loss characteristic assumes the form shown in the
gure.
The frequency range 0 to
c
is the passband.
The frequency range
c
to is the stopband.
The boundary between the passband and stopband, namely,

c
, is the cuto frequency.

c
A()
(a)

Frame # 9 Slide # 22 A. Antoniou Digital Signal Processing Secs. 10.1-10.7


Introduction Contd
t
As in digital lters, the approximation step for the design of
analog lters is the process of obtaining a realizable transfer
function that would satisfy certain desirable specications.
Frame # 10 Slide # 23 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
As in digital lters, the approximation step for the design of
analog lters is the process of obtaining a realizable transfer
function that would satisfy certain desirable specications.
t
In the classical solutions of the approximation problem, an
ideal normalized lowpass loss characteristic is assumed with a
cuto frequency of order unity, i.e.,
c
1.
Frame # 10 Slide # 24 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
As in digital lters, the approximation step for the design of
analog lters is the process of obtaining a realizable transfer
function that would satisfy certain desirable specications.
t
In the classical solutions of the approximation problem, an
ideal normalized lowpass loss characteristic is assumed with a
cuto frequency of order unity, i.e.,
c
1.
t
A set of formulas are then derived that yield the zeros and
poles or coecients of the transfer function for a specied
lter order.
Frame # 10 Slide # 25 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
Classical approximations such as the Butterworth, Chebyshev,
inverse-Chebyshev, and elliptic approximations lead to a loss
characteristic where
Frame # 11 Slide # 26 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
Classical approximations such as the Butterworth, Chebyshev,
inverse-Chebyshev, and elliptic approximations lead to a loss
characteristic where
the loss is equal to or less than A
p
dB over the frequency
range 0 to
p
;
Frame # 11 Slide # 27 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
Classical approximations such as the Butterworth, Chebyshev,
inverse-Chebyshev, and elliptic approximations lead to a loss
characteristic where
the loss is equal to or less than A
p
dB over the frequency
range 0 to
p
;
the loss is equal to or greater than A
a
dB over the frequency
range
a
to .
Frame # 11 Slide # 28 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
Classical approximations such as the Butterworth, Chebyshev,
inverse-Chebyshev, and elliptic approximations lead to a loss
characteristic where
the loss is equal to or less than A
p
dB over the frequency
range 0 to
p
;
the loss is equal to or greater than A
a
dB over the frequency
range
a
to .
t
Parameters
p
and
a
are the passband and stoband edges,
A
p
is the maximum passband loss (or attenuation), and A
a
is
the minimum stopband loss (or attenuation), respectively.
Frame # 11 Slide # 29 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
The quality of an approximation depends on the values of A
p
and A
a
for a given lter order, i.e., a lower A
p
and a larger A
a
correspond to a better lter.

p
A
p
A
a

c
A()
Frame # 12 Slide # 30 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
In practice, the cuto frequency of the lowpass lter depends
on the application at hand.
Frame # 13 Slide # 31 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
In practice, the cuto frequency of the lowpass lter depends
on the application at hand.
Furthermore, other types of lters are often required such as
highpass, bandpass, and bandstop lters.
Frame # 13 Slide # 32 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
In practice, the cuto frequency of the lowpass lter depends
on the application at hand.
Furthermore, other types of lters are often required such as
highpass, bandpass, and bandstop lters.
t
Approximations for arbitrary lowpass, highpass, bandpass, and
bandstop lters can be obtained through a process known as
denormalization.
Frame # 13 Slide # 33 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
In practice, the cuto frequency of the lowpass lter depends
on the application at hand.
Furthermore, other types of lters are often required such as
highpass, bandpass, and bandstop lters.
t
Approximations for arbitrary lowpass, highpass, bandpass, and
bandstop lters can be obtained through a process known as
denormalization.
t
Filter denormalization can be applied through the use of a
class of analog-lter transformations.
Frame # 13 Slide # 34 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Introduction Contd
t
In practice, the cuto frequency of the lowpass lter depends
on the application at hand.
Furthermore, other types of lters are often required such as
highpass, bandpass, and bandstop lters.
t
Approximations for arbitrary lowpass, highpass, bandpass, and
bandstop lters can be obtained through a process known as
denormalization.
t
Filter denormalization can be applied through the use of a
class of analog-lter transformations.
t
These transformations will be discussed in the next
presentation.
Frame # 13 Slide # 35 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
Frame # 14 Slide # 36 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
t
The coecients must be real.
Frame # 14 Slide # 37 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
t
The coecients must be real.
This requirement is imposed by the fact that inductances,
capacitances, and resistances are required to be real quantities.
Frame # 14 Slide # 38 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
t
The coecients must be real.
This requirement is imposed by the fact that inductances,
capacitances, and resistances are required to be real quantities.
t
The degree of the numerator polynomial must be equal to or less
than the degree of the denominator polynomial.
Frame # 14 Slide # 39 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
t
The coecients must be real.
This requirement is imposed by the fact that inductances,
capacitances, and resistances are required to be real quantities.
t
The degree of the numerator polynomial must be equal to or less
than the degree of the denominator polynomial.
Otherwise, the transfer function would represent a noncausal system
which would not be realizable as a real-time analog lter.
Frame # 14 Slide # 40 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
t
The coecients must be real.
This requirement is imposed by the fact that inductances,
capacitances, and resistances are required to be real quantities.
t
The degree of the numerator polynomial must be equal to or less
than the degree of the denominator polynomial.
Otherwise, the transfer function would represent a noncausal system
which would not be realizable as a real-time analog lter.
t
The poles must be in the left half s plane.
Frame # 14 Slide # 41 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Realizability Constraints
t
Realizability contraints are constraints that must be satised by a
transfer function in order to be realizable in terms of an analog-lter
network.
t
The coecients must be real.
This requirement is imposed by the fact that inductances,
capacitances, and resistances are required to be real quantities.
t
The degree of the numerator polynomial must be equal to or less
than the degree of the denominator polynomial.
Otherwise, the transfer function would represent a noncausal system
which would not be realizable as a real-time analog lter.
t
The poles must be in the left half s plane.
Otherwise, the transfer function would represent an unstable system.
Frame # 14 Slide # 42 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
Frame # 15 Slide # 43 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
The underlying assumptions in the derivation.
Frame # 15 Slide # 44 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
The underlying assumptions in the derivation.
Typical loss characteristics.
Frame # 15 Slide # 45 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
The underlying assumptions in the derivation.
Typical loss characteristics.
Available independent parameters.
Frame # 15 Slide # 46 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
The underlying assumptions in the derivation.
Typical loss characteristics.
Available independent parameters.
Formula for the loss as a function of the independent
parameters.
Frame # 15 Slide # 47 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
The underlying assumptions in the derivation.
Typical loss characteristics.
Available independent parameters.
Formula for the loss as a function of the independent
parameters.
Minimum lter order to achieved prescribed specications.
Frame # 15 Slide # 48 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Classical Analog-Filter Approximations
t
In the slides that follow, the basic features of the various
classical analog-lter approximations will be presented such as:
The underlying assumptions in the derivation.
Typical loss characteristics.
Available independent parameters.
Formula for the loss as a function of the independent
parameters.
Minimum lter order to achieved prescribed specications.
Formulas for the parameters of the transfer function (e.g.,
zeros, poles, coecients, multiplier constant).
Frame # 15 Slide # 49 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation
t
The Butterworth approximation is derived on the assumption that
the loss function L(s
2
) is a polynomial. Since
lim
s
L(s
2
) = lim

L(
2
) = a
0
+ a
2

2
+ + a
2n

2n

in such a case, a lowpass approximation is obtained.


Frame # 16 Slide # 50 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation
t
The Butterworth approximation is derived on the assumption that
the loss function L(s
2
) is a polynomial. Since
lim
s
L(s
2
) = lim

L(
2
) = a
0
+ a
2

2
+ + a
2n

2n

in such a case, a lowpass approximation is obtained.


t
For an n-order approximation, L(
2
) is assumed to be maximally
at at zero frequency.
Frame # 16 Slide # 51 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation
t
The Butterworth approximation is derived on the assumption that
the loss function L(s
2
) is a polynomial. Since
lim
s
L(s
2
) = lim

L(
2
) = a
0
+ a
2

2
+ + a
2n

2n

in such a case, a lowpass approximation is obtained.


t
For an n-order approximation, L(
2
) is assumed to be maximally
at at zero frequency.
This is achieved by letting
L(0) = 1,
d
k
L(x)
dx
k

x=0
= 0 for k n 1
where x =
2
, i.e., n derivatives of the loss are set to zero at zero
frequency.
Frame # 16 Slide # 52 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation
t
The Butterworth approximation is derived on the assumption that
the loss function L(s
2
) is a polynomial. Since
lim
s
L(s
2
) = lim

L(
2
) = a
0
+ a
2

2
+ + a
2n

2n

in such a case, a lowpass approximation is obtained.


t
For an n-order approximation, L(
2
) is assumed to be maximally
at at zero frequency.
This is achieved by letting
L(0) = 1,
d
k
L(x)
dx
k

x=0
= 0 for k n 1
where x =
2
, i.e., n derivatives of the loss are set to zero at zero
frequency.
t
Assuming that L(1) = 2, the loss function in dB can be expressed as
L(
2
) = 1 +
2n
and A() = 10 log(1 +
2n
)
Frame # 16 Slide # 53 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Typical loss characteristics:
0 0.5 1.0 1.5
10
5
0
15
20
25
30
A
(

)
,

d
B
, rad/s
n = 3
n = 6
n = 9
Frame # 17 Slide # 54 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
The loss function for the normalized Butterworth
approximation (3-dB frequency at 1 rad/s) is given by
L(s
2
) = 1 + (s
2
)
n
=
2n

i =1
(s z
i
)
where z
i
=
_
e
j (2i 1)/2n
for even n
e
j (i 1)/n
for odd n
Frame # 18 Slide # 55 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
The loss function for the normalized Butterworth
approximation (3-dB frequency at 1 rad/s) is given by
L(s
2
) = 1 + (s
2
)
n
=
2n

i =1
(s z
i
)
where z
i
=
_
e
j (2i 1)/2n
for even n
e
j (i 1)/n
for odd n
t
Since |z
k
| = 1, the zeros of L(s
2
) are located on the unit
circle |s| = 1.
Frame # 18 Slide # 56 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Zero-pole plots for loss function:
s plane n = 6 s plane n = 5
(a) (b)
Re s
j
I
m

s

Re s
j
I
m

s

Frame # 19 Slide # 57 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
The zeros of the loss function are the poles of the transfer
function and their negatives.
Frame # 20 Slide # 58 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
The zeros of the loss function are the poles of the transfer
function and their negatives.
t
For stability, the poles of the transfer function must be
located in the left-half s plane.
Frame # 20 Slide # 59 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
The zeros of the loss function are the poles of the transfer
function and their negatives.
t
For stability, the poles of the transfer function must be
located in the left-half s plane.
Therefore, they are identical with the zeros of the loss
function located in the left-half s plane.
Frame # 20 Slide # 60 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Typically in practice, the required lter order is unknown.
Frame # 21 Slide # 61 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Typically in practice, the required lter order is unknown.
For Butterworth, Chebyshev, inverse-Chebyshev, and elliptic lters,
it can by easily deduced if the required specications are known.
Frame # 21 Slide # 62 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Typically in practice, the required lter order is unknown.
For Butterworth, Chebyshev, inverse-Chebyshev, and elliptic lters,
it can by easily deduced if the required specications are known.
t
Let us assume that we need a normalized Butterworth lter with a
maximum passband loss A
p
, minimum stopband loss A
a
, passband
edge
p
, and stopband edge
a
.
Frame # 21 Slide # 63 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Typically in practice, the required lter order is unknown.
For Butterworth, Chebyshev, inverse-Chebyshev, and elliptic lters,
it can by easily deduced if the required specications are known.
t
Let us assume that we need a normalized Butterworth lter with a
maximum passband loss A
p
, minimum stopband loss A
a
, passband
edge
p
, and stopband edge
a
.
The minimum lter order that will satisfy the required specications
must be large enough to satisfy both of the following inequalities:
n
[log(10
0.1A
p
1)]
(2 log
p
)
and n
log(10
0.1A
a
1)
2 log
a
Frame # 21 Slide # 64 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd
t
Typically in practice, the required lter order is unknown.
For Butterworth, Chebyshev, inverse-Chebyshev, and elliptic lters,
it can by easily deduced if the required specications are known.
t
Let us assume that we need a normalized Butterworth lter with a
maximum passband loss A
p
, minimum stopband loss A
a
, passband
edge
p
, and stopband edge
a
.
The minimum lter order that will satisfy the required specications
must be large enough to satisfy both of the following inequalities:
n
[log(10
0.1A
p
1)]
(2 log
p
)
and n
log(10
0.1A
a
1)
2 log
a
(See textbook for derivations and examples.)
Frame # 21 Slide # 65 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd

n
[log(10
0.1A
p
1)]
(2 log
p
)
and n
log(10
0.1A
a
1)
2 log
a
t
The right-hand sides in the above inequalities will normally yield a
mixed number but since the lter order must be an integer, the
value obtained must be rounded up to the next integer.
Frame # 22 Slide # 66 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd

n
[log(10
0.1A
p
1)]
(2 log
p
)
and n
log(10
0.1A
a
1)
2 log
a
t
The right-hand sides in the above inequalities will normally yield a
mixed number but since the lter order must be an integer, the
value obtained must be rounded up to the next integer.
As a result, the required specications will usually be slightly
oversatised.
Frame # 22 Slide # 67 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Butterworth Approximation Contd

n
[log(10
0.1A
p
1)]
(2 log
p
)
and n
log(10
0.1A
a
1)
2 log
a
t
The right-hand sides in the above inequalities will normally yield a
mixed number but since the lter order must be an integer, the
value obtained must be rounded up to the next integer.
As a result, the required specications will usually be slightly
oversatised.
t
Once the required lter order is determined, the actual maximum
passband loss and minimum stopband loss can be calculated as
A
p
= A(
p
) = 10 log(1 +
2n
p
) and A
a
= A(
a
) = 10 log(1 +
2n
a
)
respectively.
Frame # 22 Slide # 68 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation
t
In the Butterworth approximation, the loss is an increasing
monotonic function of , and as a result the passband loss is
very small at low frequencies and very large at frequencies
close to the bandpass edge.
Frame # 23 Slide # 69 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation
t
In the Butterworth approximation, the loss is an increasing
monotonic function of , and as a result the passband loss is
very small at low frequencies and very large at frequencies
close to the bandpass edge.
On the other hand, the stopband loss is very small at
frequencies close to the stopand edge and very large at very
high frequencies.
Frame # 23 Slide # 70 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation
t
In the Butterworth approximation, the loss is an increasing
monotonic function of , and as a result the passband loss is
very small at low frequencies and very large at frequencies
close to the bandpass edge.
On the other hand, the stopband loss is very small at
frequencies close to the stopand edge and very large at very
high frequencies.
t
A more balanced characteristic with respect to the passband
can be achieved by employing the Chebyshev approximation.
Frame # 23 Slide # 71 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
As in the Butterworth approximation, the loss function in the
Chebyshev approximation is assumed to be a polynomial in s,
which would assure a lowpass characteristic.
Frame # 24 Slide # 72 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
As in the Butterworth approximation, the loss function in the
Chebyshev approximation is assumed to be a polynomial in s,
which would assure a lowpass characteristic.
t
The derivation of the Chebyshev approximation is based on
the assumption that the passband loss oscillates between zero
and a specied maximum loss A
p
.
Frame # 24 Slide # 73 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
As in the Butterworth approximation, the loss function in the
Chebyshev approximation is assumed to be a polynomial in s,
which would assure a lowpass characteristic.
t
The derivation of the Chebyshev approximation is based on
the assumption that the passband loss oscillates between zero
and a specied maximum loss A
p
.
On the basis of this assumption, a dierential equation is
constructed whose solution gives the zeros of the loss function.
Frame # 24 Slide # 74 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
As in the Butterworth approximation, the loss function in the
Chebyshev approximation is assumed to be a polynomial in s,
which would assure a lowpass characteristic.
t
The derivation of the Chebyshev approximation is based on
the assumption that the passband loss oscillates between zero
and a specied maximum loss A
p
.
On the basis of this assumption, a dierential equation is
constructed whose solution gives the zeros of the loss function.
t
Then by neglecting the zeros of the loss function in the
right-half s plane, the poles of the transfer function can be
obtained.
Frame # 24 Slide # 75 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
In the case of a fourth-order Chebyshev lter the passband
loss is assumed to be zero at =
01
,
02
and equal to A
p
at = 0,

1
, 1 as shown in the gure:
0 0.2 0.4 0.6 0.8 1.0 1.2
0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
A
(

)
,

d
B
, rad/s

A
p

01

02

p
Frame # 25 Slide # 76 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
On using all the information that can be extracted from the gure
shown, a dierential equation of the form
_
dF()
d
_
2
=
M
4
[1 F
2
()]
1
2
can be constructed.
Frame # 26 Slide # 77 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
On using all the information that can be extracted from the gure
shown, a dierential equation of the form
_
dF()
d
_
2
=
M
4
[1 F
2
()]
1
2
can be constructed.
t
The solution of this dierential equation gives the loss as
L(
2
) = 1 +
2
F
2
()
where
2
= 10
0.1A
p
1
and F() = T
4
() = cos(4 cos
1
)
Frame # 26 Slide # 78 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
On using all the information that can be extracted from the gure
shown, a dierential equation of the form
_
dF()
d
_
2
=
M
4
[1 F
2
()]
1
2
can be constructed.
t
The solution of this dierential equation gives the loss as
L(
2
) = 1 +
2
F
2
()
where
2
= 10
0.1A
p
1
and F() = T
4
() = cos(4 cos
1
)
t
The function cos(4 cos
1
) is actually a polynomial known as the
4th-order Chebyshev polynomial.
Frame # 26 Slide # 79 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
Similarly, for an nth-order Chebyshev approximation, one can
show that
A() = 10 log L(
2
) = 10 log[1 +
2
T
2
n
()]
where
2
= 10
0.1A
p
1
and T
n
() =
_
cos(n cos
1
) for || 1
cosh(n cosh
1
) for || > 1
is the nth-order Chebyshev polynomial.
Frame # 27 Slide # 80 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
Typical loss characteristics for Chebyshev approximation:
1.0
0
0
0
L
o
s
s
,

d
B
1.6
, rad/s
L
o
s
s
,

d
B
(a)
1.2 0.8 0.4
10
20
30
n = 4
n = 7
Frame # 28 Slide # 81 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The zeros of the loss function for a normalized nth-order Chebyshev
approximation (
p
= 1 rad/s) are given by s
i
=
i
+ j
i
where

i
= sinh
_
1
n
sinh
1
1

_
sin
(2i 1)
2n

i
= cosh
_
1
n
sinh
1
1

_
cos
(2i 1)
2n
for i = 1, 2, . . . , n.
Frame # 29 Slide # 82 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The zeros of the loss function for a normalized nth-order Chebyshev
approximation (
p
= 1 rad/s) are given by s
i
=
i
+ j
i
where

i
= sinh
_
1
n
sinh
1
1

_
sin
(2i 1)
2n

i
= cosh
_
1
n
sinh
1
1

_
cos
(2i 1)
2n
for i = 1, 2, . . . , n.
t
From these equations, we note that

2
i
sinh
2
u
+

2
i
cosh
2
u
= 1 where u =
1
n
sinh
1
1

i.e., the zeros of L(s


2
) are located on an ellipse.
Frame # 29 Slide # 83 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
Typical zero-pole plots for Chebyshev approximation:
(a) n = 5 A
p
= 1 dB; (b) n = 6 A
p
= 1 dB.
0.5
1.2 1.2
0 0.5
1.0
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1.0
1.2
Re s
j
I
m

s
(a)
0.5 0 0.5
1.0
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1.0
1.2
Re s
j
I
m

s

(b)
Frame # 30 Slide # 84 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
An nth-order normalized Chevyshev transfer function with a
passband edge
p
= 1 rad/s and a maximum passband loss of A
p
dB can be determined as follows:
H
N
(s) =
H
0
D
0
(s)

r
i
(s p
i
)(s p

i
)
=
H
0
D
0
(s)

r
i
[s
2
2Re (p
i
)s +|p
i
|
2
]
where
r =
_
_
_
n1
2
for odd n
n
2
for even n
and D
0
(s) =
_
s p
0
for odd n
1 for even n
Frame # 31 Slide # 85 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The poles and multiplier constant, H
0
, can be calculated by using
the following formulas in sequence:
=
_
10
0.1A
p
1
p
0
=
(n+1)/2
with
(n+1)/2
= sinh
_
1
n
sinh
1
1

_
p
i
=
i
+ j
i
for i = 1, 2, . . . , r
where
i
= sinh
_
1
n
sinh
1
1

_
sin
(2i 1)
2n

i
= cosh
_
1
n
sinh
1
1

_
cos
(2i 1)
2n
H
0
=
_
p
0

r
i =1
|p
i
|
2
for odd n
10
0.05A
p

r
i =1
|p
i
|
2
for even n
Frame # 32 Slide # 86 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1

a
where D =
10
0.1A
a
1
10
0.1A
p
1
Frame # 33 Slide # 87 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1

a
where D =
10
0.1A
a
1
10
0.1A
p
1
t
As in the Butterworth approximation, the value at the right-hand
side of the inequality must be rounded up to the next integer. As a
result, the minimum stopband loss will usually be slightly
oversatised.
Frame # 33 Slide # 88 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1

a
where D =
10
0.1A
a
1
10
0.1A
p
1
t
As in the Butterworth approximation, the value at the right-hand
side of the inequality must be rounded up to the next integer. As a
result, the minimum stopband loss will usually be slightly
oversatised.
The actual minimum stopband loss can be calculated as
A
a
= A(
a
) = 10 log L(
2
a
) = 10 log[1 +
2
T
2
n
(
a
)]
Frame # 33 Slide # 89 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1

a
where D =
10
0.1A
a
1
10
0.1A
p
1
t
As in the Butterworth approximation, the value at the right-hand
side of the inequality must be rounded up to the next integer. As a
result, the minimum stopband loss will usually be slightly
oversatised.
The actual minimum stopband loss can be calculated as
A
a
= A(
a
) = 10 log L(
2
a
) = 10 log[1 +
2
T
2
n
(
a
)]
t
In the Chebyshev approximation, the actual maximum passband loss
will be exactly as specied,i.e., A
p
.
Frame # 33 Slide # 90 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation
t
The inverse-Chebyshev approximation is closely related to the
Chebyshev approximation, as may be expected, and it is
actually derived from the Chebyshev approximation.
Frame # 34 Slide # 91 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation
t
The inverse-Chebyshev approximation is closely related to the
Chebyshev approximation, as may be expected, and it is
actually derived from the Chebyshev approximation.
t
The passband loss in the inverse-Chebyshev is very similar to
that of the Butterworth approximation, i.e., it is an increasing
monotonic function of , while the stopband loss oscillates
between innity and a prescribed minimum loss A
a
.
Frame # 34 Slide # 92 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
Typical loss characteristics for inverse-Chebyshev
approximation:
1.0
0
0
L
o
s
s
,

d
B
0.2 0.4 1.0 2.0 4.0 10.0
, rad/s
20
40
60
n = 4
L
o
s
s
,

d
B
n = 7
0
(b)
Frame # 35 Slide # 93 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The loss for the inverse-Chebyshev approximation is given by
A() = 10 log
_
1 +
1

2
T
2
n
(1/)
_
where

2
=
1
10
0.1A
a
1
and the stopband extends from = 1 to .
Frame # 36 Slide # 94 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The normalized transfer function for a specied order, n, stopband
edge of
a
= 1 rad/s, and minimum stopband loss, A
a
, is given by
H
N
(s) =
H
0
D
0
(s)
r

i =1
(s 1/z
i
)(s 1/z

i
)
(s 1/p
i
)(s 1/p

i
)
=
H
0
D
0
(s)
r

i =1
s
2
+
1
|z
i
|
2
s
2
2Re
_
1
p
i
_
s +
1
|p
i
|
2
where
r =
_
n1
2
for odd n
n
2
for even n
and D
0
(s) =
_
s
1
p
0
for odd n
1 for even n
Frame # 37 Slide # 95 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The normalized transfer function for a specied order, n, stopband
edge of
a
= 1 rad/s, and minimum stopband loss, A
a
, is given by
H
N
(s) =
H
0
D
0
(s)
r

i =1
(s 1/z
i
)(s 1/z

i
)
(s 1/p
i
)(s 1/p

i
)
=
H
0
D
0
(s)
r

i =1
s
2
+
1
|z
i
|
2
s
2
2Re
_
1
p
i
_
s +
1
|p
i
|
2
where
r =
_
n1
2
for odd n
n
2
for even n
and D
0
(s) =
_
s
1
p
0
for odd n
1 for even n
t
The parameters of the transfer function can be calculated by using
the formulas in the next slide.
Frame # 37 Slide # 96 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
=
1

10
0.1A
a
1
, z
i
= j cos
(2i 1)
2n
for 1, 2, . . . , r
p
0
=
(n+1)/2
with
(n+1)/2
= sinh
_
1
n
sinh
1
1

_
p
i
=
i
+ j
i
for 1, 2, . . . , r
with
i
= sinh
_
1
n
sinh
1
1

_
sin
(2i 1)
2n

i
= cosh
_
1
n
sinh
1
1

_
cos
(2i 1)
2n
and
H
0
=
_

_
1
p
0

r
i =1
|z
i
|
2
|p
i
|
2
for odd n

r
i =1
|z
i
|
2
|p
i
|
2
for even n
Frame # 38 Slide # 97 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1
(1/
p
)
where D =
10
0.1A
a
1
10
0.1A
p
1
Frame # 39 Slide # 98 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1
(1/
p
)
where D =
10
0.1A
a
1
10
0.1A
p
1
t
The value of the right-hand side of the above inequality is rarely an
integer and, therefore, it must be rounded up to the next integer.
This will cause the actual maximum passband loss to be slightly
overspecied.
Frame # 39 Slide # 99 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1
(1/
p
)
where D =
10
0.1A
a
1
10
0.1A
p
1
t
The value of the right-hand side of the above inequality is rarely an
integer and, therefore, it must be rounded up to the next integer.
This will cause the actual maximum passband loss to be slightly
overspecied.
The actual maximum passband loss can be calculated as
A
p
= A(
p
) = 10 log
_
1 +
1

2
T
2
n
(1/
p
)
_
where
2
=
1
10
0.1A
a
1
Frame # 39 Slide # 100 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Inverse-Chebyshev Approximation Contd
t
The minimum lter order required to achieve a maximum passband
loss of A
p
and a minimum stopband loss of A
a
must be large
enough to satisfy the inequality
n
cosh
1

D
cosh
1
(1/
p
)
where D =
10
0.1A
a
1
10
0.1A
p
1
t
The value of the right-hand side of the above inequality is rarely an
integer and, therefore, it must be rounded up to the next integer.
This will cause the actual maximum passband loss to be slightly
overspecied.
The actual maximum passband loss can be calculated as
A
p
= A(
p
) = 10 log
_
1 +
1

2
T
2
n
(1/
p
)
_
where
2
=
1
10
0.1A
a
1
t
In this approximation, the actual minimum stopband loss will be
exactly as specied, i.e., A
a
.
Frame # 39 Slide # 101 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation
t
The Chebyshev approximation yields a much better passband
and the inverse-Chebyshev approximation yields a much better
stopband than the Butterworth approximation.
Frame # 40 Slide # 102 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation
t
The Chebyshev approximation yields a much better passband
and the inverse-Chebyshev approximation yields a much better
stopband than the Butterworth approximation.
t
A lter with improved passband and stopband can be
obtained by using the elliptic approximation.
Frame # 40 Slide # 103 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation
t
The Chebyshev approximation yields a much better passband
and the inverse-Chebyshev approximation yields a much better
stopband than the Butterworth approximation.
t
A lter with improved passband and stopband can be
obtained by using the elliptic approximation.
t
The elliptic approximation is more ecient that all the other
analog-lter approximations in that the transition between
passband and stopband is steeper for a given approxima-
tion order.
Frame # 40 Slide # 104 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation
t
The Chebyshev approximation yields a much better passband
and the inverse-Chebyshev approximation yields a much better
stopband than the Butterworth approximation.
t
A lter with improved passband and stopband can be
obtained by using the elliptic approximation.
t
The elliptic approximation is more ecient that all the other
analog-lter approximations in that the transition between
passband and stopband is steeper for a given approxima-
tion order.
In fact, this is the optimal approximation for a given piecewise
constant approximation.
Frame # 40 Slide # 105 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
Loss characteristic for a 5th-order elliptic approximation:
A
(

A
a
A
p

p

a

01

02
Frame # 41 Slide # 106 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
The passband loss is assumed to oscillate between zero and a
prescribed maximum A
p
and the stopband loss is assumed to
oscillate between innity and a prescribed minimum A
a
.
Frame # 42 Slide # 107 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
The passband loss is assumed to oscillate between zero and a
prescribed maximum A
p
and the stopband loss is assumed to
oscillate between innity and a prescribed minimum A
a
.
t
On the basis of the assumed structure of the loss
characteristic, a dierential equation is derived, as in the case
of the Chebyshev approximation.
Frame # 42 Slide # 108 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
The passband loss is assumed to oscillate between zero and a
prescribed maximum A
p
and the stopband loss is assumed to
oscillate between innity and a prescribed minimum A
a
.
t
On the basis of the assumed structure of the loss
characteristic, a dierential equation is derived, as in the case
of the Chebyshev approximation.
t
After considerable mathematical complexity, the dierential
equation obtained is solved through the use of elliptic
functions, and the parameters of the transfer function are
deduced.
Frame # 42 Slide # 109 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
The passband loss is assumed to oscillate between zero and a
prescribed maximum A
p
and the stopband loss is assumed to
oscillate between innity and a prescribed minimum A
a
.
t
On the basis of the assumed structure of the loss
characteristic, a dierential equation is derived, as in the case
of the Chebyshev approximation.
t
After considerable mathematical complexity, the dierential
equation obtained is solved through the use of elliptic
functions, and the parameters of the transfer function are
deduced.
The approximation owes its name to the use of elliptic
functions in the derivation.
Frame # 42 Slide # 110 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
The passband and stopband edges and cuto frequency of a
normalized elliptic approximation are dened as follows:

p
=

k,
a
=
1

k
,
c
=

p
= 1
Constants k and k
1
given by
k =

p

a
and k
1
=
_
10
0.1A
p
1
10
0.1A
a
1
_
1/2
are known as the selectivity and discrimination constants.
Frame # 43 Slide # 111 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
A normalized elliptic lowpass lter with a selectivity factor k,
passband edge
p
=

k, stopband edge
a
= 1/

k, a maximum
passband loss of A
p
dB, and a minimum stopband loss equal to or
in excess of A
a
dB has a transfer function of the form
H
N
(s) =
H
0
D
0
(s)
r

i =1
s
2
+ a
0i
s
2
+ b
1i
s + b
0i
where r =
_
_
_
n1
2
for odd n
n
2
for even n
and D
0
(s) =
_
s +
0
for odd n
1 for even n
Frame # 44 Slide # 112 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
A normalized elliptic lowpass lter with a selectivity factor k,
passband edge
p
=

k, stopband edge
a
= 1/

k, a maximum
passband loss of A
p
dB, and a minimum stopband loss equal to or
in excess of A
a
dB has a transfer function of the form
H
N
(s) =
H
0
D
0
(s)
r

i =1
s
2
+ a
0i
s
2
+ b
1i
s + b
0i
where r =
_
_
_
n1
2
for odd n
n
2
for even n
and D
0
(s) =
_
s +
0
for odd n
1 for even n
t
The parameters of the transfer function can be obtained by using
the formulas in the next three slides in sequence in the order shown.
Frame # 44 Slide # 113 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
k

=
_
1 k
2
q
0
=
1
2
_
1

1 +

_
q = q
0
+ 2q
5
0
+ 15q
9
0
+ 150q
13
0
D =
10
0.1A
a
1
10
0.1A
p
1
n
log 16D
log(1/q)
(round up to the next integer)
=
1
2n
ln
10
0.05A
p
+ 1
10
0.05A
p
1

0
=

2q
1/4

m=0
(1)
m
q
m(m+1)
sinh[(2m + 1)]
1 + 2

m=1
(1)
m
q
m
2
cosh 2m

Frame # 45 Slide # 114 A. Antoniou Digital Signal Processing Secs. 10.1-10.7


Elliptic Approximation Contd
W =

_
1 + k
2
0
_
_
1 +

2
0
k
_

i
=
2q
1/4

m=0
(1)
m
q
m(m+1)
sin
(2m+1)
n
1 + 2

m=1
(1)
m
q
m
2
cos
2m
n
where =
_
_
_
i for odd n
i
1
2
for even n
i = 1, 2, . . . , r
V
i
=

_
1 k
2
i
_
_
1

2
i
k
_
Frame # 46 Slide # 115 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
a
0i
=
1

2
i
b
0i
=
(
0
V
i
)
2
+ (
i
W)
2
_
1 +
2
0

2
i
_
2
b
1i
=
2
0
V
i
1 +
2
0

2
i
H
0
=
_

r
i =1
b
0i
a
0i
for odd n
10
0.05A
p

r
i =1
b
0i
a
0i
for even n
Frame # 47 Slide # 116 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
Because of the fact that the lter order is rounded up to the
next integer, the minimum stopband loss is usually
oversatised.
Frame # 48 Slide # 117 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
Because of the fact that the lter order is rounded up to the
next integer, the minimum stopband loss is usually
oversatised.
t
The actual minimum stopband loss is given by the following
formula:
A
a
= A(
a
) = 10 log
_
10
0.1A
p
1
16q
n
+ 1
_
Frame # 48 Slide # 118 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
Because of the fact that the lter order is rounded up to the
next integer, the minimum stopband loss is usually
oversatised.
t
The actual minimum stopband loss is given by the following
formula:
A
a
= A(
a
) = 10 log
_
10
0.1A
p
1
16q
n
+ 1
_
t
The loss of an elliptic lter is usually calculated by using the
transfer function, i.e.,
A() = 20 log
1
|H(j )|
Frame # 48 Slide # 119 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Elliptic Approximation Contd
t
Because of the fact that the lter order is rounded up to the
next integer, the minimum stopband loss is usually
oversatised.
t
The actual minimum stopband loss is given by the following
formula:
A
a
= A(
a
) = 10 log
_
10
0.1A
p
1
16q
n
+ 1
_
t
The loss of an elliptic lter is usually calculated by using the
transfer function, i.e.,
A() = 20 log
1
|H(j )|
(See textbook for details.)
Frame # 48 Slide # 120 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation
t
Ideally, the group delay of a lter should be constant; equivalently,
the phase shift should be a linear function of frequency to minimize
delay distortion (see Sec. 5.7).
Frame # 49 Slide # 121 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation
t
Ideally, the group delay of a lter should be constant; equivalently,
the phase shift should be a linear function of frequency to minimize
delay distortion (see Sec. 5.7).
t
Since the only objective in the approximations described so far is to
achieve a specic loss characteristic, there is no reason for the phase
characteristic to turn out to be linear.
Frame # 49 Slide # 122 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation
t
Ideally, the group delay of a lter should be constant; equivalently,
the phase shift should be a linear function of frequency to minimize
delay distortion (see Sec. 5.7).
t
Since the only objective in the approximations described so far is to
achieve a specic loss characteristic, there is no reason for the phase
characteristic to turn out to be linear.
In fact, it turns out to be highly nonlinear, as one might expect,
particularly in the elliptic approximation.
Frame # 49 Slide # 123 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation
t
Ideally, the group delay of a lter should be constant; equivalently,
the phase shift should be a linear function of frequency to minimize
delay distortion (see Sec. 5.7).
t
Since the only objective in the approximations described so far is to
achieve a specic loss characteristic, there is no reason for the phase
characteristic to turn out to be linear.
In fact, it turns out to be highly nonlinear, as one might expect,
particularly in the elliptic approximation.
t
The last approximation in Chap. 10, namely, the Bessel-Thomson
approximation, is derived on the assumption that the group delay is
maximally at at zero frequency.
Frame # 49 Slide # 124 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation
t
Ideally, the group delay of a lter should be constant; equivalently,
the phase shift should be a linear function of frequency to minimize
delay distortion (see Sec. 5.7).
t
Since the only objective in the approximations described so far is to
achieve a specic loss characteristic, there is no reason for the phase
characteristic to turn out to be linear.
In fact, it turns out to be highly nonlinear, as one might expect,
particularly in the elliptic approximation.
t
The last approximation in Chap. 10, namely, the Bessel-Thomson
approximation, is derived on the assumption that the group delay is
maximally at at zero frequency.
t
As in the Butterworth and Chebyshev approximations, the loss
function is a polynomial. Hence the Bessel-Thomson approximation
is essentially a lowpass approximation.
Frame # 49 Slide # 125 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation Contd
t
The transfer function for a normalized Bessel-Thomson
approximation is give by
H(s) =
b
0

n
i =0
b
i
s
i
=
b
0
s
n
B(1/s)
where
b
i
=
(2n i )!
2
ni
i !(n i )!
Frame # 50 Slide # 126 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation Contd
t
The transfer function for a normalized Bessel-Thomson
approximation is give by
H(s) =
b
0

n
i =0
b
i
s
i
=
b
0
s
n
B(1/s)
where
b
i
=
(2n i )!
2
ni
i !(n i )!
t
The group-delay is 1 s. An arbitrary delay can be obtained by
replacing s by
0
s where
0
is a constant.
Frame # 50 Slide # 127 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation Contd
t
The transfer function for a normalized Bessel-Thomson
approximation is give by
H(s) =
b
0

n
i =0
b
i
s
i
=
b
0
s
n
B(1/s)
where
b
i
=
(2n i )!
2
ni
i !(n i )!
t
The group-delay is 1 s. An arbitrary delay can be obtained by
replacing s by
0
s where
0
is a constant.
t
Function B() is a Bessel polynomial , and s
n
B(1/s) can be
shown to have zeros in the left-half s plane, i.e., the
Bessel-Thomson approximation represents stable analog lters.
Frame # 50 Slide # 128 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation Contd
t
Typical loss characteristics:
0 1 2 3 4 5 6
0
5
10
15
20
25
30
L
o
s
s
,

d
B
, rad/s
n = 3
n = 6
n = 9
Frame # 51 Slide # 129 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
Bessel-Thomson Approximation Contd
t
Typical delay characteristics:
0 1 2 3 4 5 6
0
0.2
0.4
0.6
0.8
1.0
1.2

,


s
, rad/s
n = 3
n = 6
n = 9
Frame # 52 Slide # 130 A. Antoniou Digital Signal Processing Secs. 10.1-10.7
This slide concludes the presentation.
Thank you for your attention.
Frame # 53 Slide # 131 A. Antoniou Digital Signal Processing Secs. 10.1-10.7