Vous êtes sur la page 1sur 11

INFINITE IMPULSE RESPONSE FILTER

5.1. Review of Analog Filter Design


5.1.1. Introduction

The theory of design of IIR digital filters depends on understanding


continuous-time (analog) filter design techniques. The design of a digital filter is
reduced to ~i=inCT an a prouriate analo filter ana then pertormm the conversion
n:om H (3) to H (Z), that is, mapping from s-plane to z-p ane. So, we have to reVIeW
the aesign IOrmUlas tor the standard types of analog fliters. Basically, there are four
~portant types of analog filters. They are, '

1) Butterworth
~--~ filters

2) Chebyshev filters
3) Bessel filters

and 4) Elliptic filters.

It is known that the lowpass, highpass, bandbass, and bandstop filters can be
obtained from a normalIzed lowpass !lIter VIa speclllc transformatIOn ill the s-plane.
So our discussiOn IS limited to lowpass fIlters. ~ubsequently, we descrIbe several
nequency transformatIOns, that convert a lowpass prototype filter into either a
highpass, bandpass or bandstop filter.

Butterworth lowpass filters are characterized by the property that the magnitude
chara'cterIstIc IS maxImally flat at the OrIgIn of the s-plane. Details of Butterworth
filters and the design procedure are explairieii in the following three sections.
Chebyshev filters are characterized by the property that over a prescribed band of
equencies, the peale magnitude of the approximation error is minimized. The
l~;gilltude error IS, ill tact, eqUlripple over the band of frequencies, that is, the error
~cillates between maxima and mmIma of equal amplItude. lJependmg on whether
the band frequencies over which the error is minimized is the passband or the
:;t;jpband, the filter designs are called Type-I or Type-IT These detaIls are proVldecr
'is Section 5.13.

Bessellov.rpass filters are a class of all-pole filters that are characterized by the
property that the group delay is maximally flat at the origin of the s-plane. \Ve note
that the Bessel fliter has a larger transition bandwidth, but its phase is linear within
the. passband. The maximally flat group delay or the linear phase characteristics of
t~ssel filters are destroyed by the methods of digitization to be discussed
in this chapter.
~lliptic filters are characterized by a magnitude response that iseql!LIj~
both the passband and the stopband This class of filters ~~s both poles and
zero>;. The elliptic filters are optimum in theSeIiSetnat for a given order and for
, ,given ri{>ple specifications no other filter achieves a &ster tran1?ition between the
passband and stopband, that is, has a narrower transition bandv.'idth.

As we are going to restrict our discussion to the properties and design


procedures for the analog Butterworth and Chebyshev filters, readers interested in
the other' two. types of filters may refer the books [2] and [4] by L.R.Rabiner &
B.Gold and L.C.Ludeman respectively.
5.1.2. Butterworth Filters
A linear time invariant analog filter can be characterized by its system function
H (s) or its corresponding frequency response H (j Q). The Butterworth filter of order
N is described by the magnitude squared of its frequency response as follows.

I HN (j 1'2) 12 = 11[1 + (1'2/ilc)2l'lJ (5.1)


The ma/o,'llitude squared frequency response of the Butterworth filter is shown in Fig
5.1 for various values of N.

I H, 01'2)12

N = 100
0.5 N=2
N,= 1

0 ilc Q

Fig. 5.1 The magnitude squared frequency response for a ButtelWorth filter.

\Ve can observe the followmg properties of Butterworth filters:


2
1. i HN (j U)./u=o
. I =1 for all N.

2 --~ for all finite N.


2. I HN(jQ)
I 1
u=u,-2
.It means that, 1 HN (j il) I = v; = 0.707and 20 log iHN (j il) I = - 3.0103dB at
cutoff frequency.

3. : Hr-; (j il) 12 is a monotonically decreasing function of il.


lowpass frequency
4. As N becomes larger, I HN (j il) 12 approaches an ideal
response.

5. : HN (j il) 12 is called maximally flat at the origin since all other derivaties exist
and are zero.

For an analog system we remember that the frequency response is obtained by


setting S =jilin the transfer function H (s). Therefore, replacement of !2 by S/j in
Eqn (5.1) resulis in the system transfer function. Setting ilc = 1 in Eqn (5.1) gives
Hx li il) i2 for the normalized filter as follows:
I . I

I HNO!2J 12=HN(jU).HN(-jil)=I/(I+!22N)
Therefore, HN (S) . HN ( - S) = 1/[1+ (S/j)2N]

= 1/11 + (- S2)N] (1),2)

The poles of HN (S) , HN ( - S) are given by the roots of the denominator. that is.
2N
1+ ( - S2)N =11 (S - SK) ~
K=l
where.

Jei (2K - 1) 11/21\' for even N. (f).3)


SK = 1 pi 1~-1)1I/N for odd N

Since! SK I =ITI, the poles of HN (S) , HI\' \ - S) are located on the unit-circle
I S ' 1 .t '
,= WI Jl N sepr\ration.
,

For a stable and causal filter, the poles of HI\' IS) are selected to be those in
the left-half plane and the normalized transfer function HN (S) can be formed as,
1 1
(5.4)
HN(S)=N BN(S)
11 (S- Pi)
i=1

where Pi are all the left half planet poles of HN (5) , HN ( - 5). The denominator
BJ\' (S). can be shown to be a Butterworth polynomial of order N. Table ,5.1. gives
the first eight Butterworth polynomials in the real factored form.
Table 5.1. Butterworth polynomials in factored form and I}ormalized
Lowpass Butterworth illter transfer functio~

Order N Butterworth Polynomials BN (8)

1. 8+1

2. 82 + v2 8 + 1.

3. (82 + 8 + 1) (8 + 1)

4. (82 + 0.76538 + 1) (82 + 1.8,17768 + 1)

5. (S + 1) (S2 + 0.6180S + 1) (S2 + 1.6180S + 1)


6. (S2... 0.5176S + 1) (S2 + \'2 S + 1) (82 + 1.93183 + 1)
7. (S + 1) (82 + 0.44508 + 1) (82 + 1.24568 + 1) (82 + 1.80228 + 1)

8. (S2 + 0.39868 + 1) (S2 + 1.11108 + 1) (S2 + 1.66308 + 1) (S2 + 1.9622S + 1)

Normalized Lowpass Butterworth filters, H!\ (S) = 1/BN (S)

The folJowing example illustrates the way of fmding the first and second order
Butterworth polynomials and corresponding lowpass transfer functions
HI (S) and H2 (8). '

Example 5.1: Find the transfer function for normalized first and second order
Butterworth filters.
Solution:
First - Order Case:
8ince N = 1, from Eqn (5.3), we have,
S} = ~o = 1.
and, 82 = ~ It = - 1.
Therefore, PI = - 1 (the left half plane pole)
Hence, the normalized transfer function of first order Butterworth filter is given by,
H} (s) = 1/(S + 1)
Second - Order Case:
Since N = 2, from Eqn (5.3) we have,
1 . 1
81,84 =:n:tJ:n'
-1 . 1
and S2' S3 = 72 :t J ~
-1 . 1
(the LHP poles)
Therefore,PI, P2=::r2:!:J ~

Hrooo,H, (S)=[( S +t,-j t,J( S+t,+jt, J]


1

[(s+iJ-(jiJJ 1
=(8'2 + --.f2 S + 1)
is the normalized transfer function of the second-order Butterworth filter.

5.1.3. Analog. to - Analog Transformations


From the one radian, normalized lowpass Buttterworth filters, other
non-normalized Butterworth filter could be derived by transformational methods, that
is, using a normalized lov.:pass filter as the prototype filter, we can achieve the other
types.

If we replace S of H (8), the system function for a normaliz€d lowpass filter, by


~ ' :e get a new transfer function H' (S), given by,
1.u
- ---
H' (8) = H (S) ,I =H ~ l
s ->s/uu ( ilu ) .

If we evaluate the magnitude of the transfer function H' (8) at 8 = j il to get the
frequency response we have,

I H' (j il) I = iH (j il/ilu) I

At the value a =au , we have,


I H' (j nu) I = I H (j1) I
That is, the frequency response func:tion evaluated at il =au is
for the new transfer
equal to the value of the normalized transfer function at a =
1. It means, the cutoff
. frequency is moved from 1 rad/sec to the new cutoff frequency au rad/sec and there
is a scaling of frequency axis which leads to non-normalized lowpass filter with cutoff
frequency 11" rad/sec. Similarly, transformations can be achieved for taking lowpass
transfer function to highpass, bandpass and bandstop transfer functions. These
transformations along with design equations are presented here.
Lowpass to Lowpass:

S ~ ~ where au is the new upper cutoff frequency.

0 0

-
20 log IH on)1 1
K2 t ~..- K,. L J...

n, a .Q,. n,'
1

Proto type Transformed iI


Kl and K2 are passband and stopband gains in dB. Qr is
the critical frequenCY or 1
stopband edge frequency.
I
If the transformation S ~ ~
is applied to the lowpass structure as shown, the
critical frequency Or will be transformed into .Q'r which is ilr time Qu' that is,
jj
j
-I
I
O'r=n...ilu

Low pass - to - Highpass:


I
I
I
S ~ ~I where ill is the lower cutoff frequency of the desired high-pass filter.
,
Here, the stopband edge (critical) frequency ilr will be transformed into
QI
.Q'=-
r Qr
Lowpass -to - Bandpass:
2
S + QI °u
S ~ S (a,. - ° ,) where °1 and Ou are the lower and upper cutoff frequencies of t.
the desired bandpass filter. !
I
0

20 log IH00)1

~ t ~...
11, 0 0,11, a. 0,
Proto type Transformed
Here, we have two new critical frequencies, say 01 and 02 which are defined as
follows:
-
°av = (Cu 01)/2
. 22 1/2
01 = (Or Oav+ 01 °u) - Oav a.
f 2 2 1/2
and 02 = (Or Oav + Or °u) + Oav a..

! Lowpass. to . Bandstop:
s (Ou- 01)
S -+ 2 leads to transformation from lowpass to bandstop filters where,
S+~Cu .
the two new critical frequencies, nl and n2, that is, the two stopband edge frequencies
I. are given by,
.
01 = [(Oav/Or)2 -
+ nl CuJ1/2 Oavo.

and 02 = [(nav/o.)2 + 01 CuJl/2+ °avo..


where, Oav = (Ou - 01)/2
5.1.4. Design of Lowpass Butterworth Filters
The Butterworth LP frequency response is characterized by the order N and
the cutoff frequency nc as seen in Eqn (5;1). That is,
1
HN (j n) = ~1 + (01nc>2N
The fllter specificationsare generally given in terms of the critical frequencies and
gains in the passband and the stopband say, nl, 02 andKl'~' as shown in Fig 5.2
for the lowpass response.
dB

20 log IHun)1

K2 'HHHH" '

n, n2 n rad/sec.

Fig. 5.2. Lowpass filter specifications.

A common set of conditions for the lowpass response are,

0 ~ 20 log i H (j il) i ~ Kl for all il::; ill 1 (5.5)


20 log IH (j il) i ::; K2 for all il ~ il2 J
If w(' replace H (j il) in Eqn (5.51 by the Butterworth magnitude squared function
and when equalities hold,
we have,
1
10 100'
b
[ 1 + (ill/ilc)2N ] =Kl'
1
and 10 log
[ 1 + (J:22/J:2c)2N ] = K2'
which yields,

(ill/Q/N = 10-0.1 K]- 11\


(5.6)
and (il2/P-c)2N = 10- 0.1 K2 - 1 I

respEctively. We now have an expression to relate ill' D.2' Kl' K2 and N as follows.
2N
D.l
= 10-0.l~I-1
( il2 ) 10-0.l~-1

A simple closed form expression for N is easily obtained as,


log10 [(10- 0.1 Kl - 1)/(10- 0.1 IS - 1)]
N (5.7)
2 10glO (.0.1/.0.2)

If N is an integer we use that value, otherwise we use the next larger integer. 80
1et us rewrite Eqn (5.7) as,

N>
log10 [(10- 0.1Kl - 1)/(10- 0.1 ~ - 1)] (5.8)
- 2 10glO(.0.1/.0.2)

From Eqn (5.6), it is clear that the selected value of N results in two different values
for .o.c.If we are particular about our exact requirement at .0.1, we use,
(5.9)
.o.c= .0.1/(10- 0.1K,- 1)1I2N

which yields better than our requirement at .0.2' Otherwise, we use,


(5.10)
.o.c= .0.2/(10- 0.1IS - 1)1I2N

which results in our exact requirement at .0.2 and exceeds our requirement at .0.]. A
value of nc in between these values, gives an H (j .0.) that exceeds both requirements.

Let us illustrate the above procedure to design a lowpass Butterworth filter


through an example.

Example 5.2: Design an analog Butterworth filter that has,


0 ~ 20 log I H (j .0.)"1~ - 2dB for all.o. S;20 rad/sec
20 log 1 H (j .0.) S; - 10 dB
1 for all .0.~ 30 rad/sec
Solution: Using Eqn (5.8), we have,

log10 [(10°.2- 1)/(10 - l)j


N~ 2 log10 (20/30) =3.3709
We choose the next larger integer, that is, N = 4.
Next, using Eqn '(5.9), to exactly satisfy the - 2 dB requirement, we obtain,

.o.c= 20/(10°.2 - 1)1/8 = 21.3868.

The normalized transfer function of loy/pass Butterworth filter (.o.c= 1) for N =4 is


given by,

H4 (8) = 1/[(S2 + 0.76536S + 1) (S2 + 1.847768 + 1)]

Now, applying a lowpass - to - lowpass transformation, that is, 8 -- ~c ' with


.o.c= 21.3868, yields the desired transfer function.
154 INFINITE IMPULSE RESPONSE FILTER

H (8) = H4 (8) 18-+ 8/21.3868


1
= [(8/21.3868)2 + 0.76536 (8n1.3868) + 1]
1
x
[(8/21.3868)2 + 1.84776 (8/21.3868) + 1]
0.209210 x 106
- (82 + 16.38688 + 457.395) (82 + 39.5178 + 457.395)

The procedure for designing other types of filters like highpass, bandpass and
bandstop filters, to satisfy the given set of specification is also composed in two steps.'
First, design the prototype lowpass filter with determined N value and then, apply
the lowpass to required band transformation using the desired ill for HPF, and
UI and ilu for BPF and B8F filters.
The formulas for finding the N value from given specifications of other types of
~ters are provided below
Highpass filters:

K,,....................
Gain in
dB

K, ,..............

il, il, il

loglO [(10- 0.1KI- 1)/(10- 0.1K,- l)J


N>
- 2 loglO(il2/fll)

where, Kl - passband gain.

K2 - stopband gain

nl - passband critical frequency.


and il2 - stopband critical frequency.
Bandpass filters:

K,'"''''''''''
Gain in
dB

n, n, nu n, n

N ~ log10 [(10- 0.1Kj - 1)/(10-" 0.1 ~, - 1)]


where,
A =(- QI -r ill nu)/[il1 (ilu - ill)]

B =(il~ - ill ilu)/[il2 (ilu - ill)]


Note:
!2J and Qu are passband critical frequencies.
ill and il2 are stopband critical frequencies.

Bandstop filters:

Gain in
dB

~I i...

Q, Q1 ~ Qu Q

Vous aimerez peut-être aussi