Lecture 2 - Frequency Selective Filters

2.1 Applications
Low-pass : to extract short-term average or to eliminate high-frequency uctuations (eg. noise ltering, demodulation, etc.) High-pass : to follow small-amplitude high-frequency perturbations in presence of much larger slowly-varying component (e.g. recording the electrocardiogram in the presence of a strong breathing signal) Band-pass : to select a required modulated carrier frequency out of many (e.g. radio) Band-stop : to eliminate single-frequency (e.g. mains) interference (also known as notch ltering)

2.2 Design of Analogue Filters

We will start with an analysis of analogue low-pass lters, since a low-pass lter can be mathematically transformed into any other standard type. Design of a lter may start from consideration of

The desired frequency response.

The desired phase response.

11

notch | G( ) |

low-pass band-pass high-pass

Figure 2.1: Standard lters. The majority of the time we will consider the rst case. Consider some desired response, in the general form of the (squared) magnitude of the transfer function, . This response is given as i.e.
  

where denotes complex conjugation. If represents a stable lter (its poles are on the LHS of the s-plane) then is unstable (as its poles will be on the RHS). The design procedure consists then of
 



Considering some desired response of .

 

as a polynomial in even powers

"!#  

Designing the lter with the stable part of

This means that, for any given lter response in the positive frequency domain, a mirror image exists in the negative frequency domain.

2.2.1 Ideal low-pass lter

Any frequency-selective lter may be described either by its frequency response (more common) or by its impulse response. The narrower the band of frequencies transmitted by a lter, the more extended in time is its impulse response waveform. Indeed, if the support in the frequency domain is decreased by a factor of (i.e. made narrower) then the required support in the time domain is increased by a factor of (you should be able to prove this).
$ $

12

Consider an ideal low-pass lter with a brick wall amplitude cut-off and no phase shift, as shown in Fig. 2.2.

| G( ) | 1

Figure 2.2: The ideal low-pass lter. Note the requirement of response in the negative frequency domain. Calculate the impulse response as the inverse Fourier transform of the frequency response:
243 % '&(  ) 01 5 3 5 7698(@#ACBDFE98  ) 01 B G )IH 2 BG @#ACBDFE98  01 ) 6P& Q@#ACB G DSRT@ 5 ACB G D

hence,
8VU % '&(  1 "W(X`Y 8VUa& 8VU&

Figure 2.3 shows the impulse response for the lter (this is also referred to as the lter kernel). The output starts innitely long before the impulse occurs i.e. the lter is not realisable in real time. 13

2 1.5 1

g(t)
0.5 0 0.5 5

Figure 2.3: Impulse response (lter kernel) for the ILPF. The zero crossings occur at integer multiples of .
1cb 8VU

A delay of time
d

such that
8VU % F&(  1 W(X`Y d 8eU"'&eR d 8VU"F&VR

would ensure that most of the response occurred after the input (for large ). The use of such a delay, however, introduces a phase lag proportional to frequency, since . Even then, the lter is still not exactly realisable; instead the design of analogue lters involves the choice of the most suitable approximation to the ideal frequency response.
d $gf %Sh i698Ip  8 d

2.3 Practical Low-Pass Filters

Assume that the low-pass lter transfer function is a rational function in . The types of lters to be considered in the next few pages are all-pole designs, which means that will be of the form:
  $gq qsr $Pq 5ut q ) 5ut rwvvv $ t r $gx

yP

i698

 $Pq 7698 q r i698 $gq 5ut

) q 5ut  rwvvv $ 698 t r $gx CR698 8

The magnitude-squared response is nominator of is hence a polynomial in even powers of

H 76P8 

76P8 

76P8

. The de. Hence the

14

task of approximating the ideal magnitude-squared characteristic is that of choosing a suitable denominator polynomial in , i.e. selecting the function in the following expression:
8  76P8  ) r h ) B B G B  p   p

where nominal cut-off frequency and rational function of The choice of is determined by functions such that to unity for and rises rapidly after that.
848VU ) r h 8 b

8eU

B G 8VU 

. is close

2.4 Butterworth Filters

8 h 8VU  p  h 8VU 8  p q  8VU 8  q

i.e.
i698  ) r B B G )  q

where is the order of the lter. Figure 2.4 shows the response on linear (a) and log (b) scales for various orders .

2.4.1 Butterworth lter notes

1.
 t 

for

8VU

(i.e. magnitude response is 3dB down at cut-off fre-

quency) 2. For large :

in the region in the region

88VU

i698

 )

88VU

, the steepness of

is a direct function of .

3. Response is known as maximally at, because

E q EP8 d q B9 x

for

 ! )

vvv

09e

R )

15

(a)
1 0.8

|G()|

0.6 0.4 0.2 0

n=1 n=2

n=6

0.2

0.4

0.6

0.8

(b)

1.2

1.4

1.6

1.8

10

log |G()|
10
1 2

10

10

log

10

Figure 2.4: Butterworth lter response on (a) linear and (b) log scales. On a log-log scale the response, for falls off at approx -20db/decade.
84f8eU

Proof Express
i698 7698

in a binomial expansion:
8 8  q p 5hi g  R ) 0 ) 8eU  q rkjl 8VU 8 m q R n 8eU )o 8 p q rwvvv 8VU

 h

It is then easy to show that the rst origin.

R )

derivatives are all zero at the

2.4.2 Transfer function of Butterworth low-pass lter

i698 rq 7698(R6P8

Since is derived from operation can also be done, i.e.

i698



8us

R6v

using the substitution

ts

698

, the reverse

16

(Rw

 x ) r

) R6y B G  q

or Thus the poles of

CR

 ) r

) 5 A y BG  q

) ) r 5 A y BG  q

belong either to
R6v 8VU

or
R )

CR

. The poles are given by:

t~ ! d ! ) ! 0 ! ! 0 R )

 q

@#A{z |#}

vvv

Thus
R6v  8VU @#A#z |} t~ i

Since

 @

, then we have the nal result:

1 1 0 r 0 r ) 0  8VUv{6S

i.e. the poles have the same modulus and equi-spaced arguments. For example, for a fth-order Butterworth low-pass lter (LPF), :
 n 1  0 ) 0 l s r 1 0 r ) 0 1 P r ) l9 ! n ! P ! 0 ) o ! vvv

8VU

i.e. the poles are at:

) l ! ) P ! ) l ! 0 )o n ! 0 0 ! 0 lPl ! 0 j j ! o j ! o j ! 0

in L.H. s-plane therefore stable

in R.H.S. s-plane therefore unstable 17

0.5

0.5

1.5

0.5

0.5

Figure 2.5: Stable poles of 5th order Butterworth lter. We want to design a stable lter. Since each unstable pole is we can let the stable ones be in , and the unstable ones in
CR )  CR

a stable pole,
v !(8VU@ A 

Therefore the poles of are as shown in Figure 2.5. Hence,

 R y " ) )  ) r y BG h ) r 0



8VUC@

A t xa

!(8VU@ A

t mm

!(8VUC@ A

t xa

!(8VU@

A  t p

0 WS

) BG r y BG  p h ) r 0 y W j o

y BG r

BG

or, multiplying out:

  ) r j v 0 j o) B G y r n v 0 j o) B G y  r n v ) 0 j o) B G y C r j v 0 j o) B G y m r y B G

Note that the coefcients are palindromic (read the same in reverse order) this is true for all Butterworth lters. Poles are always on same radii, at angular spacing, with half-angles at each end. If is odd, one pole is real.
 q ) 0 ) 0 v 0 v v n j v j l n v ) n v 0 l ) j o j o l j j v ) v $ t 9PP $  $ $ m $ $ p $ $g

) ) 9PP 0

) v v ) j n ) v v v v

PPP PPP ) 0 0 j o P l 0 ) j ) ) v o) ) ) n v ) ) 0 o n 0 v o v ln v ) 0 v v 9PP PPP v ) v j l ) v n v 0 j o l lPl ) 0 ) v o) ) j ) v l o ) v o P j 0 ) j v ) j v l l ) v v P9P PPP ) n ) v ) n v PPP 0 l ) v PP9

o) 0 j

o) 0 j j

) o) )o

o)

18

Butterworth LPF coefcients for

$

2.4.3 Design of Butterworth LPFs

The only design variable for Butterworth LPFs is the order of the lter . If a lter is to have an attenuation at a frequency :
8e   B  ) r )  ) B B G  q

i.e.

yP 0

 yP

R ) B BG

or since usually Butterworth design example

! d )

yP yP

B B G

Design a Butterworth LPF with at least 40 dB attenuation at a frequency of 1kHz and 3dB attenuation at = 500Hz. Answer
U

40dB

= 100;

8V

 0

PP 1

and
yP y9 t x

8VU

 )

PP 1

rads/sec
0  v o ) j o

9 )  0 t x v

Therefore Hence
0 

= 7 meets the specication into the transfer function from the above table for

Check: Substitute

 6

i6 0

 l

l v j R

) v j 6 n

which gives

 0 )

. 19

2.5 Equiripple Filters

You may have noticed that the Butterworth response is monotone i.e. it has no ripple. If a certain amount of ripple is allowed in the pass-band and/or the stopband, lters which have a sharper cut-off than the Butterworth lter (for a given order ) can be designed. There are three types of equi-ripple lters:

Chebyshev Type I (equi-ripple in the pass-band)

Chevyshev Type II (also known as inverse Chebyshev equi-ripple in the stop-band)

Elliptic (equi-ripple on both pass-band and stop-band)

2.5.1 Chebyshev polynomials

The -order Chebyshev polynomial
' dSq )  y W ' dSq

can be expressed as:

y W 5ut

for

; and
F dq  y W y 5ut W(

. These two expressions are equivalent in that the latter can be derived from the former and so the result is a single Chebyshev polynomial which applies for . Alternatively, can be expressed as a polynomial in , which can be evaluated for any :
) w F dSq F d t  y W F y W 5ut 

for

' d 

y W

where

y W

5ut

i.e.
d

' 

 0

 W

R )

 0

 R

20

Obviously, the same results are also valid with the Generally,
F dq } t r F dSq 5ut  y W r ) r y W R )

function.
0 y W y W  0 dSq F

Thus:
F dq } t  0 dq FeR dSq 5ut ' dSq } t '

is a recurrence relationship which allows the computation of two previous polynomials. Using the recurrence relationship,
d F  0 d  d FR t F  0 V 0  R ) eR 

from the

F d m

 0

'hR d 

'

 0

eR 0 j

 R

 )

m R

 r

The graphs of Fig. 2.6 below show that, for large , diverges very rapidly for ; for example, . This is just the kind of behaviour we need for a good lter function. For a Chebyshev Type I lter (i.e. equi-ripple in the pass-band):
hv ) d m ) v  dq 0 n v j n 8 h 8VU d q  p   8VU 8

X v

i698  

) ) r  ds  B B G

Chebyshev type I notes sets the ripple amplitude in the ripple pass1. The parameter band which is dened as . Since, for will uctuate between and .
) 8 8VU 8 8eU!  !9i698 d q ) )

e.g. with , the amplitude will vary between and or 1dB ripple since .
n 0 yP l ) t x v  R ) v

lPl

} i t

t t  p

, ie.

2. At

d !

 d q

'

 h

5ut

{p 

21

T (x)
1

T (x)
2

8 6 4

0 2 1 0 2 2

2 2

x T (x)
3

x T (x)
4

30 20 10 0 10 20 30 2 1 0 1 2

100 80 60 40 20 0 20 2 1 0 1 2

Figure 2.6: Chebyshev functions. ie.

d q  ' 

if

is odd and 1 if

is even

3.

8VU

, in this case, is not the

j 8 (

Ev

cut-off frequency.

By denition,

is given by:

) ) r  d q  Ba BG

) 0

ie.

 d q

B BG

 )

which means that

f8eU

, since

4. For gets large and decreases monotonically, as for the Butterworth LPF, but much more rapidly.
d q

8d8VU!

i698

22

(a)
1 0.8

|G()|

0.6 0.4 0.2 0 n=4

n=1

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

Figure 2.7: Response of Chebyshev type I lter. The lower limit of ripple in the pass-band is . Here .
b ) ) r  w v lPl n

2.5.2 Transfer function of Chebyshev Type I LPF

As before, put
i698  q i698(R698

, and let
)

8Ts

R6v

(Rw

 ) r 

 ds 0

A y BG

The poles are given by to . Example

  j  t 

5 dSq

A y B G

k 6

5ut

. Of the

roots, assign the stable ones

R6v d 8VU 

R6v R j 8VU

R6v 8VU  6

8VU r 6 H j

d 6 8VU 5ut d 6 0

i.e. 2 cubics

r j a D yV G } ig 5  g { A gi g7  CR t m 6 8eU eR 0 w

r j 8VU r 0 

8eU

and

8VU

} g " A gi g7

D ye G

ig

Select as stable poles

 

 R

t 8eU#!# 

t t p

8VU

Therefore

)  y R ) t

) h ) r 0 B G y p h ) r  B G t y r y B G  p

(factorised form)

23

(Note

 t m

for complex pair it was

   ) r 0 t  y BG r 0 )

for Butterworth lter)

y BG  r 0 y BG

Hence

(polynomial form)
n

In general, the poles lie on an ellipse (this is only meaningful for ; if , an ellipse can be drawn through any set of poles). A large means a larger ripple, narrower ellipse and less damped complex poles, ie. a more oscillatory impulse or step response but a steeper cut-off.

0.3 0.2 0.1 0 0.1 0.2 0.3 0.4 0.6 0.8 1 1.2

Figure 2.8: Stable poles of 5th order Chebyshev Type I lter.

2.5.3 Design of Chebyshev Type I LPFs

If a lter is to have an attenuation
 B  ) r 

at a frequency
)   d q ) 

8V

B B

8V

8V w y  W( y 5ut W 8VU

ie.

 R

w ) d q

8eU

24

8e 8VU 

R )

i.e.

5ut W(

or

5ut W

5ut

5ut W

B B G

Example Design a Chebyshev Type I LPF to meet the following specications:

Maximum ripple of 1dB in pass-band from 0 to 50Hz

3dB cut-off frequency at

= 65 Hz

Attenuation of at least 40 dB for

250 Hz

Answer

1. As shown on page 21, 1dB ripple corresponds to a value of 2. 3dB cut-off:

8   ) j  y y 1

lPl

for .

rads/sec. Minimum
5ut W( 5ut W( ) t x x  ) v v j )

given by:

Therefore

meets the 3 dB cut-off requirement.

 P )

3. For 40dB attenuation,

. Minimum
x ~ y 5ut W n v n

given by:
lPl  0 v o o

 "

5ut

t x

5ut

ie.

meets the overall specication.

25

2.5.4 Chebyshev Type II low-pass lters

These exhibit equiripple behaviour in the stop-band but have the same behaviour in the pass-band. This beas a Butterworth lter (ie. maximally at around haviour cannot be achieved with an all-pole lter; Chebyshev Type II lters have a transfer function which includes zeroes on the imaginary axis as well as poles in the left-half -plane.
8 d

76P8  r )

)  i z # G i # z ~ ~

where

= lowest frequency at which stop-band loss attains a specied value.

8  8VU   t i t }

The magnitude-squared response for Chebyshev Type I and II lters in Fig. 2.9. For both these lters, the pass-band edge is at where and the stop-band edge is at where .
8  8   t i

|G( )| 1/(1+ 2)

|G( )| 1/(1+ 2)

1/A

1/A c r Type I

c r Type II

Figure 2.9: Response of Chebyshev type I, II lters. Thus, for

 n

:
h ) r $  y BG r y B  r $ g y BG  p h ) r r $ y B i y BG m m r $ y BG  p r ) $ 698 t y BG 698  t

with zeroes at

d

26

2.5.5 Design of Chebyshev Type II LPFs

Example Design a Chebyshev Type II LPF to meet the following specications:

3dB cut-off frequency at 50 Hz

Attenuation of at least 50dB for

100Hz.

Answer

1. For Chebyshev Type II LPFs, the pass-band behaviour is the same as that of a Butterworth LPF; ie. . Therefore which gives
8VU  8 (  B   G t   t i t } )

2. When
 d q )  ) 8  8  ) r  !  )  ) i z ~ G i z  t ~

Since

for all , the expression reduces to:

 B ) r   d q )  B B G

Therefore, from section 2.5.3,

 y 5ut W y 5ut W B BG   R ) ig  ! )

in this case

For an attenuation of 50dB,

 y

= 10 and thus:
5ut ) 5ut y W( 0 R ) ig  v l

ie.

meets the overall specication.

27

2.5.6 Elliptic low-pass lters

Elliptic lters3 have a magnitude response which is equi-ripple in both the passband and the stop-band. Elliptic lters are optimum in the sense that, for a given order and for given ripple specications, no other lter achieves a faster transition between the pass-band and the stop-band ie. has a narrower transition bandwidth. The magnitude-squared response of a low-pass elliptic lter is of the form:
i698  ) r ) # q B BG !

where is called a Chebyshev rational function and scribing the ripple properties of
q B G B !# q B G v 8

!#

is a parameter de-

Figure 2.10 shows the magnitude-squared response of a typical elliptic lowpass lter. The frequencies represent the edges of the pass- and stop-bands and it is noted that the cut-off frequency is given as . The latter, once again, is not the 3dB point though.
y 8VU  8 8 y

2.5.7 Design of Elliptic Filters

In order to design an elliptic LPF with arbitrary attenuations in both pass-band and stop-band, three of the four parameters:

lter order

in-band loss or ripple

out-of-band loss or attenuation

transition ratio

B B

C )

can be chosen and the fourth parameter is uniquely dened. The theory behind the determination of the function involves an understanding of Jacobian elliptic functions, which are beyond the scope of this
q B G B !#

also known as Cauer, or Darlington, lters

28

|G( )| 1/(1+ 2)

1/A

p s
Figure 2.10: Elliptic lter response.

course. Hence the design of elliptic lters will not be considered in detail in this lecture course. In any case, elliptic lters are usually designed with the help of graphical procedures (see, for example, Rabiner & Gold, Theory and Application of Digital Signal Processing) or computer programmes (see, for example, Daniels, Approximation methods for the design of passive, active and digital lters). You should remember, however, that for any given specication, the elliptic lter will always be of lower order than any other.

29