0% found this document useful (0 votes)
82 views18 pages

Section 5: Z-Transforms' & IIR-type Digital Filters

This document discusses digital filters and their properties. Digital filters can be represented by difference equations, which can be used to find their impulse response and frequency response. The z-transform provides a mapping between the impulse response of a digital filter and its transfer function. Poles and zeros of a filter's transfer function correspond to roots of the numerator and denominator polynomials.

Uploaded by

Nilam Niki
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
82 views18 pages

Section 5: Z-Transforms' & IIR-type Digital Filters

This document discusses digital filters and their properties. Digital filters can be represented by difference equations, which can be used to find their impulse response and frequency response. The z-transform provides a mapping between the impulse response of a digital filter and its transfer function. Poles and zeros of a filter's transfer function correspond to roots of the numerator and denominator polynomials.

Uploaded by

Nilam Niki
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 18

Comp30291 : 17 Nov09 20/05/2014 BMGC

University of Man!ester
"!oo# of Comp$ter "iene
Comp30291: %i&ita# Me'ia (roessin& 2009)10
"etion 5: *+)transforms , --.)type 'i&ita# fi#ters
5/1/ -ntro'$tion:
A general causal digital filter has the difference equation:
N M
y[n] = a
i
x[n-i] - b
k
y[n-k]
i=0 k=1
hich is of order !ax" N#M $# and is recursi%e if any of the b
&
coefficients are non-'ero( A second
order recursi%e digital filter therefore has the difference equation:
y[n] = a
0
x[n] ) a
1
x[n-1] ) a
*
x[n-*] - b
1
y[n-1] - b
*
y[n-*]
A digital filter ith a recursi%e linear difference equation can ha%e an infinite i!+ulse-res+onse(
,e!e!ber that the frequency-res+onse of a digital filter ith i!+ulse-res+onse "h[n]$ is:

-.e
&
/ = h[n]e
- & n
n=-
5/2/ 0!e +)transform:
0onsider the res+onse of a causal stable 123 digital filter to the s+ecial sequence "'
n
$ here ' is a
co!+lex( 3f "h[n]$ is the i!+ulse-res+onse# by discrete ti!e con%olution# the out+ut is a sequence
"y[n]$ here

y[n] = h[k] '
n - k
= '
n
h[k] '
- k
k=- k=-

= '
n
-.'/ ith -.'/ = h[k] '
- k
k=-
2he ex+ression obtained for -.'/ is the 4'-transfor!5 of the i!+ulse-res+onse( -.'/ is a co!+lex
nu!ber hen e%aluated for a gi%en co!+lex %alue of '(
3t !ay be shon that for a causal stable syste!# -.'/ !ust be finite hen e%aluated for a co!+lex
nu!ber ' ith !odulus greater than or equal to one(


=

= =
n
n j i
n
n
z n h e H z n h z H ] [ / . : res+onse - frequency the and ] [ / . 6ince
it is clear that re+lacing ' by e
&
in -.'/ gi%es -.e
&
/ (
5/3/ 0!e *+)p#ane:
0o!+70*81: 9igital Media :rocessing ;( * *< =ct 50> ?M@0
3t is useful to re+resent co!+lex nu!bers on an 4Argand diagra!5 as illustrated belo( 2he !ain
reason for doing this is that the !odulus of the difference beteen to co!+lex nu!bers a)&b and
c)&d say i(e( 1 .a)&b/ - .c)&d/ 1is re+resented gra+hically as the length of the line beteen the to
co!+lex nu!bers as +lotted on the Argand diagra!(
3f one of these co!+lex nu!bers# c )&d say is 'ero i(e( 0)&0# then the !odulus of the other nu!ber A
a)&bA is the distance of a)&b fro! the origin 0)&0 on the Argand diagra!(
=f course# any co!+lex nu!ber# a)&b say# can be con%erted to +olar for! ,e
&
here ,= Aa)&bA and
= tan
-1
.bBa/( :lotting a co!+lex nu!ber ex+ressed as ,e
&
on an Argand diagra! is also illustrated
abo%e( Ce dra an arro of length , starting fro! the origin and set at an angle fro! the 4real
+art5 axis .!easured anti-clockise/( ,e
&
is then at the ti+ of the arro( 3n the illustration abo%e#
is about BD or D; degrees( 3f ,=1# ,e
&
= e
&
and on the Argand diagra! ould be a +oint at a
distance 1 fro! the origin( :lotting e
&
for %alues in the range 0 to * .7<0
=/
+roduces +oints all of
hich lie on a 4unit circle5 # i(e( a circle of radius 1# ith centre at the origin(
Chere the co!+lex nu!bers +lotted on an Argand diagra! are %alues of ' for hich e are
interested in -.'/# the diagra! is referred to as 4the '-+lane5( :oints ith ' = e
&
lie on a unit circle#
as shon in Eig ;(1( ,e!e!ber that Ae
&
A = Acos./ )&sin./A = [cos
*
./ ) sin
*
./] = 1(
2herefore e%aluating the frequency-res+onse -.e
i
/ for in the range 0 to is equi%alent to
e%aluating -.'/ for ' =e
&
hich goes round the u++er +art of the unit circle as goes fro! 0 to (

Real part of z
Imaginary part of z
Fig 5.1
+ 2 e3p45 6
1

5/4/ .e#atin& 74+68 si&na#)f#o9 &rap!s an' 'ifferene e:$ations


2his is sur+risingly straightforard( 0onsider non-recursi%e and recursi%e difference equations
se+arately(
;3amp#e 5/1: Eind -.'/ for the difference equation: y[n] = x[n] ) x[n-1]
,eal +art
3!aginary +art
-7)7&
1-*&
,e
&
,

0o!+70*81: 9igital Media :rocessing ;( 7 *< =ct 50> ?M@0


"o#$tion: 2he i!+ulse res+onse is: "h[n]$ = " ((( # 0# 1# 1# 0# ((( $

1
1
0
1 ] [ ] [ / .

=

+ = = =

z z n h z n h z H
n
n
n
n
;3amp#e 5/2:
Eind -.'/ for the recursi%e difference equation: y[n] = a
0
x[n] ) a
1
x[n-1] - b
1
y[n-1]
"o#$tion:
2he !ethod used in Fxa!+le ;(1 is not so easy because the i!+ulse-res+onse can no be infinite(
Eortunately there is another ay( ,e!e!ber that if x[n] = '
n
then y[n] = -.'/ '
n
# y[n-1] = -.'/
'
n - 1
etc( 6ubstitute into the difference equation to obtain:
-.'/ '
n
= a
0
'
n
) a
1
'
n - 1
- b
1
-.'/ '
n - 1

a
0
) a
1
'
- 1
2herefore# -.'/ =
1 ) b
1
'
- 1
exce+t hen ' = - b
1
( Chen ' = -b
1
# -.'/ = (
?y the sa!e !ethod# -.'/ for a general digital filter hose difference-equation as gi%en earlier
is:
a
0
) a
1
'
- 1
) a
*
'
- *
) ((( ) a
N
'
- N
-.'/ = .ith b
0
= 1/
b
0
) b
1
'
- 1
) b
*
'
- *
) ((( ) b
M
'
- M

@i%en -.'/ in this for!# e can easily go back to its difference-equation and hence its signal-
flo gra+h# as illustrated by the folloing exa!+le(
;3amp#e 5/3: @i%e a signal flo gra+h for the second order digital filter ith:
a
0
) a
1
'
-1
) a
*
'
- *
-.'/ =
1 ) b
1
'
- 1
) b
*
'
- *
"o#$tion: 2he difference-equation is:
y[n] = a
0
x[n] ) a
1
x[n-1] ) a
*
x[n-*] - b
1
y[n-1] - b
*
y[n-*]
2he signal-flo gra+h in Eig ;(* is readily deduced fro! this difference-equation( 3t is referred to as
a second order or 4bi-quadratic5 33, section in 4direct for! 15(
0o!+70*81: 9igital Media :rocessing ;( D *< =ct 50> ?M@0
x[n]
a0
'-1
'-1
'-1
'-1
a1
a*
y[n]
-b1
-b*
Eig ;(*: G 9irect Eor! 3 H ?iquadratic section
Alternati%e signal flo gra+hs can generally be found for a gi%en difference-equation( 0onsidering
again the 49irect Eor! 35 bi-quadratic section in Eig ;(*# re-ordering the to hal%es in this signal
flo gra+h gi%es Eig ;(7 hich# by :roble! ;(8# ill ha%e the sa!e i!+ulse-res+onse as the signal-
flo gra+h in fig ;(*( No obser%e that Eig ;(7 !ay be si!+lified to the signal-flo gra+h in Eig
;(D hich is knon as a 49irect Eor! 335 i!+le!entation of a bi-quadratic section( 3t has the
!ini!u! +ossible nu!ber of delay boxes and is said to be 4canonical5( 3ts syste! function is
identical to that of the 49irect Eor! 35 signal-flo gra+h# and therefore it can i!+le!ent any second
order bi-quadratic syste! function(
x[n] a0
'-1
'-1
a1
a*
y[n]
'-1
'-1
-b1
-b*
Eig( ;(7: G 9irect Eor! 3 H rearranged
x[n] y[n]
-b1
-b*
a0
'-1
'-1
a1
a*
C
C1
C*
Eig ;(D: G 9irect Eor! 33 H ?iquadratic 6ection
;3amp#e 5/4:
0o!+70*81: 9igital Media :rocessing ;( ; *< =ct 50> ?M@0
@i%en %alues for a
1
# a
*
#a
0
# b
1
and b
*
# rite a +rogra! to i!+le!ent 9irect Eor! 33(
"o#$tion:
C1 = 0I C* = 0I "Assign %ars C1#C* to delay out+uts$
hile 1 "3nfinite JhileJ loo+$
K=in+ut.JK = J/I "Assign K to recei%e a single in+ut sa!+le$
C = K-b1LC1-b*LC*I ",ecursi%e +art of filter$
M =CLa0)C1La1)C*La*I "Non-recursi%e +art$
C* =C1I C1 =CI "6et u+ delay out+uts for next ti!e$
dis+.[J M=J nu!*str.M/]/I "Assign %ar M to out+ut$I
endI "@o back for next sa!+le$
5/5/ "ystem f$ntion: 2he ex+ression obtained for -.'/ is a ratio of +olyno!ials in '
1
( -.'/ is
the 4syste! function5( Chen ' N 1# -.'/ need not be finite(
5/</ (o#es an' +eros of 74+6:
2he ex+ression abo%e for -.'/ for a general digital filter !ay be re-ex+ressed as:
.a
0
'
N
) a
1
'
N -1
) ((( ) a
N
/
-.'/ = '
M-N

.'
M
) b
1
'
N-1
) ((( ) b
M
/
2he deno!inator and nu!erator +olyno!ials !ay no be ex+ressed in factorised for! to obtain:
.' - '
1
/.' - '
*
/.' - '
7
/(((. ' - '
N
/
-.'/ = a
0
'
M-N

.' - +
1
/. ' - +
*
/.' - +
7
/(((.' - +
M
/
2he roots of the nu!erator: '
1
# '
*
#(((# '
N
# are called the 4'eros5 of -.'/(
2he roots of the deno!inator: +
1
#+
*
#(((# +
M
# are called the 4+oles5 of -.'/
-.'/ ill be infinite hen e%aluated ith ' equal to a +ole# and ill beco!e 'ero ith ' equal to a
'ero exce+t in the s+ecial case here the 'ero coincides exactly ith one of the +oles(
Eor a causal stable syste!# -.'/ !ust be finite for ' 1( 2herefore there cannot be a +ole
hose !odulus is greater than or equal to 1( All +oles !ust satisfy ' N 1# and hen +lotted on
the Argand diagra!# this !eans that they !ust lie inside the unit circle( 2here is no restriction on
the +ositions of 'eros(
5/7/ %istane r$#e: =!itted in *008-10
5/=/ ;stimation of &ain response for ;3amp#e 5/5 from po#es an' +eros: =!itted in *008-10
;3amp#e 5/<: =!itted in *001-10
5/9/ %esi&n of a not! fi#ter >y M?0@?B: Modified in *008-10
Assu!e e ish to design a Dth order JnotchJ digital filter to eli!inate an unanted sinusoid at
O00 -' ithout se%erely affecting rest of signal( 2he sa!+ling rate is E6 = 10 k-'(
=ne si!+le ay is to use the MA21A? function 4butter5 as follos:
A"210000B
A@ 2 =00 C 25 B AU 2 =00D25B
Ea >F 2 >$tter428 EA@ AUF/4A"/268stop6B
0o!+70*81: 9igital Media :rocessing ;( < *< =ct 50> ?M@0
a = [0(8O -7(D7 D(8< -7(D7 0(8O]
b= [ 1 -7(D> D(8< -7(78 0(8<]
fre:+4a8 >6B
fre:+4a8 >8 5128 A"6B G Better &rap!
a3is4E0 A"/2 )50 5F6B G "a#es a3es
2he frequency-res+onses .gain and +hase/ +roduced by the final to MA21A? state!ents are as
follos:
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-400
-300
-200
-100
0
Frequency (Hz)
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-40
-20
0
Frequency (Hz)
M
a
g
n
i
t
u
d
e

(
d
B
)
6ince the ?utterorth band-sto+ filter ill ha%e -7d? gain at the to cut-off frequencies
E1 = O00-*; and EP=O00)*;# the notch has 4-7 d? frequency bandidth5: *; ) *; = ;0 -'(
No consider ho to i!+le!ent the D
th
order digital filter( 2he MA21A? function ga%e us:
a = [0(8O -7(D7 D(8< -7(D7 0(8O]
b= [ 1 -7(D> D(8< -7(78 0(8<]
2he transfer .6yste!/ Eunction is# therefore:
A 49irect Eor! 335 i!+le!entation of the D
th
order notch filter ould ha%e the signal-flo gra+h
belo:
( )

+ +
+ +
=


D 7 * 1
D 7 * 1
8< ( 0 78 ( 7 8< ( D D> ( 7 1
8O ( 0 D7 ( 7 8< ( D D7 ( 7 8O ( 0
z z z z
z z z z
z H
0o!+70*81: 9igital Media :rocessing ;( > *< =ct 50> ?M@0
2his i!+le!entation orks fine in MA21A?( ?ut 4direct for!5 33, i!+le!entations of order
greater than to are rarely used( 6ensiti%ity to round-off error in coefficient %alues ill be high(
Also the range of 4inter!ediate5 signals in the '
-1
boxes ill be high(
-igh ord-length floating +oint arith!etic hides this +roble!# but in fixed +oint arith!etic# great
difficulty occurs( 3nstead e use 4cascaded bi-quad sections5

@i%en a D
th
order transfer function -.'/( 3nstead of the direct for! reali'ation belo:
e +refer to arrange to bi-quad sections# ith a single leading !ulti+lier @# as follos:
2o con%ert the D
th
order transfer function -.'/ to this ne for! is definitely a &ob for MA21A?(
9o it as follos after getting a Q b for the D
th
order transfer function# -.'/# as before:
Ea >F 2 >$tter428 EA@ AUF/4A"/268stop6B
E"H" GF 2 tf2sos4a8>6
+
)1
+
)1
+
)1
+
)1
D
D
D
D
D
D
D
D
0/9=
7(D>
-D(8<
7(78
-0(8<
0/0/9=
x[n]
y[n]
)3/43
4/9<
)3/43
-.'/
x[n] y[n]
-
1
.'/ -
*
.'/
x[n] y[n]
@
0o!+70*81: 9igital Media :rocessing ;( O *< =ct 50> ?M@0
R MA21A? res+onds ith:
"H" 2 1 )1/753 1 1 )1/722 0/977<
1 )1/753 1 1 )1/744 0/97=5
G 2 0/97=
3n MA21A?# 46=65 stands for 4second order section5 .i(e( bi-quad/ and the function 4tf*6=65
con%erts the coefficients in arrays 4a5 and 4b5 to the ne set of coefficients stored in array 46=65 and
the constant @( 2he array 6=6 has to ros: one ro for the first bi-quad section and one ro for
the second bi-quad section( 3n each ro# the first three ter!s s+ecify the non-recursi%e +art and the
second three ter!s s+ecify the recursi%e +art( 2herefore -.'/ !ay no be reali'ed as follos:
2his is no a +ractical and reali'able 33, digital 4notch5 filter# though e so!eti!es i!+le!ent the
single !ulti+lier @ =0(81O by to !ulti+liers# one for each bi-quad section( More about this later(
5/10 Ca#$#ation of &ain)response of not! fi#ter:
-o good is a notch filterS Ce can start to anser this question by s+ecifying the filterJs 7d?
bandidth i(e( the difference beteen the frequencies here the gain crosses 0(>0> .-7d? /( Ce
should also ask hat is the gain at the notch frequency .O00 -' in +re%ious exa!+le/I i(e( hat is the
4de+th5 of the notch( 3f it is not dee+ enough either .i/ increase the -7 d? bandidth or .ii/ increase
the order( 9o both if necessary( 2o 4shar+en5 the notch# decrease the -7d? bandidth# but this ill
!ake the notch less dee+I so it !ay be necessary to increase the order to !aintain a dee+ enough
notch( 2his is an 4ad-hoc5 a++roach T e can surely de%elo+ so!e theory later( 3t !odifies the
!ore for!al a++roach# based on +oles and 'eroes# ado+ted last year(
3EnF
0/97=
)1/753

1/722
)0/97=
yEnF
)1/753
1/744
)0/979
Ao$rt! or'er --. not! fi#ter rea#ise' as t9o >i:$a' 4"H"6 setions
0o!+70*81: 9igital Media :rocessing ;( 8 *< =ct 50> ?M@0
(
;3amp#e: A digital filter ith a sa!+ling rate of *00 -' is required to eli!inate an unanted
;0 -' sinusoidal co!+onent of an in+ut signal ithout affecting the !agnitudes of other
co!+onents too se%erely( 9esign a Dth order UnotchU filter for this +ur+ose hose 7d? bandidth
is not greater than 7(* -'( .MA21A? calls this *
nd
order(/ -o dee+ is the notchS
"o#$tion:
A"2200B A@250)1/<B AU250D1/<B
Ea >F2>$tter428EA@8AUF/4A"/268 *stop6B
E"H" GF 2 tf2sos4a8>6
5/11/ --. 'i&ita# fi#ter 'esi&n >y >i#inear transformation
Many design techniques for 33, discrete ti!e filters ha%e ado+ted ideas and ter!inology de%elo+ed
for analogue filters# and are i!+le!ented by transfor!ing the syste! function# -
a
.s/# of an analogue
4+rototy+e5 filter into the syste! function -.'/ of a digital filter ith si!ilar# but not identical#
characteristics(
Eor analogue filters# there is a ide %ariety of techniques for deri%ing -
a
.s/ to ha%e a s+ecified ty+e
of gain-res+onse( Eor exa!+le# it is +ossible to deri%ing -
a
.s/ for an n
th
order analogue ?utterorth
lo-+ass filter# ith gain res+onse:

3t is then +ossible to transfor! -
a
.s/ into -.'/ for an equi%alent digital filter( 2here are !any ays
of doing this# the !ost fa!ous being the 4bilinear transfor!ation5( 3t is not the only +ossible
transfor!ation# but a %ery useful and reliable one(
2he bilinear transfor!ation in%ol%es re+lacing s by .*B2/ .'-1/B.')1/]# but fortunately# MA21A?
takes care of all the detail and e can design a ?utterorth lo +ass filter si!+ly by executing the
MA21A? state!ent:
[a b] = butter.N# fc/
N is the required order and fc is the required 47 d?5 cut-off frequency nor!alised .as usual ith
n
C
a
G
*
/ B . 1
1
/ .

+
=
1
5I7+
f
Gain
0/7
0
=00
=00 D =00)
0 'B
)3 'B
0o!+70*81: 9igital Media :rocessing ;( 10 *< =ct 50> ?M@0
MA21A?/ to f
S
B*( Analogue ?utterorth filters ha%e a gain hich is 'ero in the +ass-band and falls
to -7 d? at the cut-off frequency( 2hese to +ro+erties are +reser%ed by the bilinear transfor!ation#
though the traditional ?utterorth sha+e is changed( 2he sha+e change is caused by a +rocess
referred to as 4frequency ar+ing5( Although the gain-res+onse of the digital filter is consequently
rather different fro! that of the analogue ?utterorth gain res+onse it is deri%ed fro!# the ter!
4?utterorth filter5 is still a++lied to the digital filter( 2he order of -.'/ is equal to the order of
-
a
.s/
Are:$eny 9arpin&:
3t !ay be shon that the ne gain-res+onse @./ = @
a
./ here = * tan.B*// 2he gra+h of
against belo# shos ho in the range - to is !a++ed to in the range - to ( 2he
!a++ing is reasonably linear for in the range -* to * .gi%ing in the range -B* to B*/# but as
increases beyond this range# a gi%en increase in +roduces s!aller and s!aller increases in ( 2he
effect of frequency ar+ing is ell illustrated by considering the analogue gain-res+onse shon in
fig ;(1>.a/( 3f this ere transfor!ed to the digital filter gain res+onse shon in fig ;(1>.b/# the
latter ould beco!e !ore and !ore co!+ressed as (
Fig 6.1: Frequency warping
-3.14
-2.355
-1.57
-0.785
0
0.785
1.57
2.355
3.14
-12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12
Radians/second
R
a
d
i
a
n
s
/
s
a
m
p
l
e

17a45 61 174e3p45 61

/2
Ai& </24a6: ?na#o&$e &ain response Ai& </24>6: ;ffet of >i#inear transformation
*(rototype ana#o&$e transfer f$ntion: Although the sha+e changes# e ould like @./ at its
cut off
0
to the sa!e as @
a
./ at its cut-off frequency( 3f @
a
./ is ?utterorth# it is -7d? at its
cut-off frequency( 6o e ould like @./ to be -7 d? at its cut-off
0
(
Achie%ed if the analogue +rototy+e is designed to ha%e its cut-off frequency at
0
= * tan.
0
B*/(

0
is then called the 4+re-ar+ed5 cut-off frequency(
Ai& 5/1< Are:$eny Jarpin&
Ai&/ 5/174a6: ?na#o&$e Gain .esponse Ai&/ 5/174>6: ;ffet of Bi#inear 0ransformation
0o!+70*81: 9igital Media :rocessing ;( 11 *< =ct 50> ?M@0
9esigning the analogue +rototy+e ith cut-off frequency * tan.
0
B*/ guarantees that the digital
filter ill ha%e its cut-off at
0
(
%esi&n of a 2n' or'er --. #o9)pass 'i&ita# fi#ter >y t!e >i#inear transform met!o' 4*>y !an'6
1et the required cut-off frequency
0
= BD radiansBsa!+le( Ce need a +rototy+e transfer function
-
a
.s/ for a *nd order analogue ?utterorth lo-+ass filter ith 7 d? cut-off at
0
= *tan.
0
B*/ =
* tan.BO/ radiansBsecond( 2herefore#
0
= * tan.BO/ = 0(O*O( 3t is ell knon by analogue filter
designers that the transfer function for a *nd order ?utterorth lo-+ass filter ith cut-off
frequency =1 radianBsecond is:
Chen the cut-off frequency is =
0
rather than = 1# the second order ex+ression for -.s/
beco!es:
,e+lacing s by & and taking the !odulus of this ex+ression gi%es @./ = 1B[1).B
0
/
*n
] ith
n=*( 2his is the *
nd
order ?utterorth lo-+ass gain-res+onse a++roxi!ation( 9eri%ing the abo%e
ex+ression for -
a
.s/# and corres+onding ex+ressions for higher orders# is not +art of our syllabus( 3t
ill not be necessary since MA21A? ill take care of it(
6etting
0
= 0(O*O in this for!ula# then re+lacing s by *.'-1/B.')1/ gi%es us -.'/ for the required
33, digital filter( Mou can check this 4by hand5# but fortunately MA21A? does all this for us(
;3amp#e 5/7
Psing MA21A?# design a second order ?utterorth-ty+e 33, lo-+ass filter ith
c
= B D(
"o#$tion:
[a b] = butter.*# 0(*;/
a = [0(08O 0(18< 0(08O]
b = [1 -0(8D 0(77]
2he required ex+ression for -.'/ is

0(08O ) 0(18< '
-1
) 0(08O '
-*

-.'/ =
1 - 0(8D '
-1
) 0(77'
-*

( )

+
+ +
=


* 1
* 1
77 ( 0 8D ( 0 1
* 1
08O ( 0
z z
z z
z H
hich !ay be realised by the signal flo gra+h in fig ;(1O( Note the sa%ing of to !ulti+liers by
using a !ulti+ler to scale the in+ut by 0(08O(
*
/ * . 1
1
/ .
s s
s H
a
+ +
=
*
/ B . / B . * 1
1
/ .
C C
a
s s
s H
+ +
=
0o!+70*81: 9igital Media :rocessing ;( 1* *< =ct 50> ?M@0

3EnF yEnF
Ai&/ </3
0/097
2
0/94
)0/33
5/12: 7i&!er or'er --. 'i&ita# fi#ters:
,ecursi%e filters of order greater than to are highly sensiti%e to quantisation error and o%erflo( 3t
is nor!al# therefore# to design higher order 33, filters as cascades of bi-quadratic sections(
MA21A? does not do this directly as de!onstrated by Fxa!+le ;(O(
;3amp#e 5/=: 9esign a D
th
order ?utterorth-ty+e 33, lo-+ass digital filter is needed ith 7d?
cut-off at one sixteenth of the sa!+ling frequency f
S.
"o#$tion: ,elati%e cut-off frequency is BO( 2he MA21A? co!!and belo +roduces the arrays a
and b ith the nu!erator and deno!inator coefficients for the D
th
order syste! function -.'/(
Ea >F 2 >$tter448 0/1256
=ut+ut +roduced by MA21A? is:
a 2 0/00093 0/0037 0/005< 0/0037 0/00093
> 2 1 )2/97<= 3/4223 )1/7=<1 0/355<
2he syste! function is therefore as follos:
2his corres+onds to the 4D
th
order 4direct for!5 signal flo gra+h shon belo(
Eigure ;(18: A D
th
order 4direct for! 335 realisation .not co!!only used/
-igher order 33, digital filters are generally not i!+le!ented like this( 3nstead# they are
i!+le!ented as cascaded biquad or second order sections .6=6/( Eortunately MA21A? can
transfor! the 4direct for!5 coefficients to second order section .6=6/ coefficients using a 46ignal
:rocessing 2oolbox5 function 4tf*sos5 as follos:
Ea >F 2 >$tter448 0/1256
Esos GF 2 tf2sos4a8>6
Fxecuting these state!ents gi%es the folloing res+onse:
Ai&/ 5/1=
( )

+ +
+ + + +
=


D 7 * 1
D 7 * 1
7;;< ( 0 >O< ( 1 D** ( 7 8>> ( * 1
00087 ( 0 007> ( 00;< ( 007> ( 0 00087 ( 0
z z z z
z z z z
z H
)
)
)
)
)
0(0008
78
*(8>>
-7(D**
1(>8
-0(7;<
0(0008
'
-1
'
-1
'
-1
'
-1
0o!+70*81: 9igital Media :rocessing ;( 17 *< =ct 50> ?M@0
Ea >F 2 >$tter448 0/1256
a 2 E0/0009 0/0037 0/005< 0/0037 0/0009F
> 2 E1 )2/97<= 3/4223 )1/7=<1 0/355< F
Esos GF 2 tf2sos4a8>6
sos 2 E1 2 1 1 )1/3<5 0/47=
1 2 1 1 )1/<12 0/745 F
G 2 0/00093
2his +roduces a *-di!ensional array 4sos5 containing to sets of biquad coefficients and a 4gain5
constant @( A !athe!atically correct syste! function based on this data is as follos:
( )

+ +

+
+ +
=




* 1
* 1
* 1
* 1
>D; ( 0 <1* ( 1 1
* 1
D>O ( 0 7<; ( 1 1
* 1
00087 ( 0
z z
z z
z z
z z
z H
3n +ractice# es+ecially in fixed +oint arith!etic# the effect of @ is often distributed a!ong the to
sections( Noting that 0(077 x 0(0*O 0(00087# and noting also that the to sections can be in
either order# an alternati%e ex+ression for -.'/ is as follos:

( )

+
+ +

+ +
=




* 1
* 1
* 1
* 1
D>O ( 0 7<; ( 1 1
* 1
0*O ( 0
>D; ( 0 <1* ( 1 1
* 1
077 ( 0
z z
z z
z z
z z
z H
2his alternati%e ex+ression for -.'/ !ay be realised in the for! of cascaded bi-quadratic sections as
shon in fig ;(*0(
3EnF
0/033
2
1/<
)0/74
0/02=
yEnF
2
1/3<
)0/4=
Ai&/ </4: Ao$rt! or'er --. B$tter9ort! fi#ter 9it! $t)off fs/1<

-0.1
0.1
0.3
0.5
0.7
0.9
1.1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Radians/second
G
a
i
n
Eig( ;(*0 Eourth order 33, ?utterorth 1: filter ith cut-off fsB1<
Ai&/ 5/214a6 ?na#o&$e 4t! or'er B$tter9ort! @( &ain response
0(>0>=1B* at 47d? +oint5
0o!+70*81: 9igital Media :rocessing ;( 1D *< =ct 50> ?M@0

Fig. 5.21(b): Gain response o !"# order $$R il"er
-0.1
0.1
0.3
0.5
0.7
0.9
1.1
0 0.785 1.57 2.355 3.14
Radians/sample
G
a
i
n
Eig( ;(*1.a/ shos the Dth order ?utterorth lo-+ass gain res+onse:
/ 1 .
1
/ .
O

+
= G
.ith cut-off frequency nor!alised to 1/ as used by MA21A? as a +rototy+e( Eig ;(*1.b/ shos
the gain-res+onse of the deri%ed digital filter hich# like the analogue filter# is 1 at 'ero frequency
and 0(>0> .-7d?/ at the cut-off frequency .BO 0(78 radiansBsa!+le/( Note hoe%er that the
analogue gain a++roaches 0 as hereas the gain of the digital filter beco!es exactly 'ero at
= ( 2he sha+e of the ?utterorth gain res+onse is 4ar+ed5 by the bilinear transfor!ation(
-oe%er# the 7d? +oint occurs exactly at
c
for the digital filter# and the cut-off rate beco!es
shar+er and shar+er as because of the co!+ression as (
5/13: --. 'i&ita# !i&!)pass >an')pass an' >an')stop fi#ter 'esi&n:
2he bilinear transfor!ation !ay be a++lied to analogue syste! functions hich are high-+ass# band-
+ass or band-sto+ to obtain digital filter equi%alents( Eor exa!+le a 4high-+ass5 digital filter !ay be
designed as illustrated belo:
;3amp#e 5/9 9esign a D
th
order high-+ass 33, filter ith cut-off frequency fsB1<(
"o#$tion: Fxecute the folloing MA21A? co!!ands and +roceed as for lo-+ass
[a b] = butter.D#0(1*;#5high5/I
freq'.a#b/I
[sos @] = tf*sos.a#b/
Cide-band band-+ass and band-sto+ filters .f
P
VV *f
1
/ !ay be designed by cascading lo-+ass and
high-+ass sections# but Jnarro bandJ band-+assBsto+ filters .f
P
not VV *f
1
/ ill not be %ery accurate
if this cascading a++roach is used( 2he MA21A? band-+ass a++roach alays orks# i(e( for
narroband and ideband( A +ossible source of confusion is that s+ecifying an order 4*5 +roduces
hat !any +eo+le .including !e# ?arry/ ould call a Dth order 33, digital filter( 2he design +rocess
carried out by 4butter5 in%ol%es the design of a lo-+ass +rototy+e and then a++lying a lo-+ass to
band-+ass transfor!ation hich doubles the co!+lexity( 2he order s+ecified is the order of the
+rototy+e( 6o if e s+ecify *nd order for band-+ass e get a Dth order syste! function hich can be
re-ex+ressed .using tf*sos/ as 2C= biquad sections(
;3amp#e 5/10: 9esign a *nd .Dth/order band+ass filter ith
1
= BD #
u
= B*(
"o#$tion: Fxecute the folloing MA21A? state!ents:
[a b] = butter.*#[0(*; 0(;]/
freq'.a#b/I
0(D
1B*
0o!+70*81: 9igital Media :rocessing ;( 1; *< =ct 50> ?M@0
[sos @] = tf*sos.a#b/
MA21A? out+ut:is:
a = 0(08O 0 -0(18; 0 0(08O
b = 1 -1(*18 1(777 -0(<<> 0(77
sos = 1 * 1 1 -0(1<<; 0(;7DO
1 -* 1 1 -1(0;*D 0(<*7*
@ = 0(08O
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-200
-100
0
100
200
Normalized Frequency (

rad/sample)
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
-40
-30
-20
-10
Normalized Frequency (

rad/sample)
M
a
g
n
i
t
u
d
e

(
d
B
)
;3amp#e 5/11: 9esign a Dth .Oth/order band+ass filter ith
1
= BD #
u
= B*(
"o#$tion: Fxecute the folloing MA21A? state!ents
[a b] = butter.D#[0(*; 0(;]/
freq'.a#b/I axis.[0 1 -D0 0]/I
[sos @] = tf*sos.a#b/
to obtain the MA21A? out+ut:
a = 0(01 0 -0(0D1 0 0(0<1 0 -0(0D1 0 0(01
b = 1 -*(D>* D(708 -D(OO< D(D>> -*(81D 1(;18 -0(; 0(1*
sos =1 * 1 1 -0(7;1 0(D*O
1 -*( 1 1 -0(O7* 0(D8
1 *( 1 1 -0(0D< 0(>*D
1 -* 1 1 -1(*DD 0(>87
@ = 0(01
0o!+70*81: 9igital Media :rocessing ;( 1< *< =ct 50> ?M@0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-800
-600
-400
-200
0
Normalized Frequency (

rad/sample)
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-30
-20
-10
0
Normalized Frequency (

rad/sample)
M
a
g
n
i
t
u
d
e

(
d
B
)
4th (8th) order IIR Band-pass (Fs/8 - Fs/4)
;3amp#e 5/12: 9esign a Dth .Oth/order band-sto+ filter ith
1
= BD #
u
= B*(
"o#$tion: Fxecute the folloing MA21A? state!ents
[a b] = butter.D#[0(*; 0(;]# 4sto+5/
freq'.a#b/I axis.[0 1 -D0 0]/I
[sos @] = tf*sos.a#b/
to obtain the MA21A? out+ut:
a = 0(7D> -1(1D8 *(O1; -D(*7> ;(1 -D(*7> *(O1; -1(1D8 0(7D>
b = 1 -*(D>* D(708 -D(OO< D(D>> -*(81D 1(;18 -0(; 0(1*
sos = 1 -0(O*O 1 1 -0(7;1 0(D*O
1 -0(O*O 1 1 -0(O7* 0(D8
1 -0(O*O 1 1 -0(0D< 0(>*D
1 -0(O*O 1 1 -1(*DD 0(>87
@ = 0(7D>
0o!+70*81: 9igital Media :rocessing ;( 1> *< =ct 50> ?M@0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-800
-600
-400
-200
0
Normalized Frequency (

rad/sample)
P
h
a
s
e

(
d
e
g
r
e
e
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-40
-30
-20
-10
0
Normalized Frequency (

rad/sample)
M
a
g
n
i
t
u
d
e

(
d
B
)
4th (8th) order IIR Band-stop (Fs/8 - Fs/4)
5/14: Comparison of --. an' A-. 'i&ita# fi#ters:
33, ty+e digital filters ha%e the ad%antage of being econo!ical in their use of delays# !ulti+liers and
adders( 2hey ha%e the disad%antage of being sensiti%e to coefficient round-off inaccuracies and the
effects of o%erflo in fixed +oint arith!etic( 2hese effects can lead to instability or serious
distortion( Also# an 33, filter cannot be exactly linear +hase(
E3, filters !ay be realised by non-recursi%e structures hich are si!+ler and !ore con%enient for
+rogra!!ing es+ecially on de%ices s+ecifically designed for digital signal +rocessing( 2hese
structures are alays stable# and because there is no recursion# round-off and o%erflo errors are
easily controlled( A E3, filter can be exactly linear +hase( 2he !ain disad%antage of E3, filters is
that large orders can be required to +erfor! fairly si!+le filtering tasks(
(ro>#ems:
;(1 Eind -.'/ for the folloing difference equations
.a/ y[n] = *x[n] - 7x[n-1] ) <x[n-D]
.b/ y[n] = x[n-1] - y[n-1] - 0(;y[n-*]
;(* 6ho that +assing any sequence "x[n]$ through a syste! ith -.'/ = '
- 1
+roduces
"x[n-1]$ i(e( all sa!+les are delayed by 1 sa!+ling inter%al(
;(7( 0alculate the i!+ulse-res+onse of the digital filter ith
1
-.'/ =
1 - * '
- 1
;(D 9ra the signal-flo gra+h for exa!+le ;(7# and +lot its +oles and 'eros(
;(; 3f discrete ti!e 123 syste!s 1
1
and 1
*
# ith i!+ulse res+onses "h
1
[n] $ and "h
*
[n] $
res+ecti%ely# are serially cascaded as shon belo# calculate the o%erall i!+ulse
0o!+70*81: 9igital Media :rocessing ;( 1O *< =ct 50> ?M@0
res+onse( 6ho that this ill not be affected by interchanging 1
1
Q 1
*
(

@1 @2
3EnF
yEnF
K!1EnFL K!2EnFL
;(<( 9esign a Dth order band-+ass 33, digital filter ith loer Q u++er cut-off
frequencies at 700 -' Q 7D00 -' hen f
6
= O k-'(
;(>( 9esign a Dth order band-+ass 33, digital filter ith loer Q u++er cut-off
frequencies at *000 -' Q 7000 -' hen f
6
= O k-'(
;(O( Chat li!its ho good a notch filter e can i!+le!ent on a fixed +oint 96: +rocessorS
3n theory e can !ake notch shar+er Q shar+er by reducing the -7d? bandidth andBor
increasing the order( Chat li!its us in +ractice(
-o shar+ a notch can e get in 1<-bit fixed +t arith!eticS
;(8( Chat order of E3, filter ould be required to i!+le!ent a BD notch a++roxi!ately as good
as a *nd order 33, BD notch ith 7 d? bandidth 0(* radiansBsa!+leS
;(10( Chat order of E3, lo-+ass filter ould be required to be a++rox as good as the *nd order
33, lo-+ass filter .BD cut-off/ designed in these notesS

You might also like