You are on page 1of 7

Sugata Munshi

Department of Electrical Engineering Department, Jadavpur University, Kolkata.

Overview of Quantization Effects in IIR Filters

G ( z) =
x(t)

ADC

xn

vn = yn

y(t)

DAC

Ideal

1
1 az 1

Ideal

vn-1

[A] Ideal Digital Filter Implementation


(No Quantization)

x(t)

ADC

xn

QBi

xn

vn

QBo

Practical ADC

DAC

Practical DAC

QB
n 1
av

Quantizes 2B+1 bit [B] [B]


n 1 into a B+1 bit
av
number since the adder
uses (B+1) bits

y n

A More Realistic Model.

vn 1

a = QB [ a ]

y ( t )

Sugata Munshi
Department of Electrical Engineering Department, Jadavpur University, Kolkata.

Effect of Co-efficient Quantization in IIR Filters


G ( z) =
=

1
2
+
1
1 1z
1 2 z 1

(1)

1 + 2 ( 1 2 + 2 1 ) z 1
1 ( 1 + 2 ) z 1 + 1 2 z 2

(2)

Poles at : z = 1 , z = 2
Direct Form Structure
G ( z) =

a3 + a4 z 1
,
1 a1 z 1 a2 z 2

xn

a1 = 1 + 2 ; a3 = 1 + 2

a3

a2 = 1 2 ; a4 = ( 1 2 + 2 1 )

yn

a4

a1

a2

yn = a1 yn 1 + a2 yn 2 + a3 xn + a4 xn 1.

It is required to find sensitivity of pole location w.r.t co-efficient values.


1
= sensitivity of pole located at 1 w.r.t change in co-efficient value a1.
a1
Similarly,

1 2 2
,
,
.
a2 a1 a2

1
1
=
a1 1 2

(3a)

2
2
=
a1 2 1

(3b)

1
1
=
a2 1 2

(3c)

2
1
=
a2 2 1

(3d)

Sensitivity depends on proximity of nominal poles i.e. larger will be shift pole locations
close the nominal poles.
2

Sugata Munshi
Department of Electrical Engineering Department, Jadavpur University, Kolkata.

Parallel Realization
y1

xn

1 = 1

+
y2

2 = 2

Coefficients which govern the location of poles are 1 , 2


1
=1
1

....4(a)

1
= 0
2

....4(b)

2
= 0
1

....4(c)

2
=1
2

....4(d)

For this realization, pole locations are either insensitive to coefficient quantization (4b,
4c) or the sensitivity is fixed (independent of actual positions / proximity of poles). This
realization is better therefore better than the direct form.
Cascade Realization
G ( z) =

P1 + P2 z 1
( 1 1z 1 ) ( 1 2 z 1 )

P1 = 1 + 2

P2 = ( 1 2 + 2 1 )

Sugata Munshi
Department of Electrical Engineering Department, Jadavpur University, Kolkata.

P1

an

yn

P2

1 = 1

1
=1
1

(5a)

1
= 0
2

(5b)

2
= 0
1

(5c)

2
=1
2

(5d)

2= 2

Zero-Input Limit Cycle in IIR Filters


an

yn = xn 0.9 yn 1

yn

Product of yn 1 and (- 0.9) is rounded off

to nearest integer.

- 0.9

xn = 10 n , xn = 10

for n = 0, xn = 0

for n 0

Input (n)
Ideal O/P (yn)
Quantized O/P (yn)
0
10
10
1
-9
-9
2
8.1
8
3
-7.29
-7
4
6.561
6
5
-5.8945
-5
6
4.80541
5
7
-4.32469
-5
8
+3.892381
+5
From n = 5, filter starts oscillating sustained oscillation.
Zero o/p limit cycle.
Frequency of oscillation =

1
sampling frequency (in this case)
2

Sugata Munshi
Department of Electrical Engineering Department, Jadavpur University, Kolkata.

Dead Band Effect


yn = xn + 0.9 yn 1
123
product rounded off to nearest integer.
.
xn = 10 n
y o = 10
y1 = 0.9 10 = 9
y 2 = 0.9 9 = 8.1 8
y3 = 0.9 8 = 7.2 7
y 4 = 0.9 7 = 6.3 6

o/p has got locked to steady value


in absence of i/p dead band
effect

y5 = 0.9 6 = 5.4 5
y 6 = 0.9 5 = 4.5 5
y 7 = 0.9 5 = 4.5 5

Measure of Coefficient Sensitivity and Shifting of Poles for Direct Form


Structures
Let us consider the general expression for the Z-transfer function of a IIR (i.e.
recursive) filter as,
M

G ( z) =

i
Bi z

i= 0
N

1 Ai z

(1)

i= 1

As a result of coefficient quantization, the transfer function becomes,


M

G ( z ) =

i
Bi z

i= 0
N

(2)

1 Ai z i
i= 1

{ }

{ }

where Ai and Bi are the quantized coefficients and,

Sugata Munshi
Department of Electrical Engineering Department, Jadavpur University, Kolkata.

Ai = Ai + Ai
Bi = Bi + Bi
where { Ai } and

{ Bi } are coefficient quantization errors.

Let us assume that G(z) has first order poles.


The denominator polynomial is
D ( z ) = 1 Ai z i = ( 1 z j z 1 )
N

i= 1

j= 1

(3)

Poles of G(z) are at z = zj ; j = 1,2,..N.


It is to be noted that quantization of each of the coefficients { Ai } results in shifting
of all the poles
Poles of G ( z ) are at ZK + ZK , K = 1,2,3,..N.
Shift in the location of the Kth pole as a result of quantization of A1 is
ZK1 =
where

ZK
A1
A1

ZK
sensitivity of Kth pole to quantization of coefficient A1.
A1

Similarly,
ZK 2 =

ZK
A2
A2

So, the total shift in the location of the Kth pole as a result of quantization of all the coefficients of D(z) is,
N

ZK =

i= 1

ZK
Ai
Ai

(4)

Now,
D( z)
ZK D ( Z )
=

Z K Z = Z K Am Am Z = Z K

or,

(5)

D( z)
( Z Z K ) ..... ( Z Z N )
( Z Z1 ) ( Z Z 2 )
=
......

ZK
ZK
Z
Z
Z
Z

Sugata Munshi
Department of Electrical Engineering Department, Jadavpur University, Kolkata.

( Z Z1 ) ( Z Z 2 )
( Z ZN ) 1
.....

Z
Z
Z
Z
=
( Z ZK )
excluding
Z
1 N ( Z Zj)
=
Z j= 1
Z
j K

D( z)


= ( Z K Z j ) Z K N

j= 1

Z K Z = ZK
j K

(6)

Again,

N
D( z)

=
1 Ai z i
= Z K m

i= 1

A
m Z= Z
m
Z = ZK

(7)

Combining equations (5) , (6) and (7), we have


ZK
Z N m
= N K
Am ( Z Z )
K
j

(8)

j= 1
j K

( Z K Z j ) Z K N Z K = Z K m

Am
j= 1

j K

Substituting (8) in (4), we have

N
Z KN i

ZK = N
( Ai )
i= 1
j= 1 ( Z K Z j )

j K

(9)

This is the expression for the total shift in the K th pole as a result of quantization of all the
coefficients of the denominator polynomial.

You might also like