0% found this document useful (0 votes)
122 views8 pages

9-2coefficient Quantization Fir

This document discusses the effect of coefficient quantization on finite impulse response (FIR) filters. It describes how quantizing the filter coefficients results in an equivalent system that is the parallel connection of the ideal filter and an error filter. It also derives bounds on the error introduced by the quantization process and provides an example quantization to demonstrate these effects. The key points are that quantizing the coefficients introduces an error signal and degrades the frequency response according to derived bounds, but direct form implementations generally provide acceptable performance for most applications.

Uploaded by

sudeepa
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
122 views8 pages

9-2coefficient Quantization Fir

This document discusses the effect of coefficient quantization on finite impulse response (FIR) filters. It describes how quantizing the filter coefficients results in an equivalent system that is the parallel connection of the ideal filter and an error filter. It also derives bounds on the error introduced by the quantization process and provides an example quantization to demonstrate these effects. The key points are that quantizing the coefficients introduces an error signal and degrades the frequency response according to derived bounds, but direct form implementations generally provide acceptable performance for most applications.

Uploaded by

sudeepa
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

DSP: Effect of Coefficient Quantization on FIR Filters

Digital Signal Processing


Effect of Coefficient Quantization on FIR Filters
D. Richard Brown III

D. Richard Brown III

1/8

DSP: Effect of Coefficient Quantization on FIR Filters

Effect of Coefficient Quantization

General model accounting for finite-precision effects:


x
[n]
x(t)

ideal sampling

y[n]

Q()
Q()

Q()

Q()

Q()

ideal reconstruction

y(t)

z 1

Here, we focus on the effect of coefficient quantization and ignore the


other sources of quantization error.

D. Richard Brown III

2/8

DSP: Effect of Coefficient Quantization on FIR Filters

FIR Filter Coefficient Quantization


For a direct form FIR filter, we have the realization structure
x[n]
x(t)

y[n]

ideal sampling

ideal reconstruction

y(t)

0
h
z 1

Q()

h0

1
h
z 1

Q()

h1

2
h
Q()

h2

n = hn + en .
with h
D. Richard Brown III

3/8

DSP: Effect of Coefficient Quantization on FIR Filters

Coefficient Quantization Example (4 bits)


Suppose we quantize the coefficients {hn } using a 4-bit quantizer
(including a sign bit) and denote the number of fractional bits as q.
q = 0 ( = 1)

q = 1 ( = 12 )

q = 2 ( = 14 )

q = 3 ( = 81 )

hn

n
h

en

n
h

en

n
h

en

n
h

en

0.13

-0.13

-0.13

0.25

-0.12

0.125

-0.005

-0.8

-1

-0.2

-1

-0.2

-0.75

0.05

-0.75

0.05

1.3

-0.3

1.5

-0.2

1.25

-0.05

0.875

-0.425

Remarks:
1. In most implementations, we use the same number of fractional bits
for all of the coefficients.
2. The idea is to minimize the quantization errors by maximizing the
number of fractional bits while avoiding overflow.
3. Problems can occur when you have both very large coefficients and
very small coefficients (large coefficient dynamic range).
D. Richard Brown III

4/8

DSP: Effect of Coefficient Quantization on FIR Filters

FIR Filter Coefficient Quantization: Equivalent System


For a causal FIR filter with N quantized coefficients, we have

H(z)
=

N
1
X
n=0
N
1
X

n z n =
h
hn z n +

n=0

N
1
X
n=0
N
1
X

(hn + en )z n
en z n = H(z) + E(z)

n=0

Hence, the quantized FIR filter H(z)


is equivalent to a parallel connection
of H(z) and E(z):
H(z)

y[n]
y[n] = y[n] + w[n]

x[n]
E(z)

w[n]

Note E(z) is FIR and causal.


D. Richard Brown III

5/8

DSP: Effect of Coefficient Quantization on FIR Filters

FIR Filter Coefficient Quantization: Error Bounds


By definition, for causal FIR E(z) with N coefficients, we have a ROC
|z| > 0 and
N
1
X
j
E(e ) =
en ejn
n=0

With rounding quantization, each en is bounded between /2 and /2.


Hence,


1
1
1
NX
NX
jn NX
N



en e

=
|E(ej )| =
en ejn


2
2
n=0

n=0

n=0

and we can write




N +
N
j
|H(e )|
|H(ej )| |H(ej )| +
.
2
2
D. Richard Brown III

6/8

DSP: Effect of Coefficient Quantization on FIR Filters

FIR Coefficient Quantization Example ( = 28 , N = 51)


20

magnitude response (dB)

20

40

60

80

100
unquantized coefficients
quantized coefficients
upper bound
lower bound

120

140

0.1

0.2

0.3

0.4
0.5
0.6
normalized frequency (times )

D. Richard Brown III

0.7

0.8

0.9

7/8

DSP: Effect of Coefficient Quantization on FIR Filters

Final Remarks on Fixed-Point FIR Filtering

Direct form realizations of fixed-point FIR filters are commonly used


since they tend to provide acceptable performance in most cases.

Symmetry properties of direct form FIR filters are not affected by


quantization. Hence, direct form linear phase filters still have linear
phase after coefficient quantization.

Cascade/parallel forms can provide better performance by reducing


the dynamic range of the coefficients, but are not often used for FIR
filters unless the zeros of H(z) are tightly clustered.

Some additional care is usually needed to preserve linear phase with


quantized coefficients in cascade form.

D. Richard Brown III

8/8

You might also like