Sunteți pe pagina 1din 4

JOURNAL OF L

A
T
E
X CLASS FILES, VOL. 6, NO. 1, JANUARY 2007 1
Comparison of Characteristics
of FIR and IIR Filter
Johannes Innerbichler, TU Graz, Achim Schweighofer, TU Graz
AbstractIn the following sections this paper deals with the
study of two classes of lter and lists the main advantages and
disadvantages of each type. At rst the characteristics of the FIR
lter are discussed followed by the properties of the IIR lter.
In the the last part these two tyes of lters are compared and
the appropriate eld of work is assigned.
Index TermsFIR, IIR, Finite Impuls Response, Innite Im-
pulse Response, DSP, lter designs
I. INTRODUCTION
I
N the eld of digital signal processing (short DSP) each
discrete time signal is represented as a sequence of numbers
or symbols. Most of the time mathematical operations have to
be performed to an analog signal to enhance certain parts (or
frequencies). To accomplish such a modication the analog
signal is most times sampled, modied in the digital domain
by a digital lter and transformed back into the continuous
domain. In contrast to an analog lter where the signal is
manipulated directly, a digital lter offers much more op-
pertunities and advantages. To describe such a digital lter
two techniques are applied. A lter can be specied by its
impulse response and/or its difference equation. Typically two
categories of digital lters are distinguished by there impulse
response: lters with an inntite impulse response (short IIR
lter) and with an nite impulse response (short FIR lter).
Of course each categorie has its own advantages and aws. In
the following section these two categories are discussed and
compared with each other.
jin
November 23, 2011
II. TECHNIQUES FOR ANALYSING A FILTER
T
WO techniques are commonly used two describe the
behaviour of a digital lter. Each technique is described
more thourough in the next subsections.
A. Impulse Response
The impulse response describes the response of a digital
lter to the Kronecker delta function (see appendix A).
1) FIR: In the case of a linear time-invariant FIR lters
the output of the lter can be determined by convoluting the
input vector (equals digital signal) with the coefcients of the
impulse response h[k]:
y[n] =
n

k=0
x[n k] h[k] = (x h)[n] (1)
2) IIR: Because of the feedback the computation of the
output is more complicated because, the current value of the
outputsequence depends on the previous and current input as
well as previous outputs. The corresponding formula can be
seen in equation 2.
M

m=0
a[m] y[n m] =
n

k=0
x[n k] h[k] (2)
The sequences a[k] and b[k] depend on the feedback of the
lter.
B. Difference Equation
Another method to dene a lter is the difference equation
using the z-transformation. The typical structure is seen in
equation 3.
H(z) =

M
i=0
(z + b
i
)
eMi

N
i=0
(z + a
i
)
eNi
(3)
Whereas the denominator is a result of the feedback.
III. FIR FILTER
I
N nite impulse response lter the impulse response con-
sists of a sequence with limited therms.
A. Linear Delay
In our research we validated our assumption that linearity of
the phase is connected to an symmetric FIR lter. This means
the phase of the frequency response (also called delay) is the
same at all frequencies (wraps of the phase at +/- 180 degrees
are ingnored). Therefore the lter does not cause any phase or
delay distortion, which could be an important advantage. If the
coefcients of the impulse response are symmetric around the
center coefcient, a linear phase results. An example impulse
response can be found in [1] and is dened in equation 4. The
corresponding plot shown in gure 1
h
FIR
[n] = 2[n 1] + 2[n 2] 2[n 3] (4)
In the plot below the symmetric property is depicted. The
center coefcient is h[2] = 2.
JOURNAL OF L
A
T
E
X CLASS FILES, VOL. 6, NO. 1, JANUARY 2007 2
n
h[n]
1
2
1
2
1 2 3 4
Fig. 1. Example impulse response h
FIR
[n]
The corresponding frequency response is calculated in ap-
pendix B and the result presented in equation 5.
H(e
j
) = e
2j

_
2 4cos()
_
(5)
The phase alters according to the linear function 2 and is
therefore linear. With this knowledge a lter with linear delay
can be simply implemented on digital signal microprocessors
and can be calculated by looping a single instruction.
Another big advantage of this lter is the suitability to muli-
rate applications. This means the sampling rate can be reduced
(decimation) and/or increased (interpolation). With this two
properties more efcient implementations can be realised.
B. No Feedback needed
Implementations of FIR lter do not require a feedback and
therefore avoid rounding errors due summed iterations. Due
the fact that each error occurs just once in each calculation
the implementation very easy.
C. Stablility
Another result of the non-existing feedback is the location
of the poles in the z-plane. Each poles is placed in the origin
and therefore within the unit circle. According to [2] lter
with poles inside the unit circle are stable. Another fact is
that due the limited size of the lter sequence FIR lter are
BIBO (bounded-input bounded-output) stabil.
D. Memory Consumption
One major draw back of FIR lter is the high memory
consumption. Every single coefcient has to be stored on its
own register. This huge amount of needed memory requires a
lot of space. Therefore this class of lter is not suitable for
realisations on small chips.
E. Design Methods
Some certain lter types can be calculated directly. Exam-
ples would be the raised cosine lter or the windowed sinc
lter. Another method is windowing. The impulse response
is derived using the inverse fast fourier transformation (short
IDFT) of the needed frequency response. The resulting IIR
lter is rened by windowing and a FIR lter is computed.
IV. IIR-FILTER
A
S the name implies, IIR lters are systems with an
innite impulse response (IIR). This means that the
impulse response could have a sequence of an innite number
of therms.
A. Feedback
Commonly IIR lters have a feedback. Earlier output values
are used to calculate the current one.
Figure 2 shows a block diagram of a simple IIR lter.
Fig. 2. A simple block diagram of an IIR lter
The difference equation for this system of rst order can be
written as in equation 6.
y[n] = b
0
x[n] + b
1
x[n 1] + a
1
y[n 1] (6)
y[n] is calculated by adding an FIR looking lter and also
the previous value y[n-1]. This means y[n] is fed back.
Generally we dene the difference equation for an IIR lter
in equation 7.
y[n] =
N

l=1
a
l
y[n l]
. .
feedback
+
M

k=1
b
k
y[n k]
. .
input
(7)
The number of feedback terms N denes the order of the
system.
B. Stability
If we consider the system from gure 2 from above the
difference equation 6 can be used to calculate the transfer-
function by z-transform. In equation 8 the transfer-function is
presented, which is calculated in appendix D.
H(z) =
Y (z)
X(z)
=
b
0
+ b
1
z
1
1 a
1
z
1
(8)
A lter is stable, if each pole is placed inside the unit circle.
[2] For our example
|z

| = |a
1
| < 1 (9)
We determined that an IIR lter is stable, if all poles of
transfer-function are placed inside the unit circle. This is in
contrast to the FIR lter where all poles are located at the
origin, and is therefore always stable.
An IIR lter is also stable, if the impulse response of the
system is absolutly summable.
JOURNAL OF L
A
T
E
X CLASS FILES, VOL. 6, NO. 1, JANUARY 2007 3
C. Minimum phase
In comparison to FIR lters the phase of an IIR lter might
be non-linear.
In an stable IIR lter, all poles are located inside the unit
circle, where the zeroes are located anywhere. Sometimes the
inverse of such a system is needed. If there are zeroes, which
are located outside the unit circle, it results in an unstable
inverse. To avoid this problem, we created a minimum-phase
system.
In control theory and signal processing, a linear, time-
invariant system is said to be minimum-phase if the system
and its inverse are causal and stable. [4]
For example the IIR lter with transfer-function in equation
10.
H(z) =
(1 + 4z
1
)(1 + 0.8z
1
)
(1 0.5z
1
)(1 0.1z
1
)
=
(z + 2)(z + 0.8)
(z + 0.5)(z + 0.1)
(10)
This system has a zero outside the unit circle in
z
0
= 4 (11)
So the inverse of this system is not causal and stable. But it
can be divided into a miminum-phase, which has an stable
and causal inverse, and an allpass system.
H(z) = H
min
(z)H
all
(z) (12)
In equation 13 the resulting mimimum-phase system, calcu-
lated in appendix E is presented. The zero outside the unit
circle has been mirrored among the unit circle.
H
min
(z) =
(z + 0.25)(z + 0.8)
(z + 0.5)(z + 0.1)
(13)
The mangitude of the minimum-phase system is equal to the
total system.
|H
min
(z)| = |H(z)| (14)
Due the allpass system an phase error occurs. So the phase of
H
min
is different from H.
We determine, that the resulting IIR minimum-phase lter
with transfer-function H
min
has now a stable and causal
inverse.
D. Comparison to analog lters
The result of our research was, that IIR lters can either be
implemented as analog or digital lters. In the design phase
of an digital IIR lter it is possible to design an analog lter
rst, and then convert this lter to an digital lter by applying
discretization techniques such as Bilinear transform or Impulse
invariance. This is one of the biggest advantages of IIR lters.
Here is an example for the conversion of an analog low-pass
RC lter [3] to an digital lter using the Bilinear transform.
In equation 15 the analog transfer function is shown. The
corresponding Bilinear transform (digital transfer function) is
calculated in Appendix C and the result presented in equation
16.
H
a
[s] =
1
1 + RCs
(15)
H
d
[z] =
1 + z
1
(1 + 2RC/T) + (1 2RC/T)z
1
(16)
So IIR lters can be used to model a huge variety of analog
lters, such as Chebyshev lter, Butterworth lter, and the
Bessel lter in discrete domain.
V. CONCLUSION
FIR lters are much easier to implement but used a relative
high memory to store the lters specication.
IIR lters of the same order have a sharper transition from
passband to stopband and so they are more selective than FIR
lters. However the phase response of IIR lters is not linear,
therefore they are less suitable for signals whose waveform
must not be changed during ltering.
There are a few methods such as Bilinear Transform to
transform an analog lter to an digital IIR lter. IIR lters
can be used to model a huge variety of analog lters, such
as Chebyshev lter, Butterworth lter, and the Bessel lter in
discrete domain. This is one of the biggest advantages of IIR
lters.
APPENDIX A
KRONECKER DELTA FUNCTION
In the eld of digital signal processing the Kronecker delta
function is described as follows:
[n] =
_
1 : n = 0
0 : n = 1
(17)
This function is often called the unit or impulse function.
APPENDIX B
CALCULATION OF THE FREQUENCY RESPONSE
H(e
j
) =
=
+

n=
(2[n 1] + 2[n 2] 2[n 3]) e
jn
= 2e
j1
+ 2e
j2
2e
j3
= e
2j

_
2e
j
+ 2 2e
j
_
= e
2j

_
2 4cos()
_
JOURNAL OF L
A
T
E
X CLASS FILES, VOL. 6, NO. 1, JANUARY 2007 4
APPENDIX C
CALCULATION OF THE BILINEAR TRANSFORM
H
d
(z) = H
a
_
2
T
z 1
z + 1
_
=
1
1 + RC
_
2
T
z1
z+1
_
=
1 + z
(1 2RC/T) + (1 + 2RC/T)z
=
1 + z
1
(1 + 2RC/T) + (1 2RC/T)z
1
APPENDIX D
CALCULATION OF THE TRANSFER-FUNCTION
(Z-TRANSFORM)
y[n] = y[n] = b
0
x[n] + b
1
x[n 1] + a
1
y[n 1] (18)
If we apply the z-transform [2], we get the following
Y (z) = b
0
X(z) + b
1
z
1
X(z) + a
1
z
1
Y (z) (19)
We obtain
H(z) =
Y (z)
X(z)
=
b
0
+ b
1
z
1
1 a
1
z
1
(20)
APPENDIX E
CALCULATION OF MINIMUM-PHASE
According to [4] to calculate the minimum-phase of an
causal system all zeroes outside of the unit circle are mirrored
into unit circle, as shown in equation 21.
z
0,k
=
_
1
z
0,k

|z
0,k
| 1
z
0,k
|z
0,k
| < 1
(21)
The transfer-function from equation 10 has zeroes in
z
0,1
= 4; z
0,2
= 0.8 (22)
According to equation 21 the new zeroes are
z
0,1
=
1
z
0,1

=
1
4
= 0.25 (23)
and
z
0,2
= z
0,2
= 0.8 (24)
The resulting transfer-function of the minimum-phase sys-
tem is presented in equation 25.
H
min
(z) =
(z + 0.25)(z + 0.8)
(z + 0.5)(z + 0.1)
(25)
REFERENCES
[1] Monson Hayes, Schaums Outline of Digital Signal Processing, 2nd ed.
USA: Schaums Outline Series, 2011.
[2] Alan V. Oppenheim and Ronald W. Schafer, Discrete-Time Signal Pro-
cessing, 3rd ed. USA: Prentice Hall Signal Processing, 2009.
[3] Ulrich Tietze and Christoph Schenk, Halbleiter-Schaltungstechnik, 3rd ed.
Germany: Springer, 2002.
[4] Babak Hassibi; Thomas Kailath; Ali H. Sayed, Linear estimation, 1st ed.
USA: Prentice Hall, 2000.

S-ar putea să vă placă și