5 - Design of Digital Filters

Design of Digital Filters
Linear Time-Invariant Systems as Frequency-

Selective Filters
• The term filter is commonly used to describe a device that

discriminates, according to some attribute of the objects
applied at its input, what passes through it.
• For example, an air filter allows air to pass through it but
prevents dust particles that are present in the air from
passing through.
• An oil filter performs a similar function, with the exception
that oil is the substance allowed to pass through the filter,
while particles of dirt are collected at the input to the filter
and prevented from passing through.
Selective Filters
• In photography, an ultraviolet filter is often used to prevent
ultraviolet light, which is present in sunlight and which is
not a part of visible light, from passing through and
affecting the chemicals on the film.
• A linear time-invariant system also performs a type of
discrimination or filtering among the various frequency
components at its input.
• The nature of this filtering action is determined by the
frequency response characteristics H(ω), which in turn
depends on the choice of the system parameters (e.g., the
coefficients (ak) and (bk) in the difference equation
characterization of the system).
Selective Filters
• Thus, by proper selection of the coefficients, we can

design frequency-selective filters that pass signals
with frequency components in some bands while
they attenuate signals containing frequency
components in other frequency bands.
Selective Filters
• In general, a linear time-invariant system modifies the input
signal spectrum X(ω) according to its frequency response
H(ω) to yield an output signal with spectrum 𝒀(𝝎) =
𝑯(𝝎)𝑿(𝝎).
• In a sense, H(ω) acts as a weighting function or a spectral
shaping function to the different frequency components in
the input signal.
• When viewed in this context, any linear time-invariant
system can be considered to be a frequency-shaping filter,
even though it may not necessarily completely block any or
all frequency components.
• Consequently, the terms“linear time-invariant
system”and“filter” are synonymous and are often used
interchangeably.
Selective Filters
• We use the term filter to describe a linear time -invariant
system used to perform spectral shaping or frequency-
selective filtering.
• Filtering is used in digital signal processing in a variety of
ways. For example,
-removal of undesirable noise from desired signals,
-spectral shaping such as equalization of communication
channels,
-signal detection in radar, sonar, and communications, and
-for performing spectral analysis of signals, and so on.
Ideal Filter Characteristics
• Filters are usually classified according to their
frequency-domain characteristics as
- low-pass
- high-pass
- band-pass
- band-stop or band-elimination filters
- all-pass
• The ideal magnitude response characteristics of these
types of filters are illustrated in Fig. 4.43 as shown,
these ideal filters have a constant-gain (usually taken as
unity-gain) pass-band characteristic and zero gain in
their stop-band.
Figure 4.43 Magnitude responses for some ideal frequency-

selective discrete-time filters.
• Another characteristic of an ideal filter is a linear phase
response.
• To demonstrate this point, let us assume that a signal
sequence {x(n)} with frequency components confined to
the frequency range ω1 < ω < ω2 is passed through a filter
with frequency response
(4.5.1)
where C and no are constants.

The signal at the output of the filter has a spectrum
• By applying the scaling and time-shifting properties

of the Fourier transform, we obtain the time-
domain output
• Consequently, the filter output is simply a delayed and

amplitude-scaled version of the input signal.
• A pure delay is usually tolerable and is not considered a
distortion of the signal. Neither is amplitude scaling.
• Therefore, ideal filters have a linear phase characteristic
within their pass-band, that is,
𝜃 𝜔 = −𝜔𝑛0 (4.5.4)
• The derivative of the phase with respect to frequency has
the units of delay. Hence we can define the signal delay as a
function of frequency as
• τg(ω) is usually called the envelope delay or the group delay

of the filter.
• We interpret τg(ω) as the time delay that a signal
component of frequency ω undergoes as it passes from the
input to the output of the system.
• Note that when ϴ(ω) is linear as in (4.5.4),
τg(ω) = no = constant.
• In this case all frequency components of the input signal
undergo the same time delay.
• In conclusion, ideal filters have a constant magnitude
characteristic and a linear phase characteristic within their
pass-band.
• In all cases, such filters are not physically realizable but
serve as a mathematical idealization of practical filters.
• For example, the ideal low pass filter has an impulse
response
(4.5.6)
• We note that this filter is not causal and it is not

absolutely summable and therefore it is also unstable.
• Consequently, this ideal filter is physically unrealizable.
• Nevertheless, its frequency response characteristics can
be approximated very closely by practical, physically
realizable filters.
Design of simple digital filters
• We already know how the location of poles and zeros
affects the frequency response characteristics of the
system
• There is a graphical method which can be used for
computing the frequency response characteristics from the
pole-zero plot.
• This same approach can be used to design a number of
simple but important digital filters with desirable frequency
response characteristics.
• The basic principle underlying the pole-zero placement
method is to locate poles near points of the unit circle
corresponding to frequencies to be emphasized, and to
place zeros near the frequencies to be deemphasized.
Furthermore, the following constraints must be imposed:
1. All poles should be placed inside the unit circle in order for
the filter to be stable. However, zeros can be placed
anywhere in the z-plane.
2. All complex zeros and poles must occur in complex-
conjugate pairs in order for the filter coefficients to be real.
• for a given pole-zero pattern, the system function H(z)
can be expressed as
(4.5.7)
where bo is a gain constant selected to normalize the

frequency response at some specified frequency . That is,
bo is selected such that
𝐻(𝜔0 ) = 1 (4.5.8)
where ωo is a frequency in the passband of the filter.

• Usually, N is selected to equal or exceed M, so that the
filter has more nontrivial poles than zeros.
Lowpass, Highpass, and Bandpass Filters
• In the design of low pass digital filters, the poles should be
placed near the unit circle at points corresponding to low
frequencies (near ω = 0) and zeros should be placed near or
on the unit circle at points corresponding to high
frequencies (near ω = π).
• The opposite holds true for highpass filters.
• Figure 4.44 illustrates the pole-zero placement of three low
pass and three high pass filters.
Figure 4.44
• The magnitude and phase responses for the single-
pole filter with system function
are illustrated in Fig. 4.45 for a = 0.9

• The gain G was selected as 1 − 𝑎, so that the filter
has unity gain at ω = 0.
• The gain of this filter at high frequencies is relatively
small.
• The addition of a zero at z = -1 further attenuates
the response of the filter at high frequencies.
• This leads to a filter with a system function
• and a frequency response characteristic that is also

illustrated in Fig. 4.45. In this case the magnitude of
H2(ω) goes to zero at ω = π.
• Similarly, we can obtain simple highpass filters by
reflecting (folding) the pole-zero locations of the
low pass filters about the imaginary axis in the z-
plane. Thus we obtain the system function
which has the frequency response characteristics

illustrated in Fig. 4.46 for a = 0.9.
Example 4.51
• A two-pole lowpass filter has the system function
Determine the values of bo and p such that the

frequency response H(ω) satisfies the conditions
H(0) = 1 and
at 𝜔 = 𝜋/4
• Hence
or equivalently
The value of p = 0.32 satisfies this equation.

Consequently, the system function for the desired filter is
Bandpass Filters
• The same principles can be applied for the design of
band pass filters.
• Basically, the band pass filter should contain one or
more pairs of complex–conjugate poles near the
unit circle, in the vicinity of the frequency band that
constitutes the passband of the filter.
• This is illustrated in the following example
Example 4.5.2
• Design a two-pole bandpass filter that has the center of
its passband at ω = π/2, zero in its frequency response
characteristic at ω = 0 and ω = π, and its magnitude
response is 1Τ 2 at 𝜔 = 4𝜋Τ9 .
Solution
Clearly the filter must have poles at
𝑝1,2 = 𝑟𝑒 ±𝑗𝜋Τ2
and zeros at z = 1 and z = — 1.
Consequently, the system function is
• The gain factor is determined by evaluating the
frequency response H(ω) of the filter at ω = π/2. Thus
we have
• The value of r is determined by evaluating H(ω) at ω =

4π/9. Thus we have
or equivalently,
The value of 𝑟2 = 0.7 satisfies this equation.

Therefore, the system function for the desired filter is
• Its frequency response is illustrated in Fig. 4.4.7

k
Figure 4.47
Magnitude and phase
response of a simple
bandpass filter in
Example 4.5.2;
0.15[(1−𝑧 −2 )
𝐻(𝑧) =
(1+0.7𝑧 −2 )
• It should be emphasized that the main purpose of the
foregoing methodology for designing simple digital
filters by pole-zero placement is to provide insight into
the effect that poles and zeros have on the frequency
response characteristic of systems.
• The methodology is not intended as a good method for
designing digital filters with well-specified passband
and stop band characteristics.
A simple lowpass-to-highpass filter transformation
• Suppose that we have designed a prototype low pass filter
with impulse response hlp(n). By using the frequency
translation property of the Fourier transform, it is possible
to convert the prototype filter to either a bandpass or a
highpass filter.
• If hlp(n) denotes the impulse response of a low pass filter
with frequency response Hlp(ω), a highpass filter can be
obtained by translating HlP(ω) by π radians (i.e., replacing ω
by 𝜔 − 𝜋). Thus
𝐻ℎ𝑝 𝜔 = 𝐻𝑙𝑝 𝜔 − 𝜋 (4.5.12)

where HhP(ω) is the frequency response of the highpass
filter.
• Since a frequency translation of π radians is equivalent
to multiplication of the impulse response hlp(n) by 𝑒𝑗𝜋𝑛 ,
the impulse response of the highpass filter is,
(4.5.13)
• therefore , the impulse response of the highpass filter is
simply obtained from the impulse response of the low
pass filter by changing the signs of the odd-numbered
samples in hlp(n).
• Conversely,
(4.5.14)
• If the low pass filter is described by the difference equation
(4.5.15)
its frequency response is
(4.5.16)
• Now, if we replace ω by 𝜔 − 𝜋, in (4.5.16), then
(4.5.17)
which corresponds to the difference equation

Example 4.5.3
Solution
• The difference equation for the highpass filter, according to
(4.5.18), is
𝑦 𝑛 = −0.9𝑦 𝑛 − 1 + 0.1𝑥(𝑛)
and its frequency response is

Digital Resonators
• A digital resonator is a special two-pole bandpass filter with
the pair of complex conjugate poles located near the unit
circle as shown in Fig. 4.48(a). The magnitude of the
frequency response of the filter is shown in Fig. 4.48(b).
• The name resonator refers to the fact that the filter has a
large magnitude response (i.e., it resonates) in the vicinity
of the pole location.
• The angular position of the pole determines the resonant
frequency of the filter.
• Digital resonators are useful in many applications, including
simple bandpass filtering and speech generation.
Figure 4.48 (a) Pole-zero pattern and (b) the corresponding

magnitude and phase response of a digital resonator with
(1) r =0.8 and (2) r = 0.95.
• In the design of a digital resonator with a resonant peak at
or near ω = ωo , we select the complex-conjugate poles at
• In addition, we can select up to two zeros. Although there

are many possible choices, two cases are of special interest.
• One choice is to locate the zeros at the origin. The other
choice is to locate a zero at z = 1 and a zero at z = -1.
• This choice completely eliminates the response of the filter
at frequencies ω = 0 and ω = π , and it is useful in many
practical applications.
• The system function of the digital resonator with zeros at
the origin is
(4.5.19)
(4.5.20)
• Since |H(ω)| has its peak at or near ω = ωo, we select the

gain bo so that |H(ωo)| = 1. From (4.5.19) we obtain
(4.5.21)
and hence
• Thus the desired normalization factor is
• The frequency response of the resonator in (4.5.19) can be

expressed as
• where U1(ω) and U2(ω) are the magnitudes of the vectors
from p1 and p2 to the point ω in the unit circle and Φ1(ω)
and Φ2(ω) are the corresponding angles of these two
vectors.
• The magnitudes U1(ω) and U2(ω) may be expressed as
• For any value of r, U1(ω) takes its minimum value (1 − 𝑟) at

ω = ωo. The product U1(ω)U2(ω) reaches a minimum value
at the frequency
which defines precisely the resonant frequency of the filter.
• We observe that when r is very close to unity, ωr ≈ωo,
which is the angular position of the pole.
• We also observe that as r approaches unity, the resonance
peak becomes sharper because U1(ω) changes more rapidly
in relative size in the vicinity of ωo.
• A quantitative measure of the sharpness of the resonance
is provided by the 3-dB band width ∆ω of the filter.
• For values of r close to unity
• Figure 4.48 illustrates the magnitude and phase of digital

resonators with ωo = π/3 , r = 0.8 and ωo = π/3, r = 0.95.
• We note that the phase response undergoes its greatest
rate of change near the resonant frequency.
• If the zeros of the digital resonator are placed at z = 1 and z
= -1, the resonator has the system function
(4.5.27)
and a frequency response characteristic
• We observe that the zeros at z = ± 1 affect both the

magnitude and phase response of the resonator.
• For example, the magnitude response is
where N(ω) is defined as
• Due to the presence of the zero factor, the resonant

frequency is altered from that given by the expression in
(4.5.25). The bandwidth of the filter is also altered .
• Figure 4.49 illustrates the magnitude and phase
characteristics for ωo= π/3, r = 0.8 and ωo = π/3, r =
0.95.
• We observe that this filter has a slightly smaller
bandwidth than the resonator, which has zeros at
the origin.
• In addition, there appears to be a very small shift in
the resonant frequency due to the presence of the
zeros.
Figure 4.49 Magnitude and phase response of digital

resonator with zeros at ω = 0 and ω = π and (1) r = 0.8 and
(2) r = 0.95.
Notch Filters
• A notch filter is a filter that contains one or more deep
notches or, ideally, perfect nulls in its frequency response
characteristic.
• Figure 4.50 illustrates the frequency response characteristic
of a notch filter with nulls at frequencies ωo and ω1.
• Notch filters are useful in many applications where specific
frequency components must be eliminated .
• For example, instrumentation and recording systems
require that the power-line frequency of 60Hz and its
harmonics be eliminated.
Figure 4.50 Frequency response characteristic of a

notch filter.
• To create a null in the frequency response of a filter at a
frequency ωo, we simply introduce a pair of complex-
conjugate zeros on the unit circle at an angle ωo. That is,
𝑧1,2 = 𝑒 ±𝑗𝜔0
• Thus the system function for an FIR notch filter is simply
(4.5.30)
• As an illustration, Fig. 4.51 shows the magnitude response
for a notch filter having a null at ω = π/4.
Fig 4.51
All-Pass Filters
• An all-pass filter is defined as a system that has a constant
magnitude response for all frequencies, that is
• The simplest example of an all-pass filter is a pure delay

system with system function
𝐻(𝑧) = 𝑧 −𝑘
• This system passes all signals without modification except
for a delay of k samples.
• This is a trivial all-pass system that has a linear phase
response characteristic.
• A more interesting all-pass filter is described by the system
function
where all the filter coefficients {𝑎𝑘} are real.

• If we define the polynomial A(z) as
• then (4.5.43) can be expressed as
(4.5.44)
Since
the system given by (4.5.44) is an all-pass system.

FIR FILTER DESIGN
• The frequency response of an Nth-order causal FIR filter is
and the design of an FIR filter involves finding the coefficients

h(n) that result in a frequency response that satisfies a given set
of filter specifications.
FIR filters have two important advantages over IIR filters.
• First, they are guaranteed to be stable, even after the filter
coefficients have been quantized.
• Second, they may be easily constrained to have (generalized)
linear phase.
Because FIR filters are generally designed to have linear phase,
in the following we consider the design of linear phase FIR
filters.
Linear Phase FIR Design Using Windows
• Let ℎ𝑑 (𝑛) be the unit sample response of an ideal
frequency selective filter with linear phase,
𝐻𝑑 𝑒 𝑗𝜔 = 𝐴 𝑒 𝑗𝜔 𝑒 −𝑗(𝛼𝜔−𝛽)
• Because ℎ𝑑 (𝑛) will generally be infinite in length, it is
necessary to find an FIR approximation to 𝐻𝑑 (𝑒 𝑗𝜔 )
• With the window design method, the filter is designed
by windowing the unit sample response,
ℎ 𝑛 = ℎ𝑑 𝑛 𝜔(𝑛)
where ω(n) is a finite-length window that is equal to zero
outside the interval 0 ≤ n ≤ N and is symmetric about its
midpoint:
𝜔(𝑛) = 𝜔(𝑁 − 𝑛)
• The effect of the window on the frequency
response may be seen from the complex
convolution theorem,
• Thus, the ideal frequency response is smoothed by

the discrete-time Fourier transform of the window,
W(𝑒 𝑗𝜔 )
• There are many different types of windows that
may be used in the window design method, a few
of which are listed in Table 9-1 .
• How well the frequency response of a filter designed with
the window design method approximates a desired
response, 𝐻𝑑 𝑒 𝑗𝜔 is determined by two factors (see Fig. 9-
2):
1. The width of the main lobe of W(𝑒 𝑗𝜔 )
2. The peak side-lobe amplitude of W(𝑒 𝑗𝜔 )
• Ideally, the main-lobe width should be narrow, and the
side-lobe amplitude should be small.
• However, for a fixed-length window, these cannot be
minimized independently.
Fig. 9-2. The DTFT of a typical window, which is characterized by the
width of its main lobe ,∆, and the peak amplitude of its side lobes, A,
relative to the amplitude of W(𝑒 𝑗𝜔 ) at ω = 0.
Some general properties of windows are as follows:
1. As the length N of the window increases, the width of the
main lobe decreases, which results in a decrease in the
transition width between passbands and stopbands.
This relationship is given approximately by
𝑁∆𝑓 = 𝑐 9.1
where ∆f is the transition width, and c is a parameter that
depends on the window.
2. The peak side-lobe amplitude of the window is determined
by the shape of the window, and it is essentially independent
of the window length.
3. If the window shape is changed to decrease the side-lobe
amplitude, the width of the main lobe will generally increase.
• Listed in Table 9.2 are the side-lobe amplitudes of several
windows along with the approximate transition width and
stopband attenuation that results when the given window
is used to design an Nth-order low-pass filter.
EXAMPLE 9.3.1
Suppose that we would like to design an FIR linear phase
low-pass filter according to the following specifications:
• For a stopband attenuation of 20 log(0.01 ) = -40 dB.

we may use a Hanning window. Although we could also
use a Hamming or a Blackman window, these windows
would overdesign the filter and produce a larger
stopband attenuation at the expense of an increase in
the transition width. Because the specification calls for
a transition width of ∆𝜔 = 𝜔𝑠 − 𝜔𝑝 = 0.02𝜋, or ∆f =
0.01, with
𝑁∆𝑓 = 3.1
• for a Hanning window (see Table 9.2), an estimate
of the required filter order is
• The last step is to find the unit sample response of

the ideal low-pass filter that is to be windowed.
𝜔𝑠 +𝜔𝑝
With a cut-off frequency of 𝜔𝑐 = ൗ2 = 0.2𝜋,
and a delay of α = N/2 = 155, the unit sample
response is
Design of IIR Filters from Analog Filters
The design of a digital filter from an analog prototype
requires that we transform ℎ𝑎 (𝑡) to h(n) or 𝐻𝑎 (𝑠) to
H(z).
• A mapping from the s-plane to the z-plane may be
written as
• where s = m(z) is the mapping function. In order for

this transformation to produce an acceptable digital
filter, the mapping m(z) should have the following
properties:
1. The mapping from the jΩ-axis to the unit circle, 𝑧 = 1,
should be one to one and onto the unit circle in order to
preserve the frequency response characteristics of the
analog filter.
2. Points in the left-half s-plane should map to points inside
the unit circle to preserve the stability of the analog filter.
3. The mapping m(z) should be a rational function of z so
that a rational 𝐻𝑎 (𝑠) is mapped to a rational H(z).
Two approaches that are commonly used to map analog
filters into digital filters are impulse invariance method and
the Bilinear Transformation.
The Bilinear Transformation

• The bilinear transformation is a mapping from the s-plane
to the z-plane defined by
2 1−𝑧 −1
𝑠=
𝑇 1+𝑧 −1
• Given an analog filter with a system function 𝐻𝑎 (𝑠) the
digital filter is designed as follows:
2 1 − 𝑧 −1
𝐻 𝑧 = 𝐻𝑎
𝑇 1 + 𝑧 −1
• The bilinear transformation is a rational function that maps
the left-half s-plane inside the unit circle and maps the jΩ-axis
in a one-to-one manner onto the unit circle.
• However, the relationship between the jΩ-axis and the unit
circle is highly nonlinear and is given by the frequency
warping function
Ω𝑇𝑠
𝜔 = 𝑡𝑎𝑛 (9.12)
2
• As a result of this warping, the bilinear transformation will
only preserve the magnitude response of analog filters that
have an ideal response that is piecewise constant.
• Therefore, the bilinear transformation is generally only
used in the design of frequency selective filters.
• The parameter 𝑇𝑠 , in the bilinear transformation is normally
included for historical reasons. However, it does not enter
into the design process, because it only scales the jΩ-axis in
the frequency warping function, and this scaling may be
done in the specification of the analog filter.
• Therefore, 𝑇𝑠 , may be set to any value to simplify the
design procedure.
• The steps involved in the design of a digital low-pass filter
with a passband cutoff frequency 𝜔𝑝 , stopband cutoff
frequency 𝜔𝑠 , passband ripple 𝛿𝑝 and stopband ripple 𝛿𝑠 ,
are as follows:
1. Prewarp the passband and stopband cutoff frequencies of
the digital filter, 𝜔𝑝 and 𝜔𝑠 using the inverse of Eq. (9.12)
to determine the passband and cutoff frequencies of the
analog low-pass filter.
With 𝑇𝑠 = 2, the prewarping function is
𝜔
Ω = 𝑡𝑎𝑛
2
2. Design an analog low-pass filter with the cutoff
frequencies found in step 1 and a passband and
stopband ripple 𝛿𝑝 and 𝛿𝑠 respectively.
3. Apply the bilinear transformation to the filter
designed in step 2.

5 - Design of Digital Filters

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

5 - Design of Digital Filters

Загружено:

Авторское право:

Доступные форматы

Design of Digital Filters

Linear Time-Invariant Systems as Frequency-

• The term filter is commonly used to describe a device that

• Thus, by proper selection of the coefficients, we can

Figure 4.43 Magnitude responses for some ideal frequency-

where C and no are constants.

• By applying the scaling and time-shifting properties

• Consequently, the filter output is simply a delayed and

• τg(ω) is usually called the envelope delay or the group delay

• We note that this filter is not causal and it is not

where bo is a gain constant selected to normalize the

where ωo is a frequency in the passband of the filter.

are illustrated in Fig. 4.45 for a = 0.9

• and a frequency response characteristic that is also

which has the frequency response characteristics

Determine the values of bo and p such that the

The value of p = 0.32 satisfies this equation.

• The value of r is determined by evaluating H(ω) at ω =

The value of 𝑟2 = 0.7 satisfies this equation.

• Its frequency response is illustrated in Fig. 4.4.7

𝐻ℎ𝑝 𝜔 = 𝐻𝑙𝑝 𝜔 − 𝜋 (4.5.12)

which corresponds to the difference equation

and its frequency response is

Figure 4.48 (a) Pole-zero pattern and (b) the corresponding

• In addition, we can select up to two zeros. Although there

• Since |H(ω)| has its peak at or near ω = ωo, we select the

• Thus the desired normalization factor is

• The frequency response of the resonator in (4.5.19) can be

• For any value of r, U1(ω) takes its minimum value (1 − 𝑟) at

• Figure 4.48 illustrates the magnitude and phase of digital

and a frequency response characteristic

• We observe that the zeros at z = ± 1 affect both the

where N(ω) is defined as

• Due to the presence of the zero factor, the resonant

Figure 4.49 Magnitude and phase response of digital

Figure 4.50 Frequency response characteristic of a

• The simplest example of an all-pass filter is a pure delay

where all the filter coefficients {𝑎𝑘} are real.

the system given by (4.5.44) is an all-pass system.

and the design of an FIR filter involves finding the coefficients

• Thus, the ideal frequency response is smoothed by

• For a stopband attenuation of 20 log(0.01 ) = -40 dB.

• The last step is to find the unit sample response of

• where s = m(z) is the mapping function. In order for

The Bilinear Transformation

Вам также может понравиться