Digital Signal Processing

Using Matlab

Digital Filter Design

Overview
• Introduction
• Ideal Filters
• Practical Filters
• LPF Design
• FIR LPF Design using Window Method
• IIR LPF Design using Butterworth
Approximation and BLT

2
Introduction
Digital filter design involves the following three steps:
1. Specification:
Specify the desired frequency response function characteristics
to address the needs of a specific application.
2. Approximation:
Approximate the desired frequency response function by the
frequency response of a filter with a polynomial or a rational
system function. The goal is to meet the specifications with
minimum complexity, that is, by using the filter with the lowest
number of coefficients.
3. Implementation:
Realize the system obtained in hardware, software, or both
using one of the structures discussed in the previous chapter.
3
Introduction
• Filters are broadly classified according to their frequency-
domain characteristics into five types:
1. Lowpass filter (LPF)
2. Highpass filter (HPF)
3. Bandpass filter (BPF)
4. Bandstop filter (BSF)
5. Allpass filter (APF)

4
Ideal Filters
Ideal Lowpass Filter (LPF): Frequency domain

• Ideal LPF passes low-

frequency signals and
eliminates high-
frequency signals.

Transfer Function Passband Stopband

1 for 𝜔 ≤ 𝜔𝑐
𝐻𝑙𝑝 𝜔 =ቊ
0 for 𝜔𝑐 ≤ 𝜔 ≤ 𝜋

5
Ideal Filters
Ideal Lowpass Filter (LPF): Time domain

ℎ𝑙𝑝 𝑛
Impulse Response
Noncausal impulse
response

𝜔𝑐ൗ
sin(𝜔𝑐 𝑛) 𝜔𝑐 𝜔𝑐 𝑛 𝜋 for 𝑛 = 0
ℎ𝑙𝑝 𝑛 = = sinc = ൞sin(𝜔𝑐 𝑛)
𝜋𝑛 𝜋 𝜋 for 𝑛 ≠ 0
𝜋𝑛
6
Ideal Filters
Ideal Highpass Filter (HPF)

• Ideal HPF passes high-

frequency signals and
eliminates low-
frequency signals.

Ideal Bandpass Filter (BPF)

• Ideal BPF passes signals

between two specific
frequencies, but
eliminates signals at
other frequencies. 7
Ideal Filters
Ideal Bandstop Filter (BSF)

• Ideal BSF filter eliminates

signals between two
specific frequencies and
passes signals at all other
frequencies.

Ideal Allpass Filter (APF)

• Ideal APF passes all

frequencies equally in
gain, but changes their
phases.
8
Practical Filters
Ideal filters have the following characteristics:
• Unity gain in passband.
• Zero gain in stopband
• Zero-phase (no delay)
• Noncausal (infinite two-sided impulse response).
• Unstable (the sinc function is not absolutely summable)

Ideal filters cannot be implemented in practice; therefore, they

have to be approximated by practically realizable filters. Practical
filters differ from ideal filters in several respects:
(a) the passband responses are not perfectly flat,
(b) the stopband responses cannot completely reject (eliminate)
bands of frequencies, and
(c) the transition between passband and stopband regions takes
place over a finite transition band. 9
Practical Filters (Real , causal)

Lowpass Highpass
filter filter

Bandpass Bandstop
filter filter

10
LPF Design
LPF Relative Specifications (Linear)

Passband
1
Gain 𝐺𝑝𝑎𝑠𝑠
Magnitude
|H(𝜔)|

Transition
Passband band Stopband
Stopband 𝐺𝑠𝑡𝑜𝑝
Gain/Ripple
0 𝜔Wp
𝑝𝑎𝑠𝑠 𝜔𝑠𝑡𝑜𝑝
Ws 𝜋1

Normalized frequency
11
LPF Design
LPF Relative Specifications (dB)

0
𝐴𝑝𝑎𝑠𝑠
Magnitude (dB)

Passband
Ripple in dB
𝐴𝑠𝑡𝑜𝑝
Stopband
Attenuation in dB

Stopband
Passband Transition
band

0 𝜔Wp
𝑝𝑎𝑠𝑠 𝜔𝑠𝑡𝑜𝑝
Ws 𝜋1

Normalized frequency
12
LPF Design
LPF Relative Specifications

( Apass ) dB = −20 log10 (G pass )  0 for passband

( Astop ) dB = −20 log10 (Gstop )  0 for stopband

Example (1):
Given the minimum passband gain 𝐺𝑝𝑎𝑠𝑠 = 0.8 and the maximum stopband
ripple 𝐺𝑠𝑡𝑜𝑝 = 0.01, determine the passband ripple and the stopband
attenuation in dB.

Solution:
𝐴𝑝𝑎𝑠𝑠 = −20 log10 (0.8) = 1.94 dB

𝐴𝑠𝑡𝑜𝑝 = −20 log10 (0.01) = 40 dB

13
LPF Design
LPF Relative Specifications

Example (2):
In a certain filter’s specifications the passband ripple is 1 dB, and the stopband
attenuation is 50 dB. Determine 𝐺𝑝𝑎𝑠𝑠 and 𝐺𝑠𝑡𝑜𝑝 .

Solution:

𝐴𝑝𝑎𝑠𝑠 = 1 = −20 log10 (𝐺𝑝𝑎𝑠𝑠 )

𝐺𝑝𝑎𝑠𝑠 = 10−1/20 = 0.891

𝐴𝑠𝑡𝑜𝑝 = 50 = −20 log10 (𝐺𝑠𝑡𝑜𝑝 )

𝐺𝑠𝑡𝑜𝑝 = 10−50/20 = 0.00316

14
LPF Design
LPF Design Parameters

– Impulse response = FIR/IIR

– Filter order = 𝑁 (Minimum/Specified)
– Sampling frequency = 𝐹𝑠 Hz
𝐹𝑝𝑎𝑠𝑠
– Passband frequency = 𝐹𝑝𝑎𝑠𝑠 Hz or 𝜔𝑝𝑎𝑠𝑠 = 2𝜋
𝐹𝑠
𝐹𝑠𝑡𝑜𝑝
– Stopband frequency = 𝐹𝑠𝑡𝑜𝑝 Hz or 𝜔𝑠𝑡𝑜𝑝 = 2𝜋
𝐹𝑠
– Passband ripple = 𝐴𝑝𝑎𝑠𝑠 dB
– Stopband attenuation = 𝐴𝑠𝑡𝑜𝑝 dB
– Design method:
• FIR = Window, Equiripple, Interpolated, etc.
• IIR = Butterworth, Chebyshev, Elliptic, etc.
15
LPF Design
LPF Relative Specifications

Example (3):
A digital LPF filter has the following specifications:
0.8 ≤ 𝐻 𝜔 ≤1 0 ≤ 𝜔 ≤ 0.2𝜋
𝐻 𝜔 ≤ 0.2 0.6𝜋 ≤ 𝜔 ≤ 𝜋
Find its related design parameters.

Solution:
𝐴𝑝𝑎𝑠𝑠 = −20 log10 0.8 = 1.94 dB

𝐴𝑠𝑡𝑜𝑝 = −20 log10 0.2 = 13.98 dB

𝜔𝑝𝑎𝑠𝑠 = 0.2𝜋

𝜔𝑠𝑡𝑜𝑝 = 0.6𝜋
16
FIR LPF Design using Window Method
The design procedure of the FIR filter via windowing is summarized
as follows:

To become a
causal filter
17
FIR LPF Design using Window Method

18
FIR LPF Design using Window Method

Impulse response for the low-pass FIR filter:

𝜔𝑐ൗ
sin(𝜔𝑐 𝑛) 𝜔𝑐 𝜔𝑐 𝑛 𝜋 for 𝑛 = 0
ℎ 𝑛 = = sinc = ൞sin(𝜔𝑐 𝑛)
𝜋𝑛 𝜋 𝜋 for 𝑛 ≠ 0
𝜋𝑛

For 3-tap FIR Filter, 𝑛 = −1, 0, 1

19
FIR LPF Design using Window Method
Solution:

2𝜋𝐹𝑐 2𝜋(800)
𝜔𝑐 = = = 0.2𝜋
𝐹𝑠 8000

2M+1=taps=3,
therefore M=1

20
FIR LPF Design using Window Method

21
FIR LPF Design using Window Method

22
FIR LPF Design using Window Method
linear-phase response, 3 taps ( odd symmetry). There are three
coefficients and two delay cells . Note that this filter is of order 2
(M), the number of delay cells

23
FIR LPF Design using Window Method
Matlab Implementation

b = fir1(n, Wn, 'ftype', window, 'normalization')

Impulse Response

0.2

0.18

0.16

MATLAB: 0.14

Amplitude
0.12
>> b = fir1(2, .2, 'noscale') 0.1

0.08

0.06

b= 0.04

0.0150 0.2000 0.0150 0.02

0 50 100 150 200 250

Time (useconds)
Impulse Response
0.9
>> b = fir1(2, .2) 0.8

0.7

b= 0.6
Amplitude

0.5

0.0651 0.8698 0.0651 0.4

0.3

0.2

0.1

0 50 100 150 200 250

Time (useconds)
24
IIR LPF Design ***
• The most common method of designing IIR filters is based on
the conversion of analog lowpass filter prototypes (like
Butterworth, Chebyshev, and Elliptic) into digital filters by using
a mapping algorithm such as the bilinear transformation (BLT).
• Bilinear transformation maps the s-plane poles and zeros of the
analog filter into the z-plane.
• Under this mapping:
o the entire 𝑗Ω axis in the s-plane is mapped onto the unit
circle in the z-plane;
o the left-half s-plane is mapped inside the unit circle and
o the right-half s-plane is mapped outside the unit circle.
• It is also possible to design an IIR filter without any reference to
analog designs.
25
IIR LPF Design

𝐻𝑎 (𝑠) Mapping (BLT) 𝐻(𝑧)

Bilinear transformation (BLT) from the s-plane (analog)

to the z-plane (digital)
26
IIR LPF Design
▪ Butterworth filters:
• Maximally flat magnitude response in both passband and
stopband (no ripple).
• Slow roll-off rate (large transition band).
• Fairly linear phase in the passband.
• Require higher order compared to other types.
▪ Chepychev filters:
• Type I: have ripple in passband and flat in stopband.
• Type II: flat in passband and have ripple in stopband.
• Moderate roll-off rate (sharper transition).
• Phase characteristics in between Butterworth and elliptic.
• Moderate filter order.
27
IIR LPF Design
▪ Elliptic filters:
• Have ripples in both passband and stopband.
• Fast roll-off rate (steeper transition than that of a
Butterworth or Chebyshev).
• Highly nonlinear phase response in the passband.
• Lower filter order.

28
IIR LPF Design

Comparison of Butterworth, Chebyshev, and elliptic filters

29
Design of Butterworth LPF using BLT

The Butterworth
analog filter
2
𝐺𝑝𝑎𝑠𝑠

𝐴𝑝𝑎𝑠𝑠 dB
= −20 log10 𝐺𝑝𝑎𝑠𝑠

𝐴𝑠𝑡𝑜𝑝 dB
= −20 log10 𝐺𝑠𝑡𝑜𝑝
2
𝐺𝑠𝑡𝑜𝑝

Ω𝑝𝑎𝑠𝑠 Ω𝑠𝑡𝑜𝑝

2
1
𝐻𝑎 (𝑗Ω) = 2𝑁
Ω
1+
Ω𝑐 30
Design of Butterworth LPF using BLT
The design procedure of a lowpass IIR filter via Butterworth analog
prototype and bilinear transformation is summarized as follows:

1. Calculate the digital frequencies 𝜔𝑝𝑎𝑠𝑠 and 𝜔𝑠𝑡𝑜𝑝 .

2. Compute the equivalent analog frequencies Ω𝑝𝑎𝑠𝑠 and Ω𝑠𝑡𝑜𝑝
using the pre-warping equation:
2 𝜔𝑝𝑎𝑠𝑠 2 𝜔𝑠𝑡𝑜𝑝
Ω𝑝𝑎𝑠𝑠 = tan Ω𝑠𝑡𝑜𝑝 = tan
𝑇 2 𝑇 2
3. Find the filter order using:
1
൘𝐺 2 − 1
log
𝑠𝑡𝑜𝑝 100.1(𝐴𝑠𝑡𝑜𝑝 )dB − 1
1 log
൘𝐺 2 − 1 100.1(𝐴𝑝𝑎𝑠𝑠)dB − 1
𝑝𝑎𝑠𝑠
𝑁= =
Ω𝑠𝑡𝑜𝑝 Ω𝑠𝑡𝑜𝑝
log log
Ω𝑝𝑎𝑠𝑠 Ω𝑝𝑎𝑠𝑠 31
Design of Butterworth LPF using BLT
4. Select the normalized (Ω𝑐 = 1) Butterworth prototype
corresponding to N from the following table:

32
Design of Butterworth LPF using BLT
5. Calculate the analog 3-dB cutoff frequency, Ω𝑐 .
Ω𝑠𝑡𝑜𝑝 Ω𝑠𝑡𝑜𝑝
Ω𝑐 = 1 = 1
1 2𝑁 100.1(𝐴𝑝𝑎𝑠𝑠)dB − 1 2𝑁
൘𝐺 2 − 1
𝑠𝑡𝑜𝑝
Note that if 𝐴𝑝𝑎𝑠𝑠 = 3 dB then Ω𝑐 = Ω𝑝𝑎𝑠𝑠
6. Find the Transfer Function, 𝐻𝑎 (𝑠) of the analog LPF by
substituting s with 𝑠ΤΩ𝑐 in 𝐻𝑝 𝑠 .
7. Use the bilinear transformation (BLT) to obtain the digital filter
𝐻(𝑧).

𝐻 𝑧 = 𝐻𝑎 𝑠 ቚ 2 1−𝑧 −1
𝑠=𝑇
1+𝑧 −1
33
Design of Butterworth LPF using BLT
Example 1:
Design a digital lowpass Butterworth filter with the following
specifications:
• Passband ripple = 1 dB
• Stopband attenuation = 20 dB
• Passband frequency 𝐹𝑝𝑎𝑠𝑠 = 1 kHz
• Stopband frequency 𝐹𝑠𝑡𝑜𝑝 = 3 kHz
• Sampling frequency 𝐹𝑠 = 8 kHz
Solution:
1. Calculate the digital frequencies 𝜔𝑝𝑎𝑠𝑠 and 𝜔𝑠𝑡𝑜𝑝 .
2𝜋𝐹𝑝𝑎𝑠𝑠 2𝜋 1𝑘
𝜔𝑝𝑎𝑠𝑠 = = = 0.25𝜋 rad/sample
𝐹𝑠 8𝑘
2𝜋𝐹𝑠𝑡𝑜𝑝 2𝜋 3𝑘
𝜔𝑠𝑡𝑜𝑝 = = = 0.75𝜋 rad/sample
𝐹𝑠 8𝑘 34
Design of Butterworth LPF using BLT
2. Pre-warp the digital frequencies to compute the equivalent
analog frequencies Ω𝑝𝑎𝑠𝑠 and Ω𝑠𝑡𝑜𝑝 .
2 𝜔𝑝𝑎𝑠𝑠 0.25𝜋
Ω𝑝𝑎𝑠𝑠 = tan = 2 8𝑘 tan = 6627.4 rad/sec
𝑇 2 2
2 𝜔𝑠𝑡𝑜𝑝 0.75𝜋
Ω𝑠𝑡𝑜𝑝 = tan = 2 8𝑘 tan = 38627.4 rad/sec
𝑇 2 2
3. Find the filter order.

100.1(𝐴𝑠𝑡𝑜𝑝 )dB − 1
log 102 − 1
100.1(𝐴𝑝𝑎𝑠𝑠 )dB − 1 log
100.1 − 1
𝑁= = = 1.69
Ω𝑠𝑡𝑜𝑝 38627.4
log log
Ω𝑝𝑎𝑠𝑠 6627.4

We choose 𝑁 = 2
35
Design of Butterworth LPF using BLT
4. The normalized Butterworth prototype for N = 2 is given by:
1
𝐻𝑃 (𝑠) = 2
𝑠 + 1.414 𝑠 + 1

5. Calculate the analog 3-dB cutoff frequency, Ω𝑐 .

Ω𝑠𝑡𝑜𝑝 38627.4
Ω𝑐 = 1 = 1 = 12245.8 rad/sec
100.1(𝐴𝑝𝑎𝑠𝑠)dB − 1 2𝑁 102 − 1 4

6. Find the Transfer Function, 𝐻𝑎 (𝑠) of the analog LPF.

12245.8 2
𝐻𝑎 𝑠 = 𝐻𝑃 𝑠 ቚ 𝑠 = 2 2
𝑠→
Ω𝑐 𝑠 + 17318.18 𝑠 + 12245.8
36
Design of Butterworth LPF using BLT
7. Finally, use the BLT to obtain the digital filter 𝐻(𝑧).

𝐻 𝑧 = 𝐻𝑎 𝑠 ቚ 2 1−𝑧 −1
𝑠=
𝑇 1+𝑧 −1

12245.8 2
= 2
2 1 − 𝑧 −1 2 1 − 𝑧 −1 2
+ 17318.18 + 12245.8
𝑇 1 + 𝑧 −1 𝑇 1 + 𝑧 −1

Simplifying this equation yields to:

0.2195 + 0.4391 𝑧 −1 + 0.2195 𝑧 −2
𝐻 𝑧 =
1 − 0.3105 𝑧 −1 + 0.1887 𝑧 −2

In final exam, no need to find the exact function 𝑯 𝒛

37
Design of Butterworth LPF using BLT
Matlab Implementation

[N, Wc] = buttord(Wpass, Wstop, Apass, Astop)

[b, a] = butter(N, Wc, ftype)
MATLAB: >> b
Apass = 1; b=
Astop = 20; 0.2195 0.4391 0.2195
Fpass = 1000;
Fstop = 3000; >> a
Fs = 8000; a=
Wpass = 2*Fpass/Fs; % without pi 1.0000 -0.3105 0.1887
Wstop = 2*Fstop/Fs; % without pi
[N, Wc] = buttord(Wpass, Wstop, Apass, Astop);
[b, a] = butter(N, Wc, 'low');
freqz(b, a, 512, Fs);

38
Design of Butterworth LPF using BLT
Matlab Implementation

39
Design of Butterworth LPF using BLT
Example 2:
Design a second-order digital lowpass Butterworth filter with a
cut-off frequency of 1kHz and a sampling rate of 10 kHz.
Solution:
In this example, N = 2 and the cut-off frequency are already given.

1. Calculate the digital cutoff frequency 𝜔𝑐

2𝜋𝐹𝑐 2𝜋 1𝑘
𝜔𝑐 = = = 0.2𝜋 rad/sample
𝐹𝑠 10𝑘
2. Pre-warp 𝜔𝑐 to get the new analog Ω𝑐
2 𝜔𝑐 2 0.2𝜋
Ω𝑐 = tan = −4 tan = 6498.4 rad/sec
𝑇 2 10 2
40
Design of Butterworth LPF using BLT
3. The filter order is given as N = 2.
4. The second-order Butterworth prototype is given by:
1
𝐻𝑝 (𝑠) = 2
𝑠 + 1.414𝑠 + 1

5. The analog 3-dB cut-off frequency Ω𝑐 = 6498.4 rad/sec

6. Find the Transfer Function, 𝐻𝑎 (𝑠) of the analog LPF.

Ω𝑐 2
𝐻𝑎 𝑠 = 𝐻𝑝 (𝑠)ቚ 𝑠 = 2
𝑠=Ω
𝑐 𝑠 2 + 1.414Ω𝑐 𝑠 + Ω𝑐

42.23 × 106
𝐻𝑎 𝑠 = 2
𝑠 + 9.19 × 103 𝑠 + 42.23 × 106
41
Design of Butterworth LPF using BLT
7. Finally, use the BLT to obtain the digital filter 𝐻(𝑧).
𝐻 𝑧 = 𝐻𝑎 𝑠 ቚ 2 1−𝑧 −1
𝑠=
𝑇 1+𝑧 −1

42.23 × 106
𝐻 𝑧 = 2
2 1 − 𝑧 −1 2 1 − 𝑧 −1
+ 9.19 × 103 + 42.23 × 106
𝑇 1 + 𝑧 −1 𝑇 1+𝑧 −1

0.0675 + 0.1349 𝑧 −1 + 0.0675 𝑧 −2

𝐻 𝑧 =
1 − 1.143 𝑧 −1 + 0.4128 𝑧 −2

42
Design of Butterworth LPF using BLT
Matlab Implementation

[b, a] = butter(N, Wc, ftype)

MATLAB: >> b
N = 2; b=
Fc = 1000; 0.0675 0.1349 0.0675
Fs = 10000;
Wc = 2*Fc/Fs; % without pi >> a
[b, a] = butter(N, Wc, 'low'); a=
freqz(b, a, 512, Fs); 1.0000 -1.1430 0.4128

43
Design of Butterworth LPF using BLT
Matlab Implementation

44
Matlab Commands
buttord Butterworth filter order and cutoff frequency
butter Butterworth filter design
cheb1ord Chebyshev Type I filter order
cheby1 Chebyshev Type I filter design
cheb2ord Chebyshev Type II filter order
cheby2 Chebyshev Type II filter design
ellipord Minimum order for elliptic filters
fir1 Window-based FIR filter design

45