Deparment of Electrical, Computer and

Telecommunications Engineering
Botswana International University of
Science and Technology
Digital Signal Processing(Code: EEEN 414)
Lab 1: Filter Design and Simulation
Issued Date
29 August 2023
Duration:3 weeks Semester 1 Maximum Marks: 100

1. Objectives of this Lab

This lab
• introduces you to filter design and testing techniques using MATLAB/Simulink;
• gives you a deeper understanding of the filter design techniques automated in MATLAB
via the Filter Design & Analysis Tool(FDAT);
2. Prelab
Prior to beginning this lab, you must have carefully read over this lab in order to plan how to
implement the tasks assigned. Make sure you are aware of all the items you need to include in
your lab report.
3. Grading and Due Date
You should have enough time to do the lab in class, but if you do not, please note that you
can come in on your own time to complete the lab requirements. The lab will be graded on
correctness, comprehensiveness and the insight you are able to provide when answering the
questions. For full points, lab reports should be written in complete sentences with correct
grammar. Please note STRICT DEADLINE for report on the course Teams site.
4. Part 1: Filter Design
In this lab we will design digital frequency-selective filters to be used to remove or filter out
noise from a signal. Recall the concept of the ideal lowpass filter, which is simply visualized
in the frequency domain as the rectangle function (of a specified width and centered at the
origin). Also recall that this filter, although handy in theory, is not implemented in practice.
There are several reasons why ideal (lowpass, bandpass, highpass, and band-stop) filters are
not used in real-life:
1. The impulse response, h[n] is non-causal as a consequence of the Paley-Wiener theorem,
which implies these filters cannot be implemented in practice on a DSP. Another way
to think about this is that h[n] has infinite support for any of the ideal filters. Thus, it
cannot be zero for n < 0 making it necessarily non-causal.
2. When implemented in software or hardware, due to the finite number of elements employed
for processing, an ideal filter exhibits the undesirable Gibbs phenomenon. In signal and
image processing, this shows up as the infamous ringing effect, i.e. it introduces extra
unwanted artifacts.
To bypass the above inadequacies, the design of digital filters using the windowing technique
is an alternative to ideal filters. These filters can be implemented as FIR filters, and make use
of the well-known Bartlett, Blackman, Hamming, Hanning, and Kaiser windows.

Student’s name: Please go on to the next page. . .

TEL. Engineering Digital Signal Processing Page 2 of 6

The shortcoming of digital filter design via windowing techniques is that one cannot individually
control the design parameters of the filter. For example, in Figure 2, ωp (pass-band frequency)
and ωs (stop-band frequency) cannot be independently controlled using the windowing technique.

Figure 1: Filter Specification.

By making the transition band (frequencies between ωp and ωs ) narrower, you must make a
sacrifice in the form of larger undesirable ripples in the pass-band (frequencies between 0 and
ωp ) and stop-band (frequencies exceeding ωs ); that is smaller ωs − ωp necessitates δ1 and δ2
larger These tradeoffs are all due to bypassing the two inadequacies of ideal filters. The above
specifications are general and do not delineate the form of the digital filter. In practice we
would prefer either a rational system function corresponding to FIR or IIR filters discussed
in the lectures. Thus we consider using the difference equation, or ARMA (autoregressive
moving average) model given in equations (1) and (2) representing the descriptions in the
time-and frequency-domain, respectively.
N
X M
X
y[n] = − ak y[n − k] + bk x[n − k] (1)
k=1 k=0

PM
bk z −k
H(z) = Pk=0
N −k
(2)
k=1 ak

Also IIR filters are generally less complex than FIR filters (the difference being that for FIR
filters there is a restriction that ak = 0 for all k), as they require fewer parameters and
less memory for the same “quality” of filter performance. IIR filters can often be defined
analytically as a rational function in the z-transform domain, and the general description of
the filter in Figure 2 is not z-transform friendly (yet) - it’s analog. Fortunately one can design
a filter meeting Figure 2 specifications, and then transform the resulting filter to a z-transform-
friendly filter by using popular transformation techniques such as approximation of derivatives,
impulse response, bilinear transformation (most popular), and matched z-transform techniques.
Luckily, MATLAB has a nice little GUI driven filter design program, which requires we to do
no more than click a few buttons ...

5. Design and Implementation: FDA Tool

Student’s name: Question 5 continues on the next page. . .

TEL. Engineering Digital Signal Processing Page 3 of 6

Figure 2: Filter Design and Analysis Tool.

In this section, you will learn how to use MATLAB’s handy Filter Design & Analysis
Tool (FDAT). To begin, start MATLAB. Now, enter fdatool into the command window. This
should, when executed, bring up the FDAT’s graphical user interface (GUI), shown in Figure 2
below. The process of designing a filter is fairly self-explanatory: you simply set all of the filter
specifications in the lower half of the GUI. When you are satisfied with your specifications,
click on the Design Filter button. The magnitude response of the resulting filter will appear
in the Magnitude Response pane. Note that you can view the coefficients of the filter’s transfer
function in second order sections by clicking on the Filter coefficients button at the top of the
GUI (which looks like [b,a]).
One useful feature of FDAT is that you can store multiple filters at once and switch between
them as you wish. After you have designed a filter, you can store it by selecting the Store
Filter button; this will prompt you to enter a name for the filter. Once you have stored the
filter, you can begin designing a new filter by choosing new filter specifications. To access
previously stored filters, click on the Filter Manager button. Finally, you can save all of your
stored filters in one “session” by selecting File → Save Session As; this will save your session
with a .fda extension. You can always open previously saved sessions in the FDAT GUI.
Another cool feature of FDAT that we will be using is to export filters to a Simulink model as a
single-input, single-output block. To do this, make sure the filter you want is currently shown
in the GUI (if not, switch to it using the Filter Manager). Then, click on File → Export to
Simulink Model; a new set of options should appear in the lower half of the GUI. Give the filter
a good, descriptive Block Name and make sure the Destination is Current. Press the Realize
Model Button (this only works if you have a Simulink Model currently open, of course).
Designing the Filters

Student’s name: Question 5 continues on the next page. . .

TEL. Engineering Digital Signal Processing Page 4 of 6

Before implementing anything, a good engineer always makes sure not to bypass the design
stage. In FDAT, perform the following tasks, assuming a sampling rate of 8 kHz:

1. Design a minimum order, stable1 (check if the filter is stable using information ’i’ button),
lowpass Butterworth filter with a passband frequency of 1 kHz and a stopband frequency
of 1.4 kHz. Make the attenuation 1 dB at the passband frequency and 80 dB at the
stopband frequency.
2. Design a minimum order, stable, lowpass Chebyshev Type I filter with the same specifications
as the Butterworth filter.
3. Design a lowpass FIR filter using the Blackman Window with a cutoff frequency of 1 kHz.
Specify the order of the filter such that the first minimum in the stopband (preceding the
first lobe) is as close to 1.4 kHz as possible without exceeding it.

Discussion Questions

1. What is the order of the lowpass Butterworth filter you designed?

2. What is the order of the lowpass Chebyshev Type I filter you designed?
3. Compare the memory usage of each of the 3 filters assuming a Direct Form II realization
is employed. Do you see how inefficient the windowing technique is? How much more
expensive in terms of memory is the windowing technique from the best IIR filter?
Designing the Filters in Simulink

Figure 3: Filter Testing in Simulink.

In this section, you’ll simulate each one of the filters you designed for a finite duration in
non-real-time. Open a new Simulink model and construct a block diagram similar to that
shown in Figure 3. Before your begin, please note the following about the block diagram:
(a) The Digital Filter block labeled “Filter” is the result of exporting a filter from
FDAT to Simulink; this is where you’ll place the filters you designed.
(b) The input to the filter is a noisy sinusoid. The high frequency noise is created in
the same way as follows: Add a Uniform Random Number block to your model. Set
Sample time to 1/8000 and Maximum and Minimum to 5 and -5, respectively. As
its name suggests, this block outputs random numbers, which for our purposes is a
good model for noise. That is, the random number generator must generate numbers
1
You can verify that a filter is stable after you have designed it by clicking on the Filter information button

Student’s name: Question 5 continues on the next page. . .

TEL. Engineering Digital Signal Processing Page 5 of 6

between -5 and 5. The highpass filter block should have an impulse response that is
IIR, a minimum order, frequency in units of Hz, a Fs of 8000, Fstop of 3600, Fpass of
3800, Astop of 60 and Apass of 1. Please also note that the highpass filter will have to
be appropriately replaced to generate different types of noise in this lab.
(c) The Sine Wave source block is a continuous-time source. The Zero-Order Hold
block is used to convert it to discrete-time and thus acts like an analog-to-digital
converter. Make sure that the zero-order hold is set to the same sampling rate that
your filters were designed for.
(d) The Smoothing Filter is just a lowpass analog filter that interpolates the digital
output of the Digital Filter and thus acts like a digital-to-analog converter. You
can obtain the Analog Filter Design block. Recall from DSP theory that the
cutoff frequency of this filter should be half of the sampling frequency. Also, note
that you can choose whatever type of analog filter you wish (i.e. it doesn’t have
to be a Chebyshev Type II filter, as shown in the Figure). Finally, you’ll want to
choose a reasonably low filter order (like 4) because an analog filter tends to be the
computational bottleneck that impedes the real-time simulation greatly.
(e) For any scope output, to ensure that a greater number of points are stored and
hence displayed, on the scope window go to Parameters → Data History. Enter an
appropriate number for the “Limit data points to last” field.
In Simulink, go to Simulation → Configuration Parameters. Please keep in mind to be
sure that the solver type is variable-step. Also, choose ode45 for the solver.
Test each of the three filters you designed in the block diagram of Figure 3 (to change
filters, just delete the old one and import a different one from FDAT). Specifically, perform
the following tasks for each filter, noting the various filter responses in your report.
For every case, save and insert the oscilloscope displays on your report. Please provide
information on the specific frequencies used for the sinusoidal source and the filter parameters
used for generating the noise. Write down any observation that you may have about the
signals and interpret the obtained results.
(a) Generate and test sine waves with frequencies in the following regions. For each
frequency, set the stop time of the simulation so that you can see several periods of
the sine wave. For these tests, you can disconnect the noise from the input signal
and terminate it with a Terminator block.
i. Passband region
ii. Transition region
iii. Stopband region
(b) Reconnect the noise to the input signal. Change the frequency of the sine wave to
be in the passband and set the stop time of the simulation appropriately. Add noise
to the sine wave whose frequency is in the following three regions (Note: in order to
contain the noise to the correct region, you’ll have to change the filter after the noise
generator to the appropriate highpass, lowpass, or bandpass filter):
i. Passband region
ii. Transition region
iii. Stopband region
(c) To measure the effectiveness of your noise filtering approach, using basic Simulink
blocks, find a way to calculate the mean squared error between the input sine wave

Student’s name: Question 5 continues on the next page. . .

TEL. Engineering Digital Signal Processing Page 6 of 6

(without noise) and the output of the overall system (please note that if your filter
shifts the sine wave so it has a different shift/phase compared to the original source,
then you must reshift it back before you compute this error so that both the original
and filtered sinusoids are aligned). Please show how you implement this in your
report; you may need a different shift for each of your filters. Discuss the performance
of each filter for the various input sinusoid and noise considered. Explain in what
situations you have the best/worst mean square error and why.
Therefore, to test each of your filters, please make sure that the sine wave is set to a
frequency in the passband. Then, add noise to the sine wave whose frequency is in
the following three regions: passband, transition and stopband. Run a simulation for
each type of noise. Repeat this for all three filters you designed. For every case, save
and insert the oscilloscope displays on your report. Write down any observation that
you may have about the signals and interpret the your results.

End of Questions

Student’s name: End of Test