ECE 4213/5213

Homework 9
Fall 2020 Dr. Havlicek

Work the Projects and Questions in Chapter 7 of the course laboratory manual.
For your report, use the file LABEX7.doc from the course web site.
Work these questions only:
1. Q7.1
2. Q7.5 - Q7.9
3. Q7.13
4. Q7.14
5. Q7.16
6. Q7.20
7. Q7.22
8. Q7.23
9. Q7.25

Clarifications:
1. In designing practical digital FIR and IIR filters, you will often use formulas and/or
tables. They all involve approximations. Different authors make the approximations
slightly differently. So there are many similar but slightly different versions of the
formulas and tables available.
When you are doing a design, you must be careful not to mix formulas and/or tables
that are based on different versions of the approximations. For example, if you are
doing a design using formulas and tables from the notes, then don’t mix them with
formulas from the Lab manual, from the textbook, or from another book. . . unless you
are sure that all of the formulas are based on the same versions of the approximations!
In particular:
• For doing FIR window designs, the formulas in equations (7.27) through (7.29) on
page 114 of the Lab manual do not exactly match the similar versions that are
found on page 10.14 of the course notes and do not match the similar versions
that are found in Section 7.5 of the textbook.
For working this homework assignment, use the versions that are in the Lab
manual.
For the Final Exam, use the versions that are in the course notes.
 • The formula for estimating the order of a Kaiser window that is given in (7.37)
on page 115 of the Lab manual does not match the formula on page 10.20 of the
course notes (which is the same as Eq. (7.76) on page 542 of the textbook).
Again, for this homework assignment use the version that is in the Lab manual.
For the Final Exam, use the version that is in the notes.
2. In (7.5) and (7.6) on p. 110 of the Lab manual, the units of Ωp and Ωs are rad/sec,
while those for FT are samples/sec. Thus, the units of ωp and ωs are

rad/sec rad
= .
sample/sec sample

But the frequencies Wp and Ws that you need for Matlab routines like buttord are
“normalized” so that the Nyquist frequency is mapped to a “normalized” frequency of
ONE.
So to call routines like buttord, the Wp that you need can be obtained from the ωp in
(7.5) by:
• divide by 2π to convert from rad/sample to Hz/sample.
• multiply by 2 to convert the Nyquist rate, given by π rad/sample = 0.5 Hz/sample,
into ONE on the normalized frequency scale.
In other words, the overall conversion for taking ωp in (7.5) to the Wp that you need
for routines like buttord is to divide by π.
For example, in Q7.1 I’ve got FT = 40 kHz, Fp = 4 kHz, and Fs = 8 kHz. Plugging
into (7.5) I get ωp = 0.2π rad/sample and dividing by π I get Wp = 0.2 for calling
buttord.
3. If you look carefully at program P7 1 on p. 119 of the Lab manual, in the third line
you have Wp = [0.3 0.7]. This is a pair of passband edge frequencies for the bandstop
filter. Let’s call this version of Wp the “Book” version.
Now, if you look in the file P7 1.m available on the course web site, you see instead on
the third line Wp = [0.2 0.8]. Let’s call this the “Program” version of Wp .
For a bandstop filter, the Book version is more stringent. . . it allows transition bands
that are only 0.1 units of normalized frequency in width. The Program version is less
stringent, since it allows for transition bands that are 0.2 units of normalized frequency
in width.
Using either version, the program will work and it will design a Butterworth bandstop
filter that meets the specification. For the Book version (more stringent), you will get
N1=9 and the order of the filter is 18. For the Program version (less stringent), you
will get N1=5 and the order of the filter is 10.
To save typing, LET’S ALL GO WITH THE PROGRAM VERSION Wp = [0.2 0.8].
This means your numerator and denominator will have only six nonzero terms (instead
of 10).
4. In problem Q7.7 on page 120 of the Lab manual it says to modify program P7 1 to
design a Type 2 Chebyshev filter, but in the Lab report shell file it says “Type 1
 Chebyshev.” The Lab manual is CORRECT and the report file is INCORRECT. You
should modify the program to design a Type 2 Chebyshev highpass filter.
For Q7.7 (and for the rest of the designs in this assignment), you may find it helpful
to add horizontal and vertical lines to the magnitude response graph that show the
required specifications. This will make it easier to answer questions about whether or
not the obtained filter design meets the spec. If you add these lines, then you can use
the zoom feature of the Matlab figure window to carefully examine the transitions and
verify if the filter did or didn’t meet the spec.
For Q7.7, I added the following code to my program to draw the lines:

% Add lines to the plot to help determine if the spec was met.
hold on;
tmpY = -60:65/511:5;
tmpX = ones(1,length(tmpY))*Wp;
plot(tmpX,tmpY,’r-’); % vertical line at passband edge freq
tmpX = ones(1,length(tmpY))*Ws;
plot(tmpX,tmpY,’g-’); % vertical line at stopband edge freq
tmpY = ones(1,length(w))*(-Rp);
plot(w/pi,tmpY,’r-’); % horizontal line at Rp
tmpY = ones(1,length(w))*(-Rs);
plot(w/pi,tmpY,’g-’); % horizontal line at Rs

5. Question Q7.9 is a little but frustrating because it’s hard to figure out what is being
asked. Here’s some help on that.
First:
• read the question and all the material that goes with it in the text of the Lab
manual.
• read the Matlab help for sinc.
• read the page that comes up when you type “doc sinc” at the Matlab command
prompt.
At this point you are probably confused.
We are talking about an ideal lowpass filter with a cutoff frequency of 0.4π rad/sample.
That means the fundamental period of the frequency response is a BOXCAR that’s
ONE from −0.4π to 0.4π and zero outside of that interval.
Look at a table of DTFT pairs like the one on the “useful formulas” handout from the
course web site. The tenth pair in that table says that the inverse DTFT of BOXCAR
is SINC. Here, you have W = 0.4π. From this entry, the impulse response of our ideal
lowpass filter with cutoff frequency 0.4π is given by
sin(0.4πn)
h[n] = .
πn

Now, according to the Matlab help and “doc sinc,” the Matlab sinc function is
sin(πx)
sinc(x) = .
πx
 So, in terms of the Matlab sinc function, our ideal impulse response can be written as
h[n] = 0.4 * sinc(0.4*n)
to get
0.4 sin[π(0.4n)] sin(0.4πn)
h[n] = = . X
π(0.4n) πn
So you can get the length-81 zero-phase lowpass filter impulse response and frequency
response magnitude plot like this:

n = -40:40; % makes the length 81

Note: for observing the Gibbs phenomena, you need to use a linear plot as just demon-
strated, not a log plot like the Mitra supplied gain function gives you.
To get the other lengths, you need to shorten the impulse response by truncating
more samples from both ends of hn 81. You can do this by extracting CENTERED
subarrays from hn 81 like this:

hn_61 = hn_81(11:71) % drops 10 samples from each end

Even though you are shortening the length of h[n], you should still use the same 1024-
point “omega” vector to evaluate the frequency response and the same 1024-point “W”
vector to plot it.
Comparing the frequency response magnitude plots for all the lengths, you should see
that decreasing the order reduces the number of ripples and makes the transition band
wider. In other words, a higher order filter achieves a sharper transition with more
ripples. But the peak of the largest ripple is never reduced no matter what the order is.
This is the reason that so much work was done to develop windows like the Hamming
window and Kaiser window that have better properties than the rectangular “boxcar”
window.
6. You should use the Mitra-supplied function kaiord.m to estimate the filter order for
both Q7.13 and Q7.20.
In the file LABEX7.doc it says to use the Matlab signal processing toolbox routine
kaiserord to estimate the order for Q7.20. But that’s INCORRECT if Q7.13 and
Q7.20 are going to be consistent and comparable. So use kaiord.m for estimating the
order in both Q7.13 and Q7.20.
Here’s an explanation of the difference between kaiord and kaiserord and why you
should use kaiord for both Q7.13 and Q7.20:
 In these two questions, you are being asked to perform a window-based FIR design
with order estimated by the general Kaiser formula given in (7.7) on p. 110 of the Lab
manual and on p. 10.2 of the course notes. That’s what kaiord.m does.
The function kaiserord is different: it is specifically for an FIR windowed design using
a Kaiser window, e.g., (7.34) on p. 115 of the Lab manual and also on p. 10.19 of the
notes.
Now, it is true that you could use a Kaiser window to design a filter that would meet
the spec. But that’s not what questions Q7.13 and Q7.20 are asking you to do.
For my solution on Q7.20, I used kaiord to estimate the order and then called h =
fir1(N,Wn) to do an FIR design using the default window choice (Hamming window).
But this filter did not meet the spec (it failed in both the passband and the stopband).
To meet the spec, I had to increase the order up to N = 66.
7. For Q7.22, again use kaiord to estimate the order (do not use remezord, even though
that’s what LABEX7.doc tells you to do – remezord is obsolete).
This should give you N = 46, which will not meet the spec. Then, you can increase
the order by one to N = 47 and you will get a filter that does meet the spec.
Note: for doing Parks-McClellan FIR filter design with firpm, the routine firpmord
will give you a better estimate of the order than kaiord. For this problem, using the
call
[n,fo,mo,w] = firpmord([2000 2500],[1 0],[0.005 0.005],10000)
instead of kaiord would have estimated the order correctly as N = 47 and would also
have given you all the parameters that are needed to call firpm. The reason I asked
you to use kaiord instead for this problem was to get you to go through the exercise
of setting up the parameters for firpm manually without using firpmord.
8. Q7.23 tells you to use formulas (7.36) and (7.37) from page 115 of the Lab manual to
estimate β and the filter order, so do that. But in the future you will probably prefer
to instead call the signal processing toolbox function kaiserord to estimate them for
you. It will give a better estimate of the order.
But for this problem, you will need to implement formulas (7.35) and (7.37) in your
Matlab program. Here’s how I implemented (7.37):
if As > 21
N = ceil((As-7.95)*2/(14.36*(abs(Wp-Ws)))+1)
else
N = ceil(0.9222*2/abs(Wp-Ws)+1)
end
The only thing that’s a little bit tricky about that is understanding why I multiplied
by 2 in the numerator. Here’s the reason: ∆f in the formula is given by the difference
between the passband edge frequency and stop band edge frequency. In Matlab, those
are Wp and Ws. But in Matlab, Wp and Ws are specified in normalized frequency –
which is radians per sample divided by π. Now, (7.37) needs for ∆f to be in units
of Hz. Normally, to convert radian frequency to Hz you divide by 2π. But Wp and
Ws in Matlab are already divided by π. So you just need to divide them by 2. Since
abs(Wp-Ws) appears on the bottom in (7.37), I implemented that by multiplying the
numerator by 2 instead.
Submit this assignment electronically on Canvas.

DUE: 12/1/2020 11:59 PM