Solution Manual For Additional Problems For Signals and Systems Using Matlab Luis F. Chaparro and Aydin Akan

Solution Manual for Additional
Problems for
SIGNALS AND SYSTEMS
USING MATLAB
Luis F. Chaparro and Aydin Akan

Copyright 2018, Elsevier, Inc. All rights reserved.
Chaparro-Akan — Signals and Systems using MATLAB 0.2

Chapter 0
From the Ground Up
0.1 Basic Problems
0.1 Consider the following problems about trigonometric and polar forms.
(a) Let z = 6ejπ/4 find (i) Re(z), (ii) Im(z)
(b) If z = 8 + j3 and v = 9 − j2, is it true that
(i) Re(z) = 0.5(z + z ∗ )? (ii) Im(v) = −0.5j(v − v ∗ )?

(iii) Re(z + v ∗ ) = Re(z + v)? (iv) Im(z + v ∗ ) = Im(z − v)?
√ √
Answers: (a) Re(z) = 3 2; Im(z) = 3 2; (b) Yes to all.
Solution
(a) z = 6ejπ/4 = 6 cos(π/4) + j6 sin(π/4)
√
i. Re(z) = 6 cos(π/4) = 3 2
√
ii. Im(z) = 6 sin(π/4) = 3 2
(b) i. Yes, Re(z) = 0.5(z + z ∗ ) = 0.5(2Re(z)) = Re(z) = 8
ii. Yes, Im(v) = −0.5j(v − v ∗ ) = −0.5j(2jIm(v)) = Im(v) = −2
iii. Yes, Re(z + v ∗ ) = Re(Re(z) + Re(v ∗ ) + Im(z) − Im(v)) = Re(z + v) = 17
iv. Yes, Im(z + v ∗ ) = Im(17 + j5) = Im(z − v) = Im(−9 + j5) = 5
1
0.2 Using the vectorial representation of complex numbers it is possible to get some interesting inequalities.
(a) Is it true that for a complex number z = x + jy we have that |x| ≤ |z|? Show it geometrically by
representing z as a vector.
(b) The so called triangle inequality says that for any complex (or real) numbers z and v we have that
|z + v| ≤ |z| + |v|. Show this geometrically.
(c) If z = 1 + j and v = 2 + j is it true that
(i) |z + v| ≤ |z| + |v|? (ii) |z − v| ≤ |z| + |v|?
Answer: (a) |x| = |z|| cos(θ)| and since | cos(θ)| ≤ 1 then |x| ≤ |z|; (c) yes to both.
Solution
(a) Representing the complex number z = x + jy = |z|ejθ then |x| = |z|| cos(θ)| and since | cos(θ)| ≤ 1
then |x| ≤ |z|, the equality holds when θ = 0 or when z = x, i.e., it is real.
(b) Adding two complex numbers is equivalent to adding two vectors to create a triangle with two sides
the two vectors being added and the other side the vector resulting from the addition. Unless the two
vector being added have the same angle, in which case |z| + |v| = |z + v|, it holds that |z| + |v| > |z + v|.
!z + !v !v
!z
Figure 1: Problem 2: Addition of two vectors illustrating the triangular inequality.
(c) The answer to both is yes. Indeed,

√ √ √
(a) |z + v| = 13 ≤ |z| + |v| = 2 + 5
√ √
(b) |z − v| = | − 1| = 1 ≤ |z| + |v| = 2 + 5

0.3 Use Euler’s identity in the following problems.

(a) Find trigonometric identities in terms of sin(α), sin(β), cos(α), cos(α) for
(i) cos(α + β) (ii) sin(α + β)
(b) Is it true that Z 1

ej2πt dt = 0 ?
0
(c) Is it true that

i. (−1)n = cos(πn) for any integer n?
ii. ej0 + ejπ/2 + ejπ + ej3π/2 = 0 ? (sketch a figure)
Answer: (a) cos(α + β) = cos(α) cos(β) − sin(α) sin(β); (c) Yes to both.
Solution
(a) i. 2 cos(α + β) = ej(α+β) + e−j(α+β) = (ejα ejβ ) + (ejα ejβ )∗ = 2Re(ejα ejβ ) and
Re[ejα ejβ ] = Re[(cos(α) + j sin(α))(cos(β) + j sin(β))] = cos(α) cos(β) − sin(α) sin(β) so that
cos(α + β) = cos(α) cos(β) − sin(α) sin(β)
ii. 2j sin(α + β) = ejα ejβ − (ejα ejβ )∗ = 2jIm[ejα ejβ ], and the imaginary is
sin(α) cos(β) + cos(α) sin(β) = sin(α + β)
(b)
1
ej2πt 1 ej2π − 1
Z
ej2πt dt = |0 = =0
0 j2π j2π
also Z 1 Z 1 Z 1
ej2πt dt = cos(2πt)dt + j sin(2πt)dt = 0 + j0
0 0 0
since the integrals of the sinusoids are over a period.
(c) i. Yes, (−1)n = (ejπ )n = ejnπ = cos(nπ) + j sin(nπ) = cos(nπ) since sin(nπ) = 0 for any integer
n.
ii. Yes, ej0 = −ejπ and ejπ/2 = −ej3π/2 so they add to zero.

0.4 Consider the following problems related to computation with complex numbers.
(a) Find and plot all roots of
(i) z 3 = −1, (ii) z 2 = 1 (iii) z 2 + 3z + 1 = 0
(b) Suppose you want to find the natural log of a complex number z = |z|ejθ , which is
log(z) = log(|z|ejθ ) = log(|z|) + log(ejθ ) = log(|z|) + jθ
If z is negative it can be written as z = |z|ejπ and we can find log(z) by using the above derivation.
The natural log of any complex number can be obtained this way also. Justify each one of the steps
in the above equation and find
(i) log(−2), (ii) log(1 + j1), (iii) log(2ejπ/4 )
(c) Let z = −1 find (i) log(z), (ii) elog(z) .

(d) Let z = 2ejπ/4 = 2(cos(π/4) + j sin(π/4))
(i) find z 2 /4, (ii) is it true that (cos(π/4) + j sin(π/4))2 = cos(π/2) + j sin(π/2) = j?
Answers: (a) (i) zk = ejπ(2k+1)/3 ; (b) log(2ejπ/4 ) = log(2) + jπ/4; (c) elog(z) = −1; (d) (i) z 2 /4 = j.
Solution
(a) We have
i. z 3 = −1 = ejπ(2k+1) for k = 0, 1, 2, so roots are zk = ejπ(2k+1)/3 , k = 0, 1, 2, i.e., on a circle of
unit radius and separated π/3 with one at −1.
ii. z 2 = 1 = ej2πk for k = 0, ±1, ±2, · · · , so roots are zk = ejπk =, i.e., on a circle of unit radius
and separated π , one at zero and the other at −1
p
iii. z 2 + 3z + 1 = (z + 1.5)2 + (1 − 1.52 ) = 0 so the roots are z1,2 = ± (1.52 − 1) − 1.5
(b) The log (using the Naperian base) of a product is the sum of the logs of the terms in the product, and
the log and the exponential are the opposite of each other so the last term in the equation.
Using the given expression for the log of a complex number (log of real numbers is a special case):
log(−2) = log(2e±jπ ) = log(2) ± jπ

√ √
log(1 + j1) = log( 2ejπ/4 ) = log( 2) + jπ/4 = 0.5 log(2) + jπ/4
log(2ejπ/4 ) = log(2) + jπ/4
(c) z = −1 = 1ejπ thus

i. log(z) = log(1ejπ ) = log(1) + jπ = jπ
ii. From above result, elog(z) = ejπ = −1
(d) i. z 2 /4 = (2ejπ/4 )2 /4 = 4ejπ/2 /4 = j
ii. Yes, (cos(π/4) + j sin(π/4))2 = (z/2)2 , and (z/2)2 = (cos(π/4) + j sin(π/4))2 = cos(π/2) +
j sin(π/2) = j

0.2 Problems using MATLAB

0.5 Sampling — Consider a signal x(t) = 4 cos(2πt) defined for −∞ < t < ∞. For the following values of
the sampling period Ts generate a discrete-time signal x(n) = x(nTs ) = x(t)|t=nTs .
(i) Ts = 0.1, (ii) Ts = 0.5, (iii) Ts = 1
Determine for which of the given values of Ts the discrete-time signal has lost the information in the
continuous-time signal. Use MATLAB to plot x(t) (use Ts = 10−4 and the plot function) and the resulting
discrete-time signals (use the stem function). Superimpose the analog and the discrete-time signals for
0 ≤ t ≤ 3; use subplot to plot the four figures as one figure. You also need to figure out how to label the
different axes and how to have the same scales and units.
Answers: For Ts = 1 the discrete signal has lost information.
Solution
As we will see later, the sampling period of x(t) with a frequency of Ωmax = 2πfmax = 2π should satisfy
the Nyquist sampling condition
1
fs = ≥ 2fmax = 2 samples/sec
Ts
so Ts ≤ 1/2 (sec/sample). Thus when Ts = 0.1 the continuous-time and the discrete-time signals look
very much like each other, indicating the signals have the same information — such a statement will be
justified in the chapter on sampling where we will show that the continuous-time signal can be recovered
from the sampled signal. It is clear that when Ts = 1 the information is lost. Although it is not clear
from the figure that when we let Ts = 0.5 the discrete-time signal keeps the information, this sampling
period satisfies the Nyquist sampling condition and as such the original signal can be recovered from the
sampled signal. The following MATLAB script is used.
% Pr. 0.5
clear all; clf
T=3; Tss= 0.0001; t=[0:Tss:T];
xa=4*cos(2*pi*t); % continuous-time signal
xamin=min(xa);xamax=max(xa);
figure(1)
subplot(221)
plot(t,xa); grid
title(’Continuous-time Signal’); ylabel(’x(t)’); xlabel(’t sec’)
axis([0 T 1.5*xamin 1.5*xamax])
N=length(t);
for k=1:3,
if k==1,Ts= 0.1; subplot(222)
t1=[0:Ts:T]; n=1:Ts/Tss: N; xd=zeros(1,N); xd(n)=4*cos(2*pi*t1);
plot(t,xa); hold on; stem(t,xd);grid;hold off
axis([0 T 1.5*xamin 1.5*xamax]); ylabel(’x(0.1 n)’); xlabel(’t’)
elseif k==2, Ts=0.5; subplot(223)
plot(t,xa); hold on; stem(t,xd); grid; hold off
axis([0 T 1.5*xamin 1.5*xamax]); ylabel(’x(0.5 n)’); xlabel(’t’)
else,Ts=1; subplot(224)
plot(t,xa); hold on; stem(t,xd); grid; hold off
axis([0 T 1.5*xamin 1.5*xamax]); ylabel(’x(n)’); xlabel(’t’)

end
end
Analog Signal
6 6
4 4
2 2
x(0.1 n)
x(t)
0 0
−2 −2
−4 −4
−6 −6
0 1 2 3 0 1 2 3
t sec t
6 6
4 4
2 2
x(0.5 n)
x(n)
0 0
−2 −2
−4 −4
−6 −6
0 1 2 3 0 1 2 3
t t
Figure 2: Problem 5: Analog continuous-time signal (top left); continuous-time and discrete-time
signals superposed for Ts = 0.1 sec (top right) and Ts = 0.5 sec and Ts = 1 sec (bottom left to right).

0.6 Differential and difference equations — Find the ordinary differential equation relating a
current source is (t) = cos(Ω0 t) with the current iL (t) in an inductor, with inductance L = 1
Henry, connected in parallel with a resistor of R = 1 Ω (see Fig. 3). Assume a zero initial
current in the inductor.
iL (t)
is (t) 1Ω 1H
Figure 3: Problem 6: RL circuit: input is (t), output iL (t).
(a) Obtain a discrete equation from the ordinary differential equation using the trapezoidal
approximation of an integral.
(b) Create a MATLAB script to solve the difference equation for Ts = 0.01 and for three fre-
quencies for is (t), Ω0 = 0.005π, 0.05π and 0.5π. Plot the input current source is (t) and the
approximate solution iL (nTs ) in the same figure. Use the MATLAB function plot. Use the
MATLAB function filter to solve the difference equation (use help to learn about filter).
(c) Solve the ordinary differential equation using symbolic MATLAB when the input fre-
quency is Ω0 = 0.5π. Compare is (t) and iL (t).
Answer: diL (t)/dt + iL (t) = is (t)
Solution (a) According to Kirchoff’s current law
vL (t)
is (t) = iR (t) + iL (t) = + iL (t)
R
but vL (t) = LdiL (t)/dt so that the ordinary differential equation relating the input is (t) to the
output current in the inductor iL (t) is
diL (t)
+ iL (t) = is (t)
dt
after replacing L = 1 and R = 1. Notice that this d.e. is the dual of the one given in the Chapter,
so that the difference equation is
Ts 2 − Ts
iL (nTs ) = [is (nTs ) + is ((n − 1)Ts )] + iL ((n − 1)Ts ) n≥1
2 + Ts 2 + Ts
iL (0) = 0
(b)(c) The scripts to solve the difference and ordinary differential equations are the following.
% Pr. 0.6
clear all

cos(1/2 π t)
1
input current 1
is(t),iL(t) output current
0.5 filter gain 0.5
0
0
0 10 20 30 40 50 60 70 80 90 100
t −0.5
1 −1
input current
is(t),iL(t)
output current 0 10 20 30 40 50 60 70 80 90 100

0 filter gain t
−...+2 (2 cos(1/2 π t)+π sin(1/2 π t))/(4+π2)
−1 1
0 10 20 30 40 50 60 70 80 90 100
t 0.5
1
input current 0
is(t),iL(t)
output current
0 filter gain
−0.5
−1
0 10 20 30 40 50 60 70 80 90 100 −1
t 0 10 20 30 40 50 60 70 80 90 100
t
Figure 4: Problem 6: Left (top to bottom): solution of difference equation for Ω0 = 0.005, 0.05, 0.5
(rad/sec). Right: input (top), solution of ordinary differential equation (bottom).
% solution of difference equation

Ts=0.01;
t=[0:Ts:100];
figure(4)
for k=0:2;
if k==0, subplot(311)
elseif k==1, subplot(312)
else, subplot(313)
end
W0= 0.005*10ˆk*pi; % frequency of source
is=cos(W0*t); % source
a=[1 (-2+Ts)/(2+Ts)]; % coefficients of i_L(n), i_L(n-1)
b=[Ts/(2+Ts) Ts/(2+Ts)]; % coefficients of i_s(n), i_s(n-1)
il=filter(b,a,is); % current in inductor computed by
% MATLAB function ’filter’
H=1/sqrt(1+W0ˆ2)*ones(1,length(t)); % filter gain at W0
plot(t,is,t,il,’r’,t,H,’g’); xlabel(’t’); ylabel(’i_s(t),i_L(t)’)
axis([0 100 1.1*min(is) 1.1*max(is)])
legend(’input current’,’output current’,’filter gain’); grid
pause(0.1)
end
%%%%
% solution of ordinary differential equation for cosine input of frequency 0.5pi
clear all
syms t x y
x=cos(0.5*pi*t);
y=dsolve(’Dy+y=cos(0.5*pi*t)’,’y(0)=0’,’t’)
figure(5)
subplot(211)
ezplot(x,[0 100]);grid
subplot(212)
ezplot(y,[0 100]);grid
axis([0 100 -1 1])

0.7 More integrals and sums — Although sums behave like integrals, because of the discrete
nature of sums one needs to be careful with the upper and lower limits more than in the integral
case. To illustrate this consider the separation of an integral into two integrals and compare it
with the separation of a sum into two sums. For the integral we have that
Z 1 Z 0.5 Z 1
tdt = tdt + tdt
0 0 0.5
Show that this this true by computing the three integrals. Then consider the sum
100
X
S= n
n=0
find this sum and determine which of the following is equal to this sum
50
X 100
X 50
X 100
X
(i) S1 = n+ n, (ii) S2 = n+ n
n=0 n=50 n=0 n=51
Use symbolic MATLAB function symsum to verify your results.

Answers: S = 5050, S1 = 5100, S2 = 5050.
Solution
The indefinite integral equals 0.5t2 . Computing it in [0, 1] gives the same value as the sum of
the integrals computed between [0, 0.5] and [0.5, 1].
As seen before, the sum
100
X 100(101)
S= n= = 5050
n=0
2
while
S1 = S + 50 = 5100
S2 = S
the first sum has an extra term when n = 50 while the other does not. To verify this use the
following script:
% Pr. 0.7
clear all
N=100;
syms n,N
S=symsum(n,0,N)
S1=symsum(n,0,N/2)+symsum(n,N/2,N)
S2=symsum(n,0,N/2)+symsum(n,N/2+1,N)
giving
S = 5050
S1 = 5100
S2 = 5050

0.8 Complex functions of time — Consider the complex function x(t) = (1 + jt)2 for −∞ < t < ∞.
(a) Find the real and the imaginary parts of x(t) and carefully plot them with MATLAB. Try
to make MATLAB plot x(t) directly, what do you get? Does MATLAB warn you? Does it
make sense?.
(b) Compute the derivative y(t) = dx(t)/dt and plot its real and imaginary parts, how do
these relate to the real and the imaginary parts of x(t)?
(c) Compute the integral
Z 1
x(t)dt
0
(d) Would the following statement be true? (remember ∗ indicates complex conjugate)
Z 1 ∗ Z 1
x(t)dt = x∗ (t)dt
0 0
R1
Answers: 0
x(t)dt = 2/3 + j1; statement is true.
Solution
(a)(b) Since
x(t) = (1 + jt)2 = 1 + j2t + j 2 t2 = 1 − t2 +j |{z}
2t
| {z }
real imag.
its derivative with respect to t is
dx(t) dRe[x(t)] dIm[x(t)]

y(t) = = −2t + 2j = +j
dt dt dt
% Pr. 0.8
clear all
t=[-5: 0.001:5];
x=(1+j*t).ˆ2;
xr=real(x);
xi=imag(x);
figure(7)
subplot(211)
plot(t,xr); title(’Real part of x(t)’); grid
subplot(212)
plot(t,xi); title(’Imaginary part of x(t)’); xlabel(’Time’); grid
% Warning when plotting complex signals
figure(8)
disp(’Read warning. MATLAB is being nice with you, this time!’)
plot(t,x); title(’COMPLEX Signal x(t)?’); xlabel(’Time’)
When plotting the complex function x(t) as function of t, MATLAB ignores the imaginary part.
One should not plot complex functions as functions of time as the results are not clear when
using MATLAB. See Fig. 5 for plots.
(c) Using the rectangular expression of x(t) we have
Z 1 Z 1 Z 1 Z 1
t3 2t2 1 2
x(t)dt = (1 − t2 + 2jt)dt = (1 − t2 )dt + 2j t dt = t − +j 0 = + j1
0 0 0 0 3 2 3

(d) The integral

Z 1 Z 1 Z 1 Z 1
∗ 2 2 t3 2t2 1 2
x (t)dt = (1 − t − 2jt)dt = (1 − t )dt − 2j t dt = t − −j 0 = − j1
0 0 0 0 3 2 3
which is the complex conjugate of the integral calculated in (c). So yes, the expression is true.
Real part of x(t) COMPLEX Signal x(t)?

5 5
−5
0
−10
−15
−20 −5
−25
−5 −4 −3 −2 −1 0 1 2 3 4 5
−10
Imaginary part of x(t)
10
−15
5
0
−20
−5
−10 −25
−5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5
Time Time
Figure 5: Problem 8: Real and imaginary parts of x(t) (left); complex signal x(t) ignoring imaginary
part.

0.9 Hyperbolic sinusoids — In filter design you will be asked to use hyperbolic functions. In this
problem we relate these functions to sinusoids and obtain a definition of these functions so that
we can actually plot them.
(a) Consider computing the cosine of an imaginary number, i.e., use
ejx + e−jx
cos(x) =
2
let x = jθ and find cos(x). The resulting function is called the hyperbolic cosine or
cos(jθ) = cosh(θ)
(b) Consider then the computation of the hyperbolic sine sinh(θ), how would you do it? Care-
fully plot it as a function of θ.
(c) Show that the hyperbolic cosine is always positive and bigger than 1 for all values of θ
(d) Show that sinh(θ) = − sinh(−θ)
(e) Write a MATLAB script to compute and plot these functions between −10 and 10.
Answers: sinh(θ) = −j sin(jθ) and odd.
Solution
(a) If x = jθ
1 −θ
cos(jθ) = (e + eθ ) = cosh(θ)
2
(b) The hyperbolic sine is defined as
1 θ
sinh(θ) = (e − e−θ )
2
which is connected with the circular sine as follows
1 −θ
sin(jθ) = (e − eθ ) = j sinh(θ) ⇒ sinh(θ) = −j sin(jθ)
2j
(c) Since e±θ > 0 then cosh(θ) = cosh(−θ) > 0, the smallest value is for θ = 0 which gives
cosh(0) = 1
(d) Indeed,
1
sinh(−θ) = (e−θ − eθ ) = − sinh(θ)
2
% Pr. 0.9
clear all
theta=sym(’theta’);
x= 0.5*(exp(-theta)+exp(theta));
y= 0.5*(exp(theta)-exp(-theta));
figure(9)
subplot(211)
ezplot(x,[-10,10])
grid
subplot(212)
ezplot(y,[-10,10])
grid

1/2 exp(θ)+1/2 exp(−θ)

6000
5000
4000
3000
2000
1000
0
−10 −8 −6 −4 −2 0 2 4 6 8 10
θ
1/2 exp(θ)−1/2 exp(−θ)
2000
1000
−1000
−2000
−10 −8 −6 −4 −2 0 2 4 6 8 10
θ
Figure 6: Problem 9: cosh(θ) (top) and sinh(θ) (bottom).


Chapter 1
Continuous–time Signals
1.1 Basic Problems
1.1 Consider a finite support signal x(t) = t, 0 ≤ t ≤ 1, and zero elsewhere.

(a) Plot x(t + 1) and x(−t + 1). Add these signals to get a new signal y(t). Do it graphically and verify
your results analytically.
(b) How does y(t) compare to the signal Λ(t) = (1 − |t|) for −1 ≤ t ≤ 1 and zero otherwise? Plot them.
Compute the integrals of y(t) and Λ(t) for all values of t and compare them.
Answers: x(−t + 1) is x(−t) delayed by 1; y(t) is even, y(0) = 2.
Solution
(a) If x(t) = t for 0 ≤ t ≤ 1, then x(t + 1) is x(t) advanced by 1, i.e., shifted to the left by 1 so that x(0) = 0
occurs at t = −1 and x(1) = 1 occurs at t = 0.
x(t) x(t + 1)
1 1
t t
0 1 −1 0
x(−t) x(−t + 1)
1 1
t t
−1 0 0 1 2
Figure 1.1: Problem 1: Original signal x(t), shifted versions x(t + 1), x(−t) and x(−t + 1).
1
The signal x(−t) is the reversal of x(t) and x(−t + 1) would be x(−t) advanced to the right by 1. Indeed,
t x(−t + 1)
1 x(0)
0 x(1)
−1 x(2)
The sum y(t) = x(t + 1) + x(−t + 1) is such that at t = 0 it is y(0) = 2; y(t) = x(t + 1) for t < 0; and
y(t) = x(−t + 1) for t > 0. Thus,
y(t) = x(t + 1) = t + 1 0 ≤ t + 1 < 1 or − 1 ≤ t < 0

y(0) = 2
y(t) = x(−t + 1) = −t + 1 0 ≤ −t + 1 < 1 or 0 < t ≤ 1
or 
 t+1 −1 ≤ t < 0
y(t) = 2 t=0
−t + 1 0<t≤1

(b) Except for the discontinuity at t = 0, y(t) looks like the even triangle signal Λ(t), their integrals are
y(t)
−1 1 t
Figure 1.2: Problem 1: Triangular signal y(t) with discontinuity at the origin.
identical as the discontinuity of y(t) does not add any area.

1.2 Given the causal full-wave rectified signal x(t) = | sin(2πt)|u(t)

(a) Find the even component of x(t), call it xe (t) and plot it. Is xe (t) periodic? if so, what is its funda-
mental fundamental period Te ? Would the odd component of x(t) or xo (t) be periodic too?
(b) Are xe (t) and xo (t) causal signals? Explain.
Answers: xe (t) is periodic, xo (t) is not, both are non-causal.
Solution
(a) x(t) is called causal because it is zero for t < 0, it repeats every 0.5 sec. for t ≥ 0.
x(t)
1
···
0.5 1
xe (t)
0.5
··· ···
−0.5 0.5 1
xo (t)
0.5
−0.5 ···
··· 0.5 1
−0.5
Figure 1.3: Problem 2.
(b) Even component xe (t) = 0.5[x(t) + x(−t)] is periodic of fundamental period Te = 0.5. The odd
component is xo (t) = 0.5[x(t) − x(−t)] is not periodic.
(c) xe (t) and xo (t) are non-causal signals as they are different from zero for negative times.

1.3 Is it true that (if not true, give correct answer)
(a) for any positive integer k

Z k
ej2πt dt = 0?
0
(b) for a periodic signal x(t) of fundamental fundamental period T0

Z T0 Z t0 +T0
x(t)dt = x(t)dt
0 t0
for any value of t0 ? Consider, for instance, x(t) = cos(2πt).

(c) cos(2πt)δ(t − 1) = 1 ?
(d) If x(t) = cos(t)u(t) then dx(t)/dt = − sin(t)u(t) + δ(t) ?
(e) the integral
Z ∞
e−t u(t) δ(t − 2)dτ = e−2 ?

−∞
(f) If x(t) = cosh(t)u(t) = 0.5(et + e−t )u(t) is dx(t)/dt = sinh(t)u(t) + δ(t) ?

(g) that the power of the causal full-rectified signal x(t) = | sin(t)|u(t) is twice the power of its even
component xe (t) = 0.5[x(t) + x(−t)]?
Answers: (a) and (b) are true; cos(2πt)δ(t − 1) = δ(t − 1); (g) yes, Pxe = 0.5Px .
Solution
(a) Yes, expressing ej2πt = cos(2πt) + j sin(2πt), periodic of fundamental period T0 = 1, then the
integral is the area under the cosine and sine in one or more periods (which is zero) when k 6= 0 and
integer. If k = 0, the integral is also zero.
(b) Yes, whether t0 = 0 (first equation) or a value different from zero, the two integrals are equal as the
area under a period is the same. In the case x(t) = cos(2πt), both integrals are zero.
(c) It is not true, cos(2πt)δ(t − 1) = cos(2π)δ(t − 1) = δ(t − 1).
(d) It is true, considering x(t) the product of cos(t) and u(t) its derivative is
dx(t) d cos(t) du(t)

= u(t) + cos(t)
dt dt dt
= − sin(t)u(t) + cos(0)δ(t)
(e) Yes,
Z ∞ Z ∞
e−t u(t) δ(t − 2)dτ
−2
= e δ(t − 2)dτ
−∞ 0
= e−2
(f) Yes,
dx(t)
= 0.5[et u(t) + et δ(t)] + 0.5[−e−t u(t) + e−t δ(t)]
dt
= 0.5[et − e−t ]u(t) + δ(t) = sinh(t)u(t) + δ(t)

(g) The even component xe (t) is a periodic full-wave rectified signal of amplitude 1/2 and fundamental
period T1 = π.
Power of x(t)
Z π
1
Px = 0.5 x2 (t)dt
π 0
Power of xe (t)
Z π
1
Pxe = (0.5x(t))2 dt = 0.5Px
π 0

1.4 The signal

−t −2 ≤ t ≤ 0
x(t) =
t 0≤t≤2
can be written as x(t) = |t|p(t).

(a) Carefully plot x(t) and define p(t), then find y(t) = dx(t)/dt and carefully plot it.
(b) Calculate Z t
y(τ )dτ
−∞
and comment on how your result relates to x(t).

(c) Is it true that Z ∞ Z 2
x(τ )dτ = 2 x(τ )dτ ?
−∞ 0
Answers: y(t) = 2δ(t + 2) − u(t + 2) + 2u(t) − u(t − 2) − 2δ(t − 2); yes, it is true.
Solution
(a) See Fig. 4a
x(t) = |t| [u(t + 2) − u(t − 2)] Derivative
| {z }
p(t)
x(t)
t
−2 2
y(t)
1
(2)
t
−2 2
−1 (−2)
Figure 1.4: Problem 4
dx(t)
y(t) = = 2δ(t + 2) − u(t + 2) + 2u(t) − u(t − 2) − 2δ(t − 2)
dt
(b) Integral

 0 t < −2
t
Z 
−t −2 ≤ t < 0

0 0
y(t )dt =
−∞ 
 t 0≤t<2
0 t≥2

which equals x(t).

(c) Yes, because x(t) is an even function of t.


1.5 Sampling signal and impulse signal — Consider the sampling signal
∞
X
δT (t) = δ(t − kT )
k=0
which we will use in the sampling of analog signals later on.

(a) Plot δT (t). Find Z t
ssT (t) = δT (τ )dτ
−∞
and carefully plot it for all t. What does the resulting signal ss(t) look like? This signal has been
called the “stairway to the stars.” Explain.
(b) Use MATLAB function stairs to plot ssT (t) for T = 0.1. Determine what signal would be the limit as
T → 0.
(c) A sampled signal is
∞
X
xs (t) = x(t)δT (t) = x(kTs )δ(t − kTs )
k=0
Let x(t) = cos(2πt)u(t) and Ts = 0.1. Find the integral
Z t
xs (t)dt
−∞
and use MATLAB to plot it for 0 ≤ t ≤ 10. In a simple way this problem illustrates the operation of a
discrete to analog converter which converts a discrete-time into a continuous-time signal (its cousin
is the digital to analog converter or DAC).
Rt
Answers: (a) ss(t) = ∞
P P
k=0 u(t − kT ); (c) −∞ xs (t)dt = k x(kTs )u(t − kTs ).
Solution
(a) The sampling signal is a sequence of unit impulses at uniform times kT , for t > 0. The integral
Z t X ∞ ∞ Z t
X ∞
X
ssT (t) = δ(t − kT )dt = δ(t − kT )dt = u(t − kT )
−∞ k=0 k=0 −∞ k=0
This signal is called the “stairway to the stars” (ssT (t)) for obvious reasons.
(b)(c) The following script will display the signal ss(t) and the conversion to analog (just uncomment the
desired signal and comment the not desired signal).
% Pr. 1_5 parts (b) and (c)
clear all; clf
T=0.1; t=0:T:10;
%f=t; % part (b)
f=cos(2*pi*t); % part (c)
figure(6)
stairs(t,f);grid
%axis([0 10 0 10]) % part (b)
axis([0 10 -1.1 1.1]) % part (c)
hold on
plot(t,f,’r’)
hold off
When f (t) = t + 1 the output looks like a digitized line with unit slope and cut at 1 (see figure
below), similarly when f (t) = cos(2πt) the output looks like a digitized sinusoid.

10
1
9
0.8
8
0.6
7
0.4
6 0.2
5 0
4 −0.2
−0.4
3
−0.6
2
−0.8
1
−1
0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Figure 1.5: Problem 5: ‘Digitized’ ramp and cosine signals using ’stairway to stars’

Chapter 2
Continuous–time Systems
2.1 Basic Problems
2.1 The input-output relationship of a system is

(
1 if x(t) ≥ 0
y(t) = sign(x(t)) =
−1 if x(t) < 0
where x(t) is the input and y(t) the output.
(a) Let the input be x(t) = sin(2πt)u(t), plot the corresponding output y(t). What is the
output of the system if we double the input? That is, what is the output y1 (t) if the input
x1 (t) = 2x(t) = 2 sin(2πt)u(t)? Is the system linear or not?
(b) Is the given system time-invariant?
P∞
Answers: output y(t) = k=0 p(t − k) with p(t) = u(t) − 2u(t − 0.5) + u(t − 1).
Solution
(a) x(t) = sin(2πt)u(t) is zero for t < 0 and repeats periodically every T0 = 1. Thus,
∞
X
y(t) = p(t − k), p(t) = u(t) − 2u(t − 0.5) + u(t − 1)
k=0
If the input is 2x(t) the output is the same as before, so the system is non-linear.
(b) The system is time-invariant: if input x(t − τ ) the output is y(t − τ ) for any value of τ .
1
2.2 Consider the following problems about the properties of systems.
(a) A system is represented by the ordinary differential equation dz(t)/dt = w(t) − w(t − 1)
where w(t) is the input and z(t) the output.
i. How is this system related to an averager having an input/output equation
Z t
z(t) = w(τ )dτ + 2 ?
t−1
ii. Is the system represented by the given ordinary differential equation LTI?
(b) If we consider the current i(t) the input of a capacitor, with a zero initial voltage across it,
the voltage across the capacitor –or the output— is given by
Z t
vC (t) = i(τ )dτ
0
i. Is this capacitor time-invariant? If not, what conditions would you impose to make it
time-invariant?
ii. If we let i(t) = u(t) what is the corresponding voltage vC (t)? If we delay the current
source so that the input is i(t − 1), find the corresponding voltage across the capacitor
and indicate if this result shows the capacitor is time invariant.
(c) An amplitude modulation system has an input/output equation
y(t) = x(t) sin(2πt) −∞<t<∞
i. Let x(t) = u(t), plot the corresponding output y(t) = sin(2πt)u(t)

ii. If the input is delayed, i.e., the input is x1 (t) = x(t − 0.5) = u(t − 0.5), plot the
corresponding output y1 (t) = x1 (t) sin(2πt), and use this result to determine whether
the given AM system is time-invariant.
Answer: (a) z(t) solution of differential equation with initial condition of 2; (b) If i(t) = 0 for
t < 0, system is TI; (c) If x(t − 0.5) = u(t − 0.5) the output is y1 (t) = sin(2πt)u(t − 0.5).
Solution
(a) Derivative
dz(t)
= w(t) − w(t − 1)
dt
which excludes the initial condition of 2. System is LTI if initial condition is zero.
(b) i. If input is i(t − µ) then the output is letting η = τ − µ

Z t Z 0 Z t−µ
i(τ − µ)dτ = i(η)dη + i(η)dη = vc (t − µ)
0 −µ 0
that is, provided that i(t) = 0 for t < 0, the system is time-invariant.

Rt
ii. If i(t) = u(t) then vc (t) = 0
u(τ )dτ = r(t). If we shift the inputs i1 (t) = i(t − 1) =
u(t − 1) the previous output is shifted, so system is time-invariant.
(c) If x(t) = u(t) then y(t) = sin(2πt)u(t) while corresponding to x(t − 0.5) = u(t − 0.5) is
y1 (t) = sin(2πt)u(t−0.5) indicating the system is not time-invariant as y1 (t) is not y(t−0.5).
sin(2 π t) heaviside(t)
−0.5 0 0.5 1 1.5

t
heaviside(t − 1/2) sin(2 π t)
−0.5 0 0.5 1 1.5

t
Figure 2.1: Problem 2(c)

2.3 The following problems relate to linearity, time–invariance and causality of systems.
(a) A system is represented by the equation z(t) = v(t)f (t) + B where v(t) is the input, z(t)
the output, f (t) a function and B a constant.
i. Let f (t) = A, a constant. Is the system linear if B 6= 0? linear if B = 0? Explain.

ii. Let f (t) = cos(Ω0 t) and B = 0 is this system linear? time-invariant?
iii. Let f (t) = u(t)−u(t−1), the input v(t) = u(t)−u(t−1), B = 0, find the corresponding
output z(t). Let then the input be delayed by 2, i.e., the input is u(t − 2) − u(t − 3) and
f (t) and B be the same, determine the corresponding output. Using these results, is
the system time invariant?
(b) An averager is defined as

Z t
1
y(t) = x(τ )dτ
T t−T
where T > 0 is the averaging interval, x(t) and y(t) are the system input and output.
i. Determine if the averager is a linear system.

ii. Let T = 1, x(t) = u(t), calculate and plot the corresponding output, delay then the
input to get x(t − 2) = u(t − 2), and calculate and plot the corresponding output.
From this example, does the system seem time invariant? Explain. Can you show it
in general.
iii. Is this system causal? Give an example to verify your assertion.
Answer: (a) If f (t) = v(t) = u(t) − u(t − 1) then z(t) = u(t) − u(t − 1), and if input is
v1 (t) = u(t−2)−u(t−3) output is z1 (t) = v1 (t)f (t) = 0; (b) If x(t) = u(t) then y(t) = r(t)−r(t−1)
and if x1 (t) = u(t − 2) the corresponding output is y1 (t) = y(t − 2).
Solution
(a) i. z(t) = Av(t) + B, system is linear if B = 0, non-linear otherwise.

ii. z(t) = v(t) cos(Ω0 t) is linear but time-varying.
iii. If f (t) = v(t) = u(t) − u(t − 1), B = 0 then z(t) = u(t) − u(t − 1), and if we shift v(t)
so the input is v1 (t) = u(t − 2) − u(t − 3) the output is z1 (t) = v1 (t)f (t) = 0 which is
different from z(t − 2), so the system is time-varying.
(b) i. The system is linear:

Z t Z t Z t
1 A B
[Ax1 (τ ) + Bx2 (τ )]dτ = x1 (τ )dτ + x2 (τ )dτ = Ay1 (t) + By2 (t)
T t−T T t−T T t−T
where yi (t), i = 1, 2, are the corresponding outputs to xi (t),i = 1, 2.

ii. If x(t) = u(t) then y(t) = r(t) − r(t − 1) and if x1 (t) = u(t − 2) the corresponding
output is y1 (t) = y(t − 2). System is time-invariant.
In general, if x1 (t) = x(t − λ) the output is
Z t Z t−λ
1 1
x1 (τ ) dτ = x(ν)dν
T t−T | {z } T t−T −λ
x(t−λ)
(using ν = τ − λ) which is the same as y(t − λ), so time-invariant.

iii. y(t) depends on present and past inputs, and zero if input is zero, so the system is
causal.

2.4 Consider the analog averager,

Z t+T /2
1
y(t) = x(τ )dτ
T t−T /2
where x(t) is the input and y(t) the output.
(a) Find the impulse response h(t) of the averager. Is this system causal?
(b) Let x(t) = u(t), find the output of the averager.
Answer: h(t) = (1/T )[u(t + T /2) − u(t − T /2)]; non–causal system.
Solution
(a) Letting x(t) = δ(t) the impulse response is
Z t+T /2
1
h(t) = δ(τ )dτ
T t−T /2
Z t Z t+T /2
1 1
= δ(τ )dτ + δ(τ )dτ
T t−T /2 T t
If t > 0, and t − T /2 < 0 the first integral includes 0, while the second does not. Thus
Z
1 t 1
h(t) = δ(τ )dτ + 0 = t > 0 and t − T /2 < 0, or 0 < t < T /2
T t−T /2 T
Likewise when t < 0 then t − T /2 < −T /2 and t + T /2 < T /2 the reverse of the previous case
happens and so
Z t+T /2
1 1
h(t) = 0 + δ(τ )dτ = t < 0 and t + T /2 > 0, or − T /2 < t < 0
T t T
so that
1
[u(t + T /2) − u(t − T /2)]
h(t) =
T
indicating that the system is non-causal as h(t) 6= 0 for t < 0.
(b) If x(t) = u(t) then the output of the averager is
Z t+T /2
1
y(t) = u(τ )dτ
T t−T /2
If t + T /2 < 0 then y(t) = 0 since the argument of the unit step signal is negative. If t + T /2 ≥ 0
and t − T /2 < 0 then
Z t+T /2
1
y(t) = u(τ )dτ = (t + T /2)
0 T
and finally when t − T /2 ≥ 0 then
Z t+T /2
1
y(t) = u(τ )dτ = 1
T t−T /2

The unit-step response of the noncausal averager is



 0 t < −T /2
y(t) = 1
T (t + T /2) −T /2 ≤ t < T /2


1 t ≥ T /2

2.5 The input-output equation characterizing a system of input x(t) and output y(t) is
Z t
−2t
y(t) = e y(0) + 2 e−2(t−τ ) x(τ )dτ t≥0
0
and zero otherwise.
(a) Find the ordinary differential equation that also characterizes this system.
(b) Suppose x(t) = u(t) and any value of y(0),we wish to determine the steady state response
of the system/ Is the value of y(0) of any significance, i.e., do we get the same steady-state
response if y(0) = 0 or y(0) = 1? Explain.
(c) Compute the steady-state response when y(0) = 0, and x(t) = u(t). To do so first find the
impulse response of the system, h(t) using the given input–output equation, and then find
y(t) by computing the convolution graphically to determine the steady–state response.
(d) Suppose the input is zero, is the system depending on the initial condition BIBO stable?
Find the zero-input response y(t) when y(0) = 1. Is it bounded?
Answers: (a) dy(t)/dt + 2y(t) = 2x(t); (d) Yes.
Solution
(a) To find the differential equation let y(0) = 0, so that
Z t
y(t) = 2e−2t e2τ x(τ )dτ
0
and its derivative is

Z t
dy(t)
= −4e−2t e2τ x(τ )dτ + 2e−2t e2t x(t) = −2y(t) + 2x(t)
dt 0
giving the differential equation

dy(t)
+ 2y(t) = 2x(t)
dt
(b) Finding the integral after replacing x(τ ) = u(τ ) we have that
e2τ t
y(t) = y(0)e−2t + 2e−2t = y(0)e−t + (1 − e−2t )
2 0
so that in the steady state, i.e., when t → ∞, the output is 1, independent of the value of the
initial condition.
(c) From the given input-output equation, letting x(t) = δ(t) and y(0) = 0 the output is the
impulse response of the system
Z t Z t
h(t) = 2 e−2(t−τ ) δ(τ )dτ = 2e−2t δ(τ )dτ = 2e−2t u(t)
0 0

Computing the convolution integral as

Z ∞
y(t) = x(t − τ )h(τ )dτ
−∞
it can be obtained graphically by reflecting the input x(t) = u(t) and shifting it linearly from
−∞ to ∞ to get
• for t < 0 then y(t) = 0

• for t ≥ 0 Z t
y(t) = 2e−2τ dτ = 1 − e−t
0
and so in the steady state the output goes to 1.

(d) The zero-input response is
yzi (t) = y(0)e−2t u(t)
and it is bounded for any finite value of the initial condition y(0), in particular for y(0) = 1,
therefore the system that depends on the initial condition is BIBO.

2.6 The input x(t) and the corresponding output y(t) of a linear time-invariant (LTI) system are
x(t) = u(t) − u(t − 1) → y(t) = r(t) − 2r(t − 1) + r(t − 2)
Determine the outputs yi (t), i = 1, 2, 3 corresponding to the following inputs
(i) x1 (t) = u(t) − u(t − 1) − u(t − 2) + u(t − 3), (ii) x2 (t) = u(t + 1) − 2u(t) + u(t − 1),
(iii) x3 (t) = δ(t) − δ(t − 1)
Plot the inputs xi (t) and the corresponding output yi (t).

Answers: y1 (t) = y(t) − y(t − 2); y3 (t) = dy(t)/dt = u(t) − 2u(t − 1) + u(t − 2).
Solution
(a) x1 (t) = x(t)−x(t−2) so y1 (t) = y(t)−y(t−2), two triangular pulses, the second multiplied
by −1.
x1 (t) y1 (t)
1
1
2 3 t
2 4
t
1
1
x2 (t) y2 (t)
1
1
1 2
t t
1 1
1
1
(b) x2 (t) = x(t + 1) − x(t) then y2 (t) = y(t + 1) − y(t) (they overlap between 0 and 1).
(c) x3 (t) = δ(t) − δ(t − 1) so y3 (t) = dy(t)/dt = u(t) − 2u(t − 1) + u(t − 2). Considering that
the output of x(t) is y(t), i.e., y(t) = S[x(t)], and that the integrator and the differentiator
are LTI systems Fig. 2.3 shows how to visualize the result in this problem by considering
that you can change the order of the cascading of LTI systems.
dy(t) dy(t)
Z x(t) y(t) dt Z dt
d d
dx(t)
S
dt
⌘ S
dt
x3 (t) = dx(t) dy(t)
dt x3 (t) = dt
dt | {z }
1

2.7 Consider a system represented by a first-order ordinary differential equation:

dy(t)
+ ay(t) = x(t), with initial condition y(0).
dt
(a) Show first that for a function f (t)
Z
d t
f (τ )dτ = f (t)
dt 0
using the definition of the derivative.
(b) Apply the above result to show that the solution of the first-order ordinary differential
equation with initial condition y(0) given above is
Z t
y(t) = y(0)e−at + e−a(t−τ ) x(τ )dτ t ≥ 0
0
(c) Another way to show the above expression is the solution of the ordinary differential
equation is to multiply both sides of the ordinary differential equation by eat
i. show that the left term is d(eat y(t))/dt,
ii. integrate the two terms to obtain the solution.
Solution
(a) By the definition of the derivative

Z t "Z Z
t+h t
d 1
f (τ )dτ = lim f (τ )dτ − f (τ )dτ
dt 0 h→0 h 0 0
"Z #
t+h
1
= lim f (τ )dτ
h→0 h t
f (t)h
= lim = f (t)
h→0 h
(b) y(t) satisfies the ordinary differential equation, indeed
Z
dy(t) −at d −at t aτ
= −ay(0)e + e e x(τ )dτ
dt dt 0
Z t
= −a[y(0)e−at + e−at eaτ x(τ )dτ ] + x(t) = −ay(t) + x(t)
0
| {z }
y(t)
(c) Multiplying the two terms of the ordinary differential equation by eat we get
eat dy(t)/dt + aeat y(t) = eat x(t)

| {z }
d(eat y(t))/dt
Z t
at
e y(t) = eaτ x(τ )dτ + y(0)
0
solving for y(t) we obtain the solution.

2.8 The bounded-input bounded-output stability assumes that the input is always bounded, lim-
ited in amplitude. If that is not the case, even a stable system would provide an unbounded
output. Consider the analog averager, with an input-output relationship
Z t
1
y(t) = x(τ )dτ
T t−T
(a) Suppose that the input to the averager is a bounded signal x(t), i.e., there is a finite value
M such that |x(t)| < M . Find the value for the bound of the output y(t) and determine
whether the averager is BIBO stable or not.
(b) Let the input to the averager be x(t) = tu(t), i.e., a ramp signal, compute the output y(t)
and determine if it is bounded or not. If y(t) is not bounded, does that mean that the
averager is an unstable system? Explain.
Answers: h(t) = (1/T )[u(t) − u(t − T )]; if input is ramp signal the output is not bounded.
Solution
(a) We can either find the impulse response or show that the output is bounded for any input
signal that is bounded. To find the impulse response change the variable to σ = t − τ which
gives
Z 0 Z T
−1 1
y(t) = x(t − σ)dσ = x(t − σ)dσ
T T 0 T
which is a convolution integral with h(t) = (1/T )[u(t) − u(t − T )]. This impulse response is
absolutely integrable as
Z ∞ Z T
1
|h(t)|dt = dt = 1
0 0 T
Likewise, if we assume that x(t) is bounded, i.e., there is a value M < ∞ such that |x(t)| < M ,
then Z Z
t t
1 M
|y(t)| ≤ |x(τ )|dτ ≤ dτ = M < ∞
T t−T T t−T
so system is BIBO stable.

(b) The ramp is not bounded, there is no M such that |r(t)| < M < ∞. So when we compute
the output of the stable system to the ramp we do not expect it to be bounded either. Indeed
Z t
1 t2 − (t − T )2 T
y(t) = τ dτ = =t−
T t−T 2T 2
increases as t increases, so it is unbound.

2.9 An ideal low-pass filter — The impulse response of an ideal low-pass filter is h(t) = sin(t)/t.
(a) Given that the impulse response is the response of the system to an input x(t) = δ(t)
with zero initial conditions, can an ideal low-pass filter be used for real-time processing?
Explain.
(b) Is the ideal low-pass filtering bounded-input bounded-output stable? Use MATLAB to
check if the impulse response satisfies the condition for BIBO stability.
Answers: Ideal low–pass filter is non–causal.
Solution
(a) The system with the sinc impulse response is non-causal because h(t) 6= 0 for t < 0. As
such an ideal low-pass filter cannot be used for real-time processing as it would require future
inputs, not available in real-time. This can be seen by looking at the convolution with a causal
signal x(t):
Z ∞
y(t) = x(τ )h(t − τ )dτ
0
Z t Z ∞
= x(τ )h(t − τ )dτ + x(τ )h(t − τ )dτ
0 t
where the upper limit of the top integral is due to h(τ ) having an infinite support, and the
lower limit because x(t) is causal. The second bottom integral shows that the output depends
on future values of the input.
abs(sin(t))/abs(t)
1
0.8
0.6
0.4
0.2
0
−30 −20 −10 0 10 20 30
t
6
Integral
0
0 5 10 15 20 25 30
n
Figure 2.4: Problem 9: absolute value of sinc function and its integral values for t = n sec.
(b) For the low-pass filter to be BIBO stable the impulse response h(t) must be absolutely inte-
grable, i.e., Z ∞
| sin(t)/t|dt < ∞
−∞

To approximtely compute this integral we use the following script. The integral is bounded, so
the filter is BIBO stable.
%% Pr 2.9
clear all; clf
syms x t x1
x=abs(sinc(t/pi));
for k=1:30,
x1=int(x,t,0,k);
xx(k)=subs(2*x1);
end
n=1:30;
figure(1)
subplot(211)
ezplot(x,[-30,30]); axis([-30 30 -0.1 1.2]);grid
subplot(212)
stem(n,xx); axis([1 30 0 8]);grid

Chapter 3
The Laplace Transform
3.1 Basic Problems
3.1 Find the Laplace transform of the following
(a) finite support signals, and indicate their region of convergence:
(i) x(t) = δ(t − 1), (ii) y(t) = δ(t + 1) − δ(t − 1)

(iii) z(t) = u(t + 1) − u(t − 1), (iv) w(t) = cos(2πt)[u(t + 1) − u(t − 1)]
(b) causal signals, and indicate their region of convergence:
(i) x1 (t) = e−t u(t), (ii) y1 (t) = e−t u(t − 1)

(iii) z1 (t) = e−t+1 u(t − 1), (iv) w1 (t) = e−t (u(t) − u(t − 1))
Answers: (a) Y (s) = 2 sinh(s), ROC the whole s-plane; W (s) = 2s sinh(s)/(s2 + 4π 2 ), ROC the
whole s-plane; (b) Y1 (s) = e−(s+1) /(s + 1), ROC σ > −1.
Solution
(a) Finite–support signals.

R∞
i. L[x(t)] = −∞ δ(t − 1)e−st dt = e−s , ROC the whole s-plane
ii. L[y(t)] = es − e−s = 2 sinh(s), ROC the whole s-plane
iii. We have
es − e−s sinh(s)
L[z(t)] = L[u(t + 1)] − L[u(t − 1)] = =2
s s
ROC whole s-plane, pole-zero cancellation at s = 0.
1
iv. Notice that because cos(2π(t ± 1)) = cos(2πt ± 2π) = cos(2πt) then
w(t) = cos(2πt)[u(t + 1) − u(t − 1)] = cos(2π(t + 1))u(t + 1) − cos(2π(t − 1))u(t − 1)
so that
es s e−s s (es − e−s )s 2s sinh(s)
W (s) = − = = 2
s2 + 4π 2 s2 + 4π 2 s2 + 4π 2 s + 4π 2
The poles p1,2 = ±j2π are cancelled by zeros z1,2 = ±j2π ( indeed, es − e−s |s=±j2π =
1 − 1 = 0) so the ROC is the whole s-plane.
(b) Causal signals.
i. X1 (s) = L[x1 (t)] = 1/(s + 1), ROC σ > −1
ii. Notice the exponential is not delayed, so the shifting property cannot be applied.
Instead,
e−(s+1)
Y1 (s) = L[y1 (t)] = L[e−1 e−(t−1) u(t − 1)] = , ROC σ > −1
s+1
iii. L[z1 (t)] = L[e−(t−1) u(t − 1)] = e−s /(s + 1), ROC σ > −1
iv. Since w1 (t) = x1 (t) − y1 (t), where x1 (t) and y1 (t) are given before then
1 − e−(s+1)
L[w1 (t)] = X1 (s) − Y1 (s) = ,
s+1
ROC whole s-plane because pole s = −1 is cancelled by zero, i.e., 1−e−(s+1) |s=−1 = 0.

3.2 Consider the pulse x(t) = u(t) − u(t − 1). Find the zeros and poles of X(s) and plot them.
(a) Suppose x(t) is the input of a LTI system with a transfer function H(s) = Y (s)/X(s) =
1/(s2 + 4π 2 ), find and plot the poles and zeros of Y (s) = L[y(t)]), where y(t) is the output
of the system.
(b) If the transfer function of the LTI system is
∞
Y
Z(s) 1
G(s) = =
X(s) s + (2kπ)2
2
k=1
and the input is the above signal x(t), compute the output z(t).
Answers: X(s) has infinite number of zeros on jΩ–axis; z(t) = δ(t).
Solution
The transform is X(s) = (1 − e−s )/s. The pole of X(s) is s = 0, while the zeros of X(s) are
values of s that make 1 − e−s = 0 or sk = j2kπ for −∞ < k < ∞ and integer. The pole is
cancelled with the k = 0 zero, so X(s) has an infinite number of zeros on the jΩ-axis.
(a) The output of the LTI system has Laplace transform
∞
(s + j2π)(s − j2π) Y
Y (s) = X(s)H(s) = (s + j2πk)(s − j2πk)
s2 + 4π 2
k=2
∞
Y
= (s2 + (2πk)2 )
k=2
There are no poles, but an infinite number of zeros at ±j2πk for integer k ≥ 2.
(b) This is similar to the above problem, but here
Z(s) = X(s)G(s) = 1
which gives
z(t) = δ(t)

3.3 Find the Laplace transform of the following signals and their region of convergence
(a) the reflection of the unit step signal, i.e., u(−t). And then use the result together with the
Laplace transform of u(t) to see if you can obtain the Laplace transform of a constant or
x(t) = u(t) + u(−t) (assume u(0) = 0.5 so there is no discontinuity at t = 0).
(b) the non-causal signal y(t) = e−|t| u(t + 1). Carefully plot it, and find its Laplace transform
Y (s) by separating y(t) into a causal and a non-causal signals, yc (t) and yac (t), and plot
them separately. Find the ROC of Y (s), Yc (s) and Yac (s).
Answers: (a) x(t) has no Laplace transform; (b) Y (s) = (es−1 (s + 1) − 2)/(s2 − 1).
Solution
(a) The Laplace transform of u(−t) is −1/s, indeed

Z 0 Z ∞
−st esτ ∞ 1
L[u(−t)] = e dt = esτ dτ = |τ =0 = −
−∞ 0 s s
provided that σ < 0 so that the limit when τ = ∞ is zero. Thus

−1
L[u(−t)] = ROC : σ < 0
s
Thus x(t) = u(t) + u(−t) = 1 for all t would have as Laplace transform
1 1
X(s) = − =0
s s
and as ROC the intersection of σ > 0 and σ < 0 which is null. So we cannot find the
Laplace of x(t) = 1, −∞ < t < ∞.
(b) Let y(t) = yc (t) + yac (t) where the causal component yc (t) = y(t)u(t) = e−t u(t) and the
anticausal component yac (t) = y(t)u(−t) = et [u(t + 1) − u(t)]. The Laplace transform of
yac (t) is
Z 0 Z 1
e(s−1) − 1
Yac (s) = e−(s−1)t dt = e(s−1)τ dτ =
−1 0 s−1
with ROC the whole plane (pole at s = 1 is cancelled by zero at 1).

The Laplace transform of yc (t) is Yc (s) = 1/(s + 1), so that
1 es−1 − 1
Y (s) = Yc (s) + Yac (s) = +
s+1 s−1
ses−1 + es−1 − 2
= ROC σ > −1
s2 − 1
The poles of Y (s) are s = ±1, and a zero is s = 1. Thus the zero cancels the pole at s = 1
leaving the pole at s = −1, and so the ROC is σ > −1.

3.4 You are given the following Laplace transform of the output y(t) of a system with input x(t)
and Laplace transform X(s):
(s − 1)X(s) 1
Y (s) = +
(s + 2)2 + 1 (s + 2)2 + 1
(a) If x(t) = u(t), find the zero-state response yzs (t).

(b) Find the zero-input response yzi (t).
(c) Determine the steady-state solution yss (t).
Answers: yzs (t) = −(1/5)u(t) + (1/5)e−2t cos(t)u(t) + (7/5)e−2t sin(t)u(t).
Solution
(a) The following term corresponds to the zero-state
(s − 1)X(s)
(s + 2)2 + 1
and the following term corresponds to zero-input
1
.
(s + 2)2 + 1
Zero-state response for x(t) = u(t)
s−1 A B + Cs
Yzs (s) = 2
= +
s((s + 2) + 1) s (s + 2)2 + 1
−1/5 9/5 + s/5 −1/5 1 (s + 2) + 7
= + 2
= +
s (s + 2) + 1 s 5 (s + 2)2 + 1
−1/5 1 (s + 2) 7 1
= + +
s 5 (s + 2)2 + 1 5 (s + 2)2 + 1
From the table of Laplace transforms
yzs (t) = −(1/5)u(t) + (1/5)e−2t cos(t)u(t)

+(7/5)e−2t sin(t)u(t)
(b) Zero-input response
1
Yzi (s) = ⇒ yzi (t) = e−2t sin(t)u(t)
(s + 2)2 + 1
(c) As t → ∞ y(t) → −1/5.

3.5 The transfer function of a LTI system is

s
H(s) =
(s + 1)2 + 1
(a) Use the Laplace transform to find the unit-step response s(t) = (h ∗ x)(t).
(b) Find the response due to the following inputs
(i) x1 (t) = u(t) − u(t − 1), (ii) x2 (t) = δ(t) − δ(t − 1), (iii) x3 (t) = r(t)
Answer: s(t) = e−t sin(t)u(t).
Solution
(a) The Laplace transform of s(t) = (h ∗ x)(t) is S(s) = H(s)X(s) which gives
1
S(s) = ⇒ s(t) = e−t sin(t)u(t)
(s + 1)2 + 1
(b) By LTI
y1 (t) = s(t) − s(t − 1)

y2 (t) = ds(t)/dt − ds(t − 1)/dt
Z t
y3 (t) = s(t0 )dt0
−∞

3.6 Consider the impulse response of a LTI system h(t) = e−at [u(t) − u(t − 1)] a > 0.
(a) Obtain the transfer function H(s).

(b) Find the poles and zeros of H(s).
(c) Is lima→0 H(s) equal to
1 − e−s
?
s
(d) Indicate how the poles and zeros of H(s) move in the s-plane as a → 0.
Answers: H(s) = (1 − e−(s+a) )/(s + a), ROC: whole s plane, pole cancelled by zero.
Solution
(a) The transfer function is
1 e−a e−s 1 − e−(s+a)

H(s) = − = ROC: whole s plane
s+a s+a s+a
because pole/zero cancellation at s = −a.
(b) Pole s = −a; zeros are values of s such that e−(s+a) = 1 = ej2πk for k = 0, ±1, ±2, · · · , or
sk = −a − j2πk (the zero s0 = −a cancels the pole at s = −a).
(c) As a → 0
1 − e−s
H(s) =
s
(d) Pole s = −a and zeros sk = −a − j2πk; pole/zero cancellation when k = 0. As a → 0 the
remaining zeros move towards the jΩ axis.

3.7 A causal LTI system has a transfer function
Y (s) 1
H(s) = =
X(s) 1 − e−s
(a) Find the poles and zeros of H(s), and from this determine if the filter is BIBO stable or not.
(b) Draw a block diagram for such a system.
(c) Find the impulse response h(t) of the system and use it to verify your stability result.
Answer: h(t) is not absolutely integrable.
Solution
(a) Poles satisfy the identity e−s = 1 = ej2πk for k = 0, ±1, ..., so there are an infinite number
of poles, sk = −j2πk. No zeros. The system is unstable because of poles on jΩ-axis.
x(t) + y(t)
Delay
y(t − 1)
(b) Y (s)(1 − e−s ) = X(s) gives y(t) = y(t − 1) + x(t). A positive feedback system with a delay
in the feedback represents the system. See Fig. 3.1.
(c) The transfer function is
X∞
1
H(s) = = (e−s )n
1 − e−s n=0
with inverse
∞
X
h(t) = δ(t − n)
n=0
which is not absolutely integrable

Z ∞ ∞ Z
X ∞ ∞
X
|h(t)|dt = δ(t − n)dt = 1→∞
−∞ n=0 −∞ n=0

3.8 Consider the following problems related to the convolution integral.
(a) The impulse response of a LTI system is h(t) = e−2t u(t) and the system input is a pulse
x(t) = u(t) − u(t − 3). Find the output of the system y(t) by means of the convolution
integral graphically and by means of the Laplace transform.
(b) It is known that the impulse response of an analog averager is h1 (t) = u(t) − u(t − 1),
consider the input to the averager x1 (t) = u(t) − u(t − 1), determine graphically as well
as by means of the Laplace transform the corresponding output of the averager y1 (t) =
[h1 ∗ x1 ](t). Is y1 (t) smoother than the input signal x1 (t)? provide an argument for your
answer.
(c) Suppose we cascade 3 analog averagers each with the same impulse response hi (t) =
u(t) − u(t − 1), i = 1, 2, 3, determine the transfer function of this system. If the duration of
the input to the first averager is M sec., what would be the duration of the output of the
3th averager?
Answers: (a) Y (s) = (1 − e−3s )/(s(s + 2)); (c) the support of the output will be 3M .
Solution
(a) Using the Laplace transform
1
H(s) =
s+2
1 − e−3s
X(s) =
s
1 − e−3s
Y (s) = H(s)X(s) =
s(s + 2)
Letting
1 0.5 0.5
F (s) = = − ⇒ f (t) = 0.5(1 − e−2t )u(t)
s(s + 2) s s+2
then
y(t) = f (t) − f (t − 3) = 0.5(1 − e−2t )u(t) − 0.5(1 − e−2(t−3) )u(t − 3)
Graphically, we plot h(τ ) and x(t − τ ) as functions of τ and shift x(t − τ ) from the left to the
right and integrate the overlapping areas to get


 0 t<0




 R
t −2τ
y(t) = e dτ = 0.5(1 − e−2t ) 0≤t<3


0




 Rt
t−3
e−2τ dτ = 0.5(e−2(t−3) − e−2t ) t ≥ 3

which can be written as
y(t) = 0.5(1 − e−2t )[u(t) − u(t − 3)] + 0.5[e−2(t−3) − e−2t ]u(t − 3)

= 0.5(1 − e−2t )u(t) − 0.5(1 − e−2(t−3) )u(t − 3)
coinciding with the result obtained using the Laplace transform.

(b) Using Laplace transform
1 − e−s
H1 (s) = X1 (s) = H(s)
s
1 − 2e−s + e−2s
Y1 (s) =
s2
so that
y1 (t) = r(t) − 2r(t − 1) + r(t − 2)
a triangular pulse.
Graphically, letting x1 (τ ) and h1 (t − τ ) (moving from left to right) be functions of τ , and inte-
grating their overlap gives


 0 t<0





 Rt



 0 dτ = t 0≤t<1
y1 (t) =

 R1

 dτ = 1 − t + 1 = 2 − t 1 ≤ t < 2

 t−1






0 t≥2
coinciding with the result using the Laplace transform. The signal y1 (t) is smoother than x1 (t)
since it is the output of an averager filter.
(c) Using the Laplace transform, letting h(t) = hi (t), i = 1, 2, 3
1 − e−s
H(s) =
s
(1 − e−s )3 1 − 3e−s + 3e−2s − e−3s
(H(s))3 = =
s3 s3
The support of the output will be 3M .

3.9 To see the effect of the zeros on the complete response of a system, suppose you have a system
with a transfer function
Y (s) s2 + 4
H(s) = =
X(s) s((s + 1)2 + 1)
(a) Find and plot the poles and zeros of H(s). Is this BIBO stable?
(b) Find the frequency Ω0 of the input x(t) = 2 cos(Ω0 t)u(t) such that the output of the given
system is zero in the steady state. Why do you think this happens?
(c) If the input is x(t) = 2 sin(Ω0 t)u(t) would you get the same result as above? Explain why
or why not.
Answers: h(t) not absolutely integrable; cosine as input gives Y (s) = 2/((s + 1)2 + 1).
Solution
(a) The poles of H(s) are at s = 0 and s = −1 ± j1, and the zeros are at s = ±j2. This system
is not BIBO stable because of the pole at the origin, the impulse response will not be absolutely
integrable.
(b) If x(t) = 2 cos(2t)u(t) then
s
X(s) = 2
s2 + 4
and thus
2
Y (s) =
(s + 1)2 + 1
i.e., the pole at s = 0 and the zeros are cancelled by the transform of the cosine. The remaining
poles are in the left-hand s-plane and so the steady state is zero.
This is a very special case because of the cancellation of the pole at s = 0 and the numerator
due to the used input. The system is not stable, and the frequency response does not exist for
all frequencies. Indeed it is infinite whenever the input frequency is 0.
(c) The input cannot be a sine of the same frequency because its Laplace transform does not
cancel the pole at zero and so the steady state is not zero.

3.2 Problem using MATLAB
3.10 Inverse Laplace transform— Consider the following inverse Laplace transform problems.
(a) Given the Laplace transform

s4 + 2s + 1
Y (s) =
s3 + 4s2 + 5s + 2
which is not proper, determine the amplitude of the δ(t) and dδ(t)/dt terms in the inverse
signal y(t).
(b) Find the inverse Laplace transform of
s2 − 3
Z(s) =
(s + 1)(s + 2)
Can you use the initial-value theorem to check your result? Explain.
(c) The inverse Laplace transform of
3s − 4
X(s) =
s(s + 1)(s + 2)
should be of the form x(t) = [Ae−t + B + Ce−2t ]u(t). Find the values of A, B and C. Use
the MATLAB function ilaplace to get the inverse and to plot it.
Answers: y(t) = −4δ(t) + dδ(t)/dt + · · · ; z(t) = δ(t) − 2e−t u(t) − e−2t u(t).
Solution (a) By long division we have that

11s2 + 20s + 9
Y (s) = −4 + s +
s3 + 4s2 + 5s + 2
The inverse of the first two terms are −4δ(t) and dδ(t)/dt. The third term is proper rational and
its partial fraction expansion can be found.
(b) Z(s) is not proper rational so we need to divide to get
3s + 5 2 1
Z(s) = 1 − =1− −
(s + 1)(s + 2) s+1 s+2
so that
z(t) = δ(t) − 2e−t u(t) − e−2t u(t)
Applying the initial–value result we get
z(0) = lim sX(s) → ∞

s→∞
given the presence of δ(t) which is not defined at t = 0.

(c)
A B C
X(s) = + +
s s+1 s+2

where A = −2, B = 7 and C = −5 so that
x(t) = −2u(t) + 7e−t u(t) − 5e−2t u(t)
% Pr. 3_10
clear all; clf
syms s t w
disp(’>>>>> Inverse Laplace <<<<<’)
x=ilaplace((3*s-4)/(sˆ3+3*sˆ2+2*s))
figure(1)
ezplot(x,[0,10])
axis([0 10 -2.5 .5])
grid
>>>>> Inverse Laplace <<<<<

x = -5*exp(-2*t)+7*exp(-t)-2
−5 exp(−2 t)+7 exp(−t)−2

0.5
−0.5
−1
−1.5
−2
−2.5
0 1 2 3 4 5 6 7 8 9 10
t

3.11 Solving differential equations— One of the uses of the Laplace transform is the solution of
differential equations.
(a) Suppose you are given the ordinary differential equation that represents a LTI system,
y (2) (t) + 0.5y (1) (t) + 0.15y(t) = x(t), t ≥ 0
where y(t) is the output and x(t) is the input of the system, y (1) (t) and y (2) (t) are first
and second order derivatives with respect to t. The input is causal, i.e., x(t) = 0 t < 0.
What should the initial conditions be for the system to be LTI? Find Y (s) for those initial
conditions.
(b) If y (1) (0) = 1 and y(0) = 1 are the initial conditions for the above ordinary differential
equation, find Y (s). If the input to the system is doubled, i.e., the input is 2x(t) is Y (s)
doubled so that its inverse Laplace transform y(t) is doubled? Is the system linear?
(c) Use MATLAB to find the solutions of the ordinary differential equation when the input is
u(t) and 2u(t) with the initial conditions given above. Compare the solutions and verify
your response in (b).
Answers: (b) Y (s) = X(s)/(s2 + 0.5s + 0.15) + (s + 1.5)/(s2 + 0.5s + 0.15), not LTI.
Solution
(a) If the initial conditions are zero the system is LTI. The Laplace transform of the differential
equations is then
Y (s)[s2 + 0.5s + 0.15] = X(s)
and so
X(s)
Y (s) =
s2 + 0.5s + 0.15
(b) If the initial conditions are dy(0)/dt = 1 and y(0) = 1 the Laplace transform of the differen-
tial equation is
[Y (s)s2 − s − 1] + 0.5[sY (s) − 1] + 0.15Y (s) = X(s)
which gives
X(s) s + 1.5
Y (s) = + 2
s2 + 0.5s + 0.15 s + 0.5s + 0.15
which can be written, with the appropriate definitions, as Y (s) = X(s)/A(s) + I(s)/A(s), so
that when we double the input the Laplace transform of the corresponding output is
2X(s) I(s)
+ 6= 2Y (s)
A(s) A(s)
thus the system is not LTI.

(c) If x(t) = u(t) then the Laplace transform of the output (including the initial conditions is)

after replacing X(s) = 1/s:
s2 + 1.5s + 1
Y (s) =
s3 + 0.5s2 + 0.15s
when we double the input the Laplace transform of the output (again including the initial
conditions) is
s2 + 1.5s + 2
Y1 (s) = .
s3 + 0.5s2 + 0.15s
The following script is used to obtain the two responses. The second one is not the double of
the first, so the system is not LTI — due to the initial conditions not being zero.
%% Pr. 3_11
clear all; clf
% response to u(t)
den=[1 0.5 0.15 0];
num=[0 1 1.5 1];
syms s t y y1
[p,r]=pfeLaplace(num,den)
y=ilaplace((num(2)*sˆ2+num(3)*s+num(4))/(den(1)*sˆ3 +den(2)*sˆ2+den(3)*s))
% response to 2u(t)
num=[0 1 1.5 2];
pfeLaplace(num,den)
y1=ilaplace((num(2)*sˆ2+num(3)*s+num(4))/(den(1)*sˆ3 +den(2)*sˆ2+den(3)*s))
figure(1)
subplot(311)
ezplot(y,[0,40])
axis([0 40 -0.01 11]);grid;title(’Response to u(t)’)
subplot(312)
ezplot(y1,[0,40])
axis([0 40 -0.01 15]);grid;title(’Response to 2 u(t)’)
subplot(313)
ezplot(y1-2*y,[0,40]);grid;title(’Difference of two responses’)
axis([0 40 -10 10])

Response to u(t)
10
0
0 5 10 15 20 25 30 35 40
t
Response to 2 u(t)
15
10
0
0 5 10 15 20 25 30 35 40
t
Difference of two responses
10
−10
0 5 10 15 20 25 30 35 40
t
Figure 3.3: Problem 11: Responses y(t) and y1 (t) corresponding to inputs u(t) and 2u(t), the differ-
ence y1 (t) − 2y(t) 6= 0 indicates the system is not linear.

3.12 Zero steady-state response of analog averager— An analog averager can be represented by
the differential equation
dy(t) 1
= [x(t) − x(t − T )]
dt T
where y(t) is its output and x(t) the input.
(a) If the input-output equation of the averager is

Z
1 t
y(t) = x(τ )dτ
T t−T
Show how to obtain the above differential equation and that y(t) is the solution of the
differential equation.
(b) If x(t) = cos(πt)u(t), choose the value of T in the averager so that the output y(t) = 0 in
the steady state. Graphically show how this is possible for your choice of T . Is there a
unique value for T that makes this possible? How does it relate to the frequency Ω0 = π
of the sinusoid?
(c) Use the impulse response h(t) of the averager found before, to show using Laplace that
the steady state is zero when x(t) = cos(πt)u(t) and T is the above chosen value. Use
MATLAB to solve the differential equation and to plot the response for the value of T you
chose. [HINT: consider x(t)/T the input and use superposition and time-invariance to
find y(t) due to (x(t) − x(t − T ))/T .]
Answers:(c) y(t) = (0.5/π)(sin(πt)u(t) − sin(π(t − 2)u(t − 2)) for input cos(πt)u(t)
Solution
(a) The input/output equation for the analog averager is
Z
1 t
y(t) = x(τ )dτ
T t−T

dy(t) 1
= [x(t) − x(t − T )]
dt T
Likewise if we integrate this from −∞ to t it gives
Z t Z t
1
y(t) = x(τ )dτ − x(τ − T )dτ
T −∞ −∞
and changing the variable in the second integral to σ = τ − T we get

"Z Z t−T #
t
1
y(t) = x(τ )dτ − x(σ)dσ
T −∞ −∞
Z
1 t
= x(τ )dτ
T t−T

the equation for the analog averager.

(b) Graphically the convolution gives zero for t < 0; for values of 0 ≤ t < T the output is a
transient; and for t > T it is a periodic signal. If we choose T = 2, the period of the sinusoid,
(or multiples of 2) the periodic output signal becomes zero—the averager computes the area
under the input of a period, which is zero. It can also be shown analytically, for T = 2
Z
1 t
y(t) = cos(πτ )dτ
2 t−2
1 sin(πτ ) t
= |t−2
2 π
1
= [sin(πt) − sin(π(t − 2))] = 0 t≥2
2π
(c) The impulse response of the averager is h(t) = 0.5[u(t) − u(t − 2)] for the value T = 2 found
above. If the input is cos(πt)u(t), then
1 X(s) X(s) −2s

Y (s) = H(s)X(s) = [1 − e−2s ]X(s) = − e
2s 2s 2s
Replacing X(s) = s/(s2 + π 2 ) above we have
1 − e−2s
Y (s) = 0.5 = Y1 (s) − Y1 (s)e−2s
s2 + π 2
where Y1 (s) = 0.5/(s2 + π 2 ) so that
0.5
y(t) = (sin(πt)u(t) − sin(π(t − 2)u(t − 2))
π
%% Prob 3_12
clear all; clf
T=2;
num=[0 0 1/T]
den=[1 0 piˆ2]
syms s t y
figure(1)
subplot(221)
[r,p]=pfeLaplace(num,den);
y1=ilaplace(0.5/(sˆ2+piˆ2))
subplot(222)
ezplot(y1,[0,8])
axis([0 8 -0.2 0.2]); title(’y_1(t)’)
grid
y=y1-y1*heaviside(t-2)
subplot(223)
ezplot(y,[0,8])

axis([0 8 -0.2 0.2]); title(’y(t)’)

grid
y1(t)
4 0.2
2 0.1
0 0
jΩ
−2 −0.1
−4 −0.2
−1 0 1 2 3 0 2 4 6 8
σ t
y(t)
0.2
0.1
−0.1
−0.2
0 2 4 6 8
t
Figure 3.4: Problem 12: Poles/zeros of Y1 (s) (top–left) , y1 (t) and y(t) (bottom).

3.13 Polynomial multiplication— When the numerator or denominator are given in a factorized
form, we need to multiply polynomials. Although this can be done by hand, MATLAB provides
the function conv that computes the coefficients of the polynomial resulting from the product
of two polynomials.
(a) Use help in MATLAB to find how conv can be used, and then consider two polynomials
P (s) = s2 + s + 1 and Q(s) = 2s3 + 3s2 + s + 1
do the multiplication of these polynomials by hand to find Z(s) = P (s)Q(s) and use conv
to verify your results.
(b) The output of a system has a Laplace transform
N (s) (s + 2)
Y (s) = = 2
D(s) s (s + 1)((s + 4)2 ) + 9)
use conv to find the denominator polynomial and then find the inverse Laplace transform
using ilaplace.
Answers: Z(s) = P (s)Q(s) = 2s5 + 5s4 + 6s3 + 5s2 + 2s + 1
Solution
(a)(b) The following script shows how to do polynomial multiplication using conv function.
The multiplication of the polynomials
P (s) = s2 + s + 1 and Q(s) = 2s3 + 3s2 + s + 1
gives
Z(s) = P (s)Q(s) = 2s5 + 5s4 + 6s3 + 5s2 + 2s + 1
The script shows how to obtain this result in the Poly Multiplication part of the script.
In the Application part of the script we show how to find the numerator N1 (s) and the denom-
inator D1 (s) of Y (s) by multiplying X(s) and H(s) to give
X(s)N (s) (s + 2)
Y (s) = = 2
D(s) s (s + 1)((s + 4)2 + 9)
so that the poles of Y (s) are s = 0 (double), and s = −4 ± j3, and no zeros. The double pole
gives a ramp and the other poles a modulated sinusoid by a decaying exponential.
% Pr. 3_13
% Poly multiplication
% coefficients of higher to lower orders of s
P=[1 1 1];Q=[2 3 1 1];

Z=conv(P,Q);
% Application
d1=[1 1]; d2=[1 8 25]; D=conv(d1,d2);
d3=[1 0 0]; den=conv(D,d3)
N=length(den)
num=[ zeros(1,N-2) 1 2]
syms s t y
figure(1)
subplot(121)
[r,p]=pfeLaplace(num,den);
y=ilaplace((num(N-1)*s+num(N))/(den(1)*sˆ5+den(2)*sˆ4+den(3)*sˆ3+den(4)*sˆ2+den(5)*s+d
subplot(122)
ezplot(y,[0,50])
axis([0 50 -0.1 1.5]); grid
den = 1 9 33 25 0 0
num = 0 0 0 0 1 2
>>>>> Zeros <<<<<
z = -2
>>>>> Poles <<<<<
p = -4.0000 + 3.0000i
-4.0000 - 3.0000i
-1.0000
0
0
>>>>> Residues <<<<<
r = 0.0050 - 0.0026i
0.0050 + 0.0026i
0.0556
-0.0656
0.0800
>>>>> Inverse Laplace <<<<<
y = -41/625+1/18*exp(-t)+2/25*t+113/11250*exp(-4*t)*cos(3*t)
+59/11250*exp(-4*t)*sin(3*t)

−41/625+...+59/11250 exp(−4 t) sin(3 t)

4 4.5
3 4
3.5
2
3
1
2.5
0
jΩ
−1
1.5
−2 1
−3 0.5
0
−4
−4 −2 0 2 0 10 20 30 40 50
σ t
Figure 3.5: Problem 13: Poles/zeros of Y (s) and y(t).

Chapter 4
Frequency Analysis: the Fourier

Series
4.1 Basic Problems
4.1 Consider the following problems related to periodicity and Fourier series:
(a) Consider the signals
x1 (t) = 1 + cos(2πt) − cos(6πt), x2 (t) = 1 + cos(2πt) − cos(6t) −∞<t<∞
i. Determine which of these signals is periodic and the corresponding fundamental pe-
riod.
ii. For the periodic signal, find a trigonometric Fourier series representation.
iii. Indicate why it is not possible to obtain a trigonometric Fourier series representation
for the non-periodic signal.
(b) Consider the signal
sin(2t + π)
x(t) =
sin(t)
i. Is this signal periodic? If so, what is its fundamental period T0 ?
ii. Find the Fourier series of x(t).
Answers: (a) x1 (t) periodic of fundamental period T = 1; x2 (t) does not have Fourier series;
(b) periodic of fundamental period T0 = 2π; x(t) = −2 cos(t).
Solution
1
(a) i. The sinusoidal components of x1 (t) have periods T1 = 1 and T2 = 1/3, with ratio
T1 /T2 = 3 so that x1 (t) is periodic of fundamental period T1 = 3T2 = 1.
The periods of the sinusoidal components of x2 (t) are T1 = 1 and T2 = 2π/6, their
ratio is T1 /T2 = 3/π, not a rational number so x2 (t) is not periodic.
ii. Trigonometric Fourier series of x1 (t), with fundamental period T0 = 1 and fundamen-
tal frequency Ω0 = 2π, is
x1 (t) = 1 + cos(Ω0 t) − cos(3Ω0 t)
iii. For x2 (t), since Ω2 = 6 cannot be expressed as an integer multiple of Ω1 = 2π or vice

versa, there is no Fourier series for x2 (t).
(b) i. For the signal to be periodic of period T0 we must have
sin(2(t + T0 ) + π)
x(t + T0 ) = = x(t)
sin(t + T0 )
this is possible if T0 = 2π
ii. It can be shown that x(t) = −2 cos(t), indeed sin(2t + π) = − sin(2t) and
− sin(2t) = −2 cos(t) sin(t)
so
x(t) = −2 cos(t) = −ejt − e−jt
thus X1 = X−1 = −1.

4.2 Find the complex exponential Fourier series for the following signals. In each case plot the
magnitude and phase line spectra for k ≥ 0.
(i) x1 (t) = cos(5t + 45o ), (ii) x2 (t) = sin2 (t), (iii) x3 (t) = cos(3t) + cos(5t)
∗
Answers: The Fourier series coefficients for x1 (t) are X1 = X−1 = 0.5ejπ/4 .
Solution
(i) Fundamental frequency Ω0 = 5 so
x1 (t) = (0.5ejπ/4 )ej5t + (0.5e−jπ/4 )e−j5t
∗
and X1 = X−1 = 0.5ejπ/4 .
(ii) We have
2
1 jt
x2 (t) = (e − e−jt ) = 0.5 − 0.25ej2t − 0.25e−j2t
2j
∗
and so X0 = 0.5, X1 = X−1 = −0.25 and Ω0 = 2 is the fundamental frequency.
(iii) Fundamental frequency of x3 (t) is Ω0 = 1 (ratio of fundamental periods of sinusoidal
components T1 /T2 = (2π/3)/(2π/5) = 5/3 and Ω0 = 2π/3T1 = 1).
x3 (t) = 0.5ej3Ω0 t + 0.5e−j3Ω0 t + 0.5ej5Ω0 t + 0.5e−j5Ω0 t
so that X3 = X−3 = 0.5 and X5 = X−5 = 0.5.

See Fig. 4.1 for magnitude and phase line spectra.
x1(t)
0.8
0.4 0.6
|Xk|
∠Xk
0.4
0.2
0.2
0 0
0 2 4 6 0 2 4 6
x2(t)
0.6 3
0.4 2
|Xk|
∠Xk
0.2 1
0 0
0 2 4 6 8 10 0 2 4 6 8 10
x3(t)
0.6 3
0.4 2
|Xk|
∠Xk
0.2 1
0 0
0 1 2 3 4 5 0 1 2 3 4 5
kΩ0 kΩ0

4.3 Let the Fourier series coefficients of a periodic signal x(t) of fundamental frequency Ω0 =
2π/T0 ) be {Xk }. Consider the following functions of x(t).
y(t) = 2x(t) − 3, z(t) = x(t − 2) + x(t), w(t) = x(2t)
Determine if y(t), z(t) and w(t) are periodic, and if so give the corresponding Fourier coeffi-
cients in terms of those for x(t).
Answers: Zk = Xk (1 + e−2jΩ0 k ); Wk = Xk/2 for k even and 0 otherwise.
Solution
(a) If the periodic signal
X
x(t) = Xk ejΩ0 kt
k
then
X
y(t) = 2x(t) − 3 = (2X0 − 3) + 2Xk ejΩ0 kt
k6=0
is also periodic of period T0 with Fourier coefficients

(
2X0 − 3 k = 0
Yk =
2Xk k 6= 0
The signal
z(t) = x(t − 2) + x(t)

X X
= Xk ejΩ0 k(t−2) + Xk ejΩ0 kt
k k
X
−2jΩ0 k
= [Xk (1 + e )]ejΩ0 kt
k
is periodic of period T0 and with Fourier series coefficients Zk = Xk (1 + e−2jΩ0 k ).

The signal
X X
w(t) = x(2t) = Xk ejΩ0 k2t = Xm/2 ejΩ0 mt
k m even
is periodic of period T0 /2, with Fourier series coefficients
(
Xk/2 k even
Wk =
0 otherwise

4.4 The period of a periodic signal x(t) with fundamental period T0 = 2 is x1 (t) = cos(t)[u(t) −
u(t − 2)].
(a) Plot the signal x(t) and find the Fourier series coefficients Xk for x(t) using the integral
equation.
(b) Use the Laplace transform to find the Fourier series coefficients {Xk }.
(c) After considering the symmetry of the zero-mean signal x(t) − X0 , can you determine
whether the Xk should be real, purely imaginary or complex? Indicate how you make
this determination.
(d) To simplify computing the Laplace transform of x1 (t) consider the following expression:
cos(t)u(t − 2) = [A cos(t − 2) + B sin(t − 2)]u(t − 2).
Find the values of A and B and show how to compute L[cos(t)u(t − 2)].
Answers: Xk = (jkπ(1 − cos(2)) + sin(2))/(2(1 − k 2 π 2 )); A = cos(2), B = − sin(2).
Solution
(a) See Fig. 4.2. Fourier series coefficients (Ω0 = π)
x(t)
1
··· ···
2 4 6
t
cos(2)
Z 2 2
1 1 jt 1 ejt(1−kπ) e−jt(1+kπ)
Xk = (e + e−jt )e−jkπt dt = +
2 0 2 4 j(1 − kπ) −j(1 + kπ) 0
j2 −j2

1 e −1 e −1
= −
4 j(1 − kπ) j(1 + kπ)
1 j2
= 2 2
j(e − e−j2 ) + jπk(ej2 + e−j2 ) − j(1 + kπ) + j(1 − kπ)
4(k π − 1)
jkπ(1 − cos(2)) + sin(2)
=
2(1 − k 2 π 2 )

(b) X1 (s) = L[cos(t)(u(t) − u(t − 2))] = L[cos(t)u(t) − cos(t)u(t − 2)]
Z ∞ Z ∞
L[cos(t)u(t − 2)] = cos(t)e−st dt = cos(ρ + 2)e−s(ρ+2) dρ
2 0
Z ∞ Z ∞
−2s
= e cos(2) cos(ρ)e−sρ dρ − sin(2) sin(ρ)e−sρ dρ
0 0
= e−2s [cos(2)L[cos(t)u(t)] − sin(2)L[sin(t)u(t)]]

cos(2)s sin(2)
= e−2s −
s2 + 1 s2 + 1
We then have

s e−2s (cos(2)s − sin(2))
Xk = 0.5 2
−
s +1 s2 + 1 s=jkπ
jkπ(1 − cos(2)) + sin(2)
=
2(1 − k 2 π 2 )
which coincides with the expression obtained using the integral.

(c) Setting k = 0 in the above expression, we find the mean is X0 = sin(2)/2. Since the signal
x(t) − X0 is neither even nor odd the Fourier coefficients Xk should be complex, as they
are.
(d) Letting
cos(t) = A cos(t − 2) + B sin(t − 2)

p
= A2 + B 2 cos(t − 2 − tan−1 (B/A))
which gives
p
A2 + B 2 = 1
tan−1 (B/A) = −2
which are satisfied by A = cos(2) and B = − sin(2), indeed A2 + B 2 = (cos(2))2 +

(sin(2))2 = 1 and tan−1 (B/A) = tan−1 [− tan(2)] = −2. So that the Laplace transform
L[cos(t)u(t − 2)] = L[(A cos(t − 2) + B sin(t − 2))u(t − 2)]

= AL[cos(t − 2)u(t − 2)] + BL[sin(t − 2)u(t − 2)]
e−2s (cos(2)s − sin(2))
=
s2 + 1
which coincides with the result using the Laplace transform obtained before.

4.5 Consider the following problems related to steady state and frequency responses.
(a) The input x(t) and the output y(t), −∞ < t < ∞, of a filter with transfer function H(s) are
x(t) = 4 cos(2πt) + 8 sin(3πt), y(t) = −2 cos(2πt) .
Determine as much as possible about the frequency response H(jΩ) of the filter using the
above input and output.
(b) Given a LTI system with frequency response
(
−π/2 Ω≥0
|H(jΩ)| = u(Ω + 2) − u(Ω − 2), ∠H(jΩ) =
π/2 Ω<0
If the input signal x(t) is periodic with Fourier series

∞
X 2
x(t) = cos(3kt/2)
k2
k=1
calculate the steady-state response yss (t) of the system.
Answers: (a) H(j2π) = −0.5, H(j3π) = 0; (b) yss (t) = 2 cos(3t/2 − π/2).
Solution
(a) The steady state is
y(t) = 4|H(j2π)| cos(2πt + ∠H(j2π)) + 8|H(j3π)| cos(3πt − π/2 + ∠H(j3π))
so H(j2π) = 0.5ejπ = −0.5, H(j3π) = 0. Nothing else can be learned about the filter from
the input/output.
(b)
∞
X 2
yss (t) = |H(j3k/2)| cos(3kt/2 + ∠H(j3k/2))
k2
k=1
= 2|H(j3/2)| cos(3t/2 + ∠H(j3/2))
= 2 cos(3t/2 − π/2)
since for frequencies bigger than 2 the magnitude response is zero.

4.6 Consider the following problems related to filtering of periodic signals:
(a) A periodic signal x(t) of fundamental frequency Ω0 = π/4 is the input of an ideal band-
pass filter with the following frequency response
 

 1 π ≤ Ω ≤ 3π/2  −Ω
 π ≤ Ω ≤ 3π/2
|H(jΩ)| = 1 −3π/2 ≤ Ω ≤ −π ∠[H(jΩ)] = Ω −3π/2 ≤ Ω ≤ −π

 

0 otherwise 0 otherwise
∗ ∗
The non-zero Fourier series coefficients of x(t) are X1 = X−1 = j, X5 = X−5 = 2.
i. Express x(t) in the form
∞
X
x(t) = Ak cos(Ωk + θk ).
k=0
ii. Find the output y(t) of the ideal band-pass filter.

(b) The Fourier series of a periodic signal x(t) is
∞
X
x(t) = (1 − (−1)k ) cos(kπt)
k=0
If x(t) is filtered with a filter having the following frequency response


(
 1 −4.5π ≤ Ω ≤ −1.5π

π/2 Ω≤0
|H(jΩ)| = 1 1.5π ≤ Ω ≤ 4.5π ∠H(jΩ) =

 −π/2 Ω>0
0 otherwise
i. Carefully plot the frequency response, magnitude and phase, of the filter and deter-
mine the type of filter it is.
ii. Calculate the steady-state response yss (t) of the filter for the given input.
Answers: (a) y(t) = 4 cos(5π(t − 1)/4); (b) yss (t) = 2 cos(3πt − π/2).
Solution
(a)
x(t) = (jejπt/4 − je−jπt/4 ) + 2(ej5πt/4 + e−j5πt/4 )

= −2 sin(πt/4) + 4 cos(5πt/4) = 2 cos(πt/4 + π/2) + 4 cos(5πt/4)
y(t) = 2|H(jπ/4)| cos(πt/4 + π/2 + ∠H(jπ/4)) + 4|H(j5π/4)| cos(5πt/4 + ∠H(j5π/4))
= 4 cos(5πt/4 − 5π/4) = 4 cos(5π(t − 1)/4)
(b) i. The filter is band-pass:

|H(jΩ)|
1
Ω
−4.5π −1.5π 1.5π 4.5π
∠H(jΩ)
π/2
Ω
−π/2
Figure 4.3: Problem 6(b)
ii. According to the frequency response of the band-pass filter y(t) can have a second,
third and fourth harmonics, but since the magnitude of the second and fourth Fourier
coefficients are X2 = X4 = 0, the output has only the third harmonic giving
y(t) = 2|H(j3π)| cos(3πt + ∠H(j3π))

= 2 cos(3πt − π/2).

4.7 A periodic signal x(t), of fundamental frequency Ω0 = π, has a period

(
1 + t −1 ≤ t ≤ 0
x1 (t) = ,
1−t 0<t≤1
The signal x(t) is the input of an ideal low-pass filter with the frequency response H(jΩ) shown
in Fig. 4.4. Let y(t) be the output of the system.
|H(jΩ)| ∠H(jΩ)
2
π/2
Ω Ω
−1.5π 1.5π
−π/2
(a) Determine the Fourier series coefficients needed to find the output y(t) of the filter.
(b) Is the output signal y(t) periodic? If so, determine its fundamental period T0 , and its dc
value.
(c) Provide the constants A, B and C in the output: y(t) = A + B cos(πt + C).
Answers: Need X0 and X1 ; A = 1, B = 8/π 2 , C = −π/2.
Solution
(a) The fundamental frequency is Ω0 = π (T0 = 2), so only the Fourier coefficients corre-
sponding to k = 0 and k = ±1, i.e., frequencies Ω = 0, ±Ω0 , are needed as the filter deletes
any other frequency component.
es − 2 + e−s
x1 (t) = r(t + 1) − 2r(t) + r(t − 1) → X1 (s) =
s2
1 −1 1 − cos(πk)
Xk = (2 cos(πk) − 2) = k 6= 0
2 k2 π2 k2 π2
2
X0 = 0.5 (by observation), X1 = X−1 = 2
π
(b) Yes, y(t) is periodic of period 2π and dc value 2X0 = 1.
(c) The steady state output is
y(t) = H(j0)X0 + H(jπ)X1 ejπt + H(−jπ)X−1 e−jπt

8 ej(πt−π/2) + e−j(πt−π/2)
= 2 × 0.5 + 2
π 2
8
1 + 2 cos(πt −π/2)
= |{z}
π
|{z} | {z }
A C
B

4.8 Given the Fourier series representation for a periodic signal x(t), we can compute derivatives
of it just like with any other signal.
(a) Consider the periodic train of pulses shown in Fig. 4.5, compute its derivative y(t) =
dx(t)/dt and carefully plot it. Find the Fourier series of y(t).
(b) Use the Fourier series representation of x(t) and find its derivative to obtain the Fourier
series of y(t). How doe it compare to the Fourier series obtained above.
x(t)
2
··· ···
−1.25 −0.75 −0.25 0.25 0.75 1.25 t
T0 = 1
Figure 4.5: Problem 8 and 9.
P
Answer: y(t) = k 4j sin(πk/2)ej2πkt .
Solution
(a) The derivative of a period of x(t) in −0.5 ≤ t ≤ 0.5 gives a period
dx1 (t)
y1 (t) = = 2δ(t + 0.25) − 2δ(t − 0.25) − 0.5 ≤ t ≤ 0.5
dt
of the periodic signal y(t). The Laplace transform of y1 (t) is Y1 (s) = 2e0.25s − 2e−0.25s so that
the Fourier series of y(t) of period T0 = 1, and Ω0 = 2π (just like x(t)) is
1
Yk = Y1 (s)|s=j2πk = 4j sin(πk/2)
T0
giving a Fourier series expansion

X
y(t) = 4j sin(πk/2)ej2kπt
k
(b) A period of x(t) is

x1 (t) = 2u(t + 0.25) − 2u(t − 0.25)
2 0.25s
X1 (s) = (e − e−0.25s )
s
1 sin(πk/2)
Xk = X1 (s) |s=j2kπ = 2
T0 πk

so that
X
x(t) = Xk ej2πkt
k
The Fourier series

dx(t) X X
y(t) = = j2πkXk ej2πkt = Yk ej2πkt
dt
k k
where
sin(πk/2)
Yk = j2πkXk = 4jkπ = 4j sin(πk/2)
πk
which coincides with the result in part (a).

4.9 Consider the integral of the Fourier series of the pulse signal p(t) = x(t) − 1 of period T0 = 1,
where x(t) is given in Fig. 4.5.
(a) Given that an integral of p(t) is the area under the curve, find and plot the function
Z t
s(t) = p(t)dt t≤1
−∞
indicate the values of s(t) for t = 0, 0.25, 0.5, 0.75 and 1.

(b) Find the Fourier series of p(t) and s(t) and relate their Fourier series coefficients.
(c) Suppose you want to compute the integral
Z T0 /2
p(t)dt
−T0 /2
using the Fourier series of p(t). What is the integral equal to?
(d) You can also compute the integral from the plot of p(t)
Z T0 /2
p(t)dt
−T0 /2
What is it? Does it coincide with the result obtained using the Fourier series? Explain
P
Answer: p(t) = k,odd
4ej2πkt /(kπ).
Solution
(a) The pulse p(t) has an average of zero, so its integral is zero in each period, for kT0 ≤ t ≤
(k + 1)T0 , k an integer, and T0 = 1. If s1 (t) is the integral of p1 (t) for 0 ≤ t ≤ 1 we have that
Z t Z t Z t
s1 (t) = p1 (τ )dτ = [x1 (τ ) − 1]dτ = x1 (τ )dτ − t 0≤t≤1
0 0 0
Notice that x1 (t) = 2 for 0 ≤ t ≤ 0.25 and for 0.75 ≤ t ≤ 1, and zero for 0.25 ≤ t ≤ 0.75, so for
the given times,
Z 0
s(0) = 2dτ − 0 = 0
0
Z 0.25
s(0.25) = 2dτ − 0.25 = 0.25
0
Z 0.25
s(0.5) = 2dτ − 0.5 = 0
0
Z 0.25
s(0.75) = 2dτ − 0.75 = −0.25
0
Z 0.25 Z 1
s(1) = 2dτ + 2dτ − 1 = 0
0 0.75

Letting the period of p(t) between 0 and 1 be
p1 (t) = x1 (t) − [u(t) − u(t − 1)] = [2u(t) − 2u(t − 0.25) + 2u(t − 0.75) − 2u(t − 1)] − [u(t) − u(t − 1)]
= u(t) − 2u(t − 0.25) + 2u(t − 0.75) − u(t − 1)
its integral gives the period
s1 (t) = 0.25r(t) − 0.5r(t − 0.25) + 0.5r(t − 0.75) − 0.25r(t − 1)
of s(t), 0 ≤ t ≤ 1, which can be used to verify the above results.

(b) The pulse p(t) has a period between 0 and 1 equal to
p1 (t) = u(t) − 2u(t − 0.25) + 2u(t − 0.75) − u(t − 1)
so that the Fourier series coefficients of p(t) are given by (T0 = 1, Ω0 = 2π)
1 −0.5s 0.5s
Pk = e (e − 2e0.25s + 2e−0.25s − e−0.5s )|s=j2πk
T0 s
1 −jπk
= e 2(sin(πk) − 2 sin(πk/2))
πk
4
= (−1)(k+1) sin(πk/2)
πk
where we have replaced sin(πk) = 0. The value of Pk = 0 when k is even since sin(πk/2) = 0
for k even, and for odd k since sin(πk/2) = (−1)(k+1) then Pk = 4/(πk). Since the average of
p(t) is zero, we have that
X 4 j2πkt
p(t) = e
kπ
k, odd
Since
0.25 −0.5s 0.5s 0.25
S(s) = e (e − 2e0.25s + 2e−0.25s − e−0.5s ) = P (s)
s2 s
then (T1 = 0.5 and Ω1 = 4π):
1 0.25
Sk = S(s)|s=jΩ1 k = 2 P (s)|s=j4πk
T1 s
(c)(d) The integral over a period [−T0 /2, T0 /2] = [−0.5, 0.5] is
Z 0.5 X 4 Z 0.5
p(τ )dτ = ej2πkτ dτ
−0.5 kπ −0.5
k, odd
X 4 ej2πkτ
= |0.5
kπ j2πk −0.5
k, odd
X 4 sin(πk)
= =0
kπ πk
k, odd
since sin(πk) = 0 for k odd. Since p(t) has zero mean, the integral over a period is zero, coin-
ciding with the above.

4.10 Addition of Periodic Signals — Consider a saw-tooth signal x(t) with period T0 = 2 and
period (
t 0≤t<1
x1 (t) =
0 otherwise
(a) Find the Fourier coefficients Xk using the Laplace transform. Consider the cases when k
is odd and even (k 6= 0). You need to compute X0 directly from the signal.
(b) Let y(t) = x(−t), find the Fourier coefficients Yk .
(c) The sum z(t) = x(t) + y(t) is a triangular function. Find the Fourier coefficients Zk and
compare them to Xk + Yk .
(d) Use MATLAB to plot x(t), y(t), z(t) and their corresponding magnitude line spectra.
Answers: Xk = j/(2πk), k 6= 0 even, X0 = 0.25
Solution
(a) The period x1 (t) = r(t) − r(t − 1) − u(t − 1), and T0 = 2 so that Ω0 = π. The Fourier
coefficients are

1 1 − e−s e−s 1 − e−s (1 + s)
Xk = − =
2 s2 s s=jπk 2s2 s=jπk
(
1 1 j/(2πk) k 6= 0, even
= 2 2
(−1)k (1 + jπk) − 1 = 2 2
2 π k −(2 + jπk)/(2π k ) k odd
The average value which cannot be obtained from above is
Z
1 1
X0 = tdt = 0.25
2 0
(b) The Fourier series of y(t) is
X
y(t) = x(−t) = X−k ejkΩ0 t
k
so that

 −j/(2πk)
 k 6= 0, even
Yk = X−k = 0.25 k=0


−(2 − jπk)/(2π 2 k 2 ) k odd
(c) Adding the above signals z(t) = x(t)+y(t) we get triangular pulses. The Fourier coefficients
of z(t) are


 0 k 6= 0, even
Zk = Xk + Yk = 0.5 k=0

 2 2
−2/(π k ) k odd

Notice that the Fourier coefficients of x(t) and y(t) which are neither even nor odd are complex,
while the Fourier coefficients of z(t) are real given that it is an even signal.
A period of z(t) is
z1 (t) = r(t) − 2r(t − 1) + r(t − 2)
and T0 = 2, Ω0 = π. The Fourier coefficients of z(t) are

(
1 (−1)k 0 k 6= 0, even
Zk = Z1 (s)|s=jπk = 2 2 (1 − cos(πk)) =
2 π k −2/(π 2 k 2 ) k odd
and by inspection Z0 = 0.5. which coincide with the ones obtained before. The following script
gives an approximation of z(t) using 41 harmonics, and shown in Fig. 4.6.
% Pr. 4_10
clear all; clf
% magnitude line spectra
X1(1)=0.25;
for k=1:2:40,
X1(k+1)=abs(-(2+j*pi*k)/(2*piˆ2*kˆ2));
end
for k=2:2:40,
X1(k+1)=abs(1/(2*pi*k));
end
kk=-40:40;
X=[fliplr(X1) X1(2:41)];
figure(1)
subplot(311)
stem(kk,X); grid;axis([-40 40 0 1]);ylabel(’|X_k|’)
Y1(1)=0.25;
for k=1:2:40,
Y1(k+1)=abs((2+j*pi*k)/(2*piˆ2*kˆ2));
end
for k=2:2:40,
Y1(k+1)=abs(-1/(2*pi*k));
end
Y=[fliplr(Y1) Y1(2:41)];
subplot(312)
stem(kk,Y); grid; axis([-40 40 0 1]);ylabel(’|Y_k|’)
Z=[fliplr(Y1+X1) Y1(2:41)+X1(2:41)];
subplot(313)
stem(kk,Z); grid; axis([-40 40 0 1])

xlabel(’k’);ylabel(’|Z_k|’)
% approximate of z(t)
t=0:0.001:10;
z=0.5*ones(1,length(t));
for k=1:2:10,
z=z-4*cos(k*pi*t)/(pi*k)ˆ2;
end
figure(2)
plot(t,z);grid
xlabel(’t (sec)’); ylabel(’z_1(t)’)
0.4 0.9
|Xk|
0.2
0.8
0
−40 −30 −20 −10 0 10 20 30 40 0.7
0.6
0.6
0.4
z1(t)
|Yk|
0.5
0.2
0.4
0
−40 −30 −20 −10 0 10 20 30 40
0.3
0.6
0.2
0.4
|Zk|
0.2 0.1
0 0
−40 −30 −20 −10 0 10 20 30 40 0 1 2 3 4 5 6 7 8 9 10
k t (sec)
Figure 4.6: Problem 10: Line spectra for x(t), y(t) and z(t), (right) approximate ẑ(t).

4.11 Fourier series coefficients via Laplace — The computation of the Fourier series coefficients is helped
by the relation between the formula for these coefficients and the Laplace transform of a period of the
periodic signal.
(a) A periodic signal x(t) of period T0 = 2 sec., has as a period the signal x1 (t) = u(t) − u(t − 1). Use
the Laplace transform of x1 (t) to obtain the Fourier coefficients of x(t).
(b) Use MATLAB to approximate x(t) with 40 harmonics, plot the approximate signal x̂(t) and its mag-
nitude line spectrum.
Answers: Xk = sin(πk/2)e−jkπ/2 /(πk/2), X0 = 0.5.
Solution
(a) x(t) is a train of rectangular pulses.The Laplace transform of the period x1 (t) is
1 e−s/2 (es/2 − e−s/2 )

X1 (s) = (1 − e−s ) =
s s
and T0 = 2, so Ω0 = π. The Fourier series coefficients are then
1 sin(πk/2) −jkπ/2
Xk = X1 (s) |s=jπk = e X0 = 0.5
2 πk/2
(b) The following script is used to find and plot the magnitude line spectrum of x(t), and to approximate
the signal with 40 harmonics.
1.2
1
0.5
0.8
0.4
0.6
xa(t)
|Xk|
0.3
0.4
0.2
0.2
0.1
0
0 −0.2
−40 −30 −20 −10 0 10 20 30 40 0 1 2 3 4 5 6 7 8 9 10
t (sec)
Figure 4.7: Problem 11: Magnitude line spectrum and approximate of train of pulses.
% Pr. 4_11
clear all; clf
X(1)=0.5;
for k=1:40,
X(k+1)=abs(0.5*sin(pi*k/2)/(pi*k/2));
end
kk=-40:40;
X=[fliplr(X) X(2:41)];

figure(1)
stem(kk,X); grid;axis([-40 40 0 0.6]);ylabel(’|X_k|’)
% approximate of z(t)
t=0:0.001:10;
x=0.5*ones(1,length(t));
for k=1:40,
x=x+(sin(pi*k/2)/(pi*k/2))*cos(pi*k*t-k*pi/2);
end
figure(2)
plot(t,x);grid
xlabel(’t (sec)’); ylabel(’x_a(t)’)

4.12 Time-support and frequency content — The support of a period of a periodic signal relates inversely to
the support of the line spectrum. Consider two periodic signals: x(t) of period T0 = 2 and y(t) of period
T1 = 1 and with periods
x1 (t) = u(t) − u(t − 1) 0≤t≤2

y1 (t) = u(t) − u(t − 0.5) 0≤t≤1
(a) Find the Fourier series coefficients for x(t) and y(t).
(b) Use MATLAB to plot the magnitude line spectra of the two signals from 0 to 20π rad/sec. Plot them
on the same figure so you can determine which has a broader support. Indicate which signal is
smoother and explain how it relates to its line spectrum.
Answers: y1 (t) = x1 (2t), line spectrum of x(t) is compressed when compared to that of y(t).
Solution
The periods of the two signals are:
x1 (t) = u(t) − u(t − 1) 0 ≤ t ≤ T0 = 2

y1 (t) = x1 (2t) = u(2t) − u(2t − 1) = u(t) − u(t − 0.5) 0 ≤ t ≤ T1 = 1
the Fourier series coefficients of these signals are
1 −s/2 s/2 sin(πk/2) −jπk/2

Xk = e (e − e−s/2 ) |s=jπk = e
2s πk
1 sin(πk/2) −jπk/2
Yk = e−s/4 (es/4 − e−s/4 ) |s=j2πk = e
s πk
We see that the coefficients for the two signals are identical, but the fundamental frequency for x(t) is
Ω0 = 2π/T0 = π while the one for y(t) is Ω1 = 2π/T1 = 2π = 2Ω0 so that the line spectrum of x(t) is
compressed when compared to that of y(t). Thus compression in a period y(t) = x(2t) corresponds to
expansion in the frequency domain. Thus, x(t) is smoother than y(t) as this displays higher frequencies.
% Pr. 4_12
clear all; clf
X(1)=0.5;Y(1)=X(1);
for k=1:40,
X(k+1)=abs(0.5 *sin(pi*k/2)/(pi*k*0.5));
end
Y=X;
kk=-40:40;
X=[fliplr(X) X(2:41)];
Y=[fliplr(Y) Y(2:41)];
figure(1)
subplot(211)
stem(pi*kk,X); grid;axis([-40*pi 40*pi 0 0.6]);ylabel(’|X_k|’)

subplot(212)
stem(2*pi*kk,Y); grid;axis([-40*2*pi 40*2*pi 0 0.6]);ylabel(’|Y_k|’)
0.5
0.4
|Xk|
0.3
0.2
0.1
0
−100 −50 0 50 100
0.5
0.4
|Yk|
0.3
0.2
0.1
0
−250 −200 −150 −100 −50 0 50 100 150 200 250
Figure 4.8: Problem 12: The line spectrum of Yk is an expansion of that of Xk (notice the frequencies
for each).

4.13 Windowing and music sounds — In the computer generation of musical sounds, pure tones
need to be windowed to make them more interesting. Windowing mimics the way a musician
would approach the generation of a certain sound. Increasing the richness of the harmonic fre-
quencies is the result of the windowing as we will see in this problem. Consider the generation
of a musical note with frequencies around fA = 880 Hz. Assume our “musician” while playing
this note uses three strokes corresponding to a window w1 (t) = r(t)−r(t−T1 )−r(t−T2 )+r(t−
T0 ), so that the resulting sound would be the multiplication, or windowing, of a pure sinusoid
cos(2πfA t) by a periodic signal w(t) with w1 (t) a period that repeats every T0 = 5T where T is
the period of the sinusoid. Let T1 = T0 /4, and T2 = 3T0 /4.
(a) Analytically determine the Fourier series of the window w(t) and plot its line spectrum
using MATLAB. Indicate how you would choose the number of harmonics needed to
obtain a good approximation to w(t).
(b) Use the modulation or the convolution properties of the Fourier series to obtain the coeffi-
cients of the product s(t) = cos(2πfA t)w(t). Use MATLAB to plot the line spectrum of this
periodic signal and again determine how many harmonic frequencies you would need to
obtain a good approximation to s(t).
(c) The line spectrum of the pure tone p(t) = cos(2πfA t) only displays one harmonic, the one
corresponding to the fA = 880 Hz frequency, how many more harmonics does s(t) have?
To listen to the richness in harmonics use the function sound to play the sinusoid p(t) and
s(t) (use Fs = 2 × 880 Hz to play both).
(d) Consider a combination of notes in a certain scale, for instance let
p(t) = sin(2π × 440t) + sin(2π × 550t) + sin(2π × 660t).
Use the same windowing w(t), and let s(t) = p(t)w(t). Use to plot p(t) and s(t) and to
compute and plot their corresponding line spectra. Use sound to play p(nTs ) and s(nTs )
using Fs = 1000.
Answers: Since the fundamental period of the sinusoid is T = 1/880 let T0 = 5T = 5/880 sec.,
T1 = T0 /4 and T2 = 3T0 /4.
Solution
(a) The fundamental period of the sinusoid is T = 1/880 so that T0 = 5T = 5/880 sec.,
T1 = T0 /4 and T2 = 3T0 /4. The Fourier series coefficients of ω(t) are
1 1
Wk = W1 (s)|s=jkΩ0 = 2 (1 − e−sT0 /4 − e−s3T0 /4 + e−sT0 )|s=jkΩ0
T0 s
e−sT0 /2 sT0 /2
= (e − esT0 /4 − e−sT0 /4 + e−sT0 /2 )|s=jkΩ0
s2
2e−jkπ 2π
= 2 2
(cos(πk) − cos(πk/2)) Ω0 = = 352π rad/sec
−Ω0 k T0

(b) The signal s(t) = w(t) cos(2πfA t) shifts the harmonics of w(t) to a central frequency 2πfA .
The number of harmonic frequencies is increased by windowing the sinusoid or sum of sinu-
soids.
%Pr. 4_13
clear all; clc
% parameters
Fs = 880; %frequency
T = 1/Fs; %sample period
T0 = 5*T; T1 = T0/4; T2 = T0*3/4;
% a)----------------------------------------------------------
% define one period of w signal -> w1
Ts=1/17600; % overall sample period
t=0:Ts:T0-Ts; Nn=length(t);
x1=[t(1:ceil(Nn/4)) t(ceil(Nn/4)+1)*ones(1,ceil(Nn/4))]; x2=fliplr(x1);
w1=[x1 x2];
N=length(w1); %number of samples in w1
% plot line spectrum
figure(1)
FSeries(w1,Ts,T0,N);
% harmonics H to approximate w1?
w0=2*pi/T0;
DC=0.001065; H=3;
for k=1:H,
a(k)=2*(cos(k*w0*T0/4)-cos(k*w0*T0/2))/T0/kˆ2/w0ˆ2;
end
a=[DC 2*a];
figure(2)
for N1=2:H+1,
x=a(1)*ones(1,length(t));
for k=2:N1,
x=x+a(k)*cos((k-1)*w0.*(t+T0/2));
end
plot(t,x);
axis([0 max(t) 1.1*min(x) 1.1*max(x)]);
hold on;
plot(t,w1,’r’)
ylabel(’Amplitude’);xlabel(’t (sec)’);
legend(’aprox w1(t)’,’w1(t)’);grid;
hold off
pause(0.1);
clear x;

end
clear H a x1
% b)----------------------------------------------------------
s = cos(2*pi*Fs*t).*w1; %modulated signal s
figure(3);
FSeries(s,Ts,T0,N); % line spectrum of s
% how many harmonics H are necessary to approximate s?
w0=2*pi/T0;
DC=0.001065;H=3;
for k=1:H,
a(k)=2*(cos(k*w0*T0/4)-cos(k*w0*T0/2))/T0/kˆ2/w0ˆ2;
end
a=[DC 2*a];
figure(4)
for N1=2:H+1,
for k=2:N1,
x=x+a(k)*cos((k-1)*w0.*(t+T0/2));
x1=x.*cos(5*w0.*t);
end
plot(t,x1);
axis([0 max(t) 1.1*min(s) 1.1*max(s)]);
hold on;
plot(t,s,’r’)
ylabel(’Amplitude’); xlabel(’t (sec)’);
legend(’aprox s(t)’,’s(t)’); grid;
hold off
pause(0.1);
clear x;
end
% c)----------------------------------------------------------
p = cos(2*pi*Fs*t); %signal p
figure(5);
FSeries(p,Ts,T0,N); %line spectrum of p
%sound(p); pause(1); sound(w0*s);pause(2);
% d)----------------------------------------------------------
p = sin(2*pi*440*t)+sin(2*pi*550*t)+sin(2*pi*660*t); %signal p
s = p.*w1; %modulated signal s
figure(6);
FSeries(s,Ts,T0,N); %line spectrum of s
% sound(100*p); pause(1); sound(100*w0*s);

−3 −4 −3
x 10 Periodic signal x 10 x 10 Periodic signal
2.5 aprox w1(t) 2

w1(t)
2 14 1
1.5
x(t)
x(t)
0
1 12
−1
0.5 −2
0 10
Amplitude
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02

t (sec) t (sec)
−3
x 10 Magnitude Line Spectrum
−3
x 10 Phase Line Spectrum 8 −4
x 10 Magnitude Line Spectrum
−4
x 10 Phase Line Spectrum
1 5
1 5
0.5 6
0.8 4
<Xk|
<Xk|
|Xk|
|Xk|
0.6 0 3 0
4
0.4 2
−0.5
0.2 1
2
0 −1 0 −5
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 0 1 2 3 4 5 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
w (rad/sec) 4 w (rad/sec) 4 t (sec) −3 w (rad/sec) 4 w (rad/sec) 4
x 10 x 10 x 10 x 10 x 10
Figure 4.9: Problem 13. Right to left: window and line spectra, approximation of window, win-
dowed sinusoid and its line spectra.
−3 −3
x 10 Periodic signal x 10 Periodic signal
2
1.5
aprox s(t) 2
s(t) 1
0
1
x(t)
x(t)
0
−2
−1 −4
0.5
−2
Amplitude
0 0.005 0.01 0.015 0.02 0 0.005 0.01 0.015 0.02

0 t (sec) t (sec)
−16 −4 −3
Magnitude Line Spectrum x 10 Phase Line Spectrum x 10 Magnitude Line Spectrum x 10 Phase Line Spectrum
1.5
0.5 5 5
−0.5 1
0.4 4
0.5
0.3 3
<Xk|
<Xk|
|Xk|
|Xk|
0 0
−1
0.2 2 −0.5
0.1 1 −1
−1.5 −5
0 0 −1.5
0 1 2 3 4 5 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
t (sec) −3 w (rad/sec) 4 w (rad/sec) 4 w (rad/sec) 4 w (rad/sec) 4
x 10 x 10 x 10 x 10 x 10
Figure 4.10: Problem 13: from right to left: approximation of windowed signal, line spectra of
sinusoid, windowed sum of sinusoids and line spectra.

4.14 Computation of π — As you know, π is an irrational number that can only be approximated
by a number with a finite number of decimals. How to compute this value recursively is a
problem of theoretical interest. In this problem we show that the Fourier series can provide
that formulation.
(a) Consider a train of rectangular pulses x(t), with a period
x1 (t) = 2[u(t + 0.25) − u(t − 0.25)] − 1, −0.5 ≤ t ≤ 0.5
and period T0 = 1. Plot the periodic signal and find its trigonometric Fourier series.
(b) Use the above Fourier series to find an infinite sum for π.
(c) If πN is an approximation of the infinite sum with N coefficients, and π is the value given
by , find the value of N so that πN is 95% of the value of π given by MATLAB.
P∞
Answer: π = 4 k=1 sin(πk/2)/k
Solution
(a) The signal x(t) is zero-mean. The Fourier series coefficients are obtained from (Ω0 =
2π/T0 = 2π)
2 0.25s sin(πk/2)
X1 (s) = (e − e−0.25s )|s=j2kπ =
s πk/2
so that
∞
X sin(πk/2) j2kπt
x(t) = e
πk/2
k=−∞,6=0
X∞
sin(πk/2)
= 2 cos(2πkt)
πk/2
k=1
Since x(0) = 1, letting t = 0 in both sides and then multiplying by π we get that
∞
X sin(πk/2)
π=4
k
k=1
If we let
N
X sin(πk/2)
πN = 4 = 0.95π
k
k=1
we use MATLAB to find the value of N .
% Pr. 4_14
clear all; clc
N=100; x = 0; %DC value
for k=1:N

x = x + 4*sin(k*pi/2)/k;
xx(k) =x;
end
format long
display(’Error of approximation for’)
N
e = pi-x
ppi=pi*ones(1,length(xx)); n=1:N;
figure(1)
subplot(211); plot(n,xx); title(’Approximation of PI’); axis([1 N 0 4.5])
hold on; plot(n,ppi,’r’); grid; hold off; legend(’approximate’, ’pi’)
subplot(212); plot(n,xx-pi); title(’Error of approximation’); grid
% N for 95 percent of pi
sum = 0; %initial approximation
for i = 1:N,
sum = sum + 4*cos(i*pi/2-pi/2)/i;
if (sum >=(0.95*pi))&&(sum<=pi) break;
end
end
display([num2str(i) ’ coeficients needed of Fourier series for 95% approx. of PI ’])
Error of approximation for

N = 100
e = 0.019998000998784
15 coeficients needed of Fourier series for 95\% approx. of PI

Approximation of PI
2 approximate
pi
1
0
10 20 30 40 50 60 70 80 90 100
n
Error of approximation
1
0.5
−0.5
0 10 20 30 40 50 60 70 80 90 100
n
Figure 4.11: Problem 14: Approximation of π by series, error of approximation

4.15 Square error approximation of periodic signals — To understand the Fourier series consider
a more general problem, where a periodic signal x(t), of period T0 , is approximated as a finite
sum of terms
N
X
x̂(t) = X̂k φk (t)
k=−N
where {φk (t)} are ortho-normal functions. To pose the problem as an optimization problem,
consider the square error Z
ε= |x(t) − x̂(t)|2 dt
T0
and we look for the coefficients {X̂(k)} that minimize ε.
(a) Assume that x(t) as well as x̂(t) are real valued, and that x(t) is even so that the Fourier
series coefficients Xk are real. Show that the error can be expressed as
Z N
X Z N
X
2
ε= x (t)dt − 2 X̂k x(t)φk (t)dt + |X̂` |2 T0 .
T0 k=−N T0 `=−N
(b) Compute the derivative of ε with respect to X̂n and set it to zero to minimize the error.
Find X̂n .
(c) In the Fourier series the {φk (t)} are the complex exponentials and the {X̂n } coincide with
the Fourier series coefficients. To illustrate the above procedure consider the case of the
pulse signal x(t), of period T0 = 1 and a period
x1 (t) = 2[u(t + 0.25) − u(t − 0.25)]
Use MATLAB to compute and plot the approximation x̂(t) and the error ε for increasing
values of N from 1 to 100.
(d) Concentrate your plot of x̂(t) around one of the discontinuities, and observe the Gibb’s
phenomenon. Does it disappear when N is very large. Plot x̂(t) around the discontinuity
for N = 1000.
R
Answer: dε/dX̂n = −2 T0
x(t)φn (t)dt + 2T0 X̂n = 0.
Solution
(a) Assuming x(t) and x̂(t) being real and even, so that X̂k are real we have
Z Z
ε = (x(t) − x̂(t))2 dt = x2 (t) − 2x(t)x̂(t) + x̂2 (t)dt
T T0
Z 0 Z X Z XX
= 2
x (t)dt − 2 x(t) X̂k φk (t)dt + X̂k X̂`∗ φk (t)φ∗` (t)
T0 T0 k T0 k `
Z XZ XX Z
= x2 (t)dt − 2 x(t)X̂k φk (t)dt + X̂k X̂` φk (t)φ∗` (t)
T0 k T0 k ` T0

To minimize the error we set its derivative w.r.t. X̂n to zero, i.e.,
Z Z
dε 1
= −2 x(t)φn (t)dt + 2T0 X̂n = 0 so that X̂n = x(t)φn (t)dt
dX̂n T0 T0 T0
% Pr. 4_15
clear all; clc
% c)d)--------------------------------------------------------
% Fourier Series of a train of pulses of period T0=1 and dc value 1
w0 = 2*pi; T0 = 1; DC=1; N=100;
for k=1:N,
a(k)=sin(k*pi/2)/(k*pi/2);
end
a=[DC 2*a];
Ts = 0.001; %sample period

t=0:Ts:1-Ts; L=length(t); %number of samples
x1=[ones(1,L/4) zeros(1,L/2) ones(1,L/4)]; %train signal

x1=2*DC*x1; %DC gain adaptation
for N1=2:N+1,
for k=2:N1,
x=x+a(k)*cos(2*pi*(k-1).*t);
end
%plot figures
figure(1)
plot(t,x); axis([0 max(t) 1.1*min(x) 1.1*max(x)]); hold on
plot(t,x1,’r’); ylabel(’x(t), x_a(t)’); xlabel(’t (sec)’); hold off
figure(3)
plot(t,x); axis([0.175 0.275 1.75 2.25]); hold on
plot(t,x1,’r’); title(’Gibbs Phenomenon’);
ylabel(’x(t), x_a(t)’); xlabel(’t (sec)’);
pause(0.05)
%compute error
e=(x1-x).ˆ2;
error(N1-1)=0;
for k = 1:length(e)
error(N1-1) = error(N1-1) + e(k);
end
clear x

end
figure(2)
axis([1 N 0 1.1*max(error)]); hold on
plot(error,’linewidth’,2);
title(’e = int((x-x1)ˆ2)’); ylabel(’Amplitude of error’); xlabel(’t (samples)’)
hold off e = int((x−x1)2)
200
180
2
160
1.5 140
Amplitude of error
120
x(t), xa(t)
100
1
80
60
0.5
40
20
0
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10 20 30 40 50 60 70 80 90 100
t (sec) t (samples)
Gibbs Phenomenon
2.25
2.2
2.15
2.1
2.05
x(t), xa(t)
1.95
1.9
1.85
1.8
1.75
0.18 0.19 0.2 0.21 0.22 0.23 0.24 0.25 0.26 0.27
t (sec)
Figure 4.12: Problem 15: Gibb’s phenomenon: (top–left) square pulse and approximate, (bottom)
detail of approximation around the discontinuity; (top-right) approximation error.


Chapter 5
Frequency Analysis: the Fourier

Transform
5.1 Basic Problems
5.1 The Fourier transform of a signal x(t) is

2
X(Ω) = .
1 + Ω2
Use properties of the Fourier transform to
(a) find the integral Z ∞

x(t)dt,
−∞
(b) find the value x(0).

(c) Let s = jΩ or Ω = (s/j), find X(s) and from it obtain x(t).
Answers: x(0) = 1; x(t) = e−|t| .
Solution
(a) The Fourier pair is

Z ∞ Z ∞
1
x(t) = X(Ω)e jΩt
dΩ ⇔ X(Ω) = x(t)e−jΩt dt
2π −∞ −∞
For Ω = 0 the Fourier transform gives

Z ∞
x(t)dt = X(0) = 2
−∞
1
(b) From the inverse Fourier transform

Z ∞ Z ∞
1 2 2
x(0) = X(Ω)dΩ = dΩ
2π −∞ 2π 0 1 + Ω2
2 2
= tan−1 (Ω)|∞
0 = (π/2) = 1
π π
(c) Ω = s/j
2 −2 1 1
X(s) = = = + then
1 + (s/j)2 (s − 1)(s + 1) 1−s s+1
x(t) = et u(−t) + e−t u(t)

5.2 Find the Fourier transform of δ(t − τ ) and use it to find the Fourier transforms of
(a) δ(t − 1) + δ(t + 1)

(b) cos(Ω0 t)
(c) sin(Ω0 t)
Answers: F[δ(t − 1) + δ(t + 1)] = 2 cos(Ω) and apply duality for the others.
Solution F[δ(t − τ )] = L[δ(t − τ )]s=jΩ = e−jΩτ so
(a) By linearity and time–shift
F[δ(t − 1) + δ(t + 1)] = 2 cos(Ω)
(b) By duality
0.5[δ(t − τ ) + δ(t + τ )] ↔ cos(Ωτ )

cos(Ω0 t) ↔ π[δ(Ω + Ω0 ) + δ(Ω − Ω0 )]
by letting τ = Ω0 in the second equation.

(c) Considering
F[δ(t − 1) − δ(t + 1)] = 2j sin(Ω),
by duality
−0.5j[δ(t − τ ) + δ(t + τ )] ↔ sin(Ωτ )

sin(Ω0 t) ↔ − jπ[δ(Ω + Ω0 ) + δ(Ω + Ω0 )]
by letting τ = Ω0 in the second equation.

5.3 A periodic signal x(t) has a period
x1 (t) = r(t) − 2r(t − 1) + r(t − 2), T0 = 2.
(a) Find the Fourier series of z(t) = d2 x(t)/dt2 using the Laplace transform. Then use the
derivative property to find the Fourier transform of x(t).
(b) To check your results figure out what the dc value of x(t) and z(t) should be, and whether
the Fourier coefficients should be real, purely imaginary or complex.
(c) Show that the following equation can be used to compute the Fourier transform of the
periodic signal z(t):
∞
2π X
Z(Ω) = Z1 (Ω) δ(Ω − kΩ0 )
T0
k=−∞
where Z1 (Ω) is the Fourier transform of z1 (t) = d2 x1 (t)/dt2 . Find its inverse z(t).
P
Answer: z(t) = m 2ej(2m+1)πt .
Solution
(a) A period of z(t) is z1 (t) = d2 x1 (t)/dt2 = δ(t) − 2δ(t − 1) + δ(t − 2), with period T0 = 2
(Ω0 = π), its Fourier series coefficients are
1 1
Zk = Z1 (s)|s=jπk = 2e−s (cosh(s) − 1)|s=jπk
2 2 (
k+1 k 2 k odd
= (−1) (1 − (−1) ) =
0 k even
then
∞
X ∞
X
z(t) = 2ejkπt = 2ej(2m+1)πt
k=−∞,odd m=−∞
by letting k = 2m + 1. By the derivative property
Zk −2
Xk = = 2 2 , k odd
(jkΩ0 )2 k Ω0
and
∞
X ∞
X
2(−1)k jkπt 2(−1)2m+1 j(2m+1)πt
x(t) = e = e
k 2 Ω20 m=−∞
(2m + 1)2 Ω20
k=−∞,odd
(b) The average X0 = 0.5, and x(t) − 0.5 is even so Xk , k 6= 0 are real and
z(t) is even and zero mean, so Zk are real and Z0 = 0

(c) The Fourier transform is

∞ ∞
2π X X Z1 (Ω0 )
Z(Ω) = Z1 (Ω) δ(Ω − kΩ0 ) = 2π δ(Ω − kΩ0 )
T0 T0
k=−∞ k=−∞
∞
X
= 2πZk δ(Ω − kΩ0 )
k=−∞
and the inverse

X X
z(t) = F −1 [2πZk δ(Ω − kΩ0 )] = Zk ejkΩ0
k k

5.4 A sinc signal x(t) = sin(0.5t)/(πt) is passed through an ideal low-pass filter with a frequency
response H(jΩ) = u(Ω + 0.5) − u(Ω − 0.5)
(a) Find the Fourier transform X(Ω) and carefully plot it.
(b) Find the output y(t) of the filter by calculating first Y (Ω). What is the convolution integral
of x(t) with x(t)? Explain.
Answers: X(Ω) = u(Ω + 0.5) − u(Ω − 0.5); y(t) = (x ∗ x)(t) = x(t).
Solution
(a) Let X(Ω) = A[u(Ω + Ω0 ) − u(Ω − Ω0 )] its inverse Fourier transform is

Z Ω0
1 A sin(Ω0 t)
x(t) = AejΩt dΩ =
2π −Ω0 πt
so A = 1, Ω0 = 0.5 and X(Ω) = u(Ω + 0.5) − u(Ω − 0.5).

(b) Y (Ω) = H(Ω)X(Ω) = X(Ω) so that y(t) = (x ∗ x)(t) = x(t), or convolution of a sinc
function with itself is a sinc.

5.5 The Fourier transform of a signal x(t) is
|Ω|
X(Ω) = [u(Ω + π) − u(Ω − π)]
π
(a) Carefully plot X(Ω) as function of Ω.
(b) Determine the value of x(0)
Answer: x(0) = 0.5.
Solution
(a) Plot of X(Ω) as function of Ω:
X(Ω)
Ω
−π π
(b)
Z π Z π
1 |Ω| 2 Ω 1
x(0) = dΩ = dΩ =
2π −π π 2π 0 π 2

5.6 The Fourier series coefficients of a periodic signal x(t), with fundamental frequency Ω0 = π/4
∗
are X1 = X−1 = j, X5 = X5∗ = 2 and the rest are zero. Suppose x(t) is the input to a band-pass
filter with the following magnitude and phase responses


 1 π ≤ Ω ≤ 1.5π
|H(jΩ)| = 1 −1.5π ≤ Ω ≤ −π ∠H(jΩ) = −Ω.


0 otherwise
Let y(t) be the output of the filter.
(a) Determine the Fourier series of x(t) in the trigonometric form

∞
X
x(t) = Xk cos(kΩ0 t + ∠Xk )
k=0
and then find the steady-state response of y(t).

(b) Find the Fourier transform of x(t) and y(t) and carefully plot their magnitudes and phases.
Answers: x(t) = −2 sin(πt/4) + 4 cos(5πt/4); y(t) = 4 cos(5πt/4 − 5π/4).
Solution
(a) We have
x(t) = j(ejπt/4 − e−jπt/4 ) + 2(ej5πt/4 + e−j5πt/4 )

= −2 sin(πt/4) + 4 cos(5πt/4)
The steady state response of the output is
y(t) = 4|H(j5π/4)| cos(5πt/4 + ∠H(j5π/4)

= 4 cos(5πt/4 − 5π/4)
because the frequency component at π/4 is filtered out.

(b) The Fourier transforms are
X(Ω) = 2π [X1 δ(Ω − π/4) + X−1 δ(Ω + π/4) + X5 δ(Ω − 5π/4) + X−5 δ(Ω + 5π/4)]
= 2π [jδ(Ω − π/4) − jδ(Ω + π/4) + 2δ(Ω − 5π/4) + 2δ(Ω + 5π/4)]
Y (Ω) = 4πe−j5π/4 δ(Ω − 5π/4) + 4πej5π/4 δ(Ω + 5π/4)

5.7 An analog averager is characterized by the following relationship
dy(t)
= 0.5[x(t) − x(t − 2)]
dt
where x(t) is the input and y(t) the output. If x(t) = u(t) − 2u(t − 1) + u(t − 2)
(a) Find the Fourier transform of the output Y (Ω).

(b) Find y(t) from Y (Ω).
Answer: y(t) = 0.5r(t) − r(t − 1) + r(t − 3) − 0.5r(t − 4).
Solution
(a) The FT of the equation characterizing the averager is
jΩY (Ω) = 0.5[X(Ω) − X(Ω)e−j2Ω ]
so that
sin(Ω)e−jΩ
Y (Ω) = X(Ω)
Ω
where the FT of x(t) is
e−jΩ sin2 (Ω/2) −jΩ

X(Ω) = [2(cos(Ω) − 1)] = −4 e
jΩ jΩ
(b) Calling Z(Ω) = X(Ω)/(jΩ) we have that y(t) = 0.5z(t) − 0.5z(t − 2) and
2
X(Ω) sin(Ω/2)
Z(Ω) = = e−jΩ
jΩ Ω/2
so that z(t) is the integral of x(t) = u(t) − 2u(t − 1) + u(t − 2), or z(t) = r(t) − 2r(t − 1) + r(t − 2),
replacing it in the above expression for y(t) we get
y(t) = 0.5r(t) − r(t − 1) + 0.5r(t − 2) − 0.5r(t − 2) + r(t − 3) − 0.5r(t − 4)

= 0.5r(t) − r(t − 1) + r(t − 3) − 0.5r(t − 4)

5.8 Fourier transform from Laplace transform of infinite support signals — For signals with
infinite support, their Fourier transforms cannot be derived from the Laplace transform unless
they are absolutely integrable or the region of convergence of the Laplace transform contains
the jΩ axis. Consider the signal x(t) = 2e−2|t|
(a) Plot the signal x(t) for −∞ < t < ∞.

(b) Use the evenness of the signal to find the integral
Z ∞
|x(t)|dt
−∞
and determine whether this signal is absolutely integrable or not.

(c) Use the integral definition of the Fourier transform to find X(Ω).
(d) Use the Laplace transform of x(t) to verify the above found Fourier transform.
(e) Use MATLAB’s symbolic function fourier to compute the Fourier transform of x(t). Plot
the magnitude spectrum corresponding to x(t).
Answers: X(Ω) = 8/(4 + Ω2 ).
Solution
(a) This is a decaying exponential for both negative and positive times.
(b) The signal is absolutely integrable,
Z ∞ Z ∞
e−2t ∞
|x(t)|dt = 2 2e−2t dt = 4 | =2
−∞ 0 −2 t=0
(c) Using the integral definition of the Fourier transform
Z 0 Z ∞
X(Ω) = 2 e2t e−jΩt dt + 2 e−2t e−jΩt dt
−∞ 0
t(2−jΩ) −t(2+jΩ)
2e 2e
= |0 − |∞
2 − jΩ −∞ 2 + jΩ 0
2 2
= +
2 − jΩ 2 + jΩ
8
=
4 + Ω2
(d) The Laplace transform of X(s) is
2 2 8
X(s) = + =
−s + 2 s + 2 4 − s2
with ROC: −2 ≤ σ ≤ 2, which includes the jΩ-axis. so that
8
X(Ω) = X(s)|s=jΩ =
4 + Ω2

%% Pr. 5_8
syms t w x
x=2*exp(-2*abs(t)); X=fourier(x)
figure(1)
subplot(211)
ezplot(x,[-2,2]); grid; axis([-2 2 0 2.2])
subplot(212)
ezplot(abs(X),[-30,30]);grid;axis([-20 20 0 2.2]);
xlabel(’\Omega (rad/sec)’); ylabel(’|X(\Omega)|’)
2/exp(2 abs(t))
2
1.5
0.5
0
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
t
8/abs(w2 + 4)
2
1.5
|X(Ω)|
0.5
0
−20 −15 −10 −5 0 5 10 15 20
Ω (rad/sec)

5.9 Smoothness and frequency — Let the signal x(t) = r(t+1)−2r(t)+r(t−1) and y(t) = dx(t)/dt.
(a) Plot x(t) and y(t)

(b) Find X(Ω) and carefully plot its magnitude spectrum. Is X(Ω) real? Explain.
(c) Find Y (Ω) (use properties of Fourier transform) and carefully plot its magnitude spec-
trum. Is Y (Ω) real? Explain.
(d) Determine from the above spectra which of these two signals is smoother. Use MATLAB
integration function int to find the Fourier transforms. Plot 20 log10 |Y (Ω)| and 20 log10 |X(Ω)|
and decide. Would you say in general that computing the derivative of a signal generates
high frequencies, or possible discontinuities?
Answers: |Y (Ω)| = |X(Ω)||Ω|.
Solution
(a) The triangular pulse x(t) is even
x(t) = r(t + 1) − 2r(t) + r(t − 1)

y(t) = u(t + 1) − 2u(t) + u(t − 1)
which is an odd signal composed of two pulses of duration 1.

(b) The Fourier transform of x(t) is computed from its Laplace transform
1 s es es
X(s) = 2
[e − 2 + e−s ] = 2 [1 − 2e−s + e−2s ] = 2 (1 − e−s )2
s s s
which cancels out the poles at s = 0 so that the region of convergence is the whole s-plane. By
letting s = jΩ we have that
2 sin2 (Ω/2)
X(Ω) = [cos(Ω) − 1] =
−Ω2 (Ω/2)2
which is real given that x(t) is even.

(c) Because of the derivative
sin2 (Ω/2)
Y (Ω) = jΩX(Ω) = jΩ
(Ω/2)2
which is imaginary given that y(t) is odd. The magnitude
|Y (Ω)| = |X(Ω)||Ω|
which is the product of the sinc square and the absolute value of Ω. Thus x(t) is smoother than
its derivative y(t).

%% Pr. 5_9
syms x t w y X Y
x=(t+1)*(heaviside(t+1)-heaviside(t))+(1-t)*(heaviside(t)-heaviside(t-1));
y=heaviside(t+1)-2*heaviside(t)+heaviside(t-1);
figure(1)
subplot(211)
ezplot(x,[-1.5,1.5]);grid
subplot(212)
ezplot(y,[-1.5,1.5]);grid
X=int(x*exp(-j*w*t),t,-Inf,-Inf);
Y=int(y*exp(-j*w*t),t,-Inf,-Inf);
figure(2)
subplot(211)
XX=X*conj(X);
ezplot(10*log10(XX),[-100,100]);grid;axis([-100 100 -80 1]);ylabel(’X(\Omega)’)
subplot(212)
YY=Y*conj(Y);
ezplot(10*log10(YY),[-100,100]);grid;axis([-100 100 -80 1]);ylabel(’Y(\Omega)’)
(t+1) (heaviside(t+1)−heaviside(t))+(1−t) (heaviside(t)−heaviside(t−1))

10 log(16/w2 sin(1/2 w)2 conj(1/w2 sin(1/2 w)2))/log(10)
1
0
0.8
0.6 −20
X(Ω)
0.4 −40
0.2
−60
0
−80
−1.5 −1 −0.5 0 0.5 1 1.5 −100 −80 −60 −40 −20 0 20 40 60 80 100
t w
heaviside(t+1)−2 heaviside(t)+heaviside(t−1) 10 log(16/w sin(1/2 w)2 conj(1/w sin(1/2 w)2))/log(10)
1 0
0.5 −20
Y(Ω)
0 −40
−0.5 −60
−1 −80
−100 −80 −60 −40 −20 0 20 40 60 80 100
−1.5 −1 −0.5 0 0.5 1 1.5 w
t
Figure 5.3: Problem 9: signals and their magnitude spectra (in dB).


Chapter 6
Application of Laplace Analysis to

Control
6.1 Basic Problems
6.1 Consider a series RC circuit with input a voltage source vi (t) and output the voltage across the
capacitor vo (t).
(a) Draw a negative feedback system for the circuit using an integrator, a constant multiplier
and an adder.
(b) Let the input be a battery, i.e., vi (t) = Au(t), find the steady state error e(t) = vi (t) − vo (t).
Answer: Vo (s)/Vi (s) = G(s)/(1 + G(s)) with G(s) = 1/(RCs).
Solution
(a) The transfer function of the RC circuit is
Vo (s) 1/RC
=
Vi (s) s + 1/(RC)
An equivalent negative feedback with a feed–forward transfer function G(s), and feedback
transfer function H(s) = 1 gives
Vo (s) G(s)
=
Vi (s) 1 + G(s)
Comparing the two we get that G(s) = 1/(RCs). Considering an integrator has a transfer
function 1/s then we have the equivalent representations for the RC circuit shown in Fig. 6.1.
(b) If vi (t) = Au(t), Vi (s) = A/s, then
A/RC B D
Vo (s) = = +
s(s + 1/(RC)) s s + 1/RC
1
+ e(t) 1 !
vi (t) (.)dt vo (t)
RC
−
Figure 6.1: Problem 1: negative feedback equivalent of RC circuit.
Since 1/RC > 0, i.e., the pole of Vo (s) is in the left-hand s-plane, then in the steady state
vo (t) = B, where
A/RC
B = Vo (s)s|s=0 = =A
1/(RC)
so that in the steady–state the error signal e(t) = vi (t) − vo (t) will be zero.

6.2 Suppose the feed–forward transfer function of a negative feedback system is G(s) = N (s)/D(s),
and the feedback transfer function is unity. Let X(s) be the Laplace transform of the input x(t)
of the feedback system.
(a) Given that the Laplace transform of the error is E(s) = X(s)[1 − F (s)] where F (s) =
G(s)/(1 + G(s)) is the overall transfer function of the feedback system, find an expres-
sion for the error in terms of X(s), N (s) and D(s). Use this equation to determine the
conditions under which the steady state error is zero for x(t) = u(t).
(b) If the input is x(t) = u(t), N (s) = 1, and D(s) = (s + 1)(s + 2) find an expression for E(s)
and from it determine the initial value e(0) and the final value limt→∞ e(t) of the error.
Answers: X(s) = 1/s, zero steady–state error: roots of N (s) + D(s) in left–hand s-plane, D(s)
of the form sD1 (s).
Solution
(a) Replacing F (s) we have

G(s) 1
E(s) = X(s) 1 − = X(s)
1 + G(s) 1 + N (s)
D(s)
X(s)D(s)
=
D(s) + N (s)
If X(s) = 1/s then for the error to go to zero in the steady state, the roots of D(s) + N (s) should
be in the left–hand s-plane and D(s) must cancel the pole of E(s) at zero contributed by X(s),
i.e., D(s) must be of the form D(s) = sD1 (s).
(b) For G(s) = 1/(s + 1)(s + 2) and X(s) = 1/s the Laplace transform of the error is
D(s) (s + 1)(s + 2)
E(s) = =
s(1 + D(s)) s(s2 + 3s + 3)
According to the initial value theorem we have that
(1 + 1/s)(1 + 2/s)
e(0) = lim sE(s) = lim =1
s→∞ s→∞ 1 + 3/s + 3/s2
p
The poles of E(s) are s = 0 and s1,2 = −3/2 ± j 3/4 since s2 + 3s + 3 = (s + 3/2)2 + 3/4. A
partial fraction expansion for E(s) is
A Bs + C
E(s) = + 2
s s + 3s + 3
where
D(0) 2
A = sE(s)|s=0 = =
1 + D(0) 3
in this case the steady state error is 2/3. The error is not zero in the steady–state because D(s)
does not cancel the pole due to X(s).

6.3 Let H(s) = Y (s)/X(s) be the transfer function of the feedback system in Fig. 6.2. The impulse
response of the plant (with transfer function Hp (s)) is hp (t) = sin(t)u(t).
(a) If we want the impulse response of the feedback system (with input x(t) and output y(t))
to be h(t) = hp (t)e−t u(t), what should be the transfer function Hc (s) of the controller?
(b) Find the unit-step response s(t) of the feedback system.
x(t) + y(t)
Hc (s) Hp (s)
−
Answers: Hc (s) = (s2 + 1)/(s + 1)2 ; s(t) = [0.5 + 0.707 cos(t + 3π/4)]u(t)
Solution
(a) h(t) = hp (t)e−t = e−t sin(t)u(t) so
1
H(s) =
(s + 1)2 + 1
For feedback system
Hc (s)Hp (s) Hc (s)

H(s) = = 2
1 + Hc (s)Hp (s) s + 1 + Hc (s)
after replacing Hp (s) = 1/(s2 + 1). Then
Hc (s) 1 s2 + 1
= ⇒ Hc (s) =
s2 + 1 + Hc (s) 2
(s + 1) + 1 (s + 1)2
(b) Unit-step response s(t)
H(s) 1
S(s) = =
s s((s + 1)2 + 1)
A B B∗
= + +
s s − (−1 + j) s − (−1 − j)
where
A = sS(s)|s=0 = 0.5
1 1
B = |s=−1+j = √ ej3π/4
s(s + 1 + j) 2 2
thus
s(t) = [0.5 + 0.707 cos(t + 3π/4)]u(t).

6.4 Consider the cascade connection of two continuous-time systems shown in Fig. 6.3 where
dw(t) dy(t)
+ w(t) = x(t), + 2y(t) = w(t)
dt dt
w(t)
x(t) Syst. A Syst. B y(t)
(a) Determine the input/output differential equation for the overall cascade connection.
(b) Suppose that w(0) = 1, y(0) = 0 and x(t) = 0 for t ≥ 0,
i. Compute w(t) for t ≥ 0.
ii. Compute then y(t) for t ≥ 0.
Answers: (a) d2 y(t)/dt2 + 3dy(t)/dt + 2y(t) = x(t); (b) y(t) = (e−t − e−2t )u(t)
Solution
(a) Laplace transforms
(s + 1)W (s) = X(s)

(s + 2)Y (s) = W (s)
replacing W (s) in the second equation gives
X(s) Y (s) 1 1
(s + 2)Y (s) = ⇒ = = 2
s+1 X(s) (s + 1)(s + 2) s + 3s + 2
so that
d2 y(t) dy(t)
+3 + 2y(t) = x(t)
dt2 dt
(b) i. For the first system, the Laplace transform of its differential equation is
1
sW (s) − w(0) + W (s) = 0 ⇒ W (s) = ⇒ w(t) = e−t u(t)
s+1
ii. For the second system with input w(t) = e−t u(t)
1 1
sY (s) − y(0) + 2Y (s) = ⇒ Y (s) =
s+1 (s + 1)(s + 2)
so that
1 1
Y (s) = − ⇒ y(t) = (e−t − e−2t )u(t)
s+1 s+2

6.5 Consider a second–order system with transfer function
Y (s) 1
H(s) = = 2
X(s) s + s/Q + 1
where Y (s) and X(s) are the Laplace transforms of output y(t) and the input x(t) of the system.
Q is called the quality factor.
(a) If the feedback gain is unity, determine the feedforward transfer function G(s) .
(b) Find the poles of H(s), and plot them expressing the poles as p1,2 = re±jφ , where φ is
an angle measured with respect to the negative axis, and r is a positive radius, give the
values of r and φ. Show that Q = 1/(2 cos(φ)).
√
(c) Consider the cases when Q = 0.5, Q = 2/2 and Q → ∞, and determine the correspond-
ing poles. Find the impulse response h(t) for these cases. What happens as Q increases?
Explain.
Answers: G(s) = 1/(s(s + 1/Q)); Q → ∞ → H(s) = 1/(s2 + 1).
Solution:
(a) Let G(s) = 1/(s(s + 1/Q)) be the Laplace transform in the feedforward loop and with unit
gain in the feedback then
G(s) 1/(s(s + 1/Q)) 1

H(s) = = = 2
1 + G(s) 1 + 1/(s(s + 1/Q)) s + s/Q + 1
(b) Writing s2 + s/Q + 1 = (s + 1/(2Q))2 + (1 − 1/(4Q2 )) the poles are
1 p
p1,2 = − ± j 1 − 1/(4Q2 )
2Q
which have a magnitude of one and the angle φ measured with respect to the negative
real–axis is p
1 − 1/(4Q2 ) p
−1
φ = tan = tan−1 ( 4Q2 − 1)
1/2Q
The real part
1 1
= 1 cos(φ) → Q =
2Q 2 cos(φ))
(c) When Q = 0.5, 0.707, ∞ the angle φ = 0, π/4, π/2. For
1
H(s) = when
(s + 1/(2Q))2 + (1 − 1/(4Q2 ))
1
Q = 0.5 → H(s) =
(s + 1)2
1
Q = 0.707 → H(s) = √
(s + 1/ 2)2 + 1/2
1
Q → ∞ → H(s) = 2
s +1

and the corresponding impulse responses are
h(t) = te−t u(t)

1 √ √
h(t) = √ e−t/ 2 sin(t/ 2)u(t)
2
h(t) = sin(t)u(t)
As Q increases the impulse response becomes more oscillatory.

6.6 Given the two realizations in Fig. 6 obtain the corresponding transfer functions
Y1 (s) Y2 (s)
H1 (s) = , and H2 (s) = .
X1 (s) X2 (s)
Answer: H1 (s) = Y1 (s)/X1 (s) = (s − 2)/(s2 − s − 2)
Realization 1 +
y1 (t)
−2
� v1 (t) � v2 (t)
+
x1 (t)
2
Realization 2
x2 (t)
−2
w2 (t) w1 (t)
� �
+ +
y2 (t)
2
Solution: For the realization (1) we have the following equations
y1 (t) = v1 (t) − 2v2 (t)

v̇1 (t) = x1 (t) + v1 (t) + 2v2 (t)
v̇2 (t) = v1 (t)
which in the Laplace transform are
(i) Y1 (s) = V1 (s) − 2V2 (s)

(ii) V1 (s)(s − 1) = 2V2 (s) + X1 (s)
(iii) sV2 (s) = V1 (s)
Replacing V2 (s) = V1 (s)/s from equation (iii) in equation (ii) we have that
V1 (s)(s − 1 − 2/s) = X1 (s)

so that V1 (s) = X1 (s)/(s − 1 − 2/s) and and using (iii) 2V2 (s) = 2V1 (s)/s = X1 (s)(2/s)/(s − 1 −
2/s) so that equation (i) becomes
1 − 2/s
Y1 (s) = X1 (s)
s − 1 − 2/s
so that
Y1 (s) s−2
H1 (s) = = 2
X1 (s) s −s−2
For the realization (2) we have the following equations
y2 (t) = w1 (t)
ẇ1 (t) = x2 (t) + y2 (t) + w2 (t)
ẇ2 (t) = −2x2 (t) + 2y2 (t)
or in the Laplace domain
(i) Y2 (s) = W1 (s)

(ii) sW1 (s) = X2 (s) + Y2 (s) + W2 (s)
(iii) sW2 (s) = −2X2 (s) + 2Y2 (s)
Replacing W2 (s) from equation (iii) into equation (ii) we get
W1 (s) = (1/s − 2/s2 )X2 (s) + (1/s + 2/s2 )Y2 (s)
which replaced in equation (i) gives
(1 − 1/s − 2/s2 )Y2 (s) = (1/s − 2/s2 )X2 (s)
so that the system transfer function is
Y2 (s) s−2
H2 (s) = = 2
X2 (s) s −s−2
Indicating that the two realizations are different realizations of the same system.


6.7 Diagonalization of state variables — Suppose you are given the observer space representation
with matrix and vectors

−1 1 1
Ao = , bo = , cTo = 1 0
2 0 −1
To find a transformation that diagonalizes Ao use MATLAB function eigs which calculates the
eigenvalues and eigenvectors corresponding to the matrix and allows us to express

λ1 0
V−1 Ao V =
0 λ2
where V is a matrix created with the eigenvectors and {λi }, i = 1, 2, are the eigenvalues.
(a) Find the characteristic equation
det(sI − Ao )
corresponding to the state variable representation, and show it is the same as the denom-
inator of the transfer function (use function ss2tf to obtain the transfer function from the
state variable representation).
(b) Use the matrix V as an invertible transform to obtain a new set of state variables with a
diagonal matrix A and vectors b and cT . Obtain these matrix and vectors.
(c) Suppose that you find the controller form by letting Ac = ATo , bc = cTo and cTc = bT0 , repeat
the above diagonalization and comment on your results.
Answers: H(s) = (s − 1)/((s − 1)(s + 2)) = 1/(s + 2)
Solution
(a)(b) To find the transfer function let the input v(t) = δ(t), y(t) = h(t) and initial conditions
equal zero, then in the Laplace domain

1 s 1 1 s−1
H(s) = cTo (sI − Ao )−1 b = 2 1 0 = 2
s +s−2 2 s+1 −1 s +s−2
The denominator of H(s) is D(s) = s2 + s − 2 = (s − 1)(s + 2) is the determinant of sI − Ao ,

or the characteristic equation. Notice the pole-zero cancellation so that the simplified transfer
function is H(s) = 1/(s + 2).
% Pr. 6.7
% Observer
Ao=[-1 1; 2 0]
bo=[1 ;-1]
co=[1 0]’
% Transfer function
[num,den]=ss2tf(Ao,bo,co’,0)
roots(den)
% Diagonalization
[V,D]=eigs(Ao)

A=(inv(V))*Ao*V
b=(inv(V))*bo
c=V’*co
[num,den]= ss2tf(A,b,c’,0)
roots(den)
Ao = -1 1
2 0
bo = 1
-1
co = 1
0
num = 0 1.0000 -1.0000
den = 1 1 -2
ans = -2
1
V =-0.7071 -0.4472
0.7071 -0.8944
D = -2 0
0 1
A = -2.0000 0.0000
0 1.0000
b = -1.4142
0
c = -0.7071
-0.4472
num = 0 1 -1
den = 1 1 -2
ans =-2
1
Showing that the two realizations correspond to the same transfer function.
(c)
%Controller
Ac=Ao’
bc=co’
cc=bo’
[V,D]=eigs(Ao)
A=(inv(V))*Ao*V
b=(inv(V))*bo
c=V’*co
[num,den]= ss2tf(A,b,c’,0)
roots(den)
Ac = -1 2
1 0
bc = 1 0
cc =1 -1
V =-0.7071 -0.4472
0.7071 -0.8944
D =-2 0
0 1

A =-2.0000 0.0000
0 1.0000
b =-1.4142
0
c =-0.7071
-0.4472
num = 0 1 -1
den =1 1 -2
ans =-2
1
Again showing that the two realizations correspond to the same transfer function and that the
observer and the controller forms represent the same system.

Chapter 7
Fourier Analysis in Communications

and Filtering
7.1 Basic Problems
7.1 Consider a first–order system with transfer function
s + z1
H(s) = K
s + p1
where K > 0 is a gain, z1 is a zero and p1 is a pole.
(a) If we want unity dc gain, i.e., |H(j0)| = 1, what should be the value of K.
(b) What types of filters (lowpass, bandpass, band-eliminating, all-pass and high-pass) can
you obtain with this first-order system?
(c) Let H1 (s) = 1/(s + 1), corresponding to a low-pass filter, determine the dc gain of this
filter and use it to find the half-power frequency of the filter.
(d) Determine the magnitude response of a filter with transfer function
H2 (s) = (s − 1)/(s + 1). Plot the poles and zeros of this filter. What type of filter is it?
(e) What type of filter has the transfer function H3 (s) = s/(s + 1) ?
Answers: Low-, high- and all-pass filters can be obtained with H(s); H2 (s) is all-pass filter.
Solution
(a) |H(j0)| = K|z1 |/|p1 | = 1, so K = |p1 |/|z1 |

(b) Low-pass, high-pass and all-pass can be obtained with this first-order system. Band-pass
and stop-band require second or higher-order systems, as poles must be complex conju-
gates.
1
(c) For the low-pass filter, |H1 (j0)| = 1, the half-power frequency Ωhp is such that
|H1 (jΩhp )|2 = |H1 (j0)|2 /2 = 1/2

1 1
2 =
Ωhp + 1 2
so Ωhp = 1.
(d) The filter with H2 (s) = (s − 1)/(s + 1) is an all-pass filter. Its pole is s = −1 and zero s = 1.
Given this symmetry the magnitude response is unity for all values of Ω.
(e) The filter with H3 (s) = s/(s + 1) is a high-pass filter, |H(j0)| = 0 and when Ω → ∞ the
magnitude response grows and becomes 1 at ∞.

7.2 The loss at a frequency Ω = 2000 (rad/sec) is α(2000) = 19.4 dBs for a fifth-order lowpass
Butterworth filter with unity dc gain. If we let α(Ωp ) = αmax = 0.35 dBs, determine
• the half-power frequency Ωhp , and
• the passband frequency Ωp
of the filter.
Answers: Ωhp = 1280.9 rad/sec; Ωp = 999.82 (rad/sec)
Solution
Using the loss function
2N !
Ω Ω
α(Ω) = 10 log10 1+ ⇒ Ωhp =
Ωhp (100.1α(Ω) − 1)1/2N
which gives for Ω = 2000
2000
Ωhp = = 1280.9 rad/sec.
(101.94 − 1)0.1
When α(Ωp ) = αmax we have
Ωp = Ωhp (100.1αmax − 1)1/2N
which gives after replacing the values on the right term Ωp = 999.82 (rad/sec).


7.3 Design of low-pass Butterworth/Chebyshev filters — The specifications for a low-pass filter
are
α(0) = 20 dBs
Ωp = 1500 rad/sec, α1 = 20.5 dBs
Ωs = 3500 rad/sec, α2 = 50 dBs
(a) Determine the minimum order of the low-pass Butteworth and Chebyshev filters, and
determine which is smaller.
(b) Give the transfer function of the designed low-pass Butterworth and Chebyshev filters
(make sure the dc loss is as specified).
(c) Determine the half-power frequency of the designed filters by letting α(Ωp ) = αmax
(d) Find the loss function values provided by the designed filters at Ωp and Ωs . How are
these values related to the αmax and αmin specifications? Explain. Which of the two filters
provides more attenuation in the stop-band?
(e) If new specifications for the passband and stopband frequencies are Ωp = 750 (rad/sec)
and Ωs = 1750 (rad/sec), are the minimum order
Answers: Minimum order: Butterworth N = 6, Ωhp = 1.787 × 103 .
Solution
Since the dc loss is not zero, the normalized loss specifications are
αmax = α1 − α(0) = 0.5
αmin = α2 − α(0) = 30
with a dc loss of 20 dB. The following script is used to find the answers
% Pr 7_3
clear all; clf
alphamax=0.5 ;alphamin=30; Wp=1500; Ws=3500;
alpha0=20;
% Butterworth
K=10ˆ(alpha0/20)
D=(10ˆ(0.1*alphamin)-1)/(10ˆ(0.1*alphamax)-1);
E=Ws/Wp;
N=ceil(log10(D)/(2*log10(E)))
Whp=Wp/(10ˆ(0.1*alphamax)-1)ˆ(1/(2*N))
alpha_p=10*log10(1+(Wp/Whp)ˆ(2*N))
alpha_s=10*log10(1+(Ws/Whp)ˆ(2*N))
% Chebyshev
N=ceil(acosh(Dˆ(0.5))/acosh(E))
eps=sqrt(10ˆ(0.1*alphamax)-1)
Whp1=Wp*cosh(acosh(1/eps)/N)
alpha_p= 10*log10(1+(epsˆ2)*(cos(N*acos(1)))ˆ2)
alpha_s=10*log10(1+(epsˆ2)*cosh(N*acosh(Ws/Wp))ˆ2)
alpha1=alpha0+10*log10(1+(epsˆ2)*cosh(N*acosh(0))ˆ2);
Kc=10ˆ(alpha1/20)

The following are the results
% Butterworth
K = 10 % dc gain
N =6 % minimum order
Whp =1.7874e+03 % half-power frequency
alpha_p = 0.5000 % loss at Wp
alpha_s =35.0228 % loss at Ws
% Chebyshev
N = 4 % min order
eps = 0.3493 % ripple factor
Whp1 = 1.6397e+03 % half-power frq
alpha_p = 0.5000 % loss at Wp
alpha_s = 36.6472 % loss at Ws
Kc =10.5925 % dc gain
Notice the computation of the dc gain in the Chebyshev filter. In this case the dc loss depends
on the order of the filter and so it is not necessarily 0 dB, so to get the dc gain Kc we use

K2 2 2
α(0) = 10 log10 2 (0) = 20 log10 K − 10 log10 (1 + CN (0))
1 + 2 C N
as indicated in the script.

(d) The minimum orders of the filters depend on the ratio of the two frequencies and since it
remains the same these do not change.

7.4 Getting rid of 60 Hz hum with different filters — A desirable signal x(t) = cos(100πt) −
2 cos(50πt) is recorded as y(t) = x(t) + cos(120πt), i.e., as the desired signal but with a a 60
Hz hum. We would like to get rid of the hum and recover the desired signal. Use symbolic
MATLAB to plot x(t) and y(t).
Consider the following three different alternatives (use symbolic MATLAB to implement the
filtering and use any method to design the filters):
• Design a band-eliminating filter to get rid of the 60 Hz hum in the signal. Plot the output
of the band-eliminating filter.
• Design a high-pass filter to get the hum signal and then subtract it from y(t). Plot the
output of the high-pass filter.
• Design a band-pass filter to get rid of the hum. Plot the output of the band-pass filter.
Is any of these alternatives better than the others? Explain.
Solution
The following script is used to compute the different signals, filters, and to plot them.
% Pr 7_4
clear all; clf
syms t w
x=cos(100*pi*t)-2*cos(50*pi*t);
y=x+cos(120*pi*t);
figure(1)
subplot(211)
ezplot(x,[0:1]) % desired signal
subplot(212)
ezplot(y,[0:1]) % signal with hum
X=fourier(x);
% butterworth filter
N=5;
%wn=[2*pi*59.9 2*pi*60.1]; [b,a]=butter(N,wn,’stop’,’s’);C=1 % band-eliminating
% [b,a]=butter(2*N,110*pi,’high’,’s’);C=2 % high-pass
[b,a]=butter(N,[20*pi 90*pi],’s’);C=2 % band-pass
% frequency responses
W=0:1:150*pi;
Hm=abs(freqs(b,a,W));
figure(2)
subplot(211)
plot(W/(2*pi),Hm); axis([0 75 0 1.1]);xlabel(’f (Hz)’); ylabel(’|H|’)
M=2*N
% generation of frequency response from coefficients
n=M:-1:0;
U=(j*w).ˆn;
num=b*conj(U’); den=a*conj(U’);
H=num/den; % Butterworth LPF
% output of filter
Y1=X*H;
y1=real(ifourier(Y1,t));
subplot(212)
if C==1,

ezplot(y1,[0:1])
else
ezplot(x-y1,[0:1])
end
title(’denoised signal’)
The most natural of the three approaches is the one using the stopband filter, the other depend
on information on the desired signal that might or might not be available. The only needed
information for the stopband filter is the frequency of the hum.
cos(100 π t)−2 cos(50 π t)
−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t
cos(100 π t)−2 cos(50 π t)+cos(120 π t)
4
3
2
1
0
−1
−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t
1 1 1
0.8 0.8 0.8

0.6 0.6 0.6
|H|
|H|
|H|
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
f (Hz) f (Hz) f (Hz)
denoised signal denoised signal denoised signal
3
3 1
2 2
0
1 1
0 0 −1
−1 −1
−2
−2
−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t t t
Figure 7.1: Problem 4: top: original and original with hum signals; bottom left to right: notch filter
and filtered signal, high-pass filter and denoised signal, an bandpass filter and denoised signal.


Chapter 8
Sampling Theory
8.1 Basic Problems
8.1 Consider the sampling of a sinc signal and related signals.

(a) For the signal x(t) = sin(t)/t, find its magnitude spectrum |X(Ω)| and determine if this
signal is band-limited or not.
(b) What would be the sampling period Ts you would use for sampling x(t)) without alias-
ing?
(c) For a signal y(t) = x2 (t) what sampling frequency fs would you use to sample it without
aliasing? How does this frequency relate to the sampling frequency used to sample x(t)?
(d) Find the sampling period Ts to sample x(t) so that the sampled signal xs (0) = 1 otherwise
x(nTs ) = 0 for n 6= 0.
Answers: x(t) band-limited with Ωmax = 1 (rad/sec); for x2 (t), Ts ≤ π/2.
Solution
(a) To find X(Ω), we use duality or find the inverse Fourier transform of a pulse of amplitude
A and bandwidth Ω0 , that is
X(Ω) = A[u(Ω + Ω0 ) − u(Ω − Ω0 )]
so that
Z Ω0
1 A jΩt Ω0
x(t) = AejΩt dΩ = e −Ω0
2π −Ω0 2πjt
A
= sin(Ω0 t)
πt
which when compared with the given x(t) = sin(t)/t gives that A = π and Ω0 = 1 or
X(Ω) = π[u(Ω + 1) − u(Ω − 1)]
1
indicating that x(t) is band-limited with a maximum frequency Ωmax = 1 (rad/sec).

(b) To sample without aliasing the sampling frequency should be chosen to be
1 Ωmax
fs = ≥2
Ts 2π
which gives a sampling period
π
Ts ≤ = π sec/sample
Ωmax
(c) The spectrum of y(t) = x2 (t) is the convolution in the frequency
1
Y (Ω) = (X(Ω) ∗ X(Ω))
2π
which would have a maximum frequency Ωmax = 2, giving a sampling frequency which is
double the one for x(t). The sampling period for y(t) should be
π
Ts ≤ .
2
(d) The signal x(t) = sin(t)/t is zero whenever t = ±kπ, for k = 1, 2, · · · so that choosing Ts = π
(the Nyquist sampling period) we obtain the desired signal xs (0) = 1 and x(nTs ) = 0.

8.2 Consider the signal x(t) = δ(t + 1) + δ(t − 1).

(a) Find its Fourier transform X(Ω). Determine if x(t) is band–limited or not. If band–limited,
give its maximum frequency.
(b) Filtering x(t) with a filter of magnitude H(jΩ) = u(Ω + 1) − u(Ω − 1) and zero phase the
output is y(t). Is y(t) band–limited? If so, what is its maximum frequency.
(c) Obtain an expression for y(t), is it time-limited?
Answers: X(Ω) = 2 cos(Ω); y(t) is not time–limited.
Solution
(a) X(Ω) = ejΩ + e−jΩ = 2 cos(Ω) which has no maximum frequency, so x(t) is not bandlim-
ited.
(b) Y (Ω) = X(Ω)H(jΩ) = 2 cos(Ω)[u(Ω + 1) − u(Ω − 1)] with maximum frequency of 1 and
so y(t) is band-limited.
(c) It is not time limited, as
Z 1 Z 1
1 1
y(t) = 2 cos(Ω)ejΩt dΩ = [ejΩ(t+1) + ejΩ(t−1) ]dΩ
2π −1 2π −1
sin(t + 1) sin(t − 1)
= +
π(t + 1) π(t − 1)

8.3 Suppose you wish to sample an amplitude modulated signal x(t) = m(t) cos(Ωc t) where m(t)
is the message signal and Ωc = 2π104 (rad/sec) is the carrier frequency.
(a) If the message is an acoustic signal with frequencies in a band of [0, 22] kHz, what would
be maximum frequency present in x(t)?
(b) Determine the range of possible values of the sampling period Ts that would allow to
sample x(t) satisfying the Nyquist sampling rate condition.
(c) Given that x(t) is a bandpass signal, compare the above sampling period with the one that
can be used to sample bandpass signals.
Answers: Nyquist: Ωs ≥ 128π × 103 .
Solution
(a) The Fourier transform of x(t) is
X(Ω) = 0.5M (Ω − Ωc ) + 0.5M (Ω + Ωc )
where M (Ω) is the Fourier transform of m(t). The maximum frequency present in x(t) is
Ωmax = Ωc + 2π × 22 × 103 = 2π(10 + 22)103 = 64π 103
(b) The sampling frequency is
2π
Ωs = ≥ 2 × 64π × 103 = 128π × 103
Ts
so that
1
Ts ≤ 10−3 sec/sample
64
(c) The bandwidth of the message is B = 2π×22×103 = 44π×103 rad/sec, using this frequency
x(t) can be sampled at 2B = 88π×103 rad/sec which is much smaller than the one found above.

8.4 The input/output relation of a non-linear system is y(t) = x2 (t), where x(t) is the input and
y(t) is the output.
(a) The signal x(t) is band-limited with a maximum frequency ΩM = 2000π (rad/sec), deter-
mine if y(t) is also band-limited and if so what is its maximum frequency Ωmax .
(b) Suppose that the signal y(t) is low-pass filtered. The magnitude of the low-pass filter
is unity and the cut-off frequency is Ωc = 5000π (rad/sec), determine the value of the
sampling period Ts according to the given information.
(c) Is there a different value for Ts that would satisfy the Nyquist sampling rate condition for
sampling both x(t) and y(t) and that is larger than the one obtained above? Explain.
Answers: y(t) is band-limited; Ts ≤ 0.25 × 10−3 sec/sample.
Solution
(a) If the signal x(t) is band-limited y(t) = x2 (t) has a Fourier transform Y (Ω) = (1/2π)(X(Ω) ∗
X(Ω)) having a bandwidth double that of x(t), or
Ωmax y = 2ΩM = 4000π
so y(t) is band-limited.
(b) Filtering with a low-pass filter of cut-off frequency 5000π would not change the maximum
frequency of Y (Ω) so that
π 1
Ts ≤ = = 0.25 × 10−3
4000π 4000
(c) No. For x(t),
π
Ts1 ≤ = 0.5 × 10−3
2000π
and if Ts = 0.25 × 10−3 then Ts1 ≤ 2Ts , so that we need to use Ts to sample both x(t) and y(t).

8.5 A sinusoid x(t) = cos(t) is a band-limited signal with maximum frequency Ωmax = 1
(a) Using Fourier transform properties determine the maximum frequency of x2 (t). What
sampling period Ts can be used to sample x2 (t) without aliasing. Verify your results using
trigonometric identities.
(b) Can you generalize the above results to xN (t), for integer N > 1? Is xN (t) band-limited
for all the given values of N ? Consider the case N = 3 and use trigonometric identities to
verify the results.
Answers: (a) Ts ≤ π/2.
Solution
(a) If F(cos(t)) = X(Ω), then F(x2 (t)) = (X ∗ X)(Ω)/(2π), i.e., the convolution of X(Ω)
with itself, having twice the bandwidth of X(Ω), so the maximum frequency of x(t) is
Ωmax = 2, and Ts ≤ π/2. Thus x(t) is band-limited with twice the maximum frequency of
cos(t).
Using cos2 (t) = 0.5(1 + cos(2t)) we see that its maximum frequency is 2 and it is band-
limited and Ts ≤ π/2.
(b) In general F(xN (t)) would be N convolutions of X(Ω) with itself, and the bandwidth
being N ×bandwith of cos(t), i.e., N , so the maximum frequency is Ωmax = N and Ts ≤
π/N .
If N = 3 , cos3 (t) = 0.25(3 cos(t) + cos(3t)) with Ωmax = 3, and Ts ≤ π/3.

8.6 Let x(t) = cos(πt)[u(t) − u(t − 2)] be the input of a zero-order hold sampler. The sampler
samples every Ts sec. starting at t = 0, and the impulse response of the zero-order hold is
h(t) = u(t) − u(t − Ts ). The quantizer and coder shown in Fig. 7.12 in the text is used to
generate a digital signal.
(a) Let Ts = 0.5 sec/sample, find the values of the sampled signal using the ideal sampler.
Determine the corresponding digital signal.
(b) Suppose now that we let Ts = 0.25 sec/sample. Carefully plot the sampled signal using
the ideal sampler and then plot the output y(t) of the zero-order hold system. Determine
the corresponding digital signal.
(c) For the above two cases find and plot the quantization error ε(nTs ) = x(nTs ) − x̂(nTs ),
where x̂(nTs ) is the output of the quantizer.
Answer: The binary code is {01 00 10 00 01 00 · · · }.
Solution
(a) The sample values for Ts = 0.5 are
nTs x(nTs )
0 cos(0) = 1
0.5 cos(π/2) = 0
1.0 cos(π) = −1
1.5 cos(3π/2) = 0
2.0 cos(2π) = 1
x(nTs ) = 0 nTs > 2.
The binary code for nTs = 0, 0.5, 1, 1.5, 2, 2.5 · · · is
01 00 10 00 01 00 · · ·
corresponding to the quantized signal
x̂(nTs ) = ∆δ(nTs ) + 0δ((n − 1)Ts ) − 2∆δ((n − 2)Ts ) + 0δ((n − 3)Ts ) + ∆δ((n − 4)Ts )
where ∆ = 2/4 = 0.5, so that the quantization error is
ε(nTs ) = x(nTs ) − x̂(nTs )
= 0.5δ(nTs ) + 0δ((n − 1)Ts ) + 0δ((n − 2)Ts ) + 0δ((n − 3)Ts ) + 0.5δ((n − 4)Ts )
(b) The sample values for Ts = 0.25 are

nTs x(nTs )
0 cos(0) = 1
0.25 cos(π/4) = 0.707
0.5 cos(π/2) = 0
0.75 cos(3π/4) = −0.707
1.0 cos(π) = −1
1.25 cos(5π/4) = −0.707
1.5 cos(3π/2) = 0
1.75 cos(7π/4) = 0.707
2.0 cos(2π) = 1

The binary code for nTs = 0, 0.25, · · · , 1.75, 2., 2.25, · · ·
01 01 00 11 10 11 00 01 01 00, · · ·
corresponding to the quantized signal
x̂(nTs ) = ∆δ(nTs ) + ∆δ((n − 1)Ts ) + 0δ((n − 1)Ts ) − ∆δ((n − 3)Ts )

−2∆δ((n − 4)Ts ) − ∆δ((n − 5)Ts ) − 0δ((n − 6)Ts ) + ∆δ((n − 7)Ts )
+∆δ((n − 7)Ts )
where as before ∆ = 0.5, so that the quantization error is
ε(nTs ) = 0.5δ(nTs ) + 0.207δ((n − 1)Ts ) + 0δ((n − 2)Ts ) − 0.207δ((n − 3)Ts )

+0δ((n − 4)Ts ) − 0.207δ((n − 5)Ts ) + 0δ((n − 6)Ts ) + 0.207δ((n − 7)Ts )
+0.5δ((n − 8)Ts )

8.7 Suppose x(t) has a Fourier transform X(Ω) = u(Ω + 1) − u(Ω − 1)

(a) Determine possible values of the sampling frequency Ωs so that x(t) is sampled without
aliasing.
(b) Let y(t) = x2 (t), find the Fourier transform Y (Ω), indicating its maximum frequency. Is
y(t) band–limited?
(c) If z(t) = x(t) + y(t), what is the maximum frequency of z(t)? Find values for the sampling
period Ts that can be used to sample z(t) without aliasing.
Answers: For x(t): Ωs > 2Ωmax = 2; y(t) is band-limited.
Solution
(a) For x(t) to be sampled without aliasing we need Ωs > 2Ωmax = 2.
(b) If y(t) = x2 (t) then Y (Ω) = [X ∗ X](Ω)/(2π) which gives a triangular spectrum in [−2 2]
frequency band, i.e., it is band-limited.
(c) The maximum frequency of z(t) is Ωmax = 2 (rad/sec) so Ωs ≥ 4 (rad/sec) and Ts ≤ π/2
sec.


8.8 Uncertainty in time and frequency — Signals of finite time support have infinite support in
the frequency domain, and a band-limited signal has infinite time support. A signal cannot
have finite support in both domains.
2
(a) Consider the signals x(t) = e−t (the Gaussian function) and x1 (t) = u(t+1)−u(t−1). Use
MATLAB to find their Fourier transform X(Ω), and X1 (Ω), and to compute the energy of
the signals in Ω ∈ [−4, 4].
(b) The fact that a signal cannot be of finite support in both domains is expressed well by the
uncertainty principle which says that
2∆(t)∆(Ω) ≥ 1
where "R ∞ #0.5

−∞
t2 |x(t)|2 dt
∆(t) =
Ex
measures the duration of the signal for which the signal is significant in time, and
"R ∞ #0.5
−∞
Ω2 |X(Ω)|2 dΩ
∆(Ω) =
2πEx
measures the frequency support for which the Fourier representation is significant. The
energy of the signal is represented by Ex . Use MATLAB to compute ∆(t) and ∆(Ω) for
the signals x(t) and x1 (t) and verify that the uncertainty principle is satisfied. Comment
on the difference in results for the two signals; pay attention to the smoothness and the
frequency content of the signals.
Solution
(a) For the Gaussian signal we notice that the signal and its spectrum look very similar, and
that the spectrum is narrower that the signal. For x1 (t), we notice that it has discontinuities
and its spectrum has high frequencies because of them. See Figs. 8.1 and 8.2.
(b) Computing ∆(t) and ∆(Ω) we verify the uncertainty bound. It is 1 for the Gaussian signal
(the only signal that will give this value) and 1.2197 for the square pulse x1 (t). You need to
make some simple changes to the script for x1 (t), the script shown is for the Gaussian signal.
% Pr. 8.8
clear all; clf
syms x t X X1 w z
sigma=1
%x=heaviside(t+sigma)-heaviside(t-sigma);
x=exp(-tˆ2/(2*sigmaˆ2));
X=fourier(x,w);
XX=X*conj(X);
figure(1)
subplot(211)
ezplot(x,[-4*sigma,4*sigma]);grid;
subplot(212)

ezplot(XX,[-4*sigma,4*sigma]);grid;
Ex=int(xˆ2,t,-4*sigma,4*sigma)
figure(2)
subplot(211)
ezplot((xˆ2)*(tˆ2)/Ex,[-4*sigma,4*sigma]);grid;
subplot(212)
ezplot(XX*(wˆ2)/Ex,[-4*sigma,4*sigma]);grid;
z=int((xˆ2)*(tˆ2),t, -4*sigma, 4*sigma)/Ex
zz=subs(sqrt(z)) % Delta(t)
v=int((Xˆ2)*(wˆ2)/(2*pi),w, -4*sigma,4*sigma)/Ex
vv=subs(sqrt(v)) % Delta(W)
2*zz*vv
1/exp(t2/2) t2/(π1/2 erf(4) exp(t2))

1 0.2
0.8
0.15
0.6
0.1
0.4
0.2 0.05
0 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
t t
(2 π)/(exp(w2/2) exp(conj(w)2/2)) (2 π1/2 w2)/(erf(4) exp(w2/2) exp(conj(w)2/2))
6
5 1
4
3
0.5
2
1
0 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
w w
Figure 8.1: Problem 8: Gaussian signal x(t) and its spectrum (both are real) (left); integrands used
to calculate ∆(t) and ∆(Ω) (right)
heaviside(t + 1) − heaviside(t − 1) (t2 (heaviside(t − 1) − heaviside(t + 1))2)/2
0.8
1
0.8 0.6
0.6 0.4
0.4
0.2
0.2
0 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
t t
((cos(w) i − sin(w))/w − (cos(w) i + sin(w))/w) ((cos(conj(w)) i − sin(conj(w)))/conj(w) − (cos(conj(w)) i + sin(conj(w)))/conj (w2 ((cos(w) i − sin(w))/w − (cos(w) i + sin(w))/w) ((cos(conj(w)) i − sin(conj(w)))/conj(w) − (cos(conj(w)) i + sin(conj(w)))/conj
4 2
3 1.5
2 1
1 0.5
0 0
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
w w
Figure 8.2: Problem 8: Rectangular pulse x1 (t) and its spectrum (left); integrands used to calculate
∆(t) and ∆(Ω) (right


Chapter 9
Discrete–time Signals and Systems
9.1 Basic Problems
9.1 For a finite-support signal x[n] = r[n](u[n] − u[n − 11]) where r[n] is the discrete-time ramp
function,
(a) Find the energy of x[n].

(b) Find the even xe [n] and the odd xo [n] components of x[n].
(c) Show that the energy of x[n] the sum of the energies of xe [n] and xo [n].
Answers: εx = 385.
Solution
(a) Energy of x[n]

10
X 10 × 11 × 21
εx = n2 = = 385
n=0
6
Using the formula

N
X N (N + 1)(2N + 1)
k2 =
6
k=1
(b) Calling p[n] = u[n] − u[n − 11] so x[n] = r[n]p[n] we have
xe [n] = 0.5(r[n]p[n] + r[−n]p[−n])

xo [n] = 0.5(r[n]p[n] − r[−n]p[−n])
1
The energy
∞
X
εx = (xe [n] + xo [n])2
n=−∞
X∞
= x2e [n] + x2o [n] + 2xe [n]xo [n]
n=−∞
X∞ ∞
X
= x2e [n] + x2o [n]
n=−∞ n=−∞
since xe [n]xo [n] is odd its sum is zero. Also from plots of xe [n] and xo [n] we have that
X10 X10
n2
εx = 4 = n2 = 385
n=0
4 n=0

9.2 A two-bit quantizer has as input x(nTs ) and as output x̂(nTs ), or
k∆ ≤ x(nTs ) < (k + 1)∆ → x̂(nTs ) = k∆, k = −2, −1, 0, 1
(a) Is this system time-invariant? Explain.

(b) Suppose that the value of ∆ in the quantizer is 0.25, and the sampled signal is x(nTs ) =
nTs , Ts = 0.1 and −5 ≤ n ≤ 5. Use the sampled signal to determine whether the quantizer
is a linear system or not. Explain.
(c) From the results in this problem and the sampling theory, would you say that the A/D
converter is a LTI system? Explain.
Answers: Quantizer is non–linear, time–invariant; sampler is linear time–varying.
Solution
(a) Yes, the quantizer is time-invariant. Indeed, x(nTs ) and x(nTs − M Ts ) will give the same
values just shifted in time. In this case the M must be an integer as the sampled signal is dis-
crete in time.
(b) The samples are given by x(nTs ) = nTs = 0.1n. For ∆ = 0.25 the sample x[1] = 0.1 when
quantized gives x̂[1] = 0 since 0 < 0.1 < ∆. If we multiply the signal by 3 so that the value
3x[1] = 0.3 would give x̂[1] = 1 since ∆ < 0.3 < 2∆. Since the second output is not the first
output multiplied by 3, the quantizer is not linear.
We saw before that the sampler is linear but time-varying, while the quantizer is time-invariant
but non-linear, so the A/D converter which is composed of these systems is not LTI.

9.3 A continuous-time system is characterized by the ordinary differential equation
dy(t) dx(t)
+ y(t) = 2x(t) +
dt dt
This equation is discretized by approximating the derivatives for a signal ρ(t) as
dρ(t) ρ(nT + T ) − ρ(nT )

≈
dt T
around t = nT , and for a small value of T . Obtain a block diagram that represents the ob-
tained difference equation from the differential equation, with input x(nT ) and output y(nT )
and T = 1.
Answers: y(nT ) = (1 − T )y((n − 1)T ) + (2T − 1)x((n − 1)T ) + x(nT ).
Solution
Discretization of differential equation:
y(nT + T ) − y(nT ) x(nT + T ) − x(nT )

+ y(nT ) = 2x(nT ) +
T T
⇒ y((n + 1)T ) = (1 − T )y(nT ) + (2T − 1)x(nT ) + x((n + 1)T )
and letting m = n + 1 we get
y(mT ) = (1 − T )y((m − 1)T ) + (2T − 1)x((m − 1)T ) + x(mT )
Figure 9.1 shows the block diagram when T = 1.
x[n] x[n − 1]
D
y[n]
+
Figure 9.1: Problem 3: block diagram of difference equation when T = 1.

9.4 An LTI discrete-time system has the impulse response h[n] = (−1)n u[n]. Use the convolution
sum to compute the output response y[n], n ≥ 0 when the input is x[n] = u[n] − u[n − 3] and the
initial conditions are zero. In particular, find the following values of the output y[n], 0 ≤ n ≤ 4.
Answers: y[0] = 1, y[1] = 0, y[2] = 1, y[3] = −1, y[4] = 1.
Solution
The convolution sum is
n
X
y[n] = x[m]h[n − m]
m=0
y[0] = x[0]h[0] = 1
y[1] = x[0]h[1] + x[1]h[0] = −1 + 1 = 0
y[2] = x[0]h[2] + x[1]h[1] + x[2]h[0] = 1 − 1 + 1 = 1
y[3] = x[0]h[3] + x[1]h[2] + x[2]h[1] + x[3]h[0] = −1 + 1 − 1 + 0 = −1
y[4] = x[0]h[4] + x[1]h[3] + x[2]h[2] + x[3]h[1] + x[4]h[0] = 1 − 1 + 1 + 0 + 0 = 1
y[n] = x[0]h[n] + x[1]h[n − 1] + x[2]h[n − 2] = (−1)n + (−1)n−1 + (−1)n−2 = (−1)n−2 n ≥ 5

9.5 An LTI system represented by the difference equation y[n] = −y[n − 1] + x[n], n ≥ 0, is initially
at rest. The input of the system is x[n] and the output y[n].
(a) Using the difference equation find the system response y[n] for n ≥ 0 when the input is

1 n = 0, 1, 2
x[n] =
0 otherwise
Find y[n] for n = −1, 0, 1 · · · , 10

(b) Using the above result and the linearity and time-invariance of the system find the output
y1 [n] corresponding to a new input

1 n = 0, 1, 2, 3, 4, 5
x1 [n] =
0 otherwise
Answers: y[n] = 0 for n < 0, y[0] = 1, y[1] = 0, y[n] = (−1)n for n ≥ 2; y1 [n] = y[n] + y[n − 3].
Solution
(a) According to the difference equation the output is
y[0] = −y[−1] + x[0] = 0 + 1 = 1

y[1] = −y[0] + x[1] = −1 + 1 = 0
y[2] = −y[1] + x[2] = 0 + 1 = 1
y[3] = −y[2] + x[3] = −1 + 0 = −1
y[4] = −y[3] = 1
..
.
or
y[n] = 0 n<0
y[0] = 1
y[1] = 0
y[n] = (−1)n n≥2
(b) By linearity and time-invariance since x1 [n] = x[n] + x[n − 3] then



 y[n] n ≤ 2

0 n=3
y1 [n] = y[n] + y[n − 3] =

 1 n=4

0 n≥5

9.6 The output of a discrete-time system is y[n] = w[n]x[n] where x[n] is the input, and w[n] =
u[n] − u[n − 5] is a rectangular window.
(a) The input is x[n] = 4 sin(πn/2), −∞ < n < ∞, determine if x[n] is periodic, and if so
indicate its fundamental period N0 .
(b) Let z[n] = w[n]x[n − N ] for what values of N is y[n] = z[n]? According to this, is the
system time-invariant? Explain.
(c) Suppose that we let x1 [n] = x[n]u[n], i.e., x1 [n] is the causal component of x[n]. Determine
the outputs due to x1 [n] and to x1 [n − 4]. Again according to this, is the system time-
invariant? Explain.
Answers: x[n] is periodic of fundamental period N0 = 4; the system is time-varying.
Solution
(a) The discrete frequency of x[n] is ω0 = π/2 which can be written as 2πm/N0 with m = 1,
N0 = 4. So x[n] is periodic of fundamental period N0 = 4.
(b) Since x[n − kN0 ] = x[n] for any integer k and the fundamental period N0 = 4, then taking
N = kN0 we will have that
z[n] = x[n − kN0 ]w[n] = x[n]w[n] = y[n]
so that delaying the input by a multiple of N0 does not change the output. The output in
this case is just the period of x[n] between 0 and 4. However, that is not so if the delay is
different from a multiple of the fundamental period, thus the system is time-varying.
(c) The output corresponding to x1 [n] = x[n]u[n] is
y1 [n] = x1 [n]w[n] = x[n]u[n]w[n] = x[n]w[n] = y[n]
obtained before. However, the output corresponding to x1 [n − 4] is
x1 [n − 4]w[n] = x[n − 4]u[n − 4]w[n] = x[0]w[4] = 4 sin(0)w[4] = 0
which is not the previous result shifted, so that the system is time-varying.

9.7 Express the signal u12 [m, n] with support in the first and second quadrant in terms of one-
dimensional unit-step functions and determine if it is separable.
Solution
We have

1 −∞ < m < ∞, n ≥ 0
u12 [m, n] =
0 otherwise
= u1 [m, n] + u2 [m, n] − u1 [0, n] = u[m]u[n] + u[−m]u[n] − u[n]
= (u[m] + u[−m] − 1)u[n]
i.e., it is separable.

9.8 Although few signals encountered in practice are separable, any signal x[m, n], 0 ≤ m ≤ M −
1, 0 ≤ n ≤ N − 1 and zero otherwise, can be written as as the sum of finite number of separable
sequences
NX−1
x[m, n] = xi1 [m]xi2 [n].
i=0
Choose xi1 [m] and xi2 [n] to obtain this representation and show it can be expressed in terms of
δ[m, n].
Solution
A simple way to obtain such a representation is by letting the signal be expressed as the sum of
isolated columns. This is done by choosing for 0 ≤ i ≤ N − 1
xi1 [m] = x[m, i]

xi2 [n] = δ[n − i]
so that
N
X −1
x[m, n] = x[m, i]δ[n − i]
i=0
As functions of m, the signal x[m, i] can be expressed as

M
X −1
x[m, i] = x[k, i]δ[m − k]
k=0
and thus
N
X −1 N
X −1 M
X −1
x[m, n] = x[m, i]δ[n − i] = x[k, i] δ[m − k]δ[n − i]
| {z }
i=1 i=0 k=0
δ[m−k,n−i]
which is the representation sought.

9.9 Suppose both the input x[m, n] and the impulse response h[m, n] of a LSI system are separable.
(a) Obtain a simpler expression for the output y[m, n] of the system.
(b) Suppose x[m, n] = (u[m] − u[m − 2])(u[n] − u[n − 2]) and h[m, n] = (u[m] − u[m − 2])(u[n] −
u[n − 2]) where u[n] is the one-dimensional unit-step signal. Find y[m, n]
Solution
(a) Letting x[m, n] = x1 [m]x2 [n] and h[m, n] = h1 [m]h2 [n] we have that
∞
X ∞
X
y[m, n] = x1 [k]x2 [`]h1 [m − k]h2 [n − `]
k=−∞ `=−∞
X∞ ∞
X
= x1 [k]h1 [m − k] x2 [`]h2 [n − `]
k=−∞ `=−∞
= (x1 ∗ h1 )[m] (x2 ∗ h2 )[n]
i.e., the product of two one-dimensional convolutions.

(b) Letting x1 [m] = h1 [m] = u[m] − u[m − 2] we have that their convolution
(x1 ∗ h1 )[m] = δ[m] + 2δ[m − 1] + δ[m − 2]
and also letting x2 [n] = h2 [n] = u[n] − u[n − 2] we have
(x2 ∗ h1 )[n] = δ[n] + 2δ[n − 1] + δ[n − 2]
and so the response is
y[m, n] = (δ[m] + 2δ[m − 1] + δ[m − 2])(δ[n] + 2δ[n − 1] + δ[n − 2])

= (δ[m, n] + 2δ[m, n − 1] + δ[m, n − 2])
+(2δ[m − 1, n] + 4δ[m − 1, n − 1] + 2δ[m − 1, n − 2])
+(δ[m − 2, n] + 2δ[m − 2, n − 1] + δ[m − 2, n − 2]

9.10 Discrete sequence — Consider the formula
x[n] = x[n − 1] + x[n − 3] n≥3

x(0) = 0
x(1) = 1
x(2) = 2
Find the rest of the sequence for 0 ≤ n ≤ 5 and plot it using function stem.
Answers: x[3] = 2, x[4] = 3, x[5] = 5.
Solution
The equation
x[n] = x[n − 1] + x[n − 3] n≥3
with initial conditions x[0] = 0, x[1] = 1 and x[2] = 2 can be solved recursively for n ≥ 3 as
follows
n=3 x[3] = x[2] + x[0] = 2

n=4 x[4] = x[3] + x[1] = 2 + 1 = 3
n=5 x[5] = x[4] + x[2] = 3 + 2 = 5
.. ..
. .
The following script obtains the recursive solution for 0 ≤ n ≤ 49
% Pr 9_10
clear all; clf
x(1)=0; x(2)=1; x(3)=2; % initial conditions
for m=4:50,
x(m)=x(m-1)+x(m-3);
end
n=0:49;
figure(1)
stem(n,x); title(’x[n]’); xlabel(’n’)
grid

7 x[n]
x 10
10
0
0 5 10 15 20 25 30 35 40 45 50
n

9.11 Generation of periodic discrete-time signal — Periodic signals can be generated by obtaining
a period and adding shifted versions of this period. Suppose we wish to generate a train of
triangular pulses. A period of the signal is x[n] = 0.5(r[n] − 2r[n − 2] + r[n − 4]) where r[n]
is the discrete-time ramp signal. Obtain the values of x[n] for 0 ≤ n ≤ 4 and write a script to
generate x[n]and the periodic signal y[n]
∞
X
y[n] = x[n − 4k]
k=−∞
and plot 4 of the periods.

Answers: Period x[n] = 0.5δ[n − 1] + δ[n − 3]) + 0.5δ[n − 2].
Solution
(a) A period of a train of pulses is


 0 n<0

r[n] = 0.5n 0≤n≤2
x[n] = 0.5(r[n] − 2r[n − 2] + r[n − 4]) =

 r[n] − 2r[n − 2] = 0.5(4 − n) 2<n≤4

r[n] − 2r[n − 2] + r[n − 4] = 0 n>4
The following MATLAB script will generate x[n] for n ≥ 0 (it is zero before) and the periodic
signal y[n].
% Pr 9_11
x=zeros(1,5);
for m=1:3,
x(m)=m-1;
end
for m=3:5,
x(m)=4-(m-1)
end
x=x/2;
n=0:4;
figure(1)
subplot(211)
stem(n,x); axis([0 4 -0.1 1.1])
hold on; plot(n,x,’:r’); hold off; title(’x[n]’)
%x=[0 0.5 1 0.5 0] % another way to get the first period

y=[x x x x]; % four periods of y
N=length(y); n=0:N-1;
subplot(212)
stem(n,y); axis([0 N-1 -0.1 1.1])
hold on; plot(n,y,’:r’); hold off; title(’y[n]’); xlabel(’n’)

x[n]
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
y[n]
1
0.8
0.6
0.4
0.2
0
0 2 4 6 8 10 12 14 16 18
n
Figure 9.3: Problem 11: One period, 4 periods of y[n].

9.12 Envelope modulation — In the generation of music by computer, the process of modulation is
extremely important. When playing an instrument, the player typically does it in three stages:
(1) rise time, (2) sustained time and (3) decay time. Suppose we model these three stages as an
envelope continuous-time signal given by
1 1
e(t) = [r(t) − r(t − 0.025)] − [r(t − 200) + r(t − 300)]
0.025 0.1
where r(t) is the ramp signal.
(a) For a simple tone x(t) = cos(2π/T0 t), the modulated signal is y(t) = x(t)e(t). Find the
period T0 so that 100 cycles of the sinusoid occur for the duration of the envelope signal.
(b) Simulate in the modulated signal using the value of T0 and a simulation sampling time
of Ts = 0.1. Plot y(t) and e(t) (discretized with the sampling period Ts ) and listen to the
modulated signal using sound.
Answers: If e(t) has duration Te = 300Ts then T0 = Te /100 = 3Ts
Solution
(a) If the duration of e(t) is Te = 300Ts and for 100 cycles of the sinusoid 100T0 then T0 =
Te /100 = 3Ts .
(b) If we let t = nTs , then the sampled envelope is
e(nTs ) = (1/3)(r(nTs ) − r((n − 30)Ts )) − (1/10)(r((n − 200)Ts − r((n − 300)Ts )


 nTs /3 0 ≤ nTs ≤ 30Ts
= 10Ts 30Ts ≤ nTs ≤ 200Ts

3 − nTs /10 200Ts ≤ nTs ≤ 300Ts
The following script generates the window e[n] and the modulated signal y[n] = e[n]x[n]
% Pr 9_12
clear all;clf
Ts=0.1;
t1=0:Ts:3;
t2=20:Ts:30;
e=[t1/3 ones(1,200-30) 3-t2/10]
t=0:Ts:30+Ts;
figure(1)
plot(t,e,’r’);grid
n=0:length(e)-1;
x=cos(2*pi*n/3);
hold on
plot(t,x.*e,’b’)
hold off; axis([0 30 -1.2 1.2]);
xlabel(’nT_s’);ylabel(’x(nT_s)e(nT_s)’)

0.8
0.6
0.4
0.2
x(nTs)e(nTs)
0
−0.2
−0.4
−0.6
−0.8
−1
0 5 10 15 20 25 30
nTs
Figure 9.4: Problem 12: window e(nTs ) and modulated cosine y(n) = x(nTs )e(nTs ) for Ts = 0.1.

9.13 LTI and convolution sum — The impulse response of a discrete-time system is
h[n] = (−0.5)n u[n].
(a) If the input of the system is x[n] = δ[n] + δ[n − 1] + δ[n − 2], use the linearity and time-
invariance of the system to find the corresponding output y[n].
(b) Find the convolution sum corresponding to the above input, and show that your solution
coincides with the output y[n] obtained above.
(c) Use the function conv to find the output y[n] due to the given input x[n]. Plot x[n], h[n]
and y[n] using .
Answers: y[n] = (−0.5)n u[n] + (−0.5)n−1 u[n − 1] + (−0.5)n−2 u[n − 2].
Solution
(a) Using the superposition property of LTI systems and that h[n] is the response to δ[n] gives
y[n] = h[n] + h[n − 1] + h[n − 2] = (−0.5)n u[n] + (−0.5)n−1 u[n − 1] + (−0.5)n−2 u[n − 2]


 1 n=0

0.5 n=1
=

 0.75 n=2

3(−0.5)n n ≥ 3
as the output for the given input x[n].

(b) The convolution sum is given by
∞
X
y[n] = x[k]h[n − k]
k=−∞
X∞
= (δ[k] + δ[k − 1] + δ[k − 2])h[n − k]
k=−∞
X∞ ∞
X ∞
X
= δ[k]h[n − k] + δ[k − 1]h[n − k] + δ[k − 2]h[n − k]
k=−∞ k=−∞ k=−∞
= h[n] + h[n − 1] + h[n − 2]
since the multiplication by the delta functions gives zero except when their arguments are zero,
or k = 0, k = 1 and k = 2. This result coincides with the one in part (a).
(c) The following script is used to compute the convolution sum of the given x[n] and h[n]:
% Pr 9_13
clear all; clf
x=[1 1 1 zeros(1,98)];
n=0:100;
h=(-0.5).ˆn;
y=conv(h,x);
n1=0:20;
figure(1)
subplot(311)
stem(n1,x(1:21)); ylabel(’x[n]’);grid

subplot(312)
stem(n1,h(1:21)); ylabel(’h[n]’);grid
subplot(313)
stem(n1,y(1:21)); ylabel(’y[n]’);grid;xlabel(’n’)
x[n]
0.5
0
0 2 4 6 8 10 12 14 16 18 20
0.5
h[n] 0
−0.5
0 2 4 6 8 10 12 14 16 18 20
0.5
y[n]
−0.5
0 2 4 6 8 10 12 14 16 18 20
n
Figure 9.5: Problem 13: Input x[n], impulse response h[n] and convolution sum y[n].

9.14 Periodicity — Consider a 2 × 3 array

1 for m = n = 0
p[m, n] =
0 for 0 < m ≤ 2, 0 < n ≤ 1
(a) Generate a signal x[m, n] with period p[n, m] and rectangular periods N1 = 3 and N2 = 2,
i.e., x[m, n] = x[m + 3, n] = x[m, n + 2]. Use the functions imshow and stem3 to display the
signals.
(b) Use p[m, n] to create an array y[m, n] with periodic matrix

3 1
N=
0 2
i.e.,
y[m, n] = y[m + 3, n + 1] = y[m, n + 2],
and plot several periods as an image. Is this signal rectangular periodic? If so indicate the
periods.
Solution
(a) The array will be a stacking of p[m, n] horizontally and vertically to generate x[m, n] which
can be visualized for m ≥ 0, n ≥ 0 for 9 periods in first quadrant
p[m, n − 4] p[m − 3, n − 4] p[m − 6, n − 4] · · ·

p[m, n − 2] p[m − 3, n − 2] p[m − 6, n − 2] · · ·
p[m, n] p[m − 3, n] p[m − 6, n] ···
(b) In this case the [0, 0] value appears in [3, 1] and in [0, 2] as indicated by the periodicity
matrix. The overall array can be seen as in the first quadrant
p[m, n − 4] p[m − 3, n − 5] p[m − 6, n − 6] · · ·

p[m, n − 2] p[m − 3, n − 3] p[m − 6, n − 4] · · ·
p[m, n] p[m − 3, n − 1] p[m − 6, n − 2] · · ·
The generated y[m, n] is also periodic of periods N1 = 6 and N2 = 2. The period for these
values contains 12 points instead of the 9 for the given p[m, n].
% Pr 9_14
clear all;clf
p=[0 0 0;1 0 0]
X=[p;p;p];
X=[X X X;X X X]
Y=zeros(4,9);
Y(1,4)=1; Y(2,1)=1;Y(2,7)=1;
Y(3,4)=1;Y(4,1)=1;Y(4,7)=1;
Y=[Y;Y;Y]
figure(1)

subplot(121)
imshow(X)
hold on
stem3(X,’filled’);title(’x[m,n]’)
hold off
subplot(122)
imshow(Y)
hold on
stem3(Y,’filled’);title(’y[m,n]’)
hold off
Results are shown in Fig. 9.6.
x[m,n] y[m,n]
Figure 9.6: Rectangular periodic x[m, n] and block periodic y[m, n].

9.15 Image scaling system Consider the image scaling system
(x[m, n] − M )
y[m, n] = α + β
(N − M )
where the output is y[m, n] and the input is x[m, n] and α, β, M and N are parameters to be
determined for a certain image. Let the input x[m, n be the image clown in a gray scale.
(a) Suppose that we wish to change the pixel values of the output y[m, n] so that they range
from 6 to 256. Determine the values of M and N from the input image, and the parameters
α and β so that the desired scaling is possible. Use the functions colormap and imagesc to
plot the output in a gray scale with a range [0, 255].
(b) Let α = 9 and β = 246, and the input signal x[m, n] be the test image clown. Is this scaling
system linear? Explain why or why not. Plot the resulting image with a range of [0, 255]
for the pixel values.
(c) Compute and plot the histogram of the two images and verify the range of values of the
pixels of the two images.
Solution
(a) The values needed to obtain the desired image are M = 1 and N = 81 corresponding to
the minimum and maximum pixel values of clown and α = 6 and β = 200.
(b) The equation relating the input and the output for the given values of α and β is
(x[m, n] − 1)
y[m, n] = 10 + 246
80
This system is non-linear because of the term 10 which is a bias that does not change with
changes in the input.
% Pr 9_15
clear all;clf
load clown; x=X;
M1=min(x(:))
M2=max(x(:))
for m=1:200.
for n=1:320,
y(m,n)=9+246*(x(m,n)-M1)/(M2-M1);
end
end
M3=min(y(:))
M4y=max(y(:))
figure(1)
colormap(gray)
subplot(221)
imagesc(x,[0 256])
subplot(222)

hist(x,100)
subplot(223)
imagesc(y,[0 256])
subplot(224)
hist(y,100)
Results are shown in Fig. 9.7.
Figure 9.7: Image scaling of clown image.

Chapter 10
The Z–transform
10.1 Basic Problems
10.1 The sign function extracts the sign of a real valued signal, i.e.,

 −1
 x[n] < 0
s[n] = sign(x[n]) = 0 x[n] = 0


1 x[n] > 0
(a) Let s[n] = s1 [n]+s2 [n], where s1 [n] is causal and s2 [n] anti–causal; find their Z-transforms
and indicate the corresponding ROCs.
(b) Determine the Z-transform S(z).
Answers: s[n] = u[n] − u[−n] does not have Z-transform.
Solution
(a) We have s[n] = u[n] − u[−n], the Z-transform of s1 [n] = u[n] is
∞
X 1
S1 (z) = z −n =
n=0
1 − z −1
provided that |z −1 | < 1, thus |z| > 1 is the region of convergence for S1 (z). The Z-transform
of s2 [n] = −u[−n] is given by
0
X ∞
X −1
S2 (z) = − z −n = − zm =
n=−∞ m=0
1−z
1
where the last sum converges for |z| < 1.

(b) The condition for
1 1 1 + z −1
S(z) = S1 (z) + S2 (z) = −1
− =
1−z 1−z 1 − z −1
to converge is that |z| > 1 and that |z| < 1 simultaneously, which is not possible. Since there
is no region of convergence for S(z), the Z-transform of s[n] does not exist.

10.2 An analog pulse x(t) = u(t) − u(t − 1) is sampled using a sampling period Ts = 0.1.
(a) Obtain the discrete-time signal x(nTs ) = x(t)|t=nTs and plot it as a function of nTs
(b) If the sampled signal is represented as an analog signal as
N
X −1
xs (t) = x(nTs )δ(t − nT s)
n=0
determine the value of N in the above equation. Compute the Laplace transform of the
sampled signal. i.e., Xs (s) = L[xs (t]
(c) Determine the Z-transform of x(nTs ) or X(z). Indicate how to transform Xs (s) into X(z).
Answers: Xs (s) = (1 − e−1.1s )/(1 − e−0.1s ); let z = e0.1s .
Solution
(a) If Ts = 0.1 the discrete-time signal is
(
1 0 ≤ 0.1n ≤ 1 or 0 ≤ n ≤ 10
x(0.1n) = [u(t) − u(t − 1)] |t=0.1n =
0 otherwise
(b) Expressing x[n] as indicated, then N = 11.

The Laplace transform of the sampled signal is
10
X 10
X 1 − e−1.1s
Xs (s) = L[δ(t − nTs ) = e−0.1ns =
n=0 n=0
1 − e−0.1s
(c) The z-transform of the discrete-time signal is

10
X 1 − z −11
X(z) = z −n =
n=0
1 − z −1
To transform Xs (s) into X(z) we let z = e0.1s .

10.3 When finding the inverse Z-transform of a function with z −1 terms in the numerator, z −1 can
be thought of a delay operator to simplify the calculation. For
1 − z −10
X(z) =
1 − z −1
(a) Use the Z-transform of u[n] and the properties of the Z-transform to find x[n].
(b) Use your result for x[n] above to express X(z) as a polynomial in negative powers of z.
(c) Find the poles and zeros of X(z) and plot them on the Z–plane. Is there a pole or zero at
z = 1? Explain.
Answers: x[n] = u[n] − u[n − 10]; there is a pole/zero cancellation.
Solution
(a) Writing X(z) as
1 z −10
X(z) = −
1 − z −1 1 − z −1
since the inverse Z-transform of the first term is u[n], then the inverse of the second is −u[n −
10] given that z −10 indicates a delay of 10 samples. Thus,
(
1 0≤n≤9
x[n] = u[n] − u[n − 10] =
0 otherwise
(b) Although X(z) has been shown as a ratio of two polynomials, using the above representa-
tion of x[n] its Z-transform is
X(z) = 1 + z −1 + · · · + z −9
i.e., a 9th –order polynomial in z −1 .

(c) We can rewrite X(z) as
Q9 Q9
z 10 − 1 (z − 1) k=1 (z − ejπk/5 ) k=1 (z − e
jπk/5
)
X(z) = 9 = 9
= 9
z (z − 1) z (z − 1) z
which is obtained by finding that the zeros of X(z) are values zk10 = 1 or zk = ej2πk/10 for
k = 0, · · · , 9. For k = 0 the zero is z0 = 1, which cancels the pole at 1.

10.4 Consider the following problems related to LTI systems.
(a) The impulse response of an FIR filter is h[n] = αn (u[n] − u[n − M ])

i. Is it true that the transfer function for the filter is
1 − αM z −M
H(z) = for any value of α?
1 − αz −1
ii. Let M = 3, 0 ≤ α < 1 write H(z) as a polynomial, and then show that it equals
1 − α3 z −3
. Determine the region of convergence of H(z).
1 − αz −1
(b) Consider the two-sided impulse response h[n] = 0.5|n| , − (N − 1) ≤ n ≤ N − 1.

i. Determine the causal impulse response h1 [n], so that h[n] = h1 [n] + h1 [−n]
ii. Let N = 4, find the transfer function H(z) using the transfer function H1 [z], and
determine the region of convergence of H(z) using the ROC of H1 (z).
iii. Let N → ∞, find the transfer function H(z) and its ROC.
(c) Given the finite length impulse response h[n] = 0.5n (u[n] − u[n − 2]).
i. Find the Z-transforms of the even, he [n], and the odd, ho [n], components of h[n].
ii. Determine the regions of convergence of the Z-transforms of he [n] and ho [n]. How
would the region of convergence of H(z) be obtained from the ROCs of Xe (z) and
Xo (z)? Explain. Find he [n] and ho [n].
P3
Answers: (a) yes; (b) H1 (z) = n=0 0.5n z −n − 0.5, ROC |z| > 0.
Solution
(a) i. Yes. Cross-multiplying
H(z)(1 − αz −1 ) = (1 + αz −1 + · · · + αM −1 z −(M −1) )(1 − αz −1 )

= 1 + αz −1 + · · · + αM −1 z −(M −1) − αz −1 − · · · − αM z −M
= 1 − αM z −M
ii. H(z) = 1 + αz −1 + α2 z −2 and (1 − αz −1 )H(z) = 1 − α3 z −3 . Thus
1 − α3 z −3 z 3 − α3
H(z) = =
1 − αz −1 z 2 (z − α)
so the zeros of H(z) are zk = αej2πk/3 for k = 0, 1, 2, and the poles z = 0 of order
2, and z = α. The pole z = α is cancelled by the zero at k = 0 so ROC is the whole
Z–plane, except z = 0.

(b) i. If we let


 0.5 n=0
h1 [n] = 0.5n 1≤n≤N −1


0 otherwise
then h[n] = h1 [n] + h1 [−n]

ii. For N = 4, Z-transform of h1 [n] is
3
X 1 − 0.54 z −4 0.5 + 0.25z −1 − 0.54 z −4
H1 (z) = 0.5n z −n − 0.5 = −1
− 0.5 =
n=0
1 − 0.5z 1 − 0.5z −1
noticing that the initial expression for H1 (z) is a polynomial in z −1 then the ROC
is |z| > 0. The final expression being rational indicates that if z −1 = 2 or z = 1/2
is a zero and a pole and they cancel so that only poles at the origin remain. The
Z-transform
3
X 0.5 + 0.25z − 0.54 z 4
Z[h1 [−n]] = H1 (1/z) = 0.5n z n − 0.5 =
n=0
1 − 0.5z
with ROC the whole Z–plane as it is a polynomial in z. Verify that the pole at z = 2
of the final expression for Z[h1 [−n]] is cancelled by a zero at the same place. Thus
0.5 + 0.25z −1 − 0.54 z −4 0.5 + 0.25z − 0.54 z 4

H(z) = −1
+
1 − 0.5z 1 − 0.5z
with ROC: |z| > 0.
iii. If N → ∞, we have
1 1 −0.25
H(z) = + −1=
1 − 0.5z −1 1 − 0.5z (1 − 0.5z −1 )(1 − 0.5z)
0.5z
= ROC 0.5 ≤ |z| ≤ 2
(1 − 2z)(1 − 0.5z)
(c) i. H(z) = 1 + 0.5z −1 , |z| > 0, H(z −1 ) is Z-transform of h[−n] so

z + 0.25z 2 + 0.25
He (z) = 0.5[H(z) + H(z −1 )] = 0.5[2 + 0.5(z −1 + z)] = |z| > 0
z
2
0.25 − 0.25z
Ho (z) = 0.5[H(z) − H(z −1 )] = 0.5[0.5(z −1 − z))] = |z| > 0
z
The ROC of H(z) = He (z) + Ho (z) is the intersection of the ROCs of He (z) Ho (z) or
|z| > 0.
ii. The inverse Z-transforms give
he [n] = 0.25δ[n + 1] + δ[n] + 0.25δ[n − 1]

ho [n] = −0.25δ[n + 1] + 0.25δ[n − 1]

10.5 Consider a discrete-time LTI system represented by the difference equation with the given
initial condition
y[n] + 0.5y[n − 1] = 2(x[n] − x[n − 1]) n ≥ 0, y[−1] = 2
where x[n] is the input and y[n] the output.

Suppose the input of the system is x[n] = (1 + 0.5n cos(πn))u[n]. Determine the steady–state
response yss [n]. If the initial condition is changed, would you get the same steady–state re-
sponse? Explain.
Answers: H(z) = 2(1 − z −1 )/(1 + 0.5z −1 ); yss [n] = 0, initial conditions do not change this
response.
Solution
We have
Y (z) 2(1 − z −1 )
H(z) = =
X(z) 1 + 0.5z −1
Y (z) = H(z)(Z[u[n]] + Z[0.5n cos(πn)u[n]])
only the first of the above terms would give steady state due to its pole at z = 1, so
1 2(1 − z −1 ) 2
H(z) = =
1 − z −1 (1 − z −1 )(1 + 0.5z −1 ) 1 + 0.5z −1
which gives a transient, so yss [n] = 0. Since system is stable, changing the initial condition
will not change the steady state–response.

10.6 Determine the impulse response h[n] of the feedback system shown in Fig. 10.1. Determine if
the system is BIBO stable.
Answer: System is not BIBO stable.
e[n]
x[n] Delay y[n]
+
−
Solution Impulse response
y[n] = e[n − 1] = x[n − 1] − y[n − 1]

Y (z) z −1
H(z) = =
X(z) 1 + z −1
h[n] = (−1)n−1 u[n − 1]
System is unstable, h[n] is not absolutely summable:

∞
X
|h[n]| → ∞
n=1
| {z }
1
Moreover, the pole z = 1 is on the unit disc and as such the system is not BIBO stable.

10.7 Consider the following problems for LTI discrete-time systems.
(a) The input and the output of a LTI discrete-time system are

( 
1 n = 0, 1  1 n = 0, 3
Input x[n] = Output y[n] = 2 n = 1, 2
0 otherwise 

0 otherwise
Find the transfer function H(z)

(b) The transfer function of an LTI discrete-time system is
1 − 0.5z −1
H(z) = |z| > 0.5
1 − 0.25z −2
i. Is this system causal? Explain.
ii. Determine the impulse response of the system.
(c) The transfer function of an LTI system is
z+1
H(z) = |z| > 1
z(z − 1)
What are the values of h[0], h[1] and h[1000] of the impulse response.
Answer: (a) H(z) = 1 + z −1 + z −2 ; (b) Causal; (c) h[1000] = 2
Solution
(a) We have
Y (z) = 1 + 2z −1 + 2z −2 + z −3
X(z) = 1 + z −1
Y (z)
H(z) = = 1 + z −1 + z −2
X(z)
after division.
(b) i. Yes, because of the ROC.
ii. There is a pole-zero cancellation
1 − 0.5z −1
H(z) =
(1 − 0.5z −1 )(1 + 0.5z −1 )
so that h[n] = (−0.5)n u[n].

(c) By long division in negative powers of z
H(z) = z −1 + 2z −2 + 2z −3 + · · · + 2z −10000 + · · ·

so that h[0] = 0, h[1] = 1 and h[10000] = 2.

Another way is
z −1 z −2
H(z) = +
1 − z −1 1 − z −1
so that h[n] = u[n − 1] + u[n − 2] giving the same values as above.

10.8 The impulse response of a LTI discrete-time system is h[n] = u[n] − u[n − 3]. If the input to
this system is


 0 n<0
x[n] = n n = 0, 1, 2


1 n≥3
(a) Find the output y[n] of this system by calculating graphically the convolution sum.
(b) Express x[n] in terms of basic functions and determine X(z).
(c) Use the Z-transform to find the output y[n].
Answers: x[n] = δ[n − 1] + 2δ[n − 2] + u[n − 3]; y[n] = x[n] + x[n − 1] + x[n − 2]
Solution
(a) The output is
y[0] = 0 y[1] = 1
y[2] = 3 y[3] = 4
y[4] = 4 y[n] = 3 n ≥ 5
(b) x[n] can be written as
x[n] n(u[n] − u[n − 3]) + u[n − 3]) = δ[n − 1] + 2δ[n − 2] + u[n − 3]

=
z −3
X(z) = z −1 + 2z −2 +
1 − z −1
−1 −2
H(z) = 1 + z + z
thus
Y (z) = X(z)H(z) = X(z) + z −1 X(z) + z −2 X(z)
so that
y[n] = x[n] + x[n − 1] + x[n − 2]

= δ[n − 1] + 3δ[n − 2] + 4δ[n − 3] + 4δ[n − 4] + 3u[n − 5]

10.9 The transfer function of an RLC circuit is H(s) = Y (s)/X(s) = 2s/(s2 + 2s + 1).
(a) Obtain the ordinary differential equation with input x(t) and output y(t). Approximating
the derivatives by differences (let T = 1) obtain the difference equation that approximates
the differential equation.
(b) Let the input be a constant source, so that x[n] = u[n], and let the initial conditions for
the difference equation be zero. Solve the difference equation using the Z-transform.
Answers: y[n] − y[n − 1] + 0.25y[n − 2] = 0.5(x[n] − x[n − 1])
Solution
(a) Voltage division gives the transfer function
Y (s) 2s
H(s) = = 2
X(s) s + 2s + 1
thus the differential equation relating x(t) and y(t) is
d2 y(t) dy(t) dx(t)

2
+2 + y(t) = 2
dt dt dt
With T = 1,
dη(t)/dt ≈ η[n] − η[n − 1]

2 2
d η(t)/dt = d(dη(t)/dt)/dt ≈ (η[n] − η[n − 1]) − (η[n − 1] − η[n − 2])
= η[n] − 2η[n − 1] + η[n − 2]
so that the differential equation is approximated by
(y[n] − 2y[n − 1] + y[n − 2]) + 2(y[n] − y[n − 1]) + y[n] = 2(x[n] − x[n − 1])
y[n] − y[n − 1] + 0.25y[n − 2] = 0.5(x[n] − x[n − 1])
(b) If x[n] = u[n], then x[n] − x[n − 1] = δ[n] and
y[n] − y[n − 1] + 0.25y[n − 2] = 0.5δ[n]
Using Z-transform
Y (z)[1 − z −1 + 0.25z −2 ] = 0.5

0.5
Y (z) = ⇒ y[n] = (n + 1)0.5n+1 u[n]
(1 − 0.5z −1 )2

10.10 Suppose we cascade a “differentiator” and a “smoother” systems characterized by the follow-
ing input/output equations
Differentiator w[n] = x[n] − x[n − 1]

y[n − 1] w[n]
Smoother y[n] = +
3 3
where the output of the differentiator and input to the smoother is w[n], while x[n] is the input
of the differentiator (and of the overall system) and y[n] is the output of the smoother (and of
the overall system).
(a) If x[n] = u[n] and the initial conditions for the smoother are zero, find the output of the
overall system y[n].
(b) If x[n] = (−1)n , −∞ < n < ∞, find the steady-state response yss [n] of the overall system
yss [n].
Answers: The overall transfer function is H(z) = (1/3)(1−z −1 )/(1−z −1 /3); yss [n] = 0.5(−1)n .
Solution
Let Hs (z) and H(z) be the transfer functions of the smoother and of the overall system.
(a) If x[n] = u[n] then w[n] = δ[n] and y[n] = hs [n]. Thus, from
Y (z) 1/3
Hs (z) = = ⇒ y[n] = hs [n] = (1/3)n+1 u[n]
W (z) 1 − z −1 /3
(b) For the overall system
Y (z) Y (z) W (z) (1/3)(1 − z −1 )

H(z) = = =
X(z) W (z) X(z) 1 − z −1 /3
x[n] = (−1)n = cos(πn)
yss [n] = |H(ejπ )| cos(πn + ∠H(ejπ ))
|H(ejπ )| = 1/2, ∠H(ejπ ) = 0
yss [n] = 0.5(−1)n

10.11 Suppose we are given a finite-length sequence h[n] (it could be part of an infinite-length im-
pulse response from a discrete system that has been windowed) and would like to obtain a
rational approximation for it. This means that if H(z) = Z[h[n]], a rational approximation of
it would be H(z) = B(z)/A(z), from which we get
H(z)A(z) = B(z)
Letting
M
X −1 N
X −1
B(z) = bk z −k , A(z) = 1 + ak z −k
k=0 k=1
for some choice of M and N , equations from H(z)A(z) = B(z) should allow us to find the
M + N − 1 coefficients {ak bk }.
(a) Find a matrix equation that would allow us to find the coefficients of B(z) and A(z).
(b) Let h[n] = 0.5n (u[n] − u[n − 101]) be the sequence we wish to obtain a rational approx-
imation and let B(z) = b0 while A(z) = a0 + a1 z −1 , find the equations to solve for the
coefficients {b0 , a0 , a1 }. Use MATLAB to solve the equations.
PN −1
Answers: h[m] + k=1 ak h[m − k] = bm where h[m] = 0 for m < 0 and bm = 0 for m < 0 and
m > M − 1.
Solution
(a) The equation H(z)A(z) = B(z) using convolution and that a0 = 1
N
X −1
h[m] + ak h[m − k] = bm
k=1
where h[m] = 0 for m < 0 and bm = 0 for m < 0 and m > M − 1 or

N −1
(
X bm m = 0, 1, · · · , M − 1
h[m] + ak h[m − k] =
k=1
0 m = M, · · · , M + N − 2
Solving first for the denominator coefficients (the bottom equations above), we get a matrix
equation
     
h[M ] h[M − 1] ··· h[M − N + 1] 1 0
     
 h[M + 1] h[M ] ··· h[M − N + 2]   a1   0 
 .. .. .. ..   .. = .. 
     
 . . . .   .   . 
h[M + N − 2] h[M + N − 3] · · · h[M − 1] aN −1 0
To solve these equations we move the first column of the matrix on the left to replace the zero
vector on the right, using the fact that a0 = 1 is known.

Once the denominator coefficients are found, the following set of equations are solved for the
numerator coefficients:
     
h[0] 0 ··· 0 1 b0
     
 h[1] h[0] ··· 0   a1   b1 
 . .. .. ..   .. = .. 
 .     
 . . . .   .   . 
h[M − 1] h[M − 2] · · · h[M − N ] aN −1 bM −1
(b) We let a0 = 1 and then the equation for a1 is (M = 1, N = 2)
h[0]a1 = −h[1]
so that a1 = −h[1]/h[0] = −0.5/1 = −0.5. Then the b0 is obtained h[0] = b0 , so b0 = 1. The so

called “Pade approximant” is then
b0 1
Ĥ(z) = =
1 + a1 z −1 1 − 0.5z −1
with impulse response ĥ[n] = 0.5n u[n] which approximates very well the given impulse re-
sponse.

10.12 There are signals that do not have Z-transform. Consider
x[m, n] = 2m+n u12 [m, n].
(a) Carefully draw the support of x[m, n].

(b) Attempt to determine the Z-transform of x[m, n], and indicate why it does not exist.
Solution
(a) The support of x[m, n] coincides with that of u12 [m, n] which is the first and second-
quadrant in the (m, n) plane.
(b) We have
∞
" ∞
# " ∞ −1
#
X X 1 X X
X(z1 , z2 ) = 0.5m z1−m 0.5n z2−n = 0.5m z1−m + 0.5m z1−m
m=−∞ n=0
1 − 0.5z −1 m=0 m=−∞
| {z }
1/(1−0.5z −1

1 1 1
= + −1
1 − 0.5z −1 1 − 0.5z1−1 1 − 2z1−1
with the combined ROC:
ROC: |z2 | > 0.5, (|z1 | > 0.5, and |z1 | < 0.5)
and because the last two are contradictory the Z-transform does not exist.

10.13 The region of convergence of H(z1 , z2 ) = 1/(1 − (z1−1 + z2−1 ) is |z1−1 | + |z2−1 | < 1. Determine if
the unit bidisc is included in this ROC and sketch a possible ROC in a |z1 | × |z2 | plane.
Solution
If |z1 | = |z2 | = 1, the bidisc, their sum is 2 which is bigger than 1, so the bidisc is not in the
ROC. The ROC can be expressed as
|z1 |
|z2 | >
|z1 | − 1
and we have that

|z1 | |z2 |
0 >0
1 ∞
2 >2
3 > 1.5
∞ 1
The ROC is thus a curve such that |z2 | → ∞ as |z1 | = 1, it is above z2 = 2 when z1 = 2, is
above z2 = 1.5 when z1 = 3 and as z1 → ∞ it tends to z2 = 1.

10.14 For what values of α is the filter with the following transfer function not BIBO stable?
1
H(z1 , z2 ) =
1 + 0.5z −1 + αz2−1
Solution
The condition for stability
A(z1 , 1) = 1 + 0.5z1−1 + α
has as root z1 = −0.5/(1 + α) and if |z1 | = |0.5/(1 + α)| ≥ 1, i.e., on or outside the unit
circle, the filter is unstable. So for any value of α that makes |1 + α| ≤ 0.5, or any α such that
−1.5 ≤ α ≤ −0.5

10.15 Fibonacci sequence generation — Consider the Fibonacci sequence generated by the differ-
ence equation f [n] = f [n − 1] + f [n − 2], n ≥ 0 with initial conditions f [−1] = 1, f [−2] = −1.
(a) Find the Z-transform of f [n], or F (z). Find the poles φ1 , and φ2 and the zeros of F (z).
How are the poles connected? How are they related to the ”golden-ratio”?
(b) The Fibonacci difference equation has zero input, but its response is a sequence of ever-
increasing integers. Obtain a partial fraction expansion of F (z) and find f [n] in terms
of the poles φ1 and φ2 , and show that the result is always integer. Use MATLAB to
implement the inverse in term of the poles.
Answers: f [n] = {0, 1, 1, 2, 3, 5, 8, 13, 21, 34, · · · } for n ≥ 0
Solution
(a) Using the difference equation and the given initial conditions we get the values of the
Fibonacci series:
f [0] = f [−1] + f [−2] = 0,

f [1] = f [0] + f [−1] = 1,
f [2] = f [1] + f [0] = 1
..
.
or the sequence {0, 1, 1, 2, 3, 5, 8, 13, 21, 34, · · · } for n ≥ 0. The Z-transform of f [n] is then
F (z) = f [0] + f [1]z −1 + f [2]z −2 + f [3]z −3 + · · ·

= 0 + 1z −1 + 1z −2 + 2z −3 + · · ·
which is not in a closed-form. To get that, we use the Z-transform shift property in the given
difference equation:
F (z) = (z −1 F (z) + f [−1]) + (z −2 F (z) + f [−1]z −1 + f [−2])

= [z −1 + z −2 ]F (z) + z −1
after replacing the initial conditions. We then have that
z −1
F (z) =
1− z −1
− z −2
We have obtained the Z-transform of the Fibonacci sequence even though we know that the
sequence blows up, and that the region of convergence for this sequence is not obvious from
the development.

The zero of F (z) is at zero, and the poles are

√
1± 1+4
φ1,2 =
2
= 1.6183, −0.6183
The value φ1 = 1.6183 is the golden ratio which in ancient Greece was considered the most
pleasing ratio for many designs. The letter φ has been used in honor of the Greek sculptor
Phidias who is said to have used it in his work.
(b) Notice that φ1 + φ2 = 1. The partial fraction expansion is
A B
F (z) = −1
+
1 − φ1 z 1 − φ2 z −1
The coefficients are found

1 1
A = F (z)(1 − φ1 z −1 )|z−1 =1/φ1 = =√
φ1 − φ2 5
1 1
B = F (z)(1 − φ2 z −1 )|z−1 =1/φ2 = = −√
φ2 − φ1 5
so for ROC: |z| > φ1 = 1.6183, the inverse is
1
f [n] = √ (φn1 − φn2 )u[n]
5
This formula was first presented by Leonhard Euler in 1765 as a formula to obtain the Fi-
bonacci numbes. Let us verify that this equation gives the Fibonacci numbers, f [0] = 0,
f [1] = √1 (φ1 − φ2 ) = 1, and so on. For the region of convergence |z| > φ1 = 1.6183, the
5
system represented by the Fibonacci difference equation is not BIBO stable as one of its poles
is outside the unit circle.
Using the poles φi we compute the Fibonacci sequence. The obtained values do not coincide
exactly with the sequence, only if we find integers that smaller using floor we obtain it.
% Pr. 10.15
clear all; clf
phi1=1.6183; phi2=-0.6183;
n=0:100;
f=(phi1.ˆn-phi2.ˆn)/sqrt(5);
figure(1)
stem(n(1:10),f(1:10));grid;title(’Fibonacci sequence’);xlabel(’n’)
floor(f(1:10))
0 1 1 2 3 5 8 13 21 34

Fibonacci sequence
35
30
25
20
15
10
0
0 1 2 3 4 5 6 7 8 9
n
Figure 10.2: Problem 15: Approximate computation of Fibonacci sequence .

10.16 Generation of discrete-time sinusoid — Given that the Z-transform of a discrete-time cosine
A cos(ω0 n)u[n] is
A(1 − cos(ω0 )z −1 )
1 − 2 cos(ω0 )z −1 + z −2
(a) Use the given Z-transform to find a difference equation whose output y[n] is a discrete-
time cosine A cos(ω0 n) and input x[n] = δ[n]. What should you use as initial conditions?
(b) Verify your algorithm by generating a signal y[n] = 2 cos(πn/2)u[n] by implementing

your algorithm in MATLAB. Plot the input and the output signals x[n] and y[n].
(c) Indicate how to change your previous algorithm to generate a sine function y[n] =
2 sin(πn/2)u[n]. Use MATLAB to find y[n], and to plot it.
Answers: y[n] − 2 cos(ω0 )y[n − 1] + y[n − 2] = Ax[n] − A cos(ω0 )x[n − 1] n ≥ 0.
Solution
(a) Thinking of the given Z-transform as a transfer function with x[n] as the input and y[n] as
the output we have that
AX(z)(1 − cos(ω0 )z −1 ) = Y (z)(1 − 2 cos(ω0 )z −1 + z −2 )
which gives the following difference equation
y[n] − 2 cos(ω0 )y[n − 1] + y[n − 2] = Ax[n] − A cos(ω0 )x[n − 1] n≥0
with zero initial conditions and input x[n] = δ[n] that will generate the cosine function.
(b) We have that ω0 = π/2 so that cos(ω0 ) = 0 giving the difference equation
y[n] + y[n − 2] = 2δ[n] n≥0
with zero initial conditions. Recursively this equation gives
y[0] = 2
y[1] = 0
y[2] = −2
y[3] = 0
..
.
or y[n] = 2 cos(πn/2)u[n].

(c) The signal to be generated y1 [n] = 2 sin(πn/2) gives for n ≥ 0
y1 [0] = 0
y1 [1] = 2
y1 [2] = 0
y1 [3] = −2
..
.
i.e., y1 [n] = y[n − 1], so that the above difference equation shifted by 1 gives
y[n − 1] + y[n − 3] = 2δ[n − 1] n−1≥0

y1 [m] + y1 [m − 2] = 2δ[m] m≥0
which gives
y1 [0] = −y[−2] + 2 = 0
y1 [1] = −y[−1] + 0 = 2
y1 [2] = y1 [0] = 0
y1 [3] = −y[1] = 2
..
.
so that the initial conditions are y[−2] = 2 and y[−1] = −2.

Another way to obtain the initial conditions to generate 2 sin(πn/2) with the same difference
equation as before is in the z-domain
2X(z) − y[−1]z −1 − y[−2]

Y (z) =
1 + z −2
If we let X(z) = 1 (i.e. x[n] = δ[n]) and initial conditions y[−2] = 2 and y[−1] = −2 we get
2z −1
Y (z) =
1 + z −2
which is the Z-transform of the desired signal.
The following script computes and plots the two sinusoids. To generate the sine, we compute
the first two values using the initial conditions and the input, and then use the difference
equation to obtain the rest using a for loop.
% Pr. 10.16
clear all; clf
% generation of 2cos(pi n/2)
b=2;

a=[1 0 1];
x=[1 zeros(1,100)];
y=filter(b,a,x);
n=0:length(y)-1;
figure(1)
subplot(211)
stem(n,y); axis([0 100 -2.2 2.2]); grid; ylabel(’y[n]’)
hold on; plot(n,y,’:r’); hold off
% generation of 2sin(pi n/2)
y1(1)=0;y1(2)=2;
for m=3:101,
y1(m)=-y1(m-2);
end
subplot(212)
stem(n,y1); axis([0 100 -2.2 2.2]);grid; ylabel(’y_1[n]’);xlabel(’n’)
hold on; plot(n,y1,’:r’); hold off
1
y[n]
−1
−2
0 10 20 30 40 50 60 70 80 90 100
1
y1[n]
−1
−2
0 10 20 30 40 50 60 70 80 90 100
n
Figure 10.3: Problem 16: Generation of 2 cos(πn/2) (top) and 2 sin(πn/2) (bottom).

10.17 Inverse Z-transform — We are interested in the unit-step solution of a system represented by
the following difference equation y[n] = y[n − 1] − 0.5y[n − 2] + x[n] + x[n − 1]
(a) Find an expression for Y (z).

(b) Do a partial expansion of Y (z).
(c) Find the inverse Z-transform y[n]. Use MATLAB to verify your answer.
Answers y[n] = [4 + 3.16(0.707)n cos(πn/4 − 161.5o )]u[n].
Solution
(a) To find the unit–step response, the input is x[n] = u[n] and the initial conditions are zero.
We then have
1
Y (z)[1 − z −1 + 0.5z −2 ] = (1 + z −1 )
1 − z −1
1 + z −1
Y (z) =
(1 − z −1 )(1 − 0.5(1 + j)z −1 )(1 − 0.5(1 − j)z −1 )
(b) The partial fraction expansion is
A B B∗
Y (z) = + +
1 − z −1 1 − 0.5(1 + j)z −1 1 − 0.5(1 − j)z −1
A = Y (z)(1 − z −1 )|z−1 =1 = 4
o
B = Y (z)(1 − 0.5(1 + j)z −1 )|z−1 =0.5(1+j) = 1.58e−j161.5
(c) The inverse Z-transform is then

" √ !n √ !n #
2 jπ/4 2
y[n] = A+B e + B∗ e−jπ/4 u[n]
2 2
√ !n
2
= [A + |B| [ej(πn/4+∠B) + e−j(πn/4−∠B) ]]u[n]
2
√ !n
2
= [A + 2|B| cos(πn/4 + ∠B)]u[n]
2
= [4 + 3.16(0.707)n cos(πn/4 − 161.5o )]u[n]
The following script is used to verify the above result using MATLAB
%% Pr. 10.17
clear all; clf
N=[1 1 0 0];
D=conv([1 -1],[1 -1 0.5])
[r,p,k]=residuez(N,D)
abs(r(2))

angle(r(2))*180/pi
n=0:49;
y=r(1)*(p(1).ˆn)+r(2)*(p(2).ˆn)+r(3)*(p(3).ˆn);
figure(1)
subplot(311)
zplane(N,D)
subplot(312)
stem(n,y); ylabel(’y[n]’)
% verification
y1=4+3.16*(0.707).ˆn.*cos(pi.*n/4-161.5*pi/180);
subplot(313)
stem(n,y1); ylabel(’y_1[n]’); xlabel(’n’)
1
Imaginary Part
0 2
−1
−5 −4 −3 −2 −1 0 1 2 3 4 5
Real Part
6
4
y[n]
0
0 5 10 15 20 25 30 35 40 45 50
4
y1[n]
0
0 5 10 15 20 25 30 35 40 45 50
n
Figure 10.4: Problem 17: Verification of unit-step response using residues and analytic solution. Poles and
zeros of Y (z); unit–step response found using the partial fraction expansion with MATLAB (middle) and
analytically (bottom).

10.18 Prony’s Rational Approximation — The Pade approximant provides an exact matching of
M + N − 1 values of h[n], where M and N are the orders of the numerator and denominator
of the rational approximation. But there is no method for choosing the numerator and de-
nominator orders, M and N . Also, there is no guarantee on how well the rest of the signal is
matched. Prony’s rational approximation considers how well the rest of the signal is approxi-
mated. Let h[n] = 0.9n u[n], be the impulse response we wish to find a rational approximation.
Take the first 100 values of this signal as the impulse response.
(a) Assume the order of the numerator and the denominators are equal M = N = 1, use the
MATLAB function prony to obtain the rational approximation, and then use filter to verify
that the impulse response of the rational approximation is close to the given 100 values.
Plot the error between h[n] and the impulse response of the rational approximation for
the first 100 samples. Plot the poles and zeros of the rational approximation and compare
them to the poles and zeros of H(z) = Z(h[n].
(b) Suppose that h[n] = (h1 ∗ h2 )[n], i.e., the convolution of h1 [n] = 0.9n u[n] and h2 [n] =
0.8n u[n]. Use again Prony to find the rational approximation when the first 100 values
of h[n] are available. Use conv from MATLAB to compute h[n]. Compare the impulse
response of the rational approximation to h[n]. Plot the poles and zeros of H(z) = Z(h[n])
and of the rational approximation.
(c) Consider the h[n] given above, and perform the Prony approximation using orders M =
N = 3, explain your results. Plot poles and zeros.
Solution
(a) The following script obtains the Prony rational approximation of the first 100 value of
h[n] = 0.9n u[n].
% Pr. 10.18
clear all; clf
n=[0:100];
h=0.9.ˆn;
[b,a]=prony(h,1,1); % rational approximation of order (1,1) of h
delta=[1 zeros(1,100)];
hd=filter(b,a,delta); % approximate sequence
figure(1)
subplot(211)
plot(n,h,’o’,n,hd,’+’)
xlabel(’n’)
ylabel(’h[n],hd[n]’)
grid
p=roots(a) % poles
z=roots(b) % zeros

subplot(223)
stem(n,hd-h); ylabel(’error’)
subplot(224)
zplane(b,a)
h[n],hd[n] 0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90 100
n
−15
x 10
2.5 1
2
0.5
Imaginary Part
1.5
error
0
1
−0.5
0.5
−1
0
0 20 40 60 80 100 −1 −0.5 0 0.5 1
Real Part
Figure 10.5: Problem 18: Prony approximation, of order (1, 1), of finite segment of h[n] = 0.9n u[n] .
The poles and zeros corresponding to the actual and the approximate are very close.
(b) The convolution of h1 [n] = 0.9n u[n] and h2 [n] = 0.8n u[n] is done using the MATLAB func-
tion conv. The Prony approximation of h1 [n] ∗ h2 [n] gives an approximation to H1 (z)H2 (z) for
the correct orders. Using prony of order (2, 2) (in general this has to be guessed, but here we
have the advantage of knowing the sequences, a big advantage!).
clear all; clf

n=[0:99];
h1=(0.9).ˆn;
h2=(0.8).ˆn;
h=conv(h1,h2);
[b,a]=prony(h,2,2) % rational approximation of order (2,2) of h
hd=filter(b,a,delta); % approximate sequence
subplot(211)
plot(n,h(1:100),’o’,n,hd(1:100),’+’)
xlabel(’n’)
ylabel(’h(n),hd[n-’)
grid

p=roots(a) % poles
z=roots(b) % zeros
subplot(212)
zplane(b,a)
The following are the results obtained. Notice that the poles/zeros are those of H1 (z) and
H2 (z) or the product H(z) = H1 (z)H2 (z).
b = 1.0000 -0.0000 -0.0000

a = 1.0000 -1.7000 0.7200
p =0.9000
0.8000
z =1.0e-004 *
0.2674
-0.2674
2.5
2
h[n],hd[n]
1.5
0.5
0
0 10 20 30 40 50 60 70 80 90 100
n
0.5
Imaginary Part
−0.5
−1
−3 −2 −1 0 1 2 3
Real Part
Figure 10.6: Problem 18: Prony approximation of convolution h1 [n] and h2 [n].
(c) When the chosen order is higher than the actual one, it is possible to have a pole-zero can-
celations. For the h[n] = [h1 ∗h2 ][n] when using prony instead of a second-order approximation
we choose a third-order, the results has a pole/zero cancellation.
[b1,a1]=prony(h,3,3)
p=roots(a1)
z=roots(b1)
subplot(211)
zplane(z,p)
hd=filter(b1,a1,delta);

M=length(h);
n=[0:ceil(M/2)-1];
subplot(212)
plot(n,h(1:ceil(M/2)),’o’,n,hd(1:ceil(M/2)),’+’)
The results in this case are
b1 =
1.0000 0.0000 -0.0000 -0.0000
a1 =
1.0000 -1.7000 0.7200 0.0000
p =
0.9000
0.8000
-0.0000
z =
1.0e-03 *
0.9266
-0.4633 + 0.8017i
-0.4633 - 0.8017i
showing the same poles as before, which means a third one has been cancelled.

Chapter 11
Fourier Analysis of Discrete–time

Signals and Systems
11.1 Basic Problems
11.1 The frequency response of an ideal low-pass filter is

1 −π/2 < ω < π/2
H(ejω ) =
0 otherwise in(−π, π]
(a) Find the impulse response h[n] of this filter. Is this filter causal?
(b) Suppose that the input of this filter x[n] has a DTFT X(ejω ) = H(ejω ), what is the output
of the filter y[n].
(c) According to the above result, does that mean that (h ∗ h)[n] = h[n]? Explain.
Answers: h[n] is non–causal; yes, h[n] = (h ∗ h)[n].
Solution
(a) Impulse response
Z π/2
1 jωn 0.5 n=0
h[n] = e dω =
2π −π/2 6 0
sin(πn/2)/(πn) n =
h[n] is non-causal as h[n] 6= 0 for n < 0.

(b) Y (ejω ) = X(ejω )H(ejω ) = H(ejω ) so y[n] = h[n]
(c) Yes, H(ejω ) = H(ejω )H(ejω ) so h[n] = (h ∗ h)[n].
1
11.2 Find the DTFT X(ejω ) of x[n] = δ[n] − δ[n − 2].

(a) Sketch and label carefully the magnitude spectrum |X(ejω )| for 0 ≤ ω < 2π.
(b) Sketch and label carefully the magnitude spectrum |X(ejω )| for −π ≤ ω < π.
(c) Sketch and label carefully the phase spectrum ∠X(ejω ) for −π ≤ ω < π.
Answer: X(ejω ) = 2je−jω sin(ω).
Solution
(a) (a) (b) DTFT
X(ejω ) = 1 − e−j2ω = 2je−jω sin(ω)
Yes, X(ejω ) is periodic of period 2π since ω + 2kπ = ω. |X(ejω )| = 2| sin(ω)| is periodic
|X(ejω )|
ω
−π π
∠X(ejω )
ω
−π π
Figure 11.1: Problem 2: Magnitude and phase of X(ejω )
of period π, but also periodic of period 2π.

(b) Phase

−ω + π/2 if sin(ω) > 0
∠X(ejω ) =
−ω + 3π/2 = −ω − π/2 if sin(ω) ≤ 0
and because of the odd symmetry of the phase it is zero at ω = 0. See Fig. 11.1.


11.3 The transfer function of an FIR filter is H(z) = z −2 0.5z + 1.2 + 0.5z −1 .
(a) Find the frequency response H(ejω ) of this filter. Is the phase response of this filter linear?
(b) Find the impulse response h[n] of this filter. Is h[n] symmetric with respect to some n?
How does this relate to the phase?
Answers: Phase is linear; h[n] = 0.5δ[n − 1] + 1.2δ[n − 2] + 0.5δ[n − 3].
Solution
(a) Frequency response
H(ejω ) = e−j2ω (1.2 + cos(ω))

∠H(ejω ) = −2ω linear, as 1.2 + cos(ω) > 0 ω ∈ (−π, π]
(b) Impulse response
0.5 z −1 + |{z}
H(z) = |{z} 1.2 z −2 + |{z}
0.5 z −3
h[1] h[2] h[3]
h[n] is symmetric with respect to n = 2, so the phase is linear.

11.4 A periodic discrete-time signal x[n] with a fundamental period N = 3 is passed through a
filter with impulse response h[n] = (1/3)(u[n] − u[n − 3]). Let y[n] be the filter output. We
begin the filtering at n = 0, i.e., we are interested in y[n], n ≥ 0 and could assume y[n] = 0,
n < 0. For a period, x[n] is x[0] = 1, x[1] = −2, x[2] = 1.
(a) Determine the values of y[0], y[1] and y[2]. Because x[n] is periodic you can assume its
values for n < 0 are given.
(b) Use the convolution sum to calculate y[n], n ≥ 0. What is the steady state output? When
is the steady-state of y[n] attained?
(c) Compute the output of the filter using the DTFT.
Answers: y[0] = y[1] = y[2] = 0; H(ej0 ) = 1, H(ej2π/3 ) = 0.
Solution
(a) The impulse response is h[n] = (1/3)[δ[n] + δ[n − 1] + δ[n − 2] so that y[n] = (1/3)(x[n] +
x[n − 1] + x[n − 2]) for n ≥ 0, so
y[0] = (1/3)(x[0] + x[−1] + x[−2]) = 0
y[1] = (1/3)(x[1] + x[0] + x[−1]) = 0
y[2] = (1/3)(x[2] + x[1] + x[0]) = 0
(b) Convolution sum:

Impulse response h[n] = (1/3)(δ[n] + δ[n − 1] + δ[n − 2])
n
X n
1 X
y[n] = x[k]h[n − k] = x[k](δ[n − k] + δ[n − k − 1] + δ[n − k − 2]
3
k=−∞ k=−∞
1
= (x[n] + x[n − 1] + x[n − 2])
3
which is the average of three consecutive values of x[n] or 0. The steady state is y[n] = 0
which is attained at n = 0.
(c) The Fourier series coefficients of x[n] are
1
X[k] = (1 − 2z −1 + z −2 )|z=ej2πk/3
3
e−j2πk/3 j2πk/3 −2e−j2πk/3
= (e − 2 + e−j2πk/3 ) = [1 − cos(2πk/3)]
3 3
giving X[0] = 0 and X[1] = X ∗ [−1] = −(2/3)e−j2π/3 (1 − cos(2π/3)) = e−j2π/3 , and
ω0 = 2π/3. So that
1
X 1
X
jω jkω0 n
X(e ) = F[ X[k]e ]= 2πX[k]δ(ω − kω0 )
k=−1 k=−1
= 2πX[−1]δ(ω + ω0 ) + 2πX[1]δ(ω − ω0 )
The filter frequency response is
e−jω jω e−jω
H(ejω ) = (e + 1 + e−jω ) = (1 + 2 cos(ω))
3 3

giving H(ej0 ) = 1, H(ej2π/3 ) = 0. Thus the output is
Y (ejω ) = X(ejω )H(ejω ) = 2πX[−1]H(ej2π/3 ) + 2πX[1]H(e−j2π/3 ) = 0

11.5 From the definition of the DFT

N
X −1
X[k] = x[n]e−j2πnk/N k = 0, 1, · · · , N − 1
n=0
one can obtain a matrix equation X = Fx.

(a) Suppose that N = 2 and x[n] = 1 for n = 0, 1, express the DFT as a matrix equation.
(b) Check if the matrix F in your representation is invertible, and if so obtain the its inverse
so that you have a representation of the IDFT.
(c) Is it true that F∗ T F = αI, i.e., the product of the transpose of the complex conjugate F∗
and F is the identity multiplied by a constant α? What is α? Would the inverse of F be
(1/α)F∗ T ? Is this related to the orthonormality of the Fourier exponential basis? Explain.
Answers:
1 1
F = , F T F = −2I
1 −1
Solution
(a) Matrix equation

X[0] 1 1 1
=
X[1] 1 −1 1
which gives X[0] = 2 and X[1] = 0. Since det(F) = −2 6= 0 the matrix is invertible with
inverse 0.5FT = 0.5F and

−1 X[0] x[0]
F =
X[1] x[1]
(b) We have
FT
F=I
|det(F)|
| {z }
F−1
so α = 2.

11.6 The input of a discrete-time system is x[n] = u[n] − u[n − 4] and the impulse of the system is
h[n] = δ[n] + δ[n − 1] + δ[n − 2]
(a) Calculate the DFTs of h[n], x[n] of length N = 7. Call them H(k) and X(k). Calculate
X(k)H(k) = Ŷ (k)
(b) If y(n) = (x ∗ h)[n], would you expect ŷ[n] = y[n]? Explain. If you had chosen the lengths
of the DFTs to be N = 4 would you have had the circular and the linear convolutions
coincide? Explain.
P3 P2
Answers: (a) X[k] = n=0 e−j2πnk/7 , H[k] = n=0 e−j2πnk/7 ; (b) yes, they coincide.
Solution
(a) The DFTs of length 7 are
2
X
H[k] = e−j2πnk/7
n=0
3
X
X[k] = e−j2πnk/7
n=0
Y [k] = X[k]H[k] = 1 + 2e−j2πk/7 + 3e−j4πk/7

+3e−j6πk/7 + 2e−j8πk/7 + e−j10πk/7
(b) Yes, the circular convolution coincides with the linear convolution if the DFTs are of
length 7, which is larger than the length of the linear convolution (4 + 3 − 1 = 6). If N < 6
the two convolutions will not coincide.

11.7 Frequency shift of FIR filters — Consider a moving average FIR filter with an impulse re-
sponse h[n] = 13 (δ[n] + δ[n − 1] + δ[n − 2]). Let H(z) be the Z-transform of h[n].
(a) Find the frequency response H(ejω ) of the FIR filter.

(b) Let the impulse response of a new filter be given by h1 [n] = (−1)n h[n], use again the
eigenfunction property to find the frequency response H1 (ejω ) of the new FIR filter.
(c) Use the MATLAB functions freqz and abs to compute the magnitude response of the two
filters. Plot them and determine the poles and the zeros of the two filters. What type of
filters are these?
Answers: H1 (ejω ) = (1/3)e−jω (2 cos(ω) − 1), high–pass.
Solution
(a) The Z-transform of h[n] is
1
H(z) = (1 + z −1 + z −2 )
3
with ROC the whole Z–plane, except the origin, so we can let z = ejω to get
1 −jω jω 1
H(ejω ) = e (e + 1 + e−jω ) = e−jω (1 + 2 cos(ω))
3 3
(b)
X 1 −j(ω−π)
H1 (ejω ) = h[n]e−j(ω−π)n = H(ej(ω−π) ) = e (1 + 2 cos(ω − π))
n
3
1 −jω
= e (2 cos(ω) − 1)
3
(c) The frequency response of the two filters is computed and plotted using the following
script
% Pr. 11_7
clear all;clf
b1=(1/3)*[1 1 1];
b2=(1/3)*[1 -1 1];
[H1,w]=freqz(b1,1);
[H2,w]=freqz(b2,1);
figure(1)
subplot(221)
plot(w/pi,abs(H1));ylabel(’|H_1|’)
subplot(222)
plot(w/pi,angle(H1));ylabel(’<H_1’)
subplot(223)
plot(w/pi,abs(H2));ylabel(’|H_2|’);xlabel(’\omega/\pi’)
subplot(224)
plot(w/pi,angle(H2));ylabel(’<H_2’);xlabel(’\omega/\pi’)

1 2
0.8 1
0.6 0
|H1|
<H1
0.4 −1
0.2 −2
0 −3
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
1 3
0.8 2
0.6 1
|H2|
<H2
0.4 0
0.2 −1
0 −2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ω/π ω/π
Figure 11.2: Problem 7: Magnitude and phase of low–pass H1 (z) (top) and high–pass H2 (z) filters

11.8 Computations from DTFT definition — For simple signals it is possible to obtain some infor-
mation on their DTFTs without computing it. Let
x[n] = δ[n] + 2δ[n − 1] + 3δ[n − 2] + 2δ[n − 3] + δ[n − 4].
(a) Find X(ej0 ) and X(ejπ ) without computing the DTFT X(ejω ).
(b) Find Z π
|X(ejω )|2 dω
−π
jω
(c) Find the phase of X(e ). Is it linear?
Use MATLAB to verify your results.
Answers: X(ej0 ) = 9, X(ejπ ) = 1, phase is linear.
Solution
(a) Using
∞
X
X(ejω ) = x[n]e−jnω
n=−∞
we have
∞
X
X(ej0 ) = x[n] = 1 + 2 + 3 + 2 + 1 = 9
n=−∞
∞
X
X(ejπ ) = (−1)n x[n] = 1 − 2 + 3 − 2 + 1 = 1
n=−∞
(b) By Parseval’s result Z

1 π X
|X(ejω )|2 dω = |x[n]|2
2π −π n
we get that Z π
|X(ejω )|2 dω = 2π(1 + 4 + 9 + 4 + 1) = 38π
−π
(c) The Z–transform of x[n] is
X(z) = 1 + 2z −1 + 3z −2 + 2z −3 + z −4 = z −2 [z 2 + 2z 1 + 3 + 2z −1 + z −2 ]
and the DTFT is  

 
X(ejω ) = e−j2ω 3 + 2 cos(ω) + 4 cos(2ω)
| {z }
real-valued
so that the phase is −2ω when above term is positive and −2ω ± π when negative.
Some of the above results are verified using the following script.
% Pr. 11_8
clear all;clf
x=[1 2 3 2 1];
N=256; X=fft(x,N);

w=[0:N-1]*2*pi/N;
figure(1)
subplot(211)
plot(w,abs(X));axis([min(w) max(w) -0.1 10]);grid
ylabel(’|X|’)
subplot(212)
plot(w,unwrap(angle(X)));axis([min(w) max(w) -15 1]);grid
ylabel(’<X’);xlabel(’\omega’)
10
6
|X|
0
0 1 2 3 4 5 6
−5
<X
−10
−15
0 1 2 3 4 5 6
ω
Figure 11.3: Problem 8: Magnitude and phase responses for x[n].

11.9 Linear equations and Fourier series — The Fourier series of a signal x[n] and its coefficients
Xk are both periodic of some value N and as such can be written
N
X −1
(i) x[n] = Xk ej2πnk/N 0≤n≤N −1
k=0
N −1
1 X
(ii) Xk = x[n]e−j2πnk/N 0≤k ≤N −1
N n=0
(a) To find the x[n], 0 ≤ n ≤ N − 1 given Xk , 0 ≤ k ≤ N − 1, write a set of N linear equations.

Indicate how you would find the x[n] from the matrix equation. There is duality in the
Fourier series and its coefficients, so see it consider the reverse problem: how would you
solve for the Xk given the x[n]?
(b) Let x[n] = n for n = 0, 1, 2 and 0 for n = 3, be a period of a periodic signal x[n] of fun-
damental period N = 4, use the above method to solve for the Fourier series coefficients
Xk , 0 ≤ k ≤ 3. Use MATLAB to find the inverse of the complex exponential matrix.
(c) Suppose that when computing the Xk for the x[n] signal given above, you separate the
sum into 2 sums, one for the even values of n, i.e., n = 0, 2 and the other for the odd
values of n, i.e., n = 1, 3. Write an equivalent matrix expression for the Xk .
Solution
(a) Equation (ii) can be written in matrix form as
X = Sx
where X = [X0 X1 · · · XN −1 ]T is an N vector containing the FS coefficients, x = [x[0]/N x[1]/N · · · x[N −

1]/N ]T are the signal values divided by N , and the N × N matrix S contains the complex ex-
ponentials. Then
x = S−1 X
Calling Ŝ = S−1 we would then have
x = ŜX
and then
X = Ŝ−1 x
where ŜS = I, where I is the N × N unit matrix, i.e., Ŝ and S are orthonormal matrices.
(b) The following script computes the Fourier series coefficients from the first period of the
periodic signal, and computes the first period using the Fourier series. Notice the way the
matrix S is generated.
% Pr. 11_9
clear all; clf
x1=[0 1 2 0];
N=4;n=0:N-1;k=n;
S=exp(-j*2*pi.*(n’*k)/N)
Xk=(1/N)*S*x1’
x=inv(S)*N*Xk
S =1.0000 1.0000 1.0000 1.0000
1.0000 0.0000 - 1.0000i -1.0000 - 0.0000i -0.0000 + 1.0000i

1.0000 -1.0000 - 0.0000i 1.0000 + 0.0000i -1.0000 - 0.0000i

1.0000 -0.0000 + 1.0000i -1.0000 - 0.0000i 0.0000 - 1.0000i
Xk = 0.7500
-0.5000 - 0.2500i
0.2500 + 0.0000i
-0.5000 + 0.2500i
x = -0.0000 + 0.0000i
1.0000 + 0.0000i
2.0000 - 0.0000i
-0.0000
(c) For N = 4 we can write the Fourier series coefficients multiplied by 4 as
4Xk = x[0] + x[1]e−j2πk/4 + x[2]e−j2π2k/4 + x[3]e−j2π3k/4

n o n o
= x[0] + x[2]e−j2πk/2 + e−j2πk/4 x[1] + x[3]e−j2πk/2
where we have simplified some of the exponentials. Using that e−j2πk/2 = (−1)k and e−j2πk/4 =
(−j)k we finally obtain

Xk = 0.25 x[0] + x[2](−1)k + (−j)k x[1] + x[3](−1)k 0≤k≤3
We thus have using the above results
1 T 1 T
X= S x[0] x[1] x[2] x[3] = SP x[0] x[2] x[1] x[3]
4 4
where 41 SP = S1 and P is a permutation matrix
   
1 0 0 0 1 1 1 1
1  0 0 1 0  = 1
 1 −1 −j −j 
S1 = S   

4 0 1 0 0 4 1 1 −1 −1 
0 0 0 1 1 −1 j −j
| {z }
P
so that
   
X0 x[0]
 X1   x[2] 
   
 X2  = S1  x[1] 
X3 x[3]

11.10 DFT and IIR filters — A definite advantage of the FFT is that it reduces considerably the
computation in the convolution sum. Thus if x[n], 0 ≤ n ≤ N − 1, is the input of an FIR
filter with impulse response h[n], 0 ≤ n ≤ M − 1, their convolution sum y[n] = (x ∗ h)[n] will
be of length M + N − 1. Now if X[k] and H[k] are the DFTs (computed by the FFT) of x[n]
and h[n], then Y [k] = X[k]H[k] is the DFT of the convolution sum, of length M + N − 1. To
multiply the FFTs X[k] and H[k] they both should be of the same length as Y [k], i.e., M +N −1.
Consider what happen when the filter is IIR which has possibly an impulse response of very
large length. Let
y[n] − 1.755y[n − 1] + 0.81y[n − 2] = x[n] + 0.5x[n]
be the difference equation representing an IIR filter with input x[n] and output y[n]. Assume
the initial conditions are zero, and the input is x[n] = u[n] − u[n − 50]. Use MATLAB to obtain
the filter output.
(a) Compute using filter the first 40 values of the impulse response h[n], and call them ĥ[n],
an approximate of h[n]. Compute the filter output ŷ[n] using the FFT as indicated above.
In this case, we are approximating the IIR filter by and FIR filter of length 40. Plot the
input and the output. Use FFTs of length 128.
(b) Suppose now that we do not want to approximate h[n], so consider the following proce-
dure. Find the transfer function of the IIR filter, say H(z) = B(z)/A(z), and if X(z) is the
Z-transform of the input then
B(z)X(z)
Y (z) =
A(z)
Compute as before the FFT for x[n], of length 128, call it X[k], and compute the 128-length
of the coefficients of B(z) and A(z) to obtain DFTs B[k] and A[k]. Multiplying X[k] by
B[k] and dividing by A[k], all of length 128, results in a sequence of length 128 that should
correspond to Y [k], the DFT of y[n]. Compute the inverse FFT to get y[n] and plot it.
(c) Use filter to solve the difference equation and obtain y[n] for x[n] = u[n] − u[n − 50].
Considering this the exact solution, calculate the error with respect to the other responses
in (a) and (b). Comment on your results.
Answers: h[n] decreases by n = 40.
Solution
(a) The impulse response of the IIR filter appears to taper down around n = 30 so the approx-
imation of the impulse response seems good.
% Pr. 11_10
clear all;clf
x=ones(1,50);
b=[1 0.5 0]; a=[1 -1.755 0.81];
h=filter(b,a,delta);
N=128;
X=fft(x,N);H=fft(h,N);
Y=X.*H; y=real(ifft(Y));
figure(1)
subplot(311)
stem([x zeros(1,10)])
subplot(312)

stem([h zeros(1,10)])
subplot(313)
stem(y)
(b)(c) In this case we do not truncate the impulse response, and the results y1 [n] are better than
before y[n], (the control is the solution obtained using filter, yc [n]). See right figure in Fig. 11.4.
The error is clearly seen when compared to the first result, but there is no visible error when
compared to the second.
% part 2
X=fft(x,N);A=fft(a,N);B=fft(b,N);
Y1=X.*B./A;
y1=ifft(Y1);
yc=filter(b,a,x);
figure(2)
subplot(211)
plot(y)
hold on
plot(yc,’r’);grid
hold off
subplot(212)
plot(y1)
hold on
plot(yc, ’r’); grid
hold off
1 40
30
x[n]
0.5
y1[n], yc[n]
20
0 10
0 10 20 30 40 50 60
0
4
3 −10
0 20 40 60 80 100 120 140
2
h[n]
1
0 40
0 5 10 15 20 25 30 35 40 30
y1[n],yc[n]
20
30
20 10
y[n]
10
0
0
−10
0 20 40 60 80 100 120 0 20 40 60 80 100 120 140
n n
Figure 11.4: Problem 10: Input x[n], impulse response ĥ[n] and output y[n] (left); right: exact re-
sponse compared to solution when h[n] is truncated (top) and when it is not (bottom)

pwindows — Circular windows are of interest in the design of

11.11 Fourier transform of circular
2D-filters. Suppose that r = x2 + y 2 and let the circular window be defined as

 1 r<1
wc (x, y) = 0.5 r = 1

0 r>1
Its 2D-Fourier transform is given by

Z ∞
W (ω1 , ω2 ) = wc (x, y)e−2πj(ω1 x+ω2 y dxdy
−∞
Since the symmetry of the window is circular, express the above equation in terms of polar
coordinates, i.e., let x + jy = rejθ and ω1 + jω2 = qejφ .
(a) Using polar coordinates show that the 2D-Fourier transform becomes
Z ∞ Z 2π
W (q) = wc (r) exp (−2πjrq(cos(θ − φ))rdrdθ
0 0
(b) Lettingρ = θ − φ, show that the above equation for W (q) can be written
Z ∞
W (q) = rwc (r)J0 (2πqr)dr
0
where Z 2π
J0 (2πqr) = e−2πrq cos(ρ) dρ
0
is the zero-order Bessel function of the first kind.
(c) Replacing the circular window we finally obtain
Z 1 Z 2πq
1 J1 (2πq)
W (q) = rJ0 (2πqr)dr = r0 J0 (r0 )dr0 =
0 (2πq)2 0 2πq
where J1 (.) is the first order Bessel function of the first king.
(d) Use MATLAB to generate a circular discrete window wc [m, n] and compute its 2D-FFT
to get Wc (k, `). Determine what type of symmetry Wc (k, `) has. What does it look like?
Solution
%Pr. 11_11
clear all; clf
N=125; [m,n] = freqspace(N,’meshgrid’);
wc = ones(N);
r = sqrt(m.ˆ2 + n.ˆ2);
wc(r>0.05)=0;
figure(1)
subplot(311)
colormap(’gray’)

contour(wc); title(’Circular window’)

Wc=fft2(wc)
subplot(312
mesh(fftshift(abs(Wc)));title(’Magnitude response’)
subplot(313)
contour(fftshift(abs(Wc)),5);title(’Contour of magnitude response’)
Results are shown in Fig. 11.

Circular window
120
100
80
60
40
20
20 40 60 80 100 120
Magnitude response
40
20
0
150 150
100 100
50 50
0 0
Contour of magnitude response

120
100
80
60
40
20
20 40 60 80 100 120
Figure 11.5: Problem 11: Magnitude response of circular windows. Contour plot of circular win-
dow (top), magnitude response of window (middle) and its contour.

11.12 Separable and non-separable cosines —Consider the separable and non-separable cosine sig-
nals
x1 [m, n] = cos(0.3πm) cos(0.2πn)

x2 [m, n] = cos(0.25πn + 0.5πm)
computed in a support (0 ≤ m ≤ 123, 0 ≤ n ≤ 123).
(a) Use the MATLAB function fft2 to compute the FFTs of these two functions, and plot their
magnitude functions |Xi (ω1 , ω2 )|, i = 1, 2 using the function mesh (let the frequencies
ω1 , ω2 in the plots be in the domain [−π, π] × [−π, π]) and comment on their difference
(use the function contour to help you).
(b) Given a signal

1 0 ≤ m, n ≤ 99
x[m, n] =
0 100 ≤ m, n ≤ 123
Generate modulated signals y1 [m, n] = x[m, n]x1 [m, n] and y2 [m, n] = x[m, n]x2 [m, n],
and using fft2 determine the magnitude functions |Yi (ω1 , ω2 )|, i = 1, 2 and comment on
their differences.
Solution
%Pr. 11_12
% 2D-FFT of separable, non-separable and modulated cosines
clear all; clf
% (a) separable and non-separable cosines
for m=1:124,
for n=1:124,
x1(m,n)=cos(0.2*pi*(n-1))*cos(0.3*pi*(m-1)); % separable
x2(m,n)=cos(0.25*pi*(n-1)+0.5*pi*(m-1)); % non-separable
end
end
X1=fft2(x1); X2=fft2(x2);
X1a=abs(fftshift(X1)); X2a=abs(fftshift(X2));
w1=-pi:2*pi/124:pi-2*pi/124;w1=w1/pi; w2=w1;
figure(1)
subplot(221)
mesh(w1,w2,X1a);
axis([min(w1) max(w1) min(w2) max(w2) 0 max(max(X1a))])
xlabel(’\omega_1/\pi ’), ylabel(’\omega_2 /\pi’)
title(’Magnitude spectrum of x1[m,n]’)
subplot(222)
contour(w1,w2,X1a);grid
xlabel(’\omega_1/\pi’), ylabel(’\omega_2 /\pi’)
title(’Contour plot of the magnitude spectrum’)
subplot(223)
mesh(w1,w2,X2a); view(25,20);

M=max(max(X2a));
axis([min(w1) max(w1) min(w2) max(w2) 0 M])
xlabel(’\omega_1 /\pi’), ylabel(’\omega_2 /\pi’)
title(’Magnitude spectrum of x2[m,n]’)
subplot(224)
contour(w1,w2,X2a);grid
title(’Contour plot of the magnitude spectrum’)
% (b) modulated cosines
x=zeros(124,124);
for m=1:100,
for n=1:100,
x(m,n)=1;
end
end
y1=x*x1; y2=x*x2;
Y1=fft2(y1); Y1a=abs(fftshift(Y1));
Y2=fft2(y2); Y2a=abs(fftshift(Y2));
figure(2)
subplot(221)
mesh(w1,w2,Y1a);
axis([min(w1) max(w1) min(w2) max(w2) 0 max(max(Y1a))])
title(’Magnitude spectrum of y1[m,n]’)
subplot(222)
contour(w1,w2,Y1a);grid
title(’Contour plot of the |Y_1(k,l)|’)
subplot(223)
mesh(w1,w2,Y2a);
axis([min(w1) max(w1) min(w2) max(w2) 0 max(max(Y2a))])
title(’Magnitude spectrum of y2[m,n]’)
subplot(224)
contour(w1,w2,Y2a);grid
title(’Contour plot of the |Y_2(k,l)|’)
Results are shown in Fig. 12.

Magnitude spectrum of x1[m,n] Contour plot of the magnitude spectrum Magnitude spectrum of y1[m,n] Contour plot of the |Y 1 (k,l)|
10 -12
2000 0.5 0.5
6
4
1000
/
0 0
2
2
0 -0.5 0 -0.5
0.5 0.5
0 0.5 0 0.5
-0.5 0 -1 -0.5 0 -1
-0.5 -0.5
-1 -1 -1 -0.5 0 0.5 -1 -1 -1 -0.5 0 0.5
2
/ / 2
/ /
1 1
1
/ 1
/
Magnitude spectrum of x2[m,n] Contour plot of the magnitude spectrum Magnitude spectrum of y2[m,n] Contour plot of the |Y 2 (k,l)|
10 -10
0.5 0.5
1.5
4000
1
3000
/
/
0 0
2
2
2000 0.5
1000 -0.5 0 -0.5
0 0.5 0.5
0 0 0.5
-1 -0.5 -0.5 -1 -0.5 0 -1
0 0.5 -0.5
-1 -1 -0.5 0 0.5 -1 -1 -1 -0.5 0 0.5
/ / /
/ 2 2 1
1
1
/ 1
/
Figure 11.6: Problem 12: Results of parts (a) separable and non-separable and (b) modulated.

11.13 DCT vs DFT —To see the advantage of the DCT over the DFT consider a one-dimensional
edge having the values
x = [8 16 24 32 40 48 4 8]
Compute the FFT and DCT of length 8 and choose the first 4 values of the FFT and DCT
sequences and set the other 4 to zero . Compute the IFFT and IDCT of the modified FFT and
DCT sequences to obtain sequences that approximates the edge sequence. Plot the original
sequence, and the approximate sequences obtained from the IFFT and IDCT and plot the
approximation error. Find the mean square error corresponding to the two approximating
sequences and determine which gives the smallest error. Why could you say that the FFT-
based procedure has an ’unfair’ advantage over the DCT-based procedure.
Solution
% Pro 11.13
% comparison of DCt and FFT
%
x=[8 16 24 32 40 48 4 8];
X=fft(x)
Cx=dct(x)
X1=zeros(1,8); X1(1:4)=X(1:4);
x1=floor(real(ifft(X1)))
Cx1=zeros(1,8); Cx1(1:4)=Cx(1:4);
xx=floor(idct(Cx1))
figure(1)
subplot(211)
stem(x,’filled’); hold on; stem(x1,’red’,’filled’); hold on; plot(x);grid;hold on;
plot(x1,’r’); hold off; title(’DFT approximation’)
subplot(212)
stem(x,’filled’); hold on; stem(xx,’red’,’filled’);hold on; plot(x);grid; hold on
plot(xx,’r’); hold off;title(’DCT approximation’)
figure(2)
subplot(211)
stem(x-x1,’filled’);hold on; plot(x-x1);hold off;title(’DFT error’)
subplot(212)
stem(x-xx,’filled’);hold on; plot(x-xx); hold off; title(’DCT error’)
edft=sum((x-x1).*(x-x1))/8
edct=sum((x-xx).*(x-xx))/8
The FFT-based procedure is using complex values and the DCT-based procedure real values,
so the first is using twice as many values as the second one.

DFT approximation DFT error

50 15
40 10
5
30
0
20
-5
10 -10
0 -15
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
DCT approximation DCT error

50 15
40 10
5
30
0
20
-5
10 -10
0 -15
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
Figure 11.7: Problem 13:DCT vs DFT approximation and errors

Chapter 12
Introduction to the Design of Discrete

Filters
12.1 Basic Problems
12.1 The frequency response of a filter is H(ejω ) = 1.5 + cos(2ω), − π ≤ ω ≤ π.
(a) Is |H(ejω )| = H(ejω )? Is phase zero?

(b) Find the impulse response h[n] of this filter. What type of filter is it? FIR? IIR? Is it causal?
Explain.
(c) Let H1 (ejω ) = e−jN ω H(ejω ) determine the value of the positive integer N so that H1 (ejω )
has linear phase. Use the impulse response h[n] found above to find the impulse response
h1 [n].
Answers: (a) Yes to both; (b) h[n] = 0.5δ[n + 2] + 1.5 + 0.5δ[n − 2].
Solution
(a) Yes, H(ejω ) > 0 and real with zero phase.

(b) Impulse response
H(ejω ) = |{z} 0.5 e−j2ω
0.5 ej2ω + |{z}
1.5 + |{z}
h[0] h[−2] h[2]
It can also be found using the IDTFT

Z π
1
h[n] = (1.5 + cos(2ω))ejωn dω
2π −π
Z π Z π
1 jωn 1
= 1.5e dω + cos(2ω)ejωn dω
2π −π 2π −π
1
which is different from zero for n = 0 and n = ±2, and h[0] = 1.5 and h[−2] = h[2] = 0.5.
Indeed, then
H(z) = 0.5z 2 + 1.5 + 0.5z −2

H(ejω ) = 1.5 + cos(2ω)
The filter is a non-causal FIR.

(c) Since H(ejω ) has zero phase, then H1 (ejω ) = e−jN ω H(ejω ) has linear phase for any N ≥
1, and h1 [n] = h[n − N ].

12.2 An FIR filter has a transfer function H(z) = z −2 (z − ejπ/2 )(z − e−jπ/2 ).
(a) Find and plot the poles and zeros of this filter.
(b) Expressing the frequency response at some frequency ω0 as
Z~1 (ω0 )Z~2 (ω0 )
H(ejω0 ) =
P~1 (ω0 )P~2 (ω0 )
carefully draw the vectors on the pole-zero plot.
(c) Consider the case when ω0 = 0, find H(ej0 ) analytically and using the vectors.
(d) Repeat the above calculations for ω0 = π and ω0 = ±π/2 (verify your result from the
given Z-transform). Can you classify this filter?
Answers: H(ej0 ) = 2; H(ejπ ) = 2; H(e±jπ/2 ) = 0.
Solution
(a) Poles z = 0 double; zeros z1,2 = e±jπ/2 = ±j.
(b) Vectors Z~1 (ω0 ) and Z~2 (ω0 ) go from the zeros to the frequency where we are finding the
frequency response. Likewise, vectors P~1 (ω0 ) and P~2 (ω0 ) go from the poles to the fre-
quency for which we are finding the frequency response.
(c) Analytically, |H(ej0 )| = (1 − j)(1 + j)/1 = 2. Graphically
√ √
j0 |Z~1 (0)||Z~2 (0)| 2 2
|H(e )| = = =2
~
(|P1 (0)| 2 1
and the phase would be
∠Z~1 (0) + ∠Z~2 (0) − 2∠P~1 (0) = (2π − φ) + φ + 0 = 2π
so H(ej0 ) = 2. Notice that when ω = 0 then H(ej0 ) = H(z)|z=1 .
2
×
Figure 12.1: Problem 2: Poles and zeros and vectors to ω = 0 to find H(ej0 ).
(d) Magnitude response: analytically and graphically we have

H(ejπ/2 ) = 0
H(ejπ ) = 2ej0 = 2
The filter is a notch filter with notch frequency π/2.

12.3 The transfer function of an FIR filter is H(z) = z −2 (z − 2)(z − 0.5).

(a) Find the impulse response h[n] of this filter and plot it. Comment on any symmetries it
might have.
(b) Find the phase ∠H(ejω ) of the filter and carefully plot it. Is this phase linear for all
frequencies −π < ω ≤ π?
Answers: h[n] = δ[n] − 2.5δ[n − 1] + δ[n − 2], phase is linear.
Solution
(a) Transfer function can be written as
1 + (−2.5) z −1 + |{z}
H(z) = |{z} 1 z −2
| {z }
h[0] h[1] h[2]
thus the impulse response is
h[n] = δ[n] − 2.5δ[n − 1] + δ[n − 2]
(b) Frequency response
H(ejω ) = e−jω (ejω − 2.5 + e−jω )

= e−jω (2 cos(ω) − 2.5)
= ej(−ω±π) (−2 cos(ω) + 2.5)
| {z }
>0
so the phase is
∠H(ejω ) = −ω ± π
or linear. This corresponds to the impulse response being symmetric about n = 1 and
having no zeros on the unit circle.

12.4 Consider the following problems related to the specification of IIR filters
(a) The magnitude specifications for a low-pass filter are
1 − δ ≤ |H(ejω )| ≤ 1 0 ≤ ω ≤ 0.5π
jω
0 < |H(e )| ≤ δ 0.75π ≤ ω ≤ π
i. Find the value of δ that gives the following equivalent loss specifications for this filter
α(ej0 ) = 0 dB, αmax = 0.92 dB, αmin = 20 dB
ii. What are the values of ωp and ωst ?

(b) The following are specifications for a low-pass discrete IIR filter that will be used in
processing analog signals
10 ≤ α(ejω ) ≤ 10.1 dBs 0 ≤ f ≤ 2 kHz

α(ejω ) ≥ 60 dBs 4 kHz ≤ f ≤ 5 kHz
i. Find the values of αmax , αmin and the dc loss in dBs.

ii. What is the sampling frequency fs in kHz in this filter specifications?
iii. Obtain the equivalent specifications in discrete frequencies for 0 dB dc loss for this
filter.
Answers: (a) δ = 0.1; ωp = 0.5π; (b) αmax = 0.1, αmin = 50 dB and 10 dB dc loss.
Solution
(a) i. From αmin = 20 = −20 log10 (δ) then δ = 10−1 = 0.1 and it can be verified that
αmax = 0.92 = −20 log10 (1 − δ) = −20 log10 0.9.
ii. ωp = 0.5π and ωst = 0.75π
(b) i. The dc loss α(ej0 ) = 10 dB, and subtracting it we get that
αmax = 0.1 dB
αmin = 50 dB.
ii. The maximum frequency 5 kHz≤ fs /2, so the sampling frequency is fs ≥ 10 kHz.
iii. Using ω = 2πf /fs and defining α̂(ejω ) = α(ejω ) − 10 we get
0 ≤ α̂(ejω ) ≤ 0.1 dBs 0 ≤ ω ≤ 0.4π

jω
α̂(e ) ≥ 50 dBs 0.8π ≤ ω ≤ π
with dc loss of 0 dB.

12.5 A low-pass IIR discrete filter has a transfer function

∞
X
H(z) = 0.5n z −n
n=0
(a) Find the poles and zeros of this filter.

(b) Suppose that you multiply the impulse response of the low-pass filter by (−1)n so that
you obtain a new transfer function
∞
X
H1 (z) = (−0.5)n z −n
n=0
find the poles and zeros for H1 (z). What type of filter is it?
(c) Is it true, in general, that if every z in a low-pass filter transfer function is changed into
−z you obtain a high-pass filter? Explain.
Answers: H(z) = z/(z − 0.5); H1 (z) is high-pass filter.
Solution: (a), (b) In rational form

∞
X 1 z
H(z) = 0.5n z −n = =
n=0
1 − 0.5z −1 z − 0.5
with a zero at z = 0 and a pole at z = 0.5. The filter is a low-pass filter.

∞
X 1 z
H1 (z) = (−0.5)n z −n = =
n=0
1 + 0.5z −1 z + 0.5
with zero at z = 0 and pole at z = −0.5. The filter is a high-pass filter, as the frequency
response of h1 [n] = ejπn h[n] so that H1 (ejω ) = H(ej(ω−π) ) which is the low-pass filter shifted
to π.
(c) Suppose X
H(Z) = h[n]Z −n
n
is the transfer function of a LPF, if we let Z = −z = (−1)z we obtain

X X
h[n][(−1)z]−n = (h[n](−1)n )z −n
n n
so that the resulting filter has h[n](−1)n as impulse response, therefore it is high-pass.

12.6 Consider the following transfer function:
√
2(z − 1)(z 2 + 2 z + 1)
H(z) =
(z + 0.5)(z 2 − 0.9z + 0.81)
(a) Develop a cascade realization of H(z) using a first-order and a second-order sections.
Use minimal direct form to realize each of the sections.
(b) Develop a parallel realization of H(z) by considering first and second-order sections,
each realized using minimal direct form.
Answers: Parallel H(z) = −4.94 + 2.16/(1 + 0.5z −1 ) + (4.78 − 1.6z −1 )/(1 − 0.9z −1 + 0.81z −2 )
Solution: (a) Let

√
2(1 − z −1 ) 1 + 2z −1 + z −2
H(z) =
1 + 0.5z −1 1 − 0.9z −1 + 0.81z −2
| {z } | {z }
H1 (z) H2 (z)
The cascade realization of H(z) is as shown in Fig. 12.2
2
x[n] + + + + y[n]
− + + +
−1
z z −1
√
0.5 −2 0.9 2
z −1
−0.81
Figure 12.2: Problem 6: Cascade realization of H(z).
(b) To obtain the parallel realization we do a partial fraction expansion of H(z). Notice that
H(z) is not proper rational and so we have to obtain a constant term before performing the
partial fraction expansion
2.16 4.78 − 1.6z −1

H(z) = −4.94 + +
1 + 0.5z −1 1 − 0.9z −1 + 0.81z −2

−4.94
x[n] 2.16 y[n]

+ +
z −1
−0.5
4.78
+ +
z −1
0.9 −1.6
−1
z
−0.81
Figure 12.3: Problem 6: Parallel realization of H(z).


12.7 FIR filters: causality and phase — A three-point moving–average filter is of the form:
y[n] = β (αx[n − 1] + x[n] + αx[n + 1])
where α and β are constants, and x[n] is the input and y[n] is the output of the filter.
(a) Determine the transfer function H(z) = Y (z)/X(z) of the filter and from it find the fre-
quency response H(ejω ) of the filter in terms of α and β.
(b) Let α = 0.5, find then β so that the d.c. gain of the filter is unity, and the filter has a zero
phase. For the given and obtained values of α and β, sketch H(ejω ) and find the poles
and zeros of H(z) and plot them in the Z-plane.
(c) Suppose we let v[n] = y[n−1] be the output of a second filter. Is this filter causal? Find its
transfer function G(z) = V (z)/X(z). Use MATLAB to compute the unwrapped phases
of G(z) and to plot the poles and zeros of G(z) and H(z) and explain the relation between
G(z) and H(z).
Answers: H(ejω ) = β(1 + 2α cos(ω)); β = 0.5
Solution: (a) The transfer function is

Y (z)
H(z) = = β(αz −1 + 1 + αz)
X(z)
letting z = ejω
H(ejω ) = β(αe−jω + 1 + αejω ) = β(1 + 2α cos(ω))
(b) The dc gain is
H(ej0 ) = β(1 + 2α) = 1
so that
1
β=
1 + 2α
For the phase to be zero H(ejω ) ≥ 0 and so we could let β > 0, and 1 + 2α cos(ω) ≥ 0. The last
equation requires that for all ω the amplitude of the cos(ω) be 2α ≤ 1 or α ≤ 1/2.
If we let α = 0.5 then β = 0.5 and
H(ejω ) = 0.5(1 + cos(ω)) −π ≤ω <π
having a unit dc and zero phase as H(ejω ) ≥ 0.

For α = β = 0.5, the transfer function is
(1 + z)2
H(z) = 0.5(0.5z −1 + 1 + 0.5z) = 0.25z −1 (1 + 2z + z 2 ) = 0.25
z
so pole z = 0 and double zero z = −1 .
(c) We have that v[n] = y[n − 1] = β(αx[n − 2] + x[n − 1] + αx[n]) so the new filter is causal.
The previous filter was non-causal. The transfer function
V (z) 0.25(z + 1)2

G(z) = = z −1 H(z) =
X(z) z2

with double poles at z = 0 and double zero at z = −1. Since H(z) is zero phase, G(z) has
linear phase ∠G(ejω = −ω.
The following script computes and plots the frequency responses of the two filters.
% Pr. 12.7
clear all; clf
N=128; w=[0:N/2-1]*(2*pi/N);
H=0.5*(1+cos(w));
figure(1)
subplot(221)
plot(w, abs(H)); axis([0 pi 0 1]); ylabel(’|H|’)
subplot(222)
plot(w,angle(H));axis([0 pi -0.1 0.1]); ylabel(’<H’)
beta=0.5;alpha=0.5;
b=[0.25 0.5 0.25]
[G,w]=freqz(b,1);
subplot(223)
plot(w,abs(G)); axis([0 pi 0 1]); ylabel(’|G|’)
subplot(224)
plot(w,unwrap(angle(G))); axis([0 pi -pi 0]); ylabel(’<G’); xlabel(’\omega’)
1 0.1
0.8
0.05
0.6
|H|
<H
0
0.4
−0.05
0.2
0 −0.1
0 1 2 3 0 1 2 3
1 0
0.8 −0.5
−1
0.6
−1.5
<G
|G|
0.4 −2
0.2 −2.5
−3
0
0 1 2 3 0 1 2 3
ω ω
Figure 12.4: Problem 7: Frequency responses of H(e ) (top) and G(ejω ) (bottom).
jω

12.8 Effect of phase on filtering — Consider two filters with transfer functions
10
−100 1 − 2z −1
(i) H1 (z) = z , (ii) H2 (z) = 0.5
1 − 0.5z −1
(a) The magnitude response of these two filters is unity, but that they have different phases.
Find analytically the phase of H1 (ejω ) and use MATLAB to find the unwrapped phase of
H2 (ejω ) and to plot it.
(b) Consider the MATLAB signal handel.mat (a short piece of the Messiah by composer George
Handel), use the MATLAB function filter to filter it with the two given filters. Listen to
the outputs, plot them and compare them. What is the difference (look at the first 200
samples of the outputs from the two filters)?
(c) Can you recover the original signal by advancing either of the outputs? Explain.
Answers: Use MATLAB function conv to find coefficients H2 (z).
Solution: (a) Clearly H1 (z) has unit magnitude. H2 (z) is all-pass filter: let H2 (z) = (F (z))10 ,

1 − 2e−jω 1 − 2ejω 1 − 2e−jω −2ejω (1 − 0.5e−jω )
|F (ejω )|2 = 0.5 0.5 = 0.5 0.5 =1
1 − 0.5e−jω 1 − 0.5ejω 1 − 0.5e−jω −0.5ejω (1 − 2e−jω )
so |H2 (ejω )| = |F (ejω )|10 = 1. Since H2 (z) is the product of F (z) ten times, its coefficients can
be computed using the function conv to find the coefficients.
(b) The following script computes the coefficients of the filters and their frequency responses
and then processes handel.
% Pr. 12.8
clear all; clf
b1=[zeros(1,99) 1];
b=[0.5 -1];a=[1 -0.5];
b2=b;a2=a;
for k=2:10; % computing coefficients of H_2(z)
b2=conv(b,b2);
a2=conv(a,a2);
end
[H1,w]=freqz(b1,1);
[H2,w]=freqz(b2,a2);
figure(1)
subplot(221)
plot(w,abs(H1));ylabel(’|H1|’);axis([0 pi 0 1.2])
subplot(222)
plot(w,unwrap(angle(H1)));ylabel(’<H1’);axis([0 pi -320 0])
subplot(223)
plot(w,abs(H2));ylabel(’|H2|’);axis([0 pi 0 1.2])
subplot(224)
plot(w,unwrap(angle(H2)));ylabel(’<H1’);axis([0 pi -32 0])
load handel; x=y’;

y1=filter(b1,1,x);
y2=filter(b2,a2,x);
figure(2)

subplot(311)
plot(x(1:200));ylabel(’x[n]’);axis([0 199 -0.4 0.4])
subplot(312)
plot(y1(1:200));ylabel(’y_1[n]’);axis([0 199 -0.4 0.4])
subplot(313)
plot(y2(1:200));ylabel(’y_2[n]’);axis([0 199 -0.4 0.4])
xlabel(’n’)
The output of H1 (z) is the input signal delayed 100 samples, the output of H2 (z) is not due to
its phase not being exactly linear, although the filter is all-pass.
(c) If the phase is linear advancing the signal 100 samples gives the original signal, while that
is not possible for the non-linear phase.
0 0.4
1 −50 0.2
x[n]
0.8 −100 0
|H1|
−150
<H1
0.6 −0.2
−200 −0.4
0.4 0 20 40 60 80 100 120 140 160 180
−250
0.2
0.4
−300
0 0.2
0 1 2 3 0 1 2 3
y1[n]
0 −0.2
1 −5 −0.4
0 20 40 60 80 100 120 140 160 180
0.8 −10
0.4
−15
|H2|
<H1
0.6 0.2
−20
y2[n]
0.4 0
−25
0.2 −0.2
−30
0 −0.4
0 1 2 3 0 1 2 3 0 20 40 60 80 100 120 140 160 180
n
Figure 12.5: Problem 8: Magnitude and phase of filters (left). Part of handel signal (top right) and outputs
of linear–phase and non–linear phase filters.

12.9 Bilinear transformation and pole location — Find the poles of the discrete filter obtained by
applying the bilinear transformation with K = 1 to and analog second–order Butterworth
low-pass filter. Determine the half-power frequency ωhp of the resulting discrete filter. Use
the MATLAB function bilinear to verify your results. q
√ √
Answers: Double zero at z = −1 and poles at z1,2 = ±j (2 − 2)/(2 + 2).
Solution: The second order low-pass filter (normalized in frequency) is
1
H(s) = √
s2 + 2s + 1
applying the bilinear transformation s = (1 − z −1 )/(1 + z −1 ) we obtain the following discrete

filter
(1 + z −1 )2
H(z) = √
(1 − z −1 )2 + 2(1 − z −2 ) + (1 + z −1 )2
(1 + z −1 )2 (z + 1)2
= √ √ = √ √ √
(2 + 2) + (2 − 2)z −2 (2 + 2)(z 2 + (2 − 2)/(2 + 2))
q √ √
so we have a double zero at z = −1 and poles at z1,2 = ±j (2 − 2)/(2 + 2).
We have that Kb transforms the normalized half-power analog frequency Ωhp = 1 into the
normalized half-power frequency ωhp of the discrete filter, so that
1
Kb = 1 = ⇒ ωhp = π/2
tan(ωhp /2)
The following script verifies the analytic results
% Pr. 12.9
clear all;clf
n=1;d=[1 sqrt(2) 1];
[b,a]=bilinear(n,d,0.5)
[H,w]=freqz(b,a)
figure(1)
subplot(211)
zplane(b,a)
subplot(212)
plot(w/pi,abs(H));grid;ylabel(’|H(eˆ{j\omega})|’); xlabel(’\omega/\pi’)
The obtained coefficients are
b =0.2929 0.5858 0.2929

a =1.0000 0.0000 0.1716

0.5
Imaginary Part
0 2
−0.5
−1
−3 −2 −1 0 1 2 3
Real Part
0.8
|H(ejω)| 0.6
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ω/π
Figure 12.6: Problem 9: Poles/zeros of H(z) with ωhp = π/2.

12.10 Discrete Butterworth filter for analog processing — Design a Butterworth low-pass discrete
filter that satisfies the following specifications
0 ≤ α(ejω ) ≤ 3 dB for 0 ≤ f ≤ 25 Hz
jω
α(e ) ≥ 38 db for 50 ≤ f ≤ F s/2 Hz
and the sampling frequency is Fs = 2000 Hz. Express the transfer function H(z) of the de-
signed filter as a cascade of filters. Use first the design formulas and then use MATLAB to
confirm your results. Show that the designed filter satisfies the specifications, plotting the loss
function of the designed filter.
Q4
Answers: N = 7; H(z) = G i=1 Hi (z); H1 (z) = 0.04(1 + z −1 )/(1 − 0.92z −1 ).
Solution: We have that Ts = 1/Fs = 0.5×10−3 sec/sample, so Kb = cot(0.5ωhp ) = cot(2πfhp /(2Fs )) =

25.45. In this case, we use the loss function for the Butterworth to find the order
2N
α(ejωst ) = 10 log [1 + (Kb tan(2πfst /(2Fs )))] ≥ 38
The minimum order is
log(103.8 − 1)
N≥ = 6.298
2 log(0.078Kb )
so that N = 7.
The frequency normalized analog filter in factorized form is
G
H7 (s) = = GH1 (s)H2 (s)H3 (s)H4 (s)
(s + 1)(s2 + 1.8s + 1)(s2 + 1.25s + 1)(s2 + 0.45s + 1)
We have then
1 (1/(Kb − 1)(1 + z −1 )
H1 (z) = |s=Kb (1−z−1 )/(1+z−1 ) =
s+1 (Kb + 1)/(Kb − 1) − z −1
For the others which are second-order we have
1
Hi (z) = Hi (s)|s=Kb (1−z−1 )/(1+z−1 ) = |BT
s2
+ (1/Qi )s + 1
(1 + z −1 )2
=
(Kb2 + Kb /Qi + 1) − 2(Kb2 − 1)z −1 + (Kb2 − Kb2 /Qi + 1)z −2
for i = 2, 3, 4 and Q2 = 1/1.8, Q3 = 1/1.25, and Q4 = 1/0.45. Replacing Kb and Qi we get
0.04(1 + z −1 )
H1 (z) =
1 − 0.92z −1
1.44 × 10−3 (1 + z −1 )2
H2 (z) =
1 − 1.86z −1 + 0.87z −2
1.47 × 10−3 (1 + z −1 )2
H3 (z) =
1 − 1.9z −1 + 0.91z −2
1.51 × 10−3 (1 + z −1 )2
H4 (z) =
1 − 1.96z −1 + 0.91z −2
To get H7 (ej0 ) = 1 so that the dc loss is 0 db,
GH1 (1)H2 (1)H3 (1)H4 (1) = 1
which gives G = 1.

% Pr. 12.10
N = 7; fhp = 25; Fs = 2000;
whp = 2*pi*fhp/Fs;
[b, a] = butter(N, whp/pi);
% analytic results
% coefficients of H1
b1 = [0.04, 0.04]; a1 = [1, -0.92];
b2 = 1.44e-3*[1, 2, 1]; a2 = [1, -1.86, 0.87];
b3 = 1.47e-3*[1, 2, 1]; a3 = [1, -1.9, 0.91];
b4 = 1.51e-3*[1, 2, 1]; a4 = [1, -1.96, 0.91];
% comparison of analytic and MATLAB results
analyticalB = conv(conv(conv(b1, b2), b3), b4)
matlabB = b
analyticalA = conv(conv(conv(a1, a2), a3), a4)
matlabA = a
The comparison of analytic and computational results gives
analyticalB =1.0e-08 *
0.0128 0.0895 0.2685 0.4475 0.4475 0.2685 0.0895 0.0128
matlabB =1.0e-08 *
0.0121 0.0848 0.2544 0.4240 0.4240 0.2544 0.0848 0.0121
analyticalA = 1.0000 -6.6400 18.8560 -29.6888 27.9928 -15.8063 4.9490

-0.6628
matlabA = 1.0000 -6.6471 18.9442 -30.0084 28.5329 -16.2848 5.1656
-0.7025

12.11 Butterworth filtering of analog signal — We wish to design a discrete Butterworth filter that can be
used in filtering a continuous-time signal. The frequency components of interest in this signal are be-
tween 0 and 1 kHz, so we would like the filter to have a maximum passband attenuation of 3 dB within
that band. The undesirable components of the input signal occur beyond 2 kHz, and we like to attenuate
them by at least 10 dB. The maximum frequency present in the input signal is 5 kHz. Finally, we would
like the d.c. gain of the filter to be 10. Choose the Nyquist sampling frequency to process the input
signal. Use MATLAB to design the filter. Give the transfer function of the filter, plot its poles and zeros,
and its magnitude and unwrapped phase response using an analog frequency scale in kHz.
Answers: Low-pass, with fp = 1 kHz (fp = fhp ), αmax = 3 dB , fst = 2 kHz, αmin = 10 dB.
Solution: The desired filter is a low-pass, with fp = 1 kHz, αmax = 3 dB, so that fp = fhp ; fst = 2
kHz, αmin = 10 dB, and dc gain of 10 after the design with dc of 0 dB. Since fmax = 5 kHz, then we can
choose Fs = 10 kHz. The coefficients of the resulting filter are shown below.
% Pr. 12.11
Fs=10000;fp=1000;fst=2000;
wp=2*pi*fp/Fs/pi; ws=2*pi*fst/Fs/pi;
alphamax=3; alphamin=10;
[N,wn]=buttord(wp,ws,alphamax,alphamin)
[b,a]=butter(N,wp); % to guarantee that wp is half-power frequency (wn\ne wp)
G=10*sum(b)/sum(a);
b=G*b;
[H,w]=freqz(b,a);
figure(1)
subplot(221)
zplane(b,a)
subplot(222)
plot(w/pi*Fs/2, abs(H));grid; axis([0 Fs/2 -0.1 11])
subplot(223)
plot(w/pi*Fs/2,unwrap(angle(H)));grid; axis([0 Fs/2 -4 0])
b = 0.6746 1.3491 0.6746
a = 1.0000 -1.1430 0.4128
1 10
0.5 8
Imaginary Part
2 6
0
4
−0.5
2
−1 0
−1 −0.5 0 0.5 1 0 1000 2000 3000 4000 5000
Real Part
−1
−2
−3
−4
0 1000 2000 3000 4000 5000
Figure 12.7: Problem 11: Poles/zeros and magnitude and phase responses.

12.12 Butterworth vs Chebyshev filtering — If we wish to preserve low frequencies components

of the input, a low-pass Butterworth filter could perform better than a Chebyshev filter. MAT-
LAB provides a second Chebyshev filter function cheby2 that has a flat response in the pass-
band and a rippled one in the stopband. Let the signal to be filtered be the first 100 samples
from MATLAB’s train signal. To this signal add some Gaussian noise to be generated by randn,
multiply it by 0.1 and add it to the 100 samples of the train signal. Design three discrete filters,
each of order 20, and half-frequency (for Butterworth butter) and the passband frequency (for
the Chebyshev filters) of ωn = 0.5. For the design with cheby1 let the maximum passband
attenuation be 0.01 dB and for the design with cheby2 let the minimum stopband attenuation
be 60 dB. Obtain the three filters and use them to filter the noisy train signal.
Using MATLAB plot the following for each of the 3 filters:
• Using the fft function compute the DFT of the original signal, the noisy signal and the
noise, and plot their magnitudes. Is the cutoff frequency of the filters adequate to get rid
of the noise? Explain.
• Compute and plot the magnitude and the unwrapped phase, as well as the poles and ze-
ros for each of the three filters. Comment on the differences in the magnitude responses.
• Use the function filter to obtain the output of each of the filters, and plot the original
noiseless signal and the filtered signals. Compare them.
Solution: The following script designs the desired filters, computes and plots the output of
the filtered signals and different spectra:
% Pr. 12.12
clear all; clf
load train;y=y’; N=256;
noise=0.1*randn(1,N);
yc=y(1:N);
yn=yc+noise; % noisy signal
%[b,a]=butter(20,0.5); % butterworth
[b,a]=cheby2(20,60,0.5); % cheby2
%[b,a]=cheby1(20,0.01,0.5) % cheby1
figure(1)
[H,w]=freqz(b,a);
b=b/H(1);
[H,w]=freqz(b,a); % unit dc gain
subplot(311)
plot(w/pi, abs(H)); grid
hold on
plot(w/pi,0.707*ones(1,length(w)),’r’)
hold off
subplot(312)
plot(w/pi,unwrap(angle(H)))
subplot(313)
zplane(b,a)
k=0:N-1;w1=k*2*pi/N;w1=w1/pi-1;
Ns=fftshift(abs(fft(noise))); % noise spectrum
Yc=fftshift(abs(fft(yc))); % clean signal spectrum
Yn=fftshift(abs(fft(yn))); % noisy signal spectrum

figure(2)
subplot(311)
plot(w1,Yc); title(’clean signal spectrum’)
subplot(312)
plot(w1,Yn); title(’noisy signal spectrum’)
subplot(313)
plot(w1,Ns); title(’noise spectrum’)
yy=filter(b,a,yn);
n=0:N-1;
figure(3)
subplot(311)
plot(n,yn,’r’);title(’noisy/clean signal’)
hold on
plot(n,yc);legend(’noisy’,’clean’)
hold off
subplot(312)
plot(n,yy,’r’);legend(’clean signal’)
subplot(313)
plot(n,yc); legend(’filtered’,’clean’)
hold off
clean signal spectrum

1.5 15
1 10
0.5 5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
10 noisy signal spectrum

15
0 10
−10 5
−20 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
noise spectrum
1 4
Imaginary Part
0.5 3
0 2
−0.5 1
−1 0
−5 −4 −3 −2 −1 0 1 2 3 4 5 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
Real Part noisy/clean signal
0.4
0.2 noisy
0 clean
−0.2
−0.4
−0.6
0 50 100 150 200 250
0.4
0.2 clean signal
0
−0.2
−0.4
0 50 100 150 200 250
0.4
filtered
0.2
0
−0.2
−0.4
0 50 100 150 200 250
Figure 12.8: Problem 12: Chebyshev 2 case: Filter frequency response and poles zeros (top left), spectra of
different signals, signals (bottom)

12.13 Notch and all-pass filters — Notch filters is a family of filters that include the all-pass filter.
For the filter
(1 − α1 z −1 )(1 + α2 z −1 )
H(z) = K
(1 − 0.5z −1 )(1 + 0.5z −1 )
(a) Determine the values of α1 , α2 and K that would make H(z) an all-pass filter of unit
magnitude. Use MATLAB to compute and plot the magnitude response of H(z) using
the obtained values for α and K. Plot the poles and zeros of this filter.
(b) If we would like the filter H(z) to be a notch filter, of unit gain at at ω = π/2 (rad), and
notches at ω = 0 and π, determine the values of α and K to achieve this. Use MATLAB
functions to verify that the filter is a notch filter, and to plot the poles and zeros.
(c) Use MATLAB to show that when α1 = α2 = α and 1 ≤ α ≤ 2, the given H(z) correspond
to a family of notch filters with different attenuations. Determine K so that these filters
are unity gain at ω = π/2.
(d) Suppose we use the transformation z −1 = jZ −1 to obtain a filter H(Z) using H(z) ob-
tained in the previous item. Where are the notches of this new filter? What is the differ-
ence between the all-filters H(z) and H(Z)?
Answers: H(z) = K(1−α2 z −2 )/(1−0.25z −2 ), 1 ≤ α ≤ 2, notch filters with notches at ω = 0, π.
Solution: (a) For the filter to be all-pass the zeros must be the inverse of the poles 0.5 and
−0.5 or 2 and −2. Thus
(1 − 2z −1 )(1 + 2z −1 ) (1 − 4z −2 )
H(z) = K = K K>0
(1 − 0.5z −1 )(1 + 0.5z −1 ) (1 − 0.25z −2 )
As an all-pass H(z) has a constant magnitude (at any frequency) and in this case we want it
to be unity. For instance, for this filter to have unit gain at ω = π/2 we need that

1 − 4j 2
|H(ejπ/2 )| = K = K 5 = 1 ⇒ K = 1.25 = 0.25
1 − 0.25j 2 1.25 5
(b) The zeros of H(z) are α1 and −α2 which are real. For H(z) to be a notch filter |αi | = 1, and
so we let α1 = α2 = 1 for notches at ω = 0, π. Thus
(1 − z −1 )(1 + z −1 ) (1 − z −2 )
H(z) = K −1 −1
=K
(1 − 0.5z )(1 + 0.5z ) (1 − 0.25z −2 )
and for it to have unit gain at ω = π/2

1 − j2
|H(ejπ/2 )| = K = K 2 = 1 ⇒ K = 1.25 = 0.25
1 − 0.25j 2 1.25 2
(c) In general, the filter
(1 − α2 z −2 )
H(z) = K
(1 − 0.25z −2 )
for 1 ≤ α ≤ 2 is a family of notch filters with notches at ω = 0, π, including an all-pass filter
when α = 2. The gain K > 0 is

1 − α2 j 2 1 + α2
|H(e jπ/2
)| = K
=K = 1 ⇒ K = 1.25
1 − 0.25j 2 1.25 1 + α2

(d) If we let z −1 = jZ −1 the filter becomes
(1 − α2 j 2 Z −2 ) (1 + α2 Z −2 )
H(Z) = K = K
(1 − 0.25j 2 Z −2 ) (1 + 0.25Z −2 )
with poles at ±0.5ejπ/2 and zeros at ±αejπ/2 , i.e., shifted 90o with respect to the ones before.
For values of 1 ≤ α ≤ 2, H(Z) is a family of notch filters at ω = π/2, including an all-pass filter
when α = 2. When running the following script make the numerator and the denominator
correspond to H(z) or to H(Z) by deleting the comment symbol.
% Pr. 12.13
clear all;clf
%a=[1 0 -0.25]; %% denominator of H(z)
a=[1 0 0.25]; %% denominator of H(Z)
figure(1)
for alpha=2:-0.1:1;
alpha
K=1.25/(1+alphaˆ2)
%b=[K 0 -(alphaˆ2)*K]; %% numerator of H(z)
b=[K 0 (alphaˆ2)*K]; %% numerator of H(Z)
[H,w]=freqz(b,a);
subplot(211)
zplane(b,a); axis([-2 2 -2.5 2.5])
hold on
subplot(212)
plot(w/pi,abs(H));axis([0 1 0 1.5])
hold on
pause(1)
end
hold off

Imaginary Part
1
−1
−2
−2 0 2
Real Part
1.5
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
2
Imaginary Part
1
−1
−2
−2 0 2
Real Part
1.5
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 12.9: Problem 13: Loci of poles and zeros and magnitude responses of H(z) (top) and of H(Z)
(bottom).

12.14 Modulation property transformation for IIR filters — The modulation-based frequency trans-
formation of the DTFT is applicable to IIR filters. It is obvious in the case of FIR filters, but
requires a few more steps in the case of IIR filters. In fact, if we have that the transfer function
of the prototype IIR low-pass filter is H(z) = B(z)/A(z), with impulse response h[n], let the
transformed filter be Ĥ(z) = Z(2h[n] cos(ωo n) for some frequency ω0
(a) Find the transfer function Ĥ(z) in terms of H(z)

(b) Consider an IIR low-pass filter H(z) = 1/(1 − 0.5z −1 ). If ω0 = π/2 determine Ĥ(z).
(c) How would you obtain a high-pass filter from H(z) given in the previous item? Use
MATLAB to plot the resulting filters here and in the past item.
Answers: (b) Ĥ(z) = 2/(1 + 0.25z −2 ).
Solution: (a) Modulating the impulse response h[n] of an IIR filter gives a new filter with
transfer function
∞
X
Ĥ(z) = 2h[n] cos(ω0 n)z −n
n=0
X∞ ∞
X
= h[n](ejω0 z −1 )n + h[n](e−jω0 z −1 )n
n=0 n=0
= H(e−jω0 z) + H(ejω0 z)
so that if H(z) = B(z)/A(z) then
B(ejω0 z) B(e−jω0 z)
Ĥ(z) = + .
A(ejω0 z) A(e−jω0 z)
(b) For the given transfer function

1
H(z) =
1 − az −1
for 0 < a < 1. The transformed filter is
Ĥ(z) = H(e−jω0 z) + H(ejω0 z)

1 1
= +
1 − aejωo z −1 1 − ae−jωo z −1
2 − 2a cos(ω0 )z −1
=
1 − 2a cos(ω0 )z −1 + a2 z −2
If a = 0.5 and ω0 = π/2, i.e., we want to convert the prototype low-pass filter into a band-pass
filter, we get
2 2z 2
Ĥ(z) = −2
= 2
1 + 0.25z z + 0.25
with two zeros at z = 0 and two poles at z = ±j0.5, so that the filter is a bandpass filter with
center frequency π/2.
(c) To get a high–pass filter we let ω0 = π. The following script computes and plots the
different frequency responses.

% Pr. 12.14
clear all; clf
b=1;a=[1 -0.5];
[H,w]=freqz(b,a);
alpha=0.5; w0=pi/2
b1=[2 -2*alpha*cos(w0)];a1=[1 -2*alpha*cos(w0) alphaˆ2];
w0=pi
b2=[2 -2*alpha*cos(w0)];a2=[1 -2*alpha*cos(w0) alphaˆ2];
figure(1)
lot(311)
plot(w/pi,abs(H));axis([0 1 0 2]);grid
subplot(312)
plot(w/pi,abs(H1));axis([0 1 0 3]);grid
subplot(313)
plot(w/pi,abs(H2));axis([0 1 0 4]);grid
2
1.5
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
4
3
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 12.10: Problem 14: Magnitude responses of low–pass prototype (top) and band–pass and high–pass
filters (middle and bottom).

12.15 Effect of size of filter in denoising —The size of an FIR averaging filter has significant effect
on denoising. Consider as the clean image the peppers.png, use the function rgb2gray and double
to get a double precision gray-level image x[m, n]. Suppose it is distorted by additive Gaussian
noise with zero mean and 0.1 variance (use the function imnoise). Consider 3 × 3 and 9 × 9
low-pass averaging filters (normalized so that the maximum value of the impulse response
is unity) to filter the noisy image. Call y3 [m, n] and y5 [m, n] of the 3 × 3 and 9 × 9 filters.
To determine which of the two filters is more efficient in denoising the image, determine the
mean-square errors X
i = (yi [m, n] − x[m, n])2 /Mi ,
m,n
2 2
for i = 3, 9, M3 = 3 and M9 = 9 . Indicate which of the two filters, under this criterion, is
more effective in denoising the image.
Solution
% Pr. 12.15
% Effect of size of filter in denoising
% 3x3 vs 9x9 averaging filtering of
% signal with added Gaussian noise
clc; clear all; clf;

% clean and nosiy images
I=imread(’peppers.png’);
I=im2double(I);
I2=rgb2gray(I);
figure(1)
subplot(221)
imshow(I2);title(’CLEAN’)
I3=imnoise(I2,’gaussian’,0, 0.1);
subplot(222)
imshow(I3);title(’NOISY’)
% filtering
h3=1/9*ones(3,3);
h9=1/81*ones(9,9);
y3=filter2(h3,I3);
y9=filter2(h9,I3);
subplot(223)
imshow(y3); title(’3x3 FILTERING’)
subplot(224)
imshow(y9);title(’9x9 FILTERING’)
% errors
M=size(I2);M1=M(1);
e3=sum(sum(y3-I2).ˆ2);e3=e3/M1ˆ2
e9=sum(sum(y9-I2).ˆ2);e9=e9/M1ˆ2
Figure 15 shows the results of filtering with a 3 × 3 and a 9 × 9 averaging FIR filters. The mean-
square errors are ε3 = 0.4890 and ε9 = 0.4337 a slightly better result for the 9 × 9 averaging
filter.

CLEAN NOISY
3x3 FILTERING 9x9 FILTERING
Figure 12.11: Problem 15: Effect of size of filter in denoising.

12.16 Laplacian filtering Consider the mask of a Laplacian filter:

 
0 1 0
HL =  1 −4 1 
0 1 0
Generate an image of size 100 × 100 that is a block of ones of size 50 × 50 surrounded by zeros,
call it x[m, n].
(a) Filter x[m, n] and display it and the Laplacian filtered image y[m, n]. Use then Sobel’s
horizontal and vertical filters to filter x[m, n] and obtain a second image z[m, n] by adding
their outputs. Which of these two provide a better edge detection?
(b) Find the magnitude response of the Laplacian filter associated with HL .
(c) Apply the Laplacian filter mask HL to the peppers image and show the original and
filtered images.
Solution
% Pr. 12.16
% Laplacian Filtering
%
clc; clear all; close all;
x=zeros(100,100);
x(26:75,26:75)=ones(50,50);
figure(1)
subplot(221)
imagesc(x); title(’Original Image’)
% Laplacian
HL=[0 1 0; 1 -4 1; 0 1 0];
% Sobel
hx=[1 0 -1;2 0 -2;1 0 -1];
hy=hx’;
yx=imfilter(x,hx);
yy=imfilter(x,hy);
yS=sqrt(yx.ˆ2 + yy.ˆ2);
subplot(222)
imshow(yS);title(’Sobel Filtered Image’)
yHL=imfilter(x,HL);
subplot(223)
imshow(yHL);title(’Laplacian Filtered Image’)
HLf=fftshift(abs(fft2(HL,256,256)));
w=-1+1/128:1/128:1;
subplot(224)
mesh(w,w,HLf); title(’Frequency response of the Laplacian Filter’)
xlabel(’\omega_1/\pi’)
ylabel(’\omega_2/\pi’)
I=imread(’peppers.png’);
I2=im2double(I);

I3=rgb2gray(I2);
figure(2)
subplot(211)
imshow(I3);title(’Peppers Image’)
yL=imfilter(I3,HL);
subplot(212)
yL=yL>0.05;
imshow(yL);
title(’Laplacian Filtered Peppers Image’)
Peppers Image
Laplacian Filtered Peppers Image
Figure 12.12: Problem 16: Laplacian vs Sobel filtering (left), and edge detection using Laplacian
(right).

12.17 Circular filter Design a 2D FIR filter, using the function fwind1, that approximates the desired
magnitude response

jω1 jω2 1 r < 0.5
Hd (e , e ) =
0 otherwise
p
where r = f12 + f22 < 0.5 for normalized frequencies −1 ≤ f1 ≤ 1 and −1 ≤ f2 ≤ 1 and
fi = ωi /π, i = 1, 2. Use a Hamming window. Determine the magnitude response of the filter,
and plot the desired frequency magnitude response and that of the designed filter using the
function contour. Normalized the frequencies to be (−1, 1) × (−1, 1).
Solution
% Pr. 12.17
% Circular filter design using fwind1
clear all; clf
% filter specifications
N=125;
[f1,f2] = freqspace(N,’meshgrid’);
Hd = ones(N);
r = sqrt(f1.ˆ2 + f2.ˆ2);
% type of filter
Hd(r>0.5)=0; %low-pass
% Hamming window
h1 = fwind1(Hd,hamming(N));
H=freqz2(h1,124); w1=-1:2/124:1-2/124;w2=w1
figure(1)
subplot(221)
mesh(f1,f2,Hd); title(’Magnitude response of desired filter’)
subplot(222)
contour(f1,f2,Hd);grid
subplot(223)
mesh(f1,f2,Hd); title(’Magnitude response of designed filter’)
subplot(224)
contour(w1,w2,H); grid;
The results are shown in Fig. 17.

Figure 12.13: Problem 17:Magnitude responses of desired (top) and of designed filters (bottom).
Circular plots are shown of each of these.

Solution Manual For Additional Problems For Signals and Systems Using Matlab Luis F. Chaparro and Aydin Akan

Uploaded by

Copyright:

Available Formats

Solution Manual For Additional Problems For Signals and Systems Using Matlab Luis F. Chaparro and Aydin Akan

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solution Manual For Additional Problems For Signals and Systems Using Matlab Luis F. Chaparro and Aydin Akan

Uploaded by

Copyright:

Available Formats

Solution Manual for Additional

Luis F. Chaparro and Aydin Akan

Copyright 2018, Elsevier, Inc. All rights reserved.

From the Ground Up

0.1 Basic Problems

(i) Re(z) = 0.5(z + z ∗ )? (ii) Im(v) = −0.5j(v − v ∗ )?

(i) |z + v| ≤ |z| + |v|? (ii) |z − v| ≤ |z| + |v|?

Figure 1: Problem 2: Addition of two vectors illustrating the triangular inequality.

(c) The answer to both is yes. Indeed,

Copyright 2018, Elsevier, Inc. All rights reserved.

0.3 Use Euler’s identity in the following problems.

(i) cos(α + β) (ii) sin(α + β)

(b) Is it true that Z 1

(c) Is it true that

cos(α + β) = cos(α) cos(β) − sin(α) sin(β)

Copyright 2018, Elsevier, Inc. All rights reserved.

(i) z 3 = −1, (ii) z 2 = 1 (iii) z 2 + 3z + 1 = 0

log(z) = log(|z|ejθ ) = log(|z|) + log(ejθ ) = log(|z|) + jθ

(i) log(−2), (ii) log(1 + j1), (iii) log(2ejπ/4 )

(c) Let z = −1 find (i) log(z), (ii) elog(z) .

log(−2) = log(2e±jπ ) = log(2) ± jπ

(c) z = −1 = 1ejπ thus

Copyright 2018, Elsevier, Inc. All rights reserved.

0.2 Problems using MATLAB

(i) Ts = 0.1, (ii) Ts = 0.5, (iii) Ts = 1

Copyright 2018, Elsevier, Inc. All rights reserved.

Copyright 2018, Elsevier, Inc. All rights reserved.

Figure 3: Problem 6: RL circuit: input is (t), output iL (t).

Solution (a) According to Kirchoff’s current law

Copyright 2018, Elsevier, Inc. All rights reserved.

output current 0 10 20 30 40 50 60 70 80 90 100

% solution of difference equation

Copyright 2018, Elsevier, Inc. All rights reserved.

Use symbolic MATLAB function symsum to verify your results.

Copyright 2018, Elsevier, Inc. All rights reserved.

dx(t) dRe[x(t)] dIm[x(t)]

Copyright 2018, Elsevier, Inc. All rights reserved.

(d) The integral

Real part of x(t) COMPLEX Signal x(t)?

Copyright 2018, Elsevier, Inc. All rights reserved.

Copyright 2018, Elsevier, Inc. All rights reserved.

1/2 exp(θ)+1/2 exp(−θ)

Figure 6: Problem 9: cosh(θ) (top) and sinh(θ) (bottom).

Copyright 2018, Elsevier, Inc. All rights reserved.

Copyright 2018, Elsevier, Inc. All rights reserved.

1.1 Basic Problems

1.1 Consider a finite support signal x(t) = t, 0 ≤ t ≤ 1, and zero elsewhere.

y(t) = x(t + 1) = t + 1 0 ≤ t + 1 < 1 or − 1 ≤ t < 0

identical as the discontinuity of y(t) does not add any area.

Copyright 2018, Elsevier, Inc. All rights reserved.

1.2 Given the causal full-wave rectified signal x(t) = | sin(2πt)|u(t)

Figure 1.3: Problem 2.

Copyright 2018, Elsevier, Inc. All rights reserved.

1.3 Is it true that (if not true, give correct answer)

(a) for any positive integer k

(b) for a periodic signal x(t) of fundamental fundamental period T0

for any value of t0 ? Consider, for instance, x(t) = cos(2πt).

(f) If x(t) = cosh(t)u(t) = 0.5(et + e−t )u(t) is dx(t)/dt = sinh(t)u(t) + δ(t) ?

dx(t) d cos(t) du(t)