0% found this document useful (0 votes)

1K views

Signals and Systems Lecture Notes

Our goal is not to develop all the applications, but to prepare for them - and that preparation can only be prepared by understanding the theory'

Uploaded by

Drew Swindle

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

1K views

Signals and Systems Lecture Notes

Our goal is not to develop all the applications, but to prepare for them - and that preparation can only be prepared by understanding the theory'

Uploaded by

Drew Swindle

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 183

ECE 370 SIGNALS AND SYSTEMS - FALL 2012

Text: These lecture notes and board notes.

Optional Texts: H.P. Hsu, Signals & Systems. Shaum's Outline: McGraw{Hill, 1995.
This is essentially a good source of worked examples, all in the standard EE notation that
we use in ECE 370. Reasonably priced!
C.L. Phillips, J.M. Parr, and E.A. Riskin, Signals, Systems, & Transforms. 4th Edition,
Prentice-Hall, 2008. Only get this if you are inclined to read additional books!
Prerequisites: ECE 225 Electric Circuits, MA 238 Dierential Equations.
Course Goals: To provide electrical engineers with physical understanding and competence in mathematical signal analysis and systems theory.
Instructor: Robert W. Scharstein, Houser 209
phone 205-348-1761, e-mail [email protected]
Oce hours: to be announced and by appointment.
Lecture: 12:00 noon { 12:50 pm, Mon, Wed, Fri; SERC 1013.
Problem Session: Wednesday 7:00-8:00 pm, Houser 301, optional. Bring your questions!
Daily Homework: Homework problems and textbook reading will be assigned for each
class.
FFF !!!THESE ARE YOUR PRIMARY RESPONSIBILITY!!! FFF
Tests will consist of homework problems, modied homework problems, new problems,
and concepts and derivations from the text and lecture. Details and discussion of the
MATLAB homework problems will appear on tests.
Tentative Grades: All quizzes and tests are closed-book.
Eleven Friday quizzes : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1/3
Test 1, Friday September 21 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1/6
Test 2, Friday November 2 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1/6
Final Exam, Monday December 10 11:30 am { 2:00 pm : : : : : : : : : : : : : : : : : : : : : : : : : : 1/3
Instructor may determine nal course grades using any combination of the above. However,
the nal exam will not be weighted less than 1/3.
Friday quiz dates: 1: Aug 24, 2: Aug 31, 3: Sept 7, 4: Sept 14, 5: Sept 28, 6: Oct 12
7: Oct 19, 8: Oct 26, 9: Nov 9, 10: Nov 16 11: Nov 30

\Our goal is not to develop all the applications, but to prepare for them
- and that preparation can only come by understanding the theory."
Gilbert Strang Linear Algebra and Its Applications, Academic Press, 1976 page x

Typeset by AMS-TEX
0

ECE 370 SIGNALS AND SYSTEMS - ROUGH LECTURE NOTES

Robert W. Scharstein
23 July 2012
CONTENTS
BACKGROUND AND TIME-DOMAIN SYSTEMS
Complex Numbers : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1.
A Couple of Singularity Functions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7.
Symmetry : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10.
A Note on Mathematical Functions and Units/Physical Dimensions : : : : : : : : : : : : : : : : : 13.
Linear System : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 15.
Time-Invariant System : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 16.
Input/Output Relationship for a LTI System : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :17.
Convolution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 18.
Another Convolution Example: The Convolution of Two Gaussian Pulses : : : : : : : : : : : 21.
Yet Another Convolution Example : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 22.
Causal System : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 28.
Bounded Input/Bounded Output Stability : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 28.
Singular Functions and Integrals : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 29.
Homework Problems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :35.
Relationship Between Impulse Response and Step Response : : : : : : : : : : : : : : : : : : : : : : : : : 37.
Time-Domain RC Circuit Analysis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 38.
Step Response of an RC Circuit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 41.
Step Response of Another RC Circuit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 43.
Homework Problems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :45.
Step and Impulse Response of Series RLC Circuit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 46.
Homework : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :53.
FOURIER INTEGRAL
Evaluation of an Important Integral via the Laplace Transform : : : : : : : : : : : : : : : : : : : : : : 54.
Spectral Form of the Dirac-Delta Function : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 57.
Fourier Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 58.
Fourier Transform Exercises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 59.
Tables of Integral Transforms : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 60.
Fourier Transform of the Gaussian : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 63.
Fourier Transform of the Heaviside Unit Step Function : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 65.
Duality Property of the Fourier Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 68.
Half-Power Bandwidth or Pulse width : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 70.
Homework : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :73.
Uncertainty Principle for the Fourier Integral : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 74.

*Riemann-Lebesgue Lemma (two serious treatments) : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 76.

Riemann-Lebesgue Lemma (one reasonable treatment) : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 78.
Convergence of the Fourier Integral Representation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :79.
System Transfer Function : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :81.
Superposition Interpretation of the System Transfer Function : : : : : : : : : : : : : : : : : : : : : : : 83.
Transfer Function for Linear Electric Circuits : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 84.
Parseval's Theorem for the Fourier Integral : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 87.
Other Conventions for the Fourier Transform Pair : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :88.
Cascade of Two LTI Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 90.
Linear Dispersionless Filter : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 90.
Rectangular Pulse Response of Ideal Low Pass Filter : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 91.
Ideal Band-Pass Filter : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 95.
Complex Phasors for Time-Harmonic Circuit Analysis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :96.
Homework Problem on Uncertainty Principle : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 98.
*Approximate Analysis of Dispersion : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 99.
Fourier Integral Solution of Potential Problem : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 103.
FOURIER SERIES
Fourier Transform of the comb : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :106.
Fourier Transform of a Periodic Signal : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 107.
Periodic Signals and Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :110.
Orthogonality : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 110.
Kronecker delta : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 111.
Orthogonality of the Fourier basis : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 111.
Derivation of Fourier Coecients by Minimizing the Mean Square Error : : : : : : : : : : : : 112.
Fourier Coecients by a Direct Inner Product : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 114.
Trigonometric Form of the Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 114.
Two Standard Notations for the Trig Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 116.
Riemann-Lebesgue Lemma: Fourier Coecients : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 117.
Gibbs' Phenomenon: Example Square Wave : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 118.
Pointwise Convergence of the Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 119.
On the Convergence and Summation of Some Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :126.
Kummer Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 129.
Homework Problem: Convergence of Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 131.
Fourier Series of Full-Wave Rectied Sine Wave : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 132.
Parseval's Theorem for the Fourier Series : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 134.
Power Supply Performance : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 135.
Periodic Excitation of a Simple Parallel GLC Circuit : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 139.
Homework Problems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 140.
Simple Low-Pass and Hi-Pass Filter Response to a Periodic Input : : : : : : : : : : : : : : : : : : 141.
The Vibrating String : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 147.
Homework Problems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 150.
Poisson Sum Formula : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 151.

TRANSITION TO DISCRETE-TIME SYSTEMS

Sampling and Reconstruction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 153.
Discrete-Time Signals, Systems, and Transforms : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 157.
Exercises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 159.
Discrete-Time Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 159.
Linear Time-Invariant Discrete-Time Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 160.
Input/Output Relationship : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 160.
Example of a LTI Discrete-Time System : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 161.
The Z-Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 163.
Fibonacci Sequence - Dierence Equation and Z-Transform : : : : : : : : : : : : : : : : : : : : : : : : 165.
Convolution Theorem of the Z-Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 166.
Z-Transform of a Delayed Sequence : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 166.
System Transfer Function : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 166.
Discrete-Time Fourier Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 167.
Discrete Fourier Transform : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 167.
Exercises : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 169.
DFT Approximation of the Continuous-Time Fourier Transform : : : : : : : : : : : : : : : : : : : :170.
Numerical Example of Using the FFT to Evaluate the Fourier Integral : : : : : : : : : : : : : 171.
Example: DFT Approximation of a Fourier Sine Integral : : : : : : : : : : : : : : : : : : : : : : : : : : : 174.
Discrete-Time Approximation for the Step Response of Series RLC Circuit : : : : : : : : : 177.

* Sections marked by the asterisk are A+ level, so put these last in your prioritized list of
study topics.

\... mathematics is learned by doing it, not by watching other people do it ... "
M. Reed and B. Simon, Functional Analysis, Academic Press, 1980, page ix.

iii

Complex Numbers
p
j = + 1
y
...
..
..
..
..
...
...
...
...
.
..
.
.
..
...
...
....
..
..................
.
...
.
.....
.
..
.....
...
...
...
...
.
...
...
.
..
.
.
.
..
.
...
... .....
...
.. ..
...
... ....
...
... ..
...
... ...
.
.
.
.
.........................................................................................................................................................................................
...
...
..
...
...
..
..
...
..
...
..
...
...
..
..
...
.

complex z = x + jy plane

If z 2 C and x; y; r; 2 R, then we write the complex number z in rectangular and polar

form as
z = x + jy = rej :
Refzg = x

and

Imfzg = y

The modulus or magnitude of z is

jzj = r
and the argument or angle of z is
argfzg = :
The complex conjugate of z = x + jy = rej is
z = x jy = rej :
z + z = 2x

and

z z = 2jy

zz = jzj2 = r2
Euler's identity:

ej = cos + j sin
Taylor series for the exponential function, written as a function of the complex variable z
is
1
X
zn
z
e =
:
n!
n=0
1

Let z = jy and separate the real and imaginary parts via the even and odd powers in
1
1
1
X
X
X
(jy)n
(jy)2m
(jy)2m+1
=
+
ejy =
n!
(2m)!
(2m + 1)!
n=0
m=0
m=0
Note that j 2m = (1)m and j 2m+1 = j(1)m so that
1
1
X
X
(1)m y 2m
(1)m y 2m+1
jy
+j
= cos y + j sin y
e =
(2m)!
(2m
+
1)!
m=0
m=0
There was nothing special about the real number y, and we also write
ej = cos + j sin :
The sum of ej and its conjugate gives
ej + ej = 2 cos
and the dierence gives
ej ej = 2j sin

which are the important representations

ej + ej
2
j
e ej
sin =
2j
Many (all?) of the trig identities are easily seen in complex form: Consider

j 2
j
j
1 = je j = e
e
= cos + j sin cos j sin = cos2 + sin2 :
cos =

Now observe

z = x + jy = rej = r cos + j sin

Equating real and imaginary parts separately gives

x = r cos
2

and

y = r sin :

jzj = zz = r = (x + jy)(x jy) = x2 + y 2

p
r = + x2 + y 2
y
= tan1
x
Note that the two-argument arctangent function (that maintains full information about
quadrant) is required. For example, observe that

p j3=4
1
3
1
z1 = 1 + j = 2e
where tan
=
1
4
is certainly dierent from

p j=4
1

1
z2 = 1 j = 2e
where tan
=
1
4
Your calculator has the two-argument arctangent function as part of its rectangular-topolar conversion. In MATLAB, it's called atan2(y,x) just like Fortran and probably
C .
2

Consider z1 = x1 + jy1 and z2 = x2 + jy2 . Then our four basic operations are:
addition

z1 + z2 = x1 + jy1 + x2 + jy2 = (x1 + x2 ) + j(y1 + y2 )

multiplication by a real constant c:

cz1 = c(x1 + jy1 ) = cx1 + jcy1

multiplication z1 z2 = (x1 + jy1 )(x2 + jy2 ) = (x1 x2 y1 y2 ) + j(x1 y2 + x2 y1 )

and
division

z1 z2 = (r1 ej1 )(r2 ej2 ) = r1 r2 ej(1 +2 )

(x1 x2 + y1 y2 ) + j(x2 y1 x1 y2 )
z1
x1 + jy1
(x1 + jy1 )(x2 jy2 )
=
=
=
z2
x2 + jy2
(x2 + jy2 )(x2 jy2 )
x22 + y22
and

z1
r1 ej1
r1
=
= ej(1 2 )
j
2
z2
r2 e
r2

Generally, since the real number line is a subset of the complex plane, all of our ordinary
functions f (x) of a real variable x have extensions (or continuations) to a function f (z) of
a complex variable z. One notable dierence between the real number line and the more
general complex plane is the lack of ordering on z: That is, there is no inequality on the
complex plane. Inequality is only dened on the reals, i.e.
x1 > x2 ;

y1 y2 ;

7 < 13;

but something like z1 < z2 is meaningless.

Triangle inequalities.
jz1 + z2 j jz1 j + jz2 j

jz1 z2 j jz1 j jz2 j

etc

Proof of the rst triangle inequality:

jzj2 = x2 + y 2
jzj =

p
x2 + y 2 jxj

jzj jRe(z)j
jzj Re(z)

jz1 z2 j Re(z1 z2 )

(1)

jz1 + z2 j2 = (z1 + z2 )(z1 + z2 ) = z1 z1 + z2 z2 + z1 z2 + z1 z2

= jz1 j2 + jz2 j2 + 2Re(z1 z2 )

(jz1 j + jz2 j)2 = jz1 j2 + jz2 j2 + 2jz1 jjz2 j = jz1 j2 + jz2 j2 + 2jz1 z2 j
compare (2) and (3) in view of (1)
jz1 + z2 j2 (jz1 j + jz2 j)2

...........
.......... .....
........... .........
..........
.
.
.
.
.
.
.
.
.
.
.
.
.
1
2.....................
.....
.....
.
..........
.....
....
...........
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.....
..........
.....
..........
2
.....
...........
..........
.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
..
......................................................................................................................

z +z

jz1 + z2 j jz1 j + jz2 j

z1
Extensions
N

N
X
X

f
(z)

jfn (z)j

n=0

Z
Zb

f (t) dt jf (t)j dt

(Sometimes called the monotonicity property of the integral.)

F Prove the second triangle inequality.
Hint: let z1 = 1 2 and z2 = 2 in the rst triangle inequality.

(2)

(3)

Multivalued functions.
The nth roots of unity
1 = ej2k
11=n = ej2k=n

(k = 0; 1; 2; : : : )

(k = 0; 1; 2; : : : ; n 1) distinct roots
...
..
..
..
.........................
.
.
.
.
.
.
.
.
.
.
.......
...
.
.
.
.
....
.
.....
....
....
..
...
...
..
...
...
.
.
.
.
.
...
.
.
.
.
...
...
....
..
..
...
...................................................................................................................
...
.
..
.
.
.
...
...
....
...
.
.
.
.
...
..
....
....
...
.....
..
......
.....
......... ... ..............
......................
..
..
..
...
.

Examples.
1=2

11=3

ej0 = 1
(k = 0)
ej = 1 (k = 1)

8 j0
e =1
>
>
p
>
>
< j2=3 1 + j 3
=
= e
2 p
>
>
>
>
1

j 3
: ej4=3 =
2

(k = 0)

(k = 1)

(k = 2)

logarithm

ln(z) = ln rej = ln rej(+2k)

= ln(r) + j( + 2k)

(k = 0; 1; 2; : : : )

The value for k = 0 is called the principal branch of the logarithm.

powers: Let z = x + jy and w = u + jv be two complex numbers, that is z; w 2 C and

x; y; u; v 2 R.

z w = exp ln(z w ) = exp w ln z

and we know how to interpret/evaluate

ez = ex+jy = ex cos y + j sin y

Example.
ln(j) = j(=2 + 2k)
j j = ejj(=2+2k) = e(=2+2k)
principal value is j j = exp(=2) = 0:2079

Problem: Show that Hint: use the nite geometric series

n1
P

on page 164

k=0

n1
X

ej2k=n = 0

k=0

Fundamental theorem of algebra. The polynomial

Pn (z) = an z n + an1 z n1 + + a2 z 2 + a1 z + a0
with an 6
= 0 is said to be of degree n. It has exactly n roots (zeros) counting multiplicity.
Proposition: If the coecients ak (k = 0; 1; 2; : : : ; n) of the polynomial Pn (z) are all real,
then any complex roots must occur in complex conjugate pairs. That is, if Pn (z) = 0 then
Pn (z ) = 0.
proof:
an z n + an1 z n1 + + a2 z 2 + a1 z + a0 = 0
take the complex conjugate
an (z )n + an1 (z )n1 + + a2 (z )2 + a1 z + a0 = 0
6

A Couple of Singularity Functions

(or \Distributions" or \Generalized Functions")
u(t)

(t)

.................................................................
...
..
...
..
............................................................................................

......
.........
...........................................................................

0
Heaviside unit-step function

0; t < 0
u(t) =
1; t > 0

(1)
......
N
.........
...
0

u(t T ) =

0; t < T
:
1; t > T

Dirac delta-function (denition)

(t) = 0

t6
=0

AND

(t) dt = 1

From this strange denition, the sampling or sifting property follows:

Zb
a

f (t)(t t0 ) dt =

f (t0 ); a < t0 < b

0;
o.w.

if f (t) is continuous at t = t0 . This is eectively an operational denition of the Dirac-delta.

It is under the operation of integration that the Dirac delta-function enjoys meaning, and
thus it is sometimes said that \every Dirac-delta function is begging to be integrated."
Relationship:
Zt

( ) d = u(t)

()

(t) =

du(t)
dt

Dirac-Delta Sequences: There are lots (1!) of ordinary functions that satisfy, in some
limit, the two-statement denition of the Dirac-delta function. Here are several:

t
1
lim
= (t)
!0

=
lim 2
= (t)
!0 t + 2
r
t2
lim
e
= (t)
!1

sin t
1
= lim
lim
!1 t
!1 2

1 jtj < T =2
t
where
, u(t + T =2) u(t T =2) =
T
0 jtj > T =2
Cauchy or Lorenz distribution
Gaussian distribution
Z

ej!t d! = (t)

Dirichlet kernel

Study the time behavior of the above \Dirac-delta sequences" to see how they approach
the ideal delta function. Also examine the Heaviside unit-step sequence
lim 1 [1
!1 2

+ tanh t] = u(t)

and its time-derivative.

Note: we don't ever try to evaluate (0). You might want to, and some books might say
that \(0) = 1 ", but I would prefer we avoid this point. We don't ever have to attach a
value to the Dirac delta function when its argument is zero. But, in the words of Lennon
& McCartney, it's \nothin' to get hung about" from Strawberry Fields Forever.
Another note: (along the same line) If a Dirac-delta function appears in the integrand
of some integral, the integral is evaluated from the sampling property.
Warning: the integration limits cannot \split" a Dirac-delta function \down the middle,"
for example
Z1
(t) dt
0

should be either

(t) dt = 1

(t) dt = 0:

You either capture all of the delta-function or you don't get any of it. It cannot be divided.
Or you will have to agree on some scheme to cut it up while minimizing bloodshed.
Proof of the SAMPLING PROPERTY: No harm is done in taking t0 = 0, so that
we want to show

Zb
f (0); a < 0 < b
f (t)(t) dt =
0;
o.w.
a

for f (t) continuous in a neighborhood of t = 0. For deniteness, use the limit of the
rectangular pulse to represent

t
1
(t) = lim
!0

and represent the continuous f (t) by its Taylor series about t = 0:

f (t) = f (0) + tf 0 (0) +

t2 00
t3
f (0) + f 000 (0) + O(t4 ):
2!
3!

Then the integral of interest is, for a < 0 < b,

Zb
a

1
f (t)(t) dt = lim
!0

Z=2
t2 00
0
f (0) + tf (0) + f (0) + : : : dt:
2!
=2

Note
Z=2
=2

and so

Zb
a

dt = ;

Z=2

t dt = 0;

Z=2

3
t dt =
12
2

1
3 00
5
f (t)(t) dt = lim
f (0) + f (0) + O( ) = f (0):
!0
24

If, for a < b, either 0 < a or b < 0, then the integral is zero since (t) = 0 for t 6
= 0.

F Provide an alternate proof of the sampling property by appealing to the mean value
theorem of the integral.

\Heavy formalism is not required to get across fundamental ideas. If it can't be said simply,
it's not worth being said at all."
Bernard H. Lavenda, Statistical Physics, 1991, p. viii.

\The pursuit of excellence is gratifying and healthy. The pursuit of perfection is frustrating,
neurotic, and a terrible waste of time."
Edwin Bliss

Symmetry1

symmetric or even fe (x) = fe (x)

asymmetric or odd fo (x) = fo (x)

cos(x)

sin(x)

...
..
...
...
...
...
.
...
.
.
2
...
...
...
...
...
...
.
...
.
...
..
...
...
..
...
..
...
...
..
...
.
...
.
...
...
...
..
...
...
..
...
...
...
...
...
.
....
.
..
....
....
..
....
....
...
.....
....
..
....
....
.
.
.
.
..................................................................................................................................................................................................................................
.....
....
....
.....
.....
.....
..
.....
......
...
.....
.
.......
.
.
.
.
.
...
........
............................................
....
...
..
..
...
..

x 25

1
4
8 (35x

1 \Symmetry

1
3
2 (5x .

30x2 + 3)

3x)

is immunity to a possible change." J. Rosen, Symmetry Rules, Springer, 2008, p. 4

If a function f (x) is continuous to all orders (the function f (x) and all of its nth derivatives
f (1) (x); f (2) (x); : : : are continuous) then f (x) can be represented by a Taylor series about
the origin
1
X
f (n) (0) n
f (x) =
x :
n!
n=0
The Taylor series of an even function will have only even powers of x
1
X
f (2n) (0) 2n
fe (x) =
x
(2n)!
n=0
and therefore all of its odd derivatives (at the origin) vanish

d2n+1 f (x)
(2n+1)
fe
(0) =
= 0:
dx2n+1
x=0

In particular, note that

fe0 (0)

= 0 (if fe (x) is continuous and dierentiable at the origin).

Similarly, the Taylor series of an odd function will have only odd powers of x
1
X
f (2n+1) (0) 2n+1
x
fo (x) =
(2n
+
1)!
n=0
and therefore all of its even derivatives (at the origin) vanish

d2n f (x)
(2n)
fo (0) =
= 0:
dx2n
x=0

In particular, note that fo (0) = 0 (if fo (x) is continuous at the origin).

Even-Order Derivatives Preserve Symmetry and Odd-Order Derivatives Reverse Symmetry

f (t + t=2) f (t t=2)
t!0
t
f
(t
+
t=2)
f (t t=2)
f 0 (t) = lim
t!0
t
Even function fe (t) = fe (t)
f 0 (t) = lim

fe (t + t=2) fe (t t=2)
t!0
t

fe0 (t) = lim

fe (t + t=2) fe (t t=2)
fe (t t=2) fe (t + t=2)
= lim
t!0
t!0
t
t
fe (t + t=2) fe (t t=2)
d
= lim
= fe0 (t) =) fe (t) is odd.
t!0
t
dt

fe0 (t) = lim

Similarly,
d
fo (t) is even where fo (t) = fo (t) is an odd function.
dt
This is also seen by appealing to the Taylor series for even and odd functions.
11

Consider the integral of f (t) between limits that are symmetrically disposed about the
origin, that is look at
ZT
f (t) dt
T

where T is some xed, positive constant.

fo (t)
.

fo (t) dt = 0

fe (t)
.

fe (t) dt = 2

fe (t) dt

Any function f (t) (neither symmetric nor antisymmetric) can always be decomposed into
a linear combination of a symmetric plus an antisymmetric part since
f (t) =
Example.

f (t) + f (t) f (t) f (t)

+
= fe (t) + fo (t):
2
2

t = +

F Homework. Decompose the Heaviside unit step function u(t) into the sum of a
symmetric plus antisymmetric signal.
12

A Note on Mathematical Functions and Units/Physical Dimensions

Consider almost any \ordinary" or \regular" function2 of an independent variable x, such
as
f (x) = cos(x) or sin2 (x) or 3x2 7x or ex :

The argument or independent variable x is always dimensionless. To see this, examine the
Taylor series expansion, such as
f (x) = cos(x) = 1

x2
x4
+
::::
2!
4!

If x has units such as meters (m), then the sum would be nonsense since the rst term is
dimensionless, the second term has units (m)2 , the third term has units (m)4 , etc. Also
note that the function f (x) itself is dimensionless, too.
This may seem strange and somewhat contrary to your previous experience. For example,
you have dealt with a time-domain voltage signal such as
v(t) = 12 cos(t + 45 ) or 12 cos(t + =4):
If the time variable t is measured in (s), then there really is a transparent (or hidden)
frequency ! = 1 (s1 ) multiplying the t so that
v(t) = V0 cos(!t + ):
Now the argument !t + of the cosine is clearly dimensionless. The constant V0 = 12 (V)
carries the dimensions of the voltage signal v(t). Also note that we often say ! has units
of (rad/s), but a radian is not dimensioned: Its use in (rad/s) is simply a reminder that
we are using the angular frequency ! and not the common experimentalist's frequency
f = !=2 (cycle/s=Hz).
The operators dierentiation and integration impart units: Consider the timedomain current signal
i(t) = I0 cos(!t):
In our standard mksC (SI) units, the current amplitude I0 is in (A), the frequency ! is in
(s1 ), and the time t is in (s). Dierentiation with respect to the dimensioned variable t
imparts the units of 1=t or (s1 ) since
di(t)
= !I0 sin(!t):
dt
We can also see that the dierential operator has to have units of inverse time by appealing
to the denition of the derivative
df (t)
f (t + t) f (t)
= lim
:
t!0
dt
t
The (s1 ) comes from the denominator t that is in (s).
2 Not

(x) or u(x) or jxj, to name a few exceptions.

Integration with respect to the dimensioned variable t imparts the units of t or (s) since
Zt
t0

i( ) d =

I0 cos(! ) d =

I0
sin(!t) sin(!t0 ) :
!

Note the good practice of using a dummy time variable for the integration, to avoid confusion with the nal time that is the upper integration limit. We can also see that the
integral operator has to have units of time by appealing to the denition of the integral in
terms of a limit on the Riemann sum
Zb

f (t) dt = lim

N !1

N
X

f (tn )tn :

n=1

The prime notation for derivatives. When dealing with a generic function such as
f (t) or g(x), we commonly denote derivatives using the prime
df (t)
= f 0 (t)
dt

and

dg(x)
= g 0 (x):
dx

The prime denotes dierentiation to the entire argument of the function, as in

f 0 () =
Note that

df ()
d

f 0 (~) =

df (~)
:
d~

d
d
df () d
f () =
= f 0 () :
dt
d dt
dt

Linear System
.................................................
...
...
..
...
..
...
...
.
.
...........................
...........................
....
....
..
..
...
..
.
.
................................................

x(t)

y(t)

Assume the zero-state response y(t) of a system to the input x(t) is described by an
operator T , such that
y(t) = T fx(t)g:
Another name for T is system transformation rule. Examples might be
T1 fx(t)g = Ax(t);

T2 fx(t)g = x(t t0 );

d2
d
T4 fx(t)g = c2 2 + c1 + c0 x(t);
dt
dt
T6 fx(t)g = x2 (t);

T3 fx(t)g =

T5 fx(t)g =

d
x(t)
dt

x( ) d

T7 fx(t)g = Ax(t) + B:

Linear System: If the input/output relationship or system operator or system transformation rule satises the superposition principle
T fc1 x1 (t) + c2 x2 (t)g = c1 T fx1 (t)g + c2 T fx2 (t)g;
then the system is said to be linear.
F HW 1

Test the examples T1 ; T2 ; : : : ; T7 above for linearity.

Time-Invariant System:

x(t)

y(t)

x(t t0 )

y(t t0 )

Let the response of a system to the input signal x(t) be the output signal y(t). If the
response to the delayed input x(t t0 ) is the delayed response y(t t0 ), the characteristics
of the system did not change in time, and the system is said to be time-invariant.
Note: In our linear RLC circuit analysis, the equation-of-motion is typically a linear
integrodierential equation, with constant coecients, such as
R

i(t)

vg (t)

di(t)
1
L
+ Ri(t) +
dt
C

vg (t)

............................................
...
...
...
..
...............................
..............................
....
...
..
.
.............................................

i(t)

i( ) d = vg (t)

which is easily converted to an ordinary dierential equation by dierentiating w.r.t. t so

that
di(t)
1
d2 i(t)
+R
+ i(t) = vg0 (t):
L
2
dt
dt
C
Again, note the constant coecients L, R, and 1=C. If the equation-of-motion of a system
is a dierential equation with non-constant (i.e. time varying) coecients, then the system
is a time-variant system.

F HW 2

Give an example of a linear, time-varying system.

Input/Output Relationship for a LTI System

(zero-state or driven response)
x(t)

y(t) = T fx(t)g

This fundamental concept is easy to see and derive, starting with a little \trick" by way
of the notation. First, invoke the sampling property of the Dirac-delta function to write
Z1

x(t) =

x( )(t ) d:

Now exploit the linearity of the operator T and note that T fg operates only on functions
of time t. In particular, T treats x( ) as a constant, and acts only on the time-domain
function (t )
8 1
9
Z1
<Z
=
y(t) = T fx(t)g = T
x( )(t ) d =
x( )T f(t )g d:
:
;
1

The response of our system to the Dirac-delta function or impulse function is called the
system impulse response and it is denoted by the signal h(t). It is a fundamental
characterization of any linear, time-invariant system.
..................................................
..
...
..
...
.....
..
.
.
..............................
............................
....
....
..
..
...
.
....................................................

(t)

h(t)

(t )

h(t )

Since the system is also time-invariant,

T f(t )g = h(t )
and the result above is
y(t) =

Z1
1

x( )h(t ) d:

This integral operator that takes two functions of time x(t) and h(t), and produces a third
function of time y(t), is called the convolution integral; we write
y(t) = x(t) ~ h(t):

Convolution
The convolution operation is not restricted to our above LTI system application. It can
operate on any two (reasonably) arbitrary signals, say f (t) and g(t):
f (t) ~ g(t) =

Z1
1

f ( )g(t ) d:

Example. Evaluate s(t) = f (t) ~ g(t) where f and g are causal, decaying exponential
signals
f (t) = eat u(t)
and
g(t) = ebt u(t)
with 0 < a < b:
case: t < 0

f ( ) = ea u( )

)
g(t ) = eb(t
u(t )
.

f ( )g(t )

s(t) =

Z1
1

f ( )g(t ) d =

Z1
1

0 d = 0

case: t > 0

f ( ) = ea u( )

)
g(t ) = eb(t
u(t )
.

f ( )g(t )

s(t) =

Z1
1

f ( )g(t ) d
0 d +

= ebt

a b(t )

d +

e(ba) d =

eat ebt
ba

0 d

Now combine both expressions (rst one for t < 0 and second one for t > 0) into a single
equation valid for all t
eat ebt
u(t):
s(t) =
ba
s(t)

Let's look for the maximum of s(t) analytically: The derivative

bebt aeat
eat ebt
u(t) +
(t)
ba
ba
bebt aeat
=
u(t)
ba

s0 (t) =

since

eat ebt
= 0:
b a t=0

Recall that (t)(t) = (0)(t). The derivative vanishes at time t = tm , so write s0 (tm ) = 0
so that
aeatm = bebtm
b
e(ba)tm =
a
tm =

ln(b=a)
ba

Another Convolution Example: The Convolution of Two Gaussian Pulses

Evaluate z(t) = f (t) ~ g(t) with f (t) = exp(at2 ) and g(t) = exp(bt2 ).
Both f (t) and g(t) are \on" for all time, so we don't have to worry about sketching them
to see when they turn on or o.

z(t) =

Z1
1

f ( )g(t ) d =
2

= ebt

e[(a+b)

ea eb(t ) d

2bt ]

complete the square

2
p
bt
(bt)2
(a + b) 2bt =
a + b p

a+b
a+b
2

z(t) = exp

"
Z1
2 #
p
bt
b2
2
b t
exp
a + b p
d
a+b
a+b
1

let
x=

p
bt
a + b p
a+b
p
dx = a + b d

= 1

x = 1

recall or use (\and it is a trick!"see page 63)

ex dx =

Z1
2
b2
dx
z(t) = exp
b t2
ex p
a+b
a+b
1
r

b2

=
exp b
t2
a+b
a+b
r

ab 2
=
exp
t
a+b
a+b
The convolution of two Gaussians is a third Gaussian.
21

Yet Another Convolution Example

Evaluate s(t) = f (t) ~ g(t) where f (t) and g(t) are the given rectangular pulses:
f (t) = u(t) u(t a) =
g(t) = u(t) u(t b) =

1; 0 < t < a
0; t < 0 or

t>a

1; 0 < t < b
0; t < 0 or

t>b

Here the pulse width a of f (t) has been arbitrarily selected to be greater than the pulse
width b of g(t).
..
..
..
..
..................................................................................................................................
...
...
..
....
..
...
...
...
..
...
..
...
...
...
..
..
...
..
.
...........................................................................................................................................................................................................................................................................................................................................................................................................................
..
.

f (t)

g(t)

s(t) = f (t) ~ g(t) =

f ( )g(t ) d

Since the arguments of our individual signals are (and a shifted, rotated version of )
in the convolution integral, let's redraw our functions f and g as functions of the dummy
variable .
.
..
..
..
..
...................................................................................................................................
...
...
..
....
..
...
...
...
..
...
..
...
...
...
..
..
...
.
............................................................................................................................................................................................................................................................................................................................................................................................................................
..
.

f ( )

g( )

Let's leave f ( ) alone (as in the version of the convolution integral above), and shift and
rotate g( ) (as it appears in the version of the convolution integral above).
...
...
..
...
......................................................................................
..
..
...
..
...
...
..
..
...
...
..
...
...
...
..
...
..
...
...
..
.
...
...
...
...
..
.
.
.
..
.
.
...........................................................................................................................................................................................................................................................................................................................................................................................................................
...
.

g( t)

t+b

g(t )

0 tb

Now let's vary the parameter t (it survives the integration), compute (graph!) the
product of f ( ) and g(t ), and evaluate the innite range integral from = 1 to
= +1. That is, we must calculate the total area under the curve of f ( )g(t ).
case: t < 0

f ( )

g(t )

t b

f ( )g(t )

s(t) = f (t) ~ g(t) =

Z1
1

f ( )g(t ) d =

Z1
1

0 d = 0

case: 0 < t < b

f ( )

g(t )

t b

f ( )g(t )

s(t) = f (t) ~ g(t) =

Z1
1

f ( )g(t ) d =

d = t

case: b < t < a

f ( )

g(t )

0 t b

f ( )g(t )

0 t b

s(t) = f (t) ~ g(t) =

Z1
1

f ( )g(t ) d =

t a

Zt
tb

d = b

case: a < t < a + b

f ( )

a+b

g(t )

a+b

f ( )g(t )

s(t) = f (t) ~ g(t) =

Z1
1

f ( )g(t ) d =

t
Za

a+b

d = a + b t

case: t > a + b

f ( )

a+b

g(t )

tb a + b

f ( )g(t )

s(t) = f (t) ~ g(t) =

Z1
1

f ( )g(t ) d =

Z1
1

0 d = 0

All together now.

8
0;
>
>
>
>
>
< t;
s(t) = f (t) ~ g(t) = b;
>
>
>
a + b t;
>
>
:
0;

t<0
0<t<b
b<t<a
a<t<a+b
t>a+b

s(t)
....
..
..
.
........
................................................................
.......
..........
...
.........
.......
...
............
.
.
..........
.
.
...
.
.
.
.......
.
..........
.
.
..
.
.
......
........
.
..
.
.
.
..........
....
.
.
.
...
.
.
..........
......
.
.
.
..
.
.
.
..........
....
.
.
.
.
.
.
........
...
.
.....
.
.
........
...
.
.
.
.
....
.
..........
.
.
.
..
.
......
.........
.
.
..
.
.
.
.
.........
.....
.
.
.
.
.
...
.
.
.........
.....
.
.
..
.
.
.
.
...........
.......
.
.
.........
.
...
.
.
.....
..........
.
...
.
.
.
..........
.. ................
.........
.. ........
.........
..............
...
.
.
.
.
.
.
.................................................................................................................................................................................................................................................................................................................................................................................................................................
..
..

a+b

Homework Problems
1. In the above example on page 18, nd r(t) = f (t) ~ f (t), that is let b = a. Show that
your result can also be obtained from the limit of the above as b ! a
r(t) = lim s(t):
b!a

Use l'Hopital's rule to disarm the 0=0 form.

2. Show that f (t) ~ (t t0 ) = f (t t0 ).
3. Show that

f (t) ~ u(t) =

f ( ) d:

4. Evaluate the convolution of the standard rectangular pulse with itself (t=T ) ~ (t=T ).
Express the result in terms of the standard triangular pulse (t=T ).
1

(t=T )

T =2

(t=T )

1; (jtj < T =2)

0; (jtj > T =2)

T =2

(t=T ) =

(t=T ) =
t

5. Find an expression for, and graph, each of the derivatives

d
(t=T )
dt

and

d
(t=T ):
dt

1 jtj=T; (jtj < T )

0;
(jtj > T )

Causal system: If the input x(t) = 0 for t < t0 , then the zero-state response y(t) = 0
for t < t0 . The impulse response of a causal system is therefore a causal signal.
If f (t) and g(t) are both causal signals, then so is their convolution f (t) ~ g(t).

Bounded input/bounded output stability: If the input signal is bounded, such that
jx(t)j A < 1 for all time t, then the zero-state output y(t) is also bounded if the impulse
response is absolutely integrable:

Z
Z1
Z1

h( )x(t ) d =
h( )x(t ) d
h( )x(t ) d
jy(t)j =

Z1
1

h(t) dt < 1:

h( ) d < 1

Singular Functions and Integrals

Consider a real-valued function f of a real variable x.
With A some positive and nite constant, if jf (x)j A < 1 for all x, then we say the
function f (x) is bounded.
If jf (x0 )j = 1 for some particular x = x0 , then we say f is singular at x0 , and x0 is a
singular point of f .
Examples.
1=x
.... .
.. ...
.. ...
.. ...
... ...
... ..
.. ...
.. ..
... ....
..
...
...
....
..
.....
......
...
..........
...
...................
..
........................
...
....................................................................................................................................................................
..
.................................
..............
.
.
........
....
.....
...
.....
...
..
...
.
... ....
... ..
... ...
... ..
... ...
... ...
... ..
... ..
...

1
is singular at x = 0
x

sin(x)
is bounded at x = 0
x

cos(x)
is singular at x = 0
x

sin(x)=x
.

cos(x)=x

1=(x 7)2

1
is singular at x = 7
(x 7)2

p
1= x
......
.....
.. ...
... ..
.. ...
.. ...
... ...
.. ...
.. ..
... ...
.. ...
.. ..
... ...
.. ...
.. ..
...
...
....
..
....
...
.....
..
.......
..........
..
..............
...
......................
..
....................................
..
.....
...
..
...
...............................................................................................................................................
..

1
p is singular at x = 0
x

Consider now the denite integral of f (x) between the limits x = a and x = b. These
limits are arbitrary, but with the restriction that a b.

If f (x) dx A;

then the integral converges, and it exists.

If f (x) dx = 1;

then the integral diverges, and it does not exist.

From the triangle inequality

Zb
Zb

f (x) dx jf (x)j dx:

If f (x) = jf (x)j (f is non-negative, that is f (x) 0), and if b > a, then

Zb
a

f (x) dx 0:

Therefore, for simplicity, let's temporarily restrict our attention to non-negative f (x).
Basically, there are two ways to get a divergent integral:
(1) A nite range integral (1 < a < b < 1) having an integrand with a non-integrable
(too big!) singularity.
(2) An innite range integral (a = 1 and/or b = 1) having a bounded integrand
that doesn't decay fast enough at 1.
Examples.
Z1

dx
x2

diverges because the integrand has a non-integrable singularity at the origin.

With > 0, if we avoid the second-order pole at the origin, the integral evaluates to
Z1

1
dx
=
x2
x

x=1

1
1:

Clearly this integral is unbounded (singular) if we let ! 0.

Z1
0

1
dx

= ln(x)
= 0 ln(0) = 1 is divergent.
x
x=0

Z1
0

Z1
1

1
dx

p = 2x1=2
= 2 is convergent.
x
x=0

1
dx
1
=
x2
x

x=1

= 1 0 = 1 is convergent.

Z1
1

= ln(x)
= 1 0 = 1 is divergent.

x
x=1

Cauchy Principal Value

Sometimes we relax our concept of the integral to obtain a nite result and \dance around"
singularities in the integrand that are, strictly speaking, non-integrable. Here is one example that demonstrates what is meant by the Cauchy principal value which is one possible
interpretation of an otherwise divergent integral.
If f (x) is bounded everywhere and if a < c < b, then we can always break the integration
domain from a to b at the point c and write
Zb

f (x) dx =

f (x) dx +

f (x) dx:

We have already seen that the function f (x) = 1=x has a non-integrable singularity at the
origin, since
Z1
dx
is divergent.
x
0

Also observe that

dx
x

is divergent.

If we want to integrate 1=x from x = 1 to x = 1, we should not break it as

Z1
1

Z0
Z1
dx
dx
dx
6
=
+
:
x
x
x
1

But under the Cauchy principal value interpretation, if we exploit the odd symmetry of
the integrand, ignoring the singularity at the origin, then we can say

2
3
Z1
Z
Z1
dx
dx
dx
dx 5
=|
= lim 4
+
= 0:
!0
x
x
x
x
1

1=x

Z1
sin x

SHOW THAT
x dx IS DIVERGENT
0

(k + 1=2)

(k + 1)

(k+1)
Z
k

sin x
=2
1

x dx > (k + 1=2) = 2k + 1

(k+1)

Z1
Z
Z
Z
1
1
X
X
sin x
sin x
sin x
sin x
1

dx =
dx +

dx >
dx +
=1
x
x
x
x
2k + 1
k=1 k
k=1
0
0
0
|
{z
} | {z }
nite

divergent

\... better to say something new and stimulating to the interested students than to show
routine steps to the uninterested ones."
E.G. Peter Rowe Geometrical Physics in Minkowski Spacetime, Springer, 2001, p. v

SHOW THAT

sin x
dx IS CONVERGENT
x

0
... ...
... ...
... ...
... ...
... ... .
.. ... ...
... ... ...
... ... ...
... ... ...
.....
............... ........................ ... ... ... .
.. ... ... ... ..
......
......
. ... ... ... ...
......
.....
.
.
.
... ... ... ... ...
.....
..
.
.
.
... ... ... ... ... ...
....
.
.
.
... ... ... ... ... ... ..
.
.
....
.
.
. ... ... ... ... ... ... ...
.
.
.
....
... ... ... ... ... ... ... ... ...
..
.
.
....
.
.
.
.
....
.
.
.
.
....
..
.
.
....
..
...
...
...
....
...
...
...
.
....
..
.
.... .
..
......
..
.............................................................................................................................................................................................................................................................................................................................................................................
..
.. .....
..... ..
.
.
....
.
.
....
.....
....
....
....
.....
....
....
.
.
.
.
.....
.
.....
.....
.....
.....
.......
......
.......
.......
.
.
.
.
.
.
.
..........
... ...... ... ... ... ... ... ...
.............. ..................................... ... ... ... ... ... ... ...
. ... ... ... ..............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
. ... ... ... ...
.. ... ... ... ..
. ... ...... .
... ... ... ..
.
.
.
.
.
.
.
.
.
...
... ... ...
. ... ...
... ... ..
. ... ...
.
.
.
.
.
...
......
... ...
... ...

(2k + 2)

(2k+2)
Z

(2k + 1)

sin x
dx =
x

(2k+1)
Z

sin x
dx +
x

(2k+1)

(2k+1)
Z

sin x
dx
x

(2k+2)
Z

sin x
dx
x

let t=x

(2k+1)
Z

sin t
dt
t+

(2k+1)
Z

sin x

1
1

x x+

(2k+1)
Z

sin x
dx <
x2 + x

Z1
0

sin x
dx =
x

2N
Z
0

1
X
sin x
dx +
x

k=N

(2k+1)
Z

dx
1
=
x2
4k2 + 2k

(2k+2)
Z

sin x
dx <
x

2N
Z
0

1
X
sin x
1
dx +
2
x
4k + 2k
k=N
|
{z
}
{z
}

nite

Homework Problems

1. Express each of these four functions (signals) in terms of shifted, modulated, unit steps.
f (t)

x(t)

g(t)

2. Discuss the existence and evaluate (if appropriate!) the following integrals. First sketch
the integrand (except in an extreme case!).
Z1

u(t) dt;

u(t)et dt ( > 0);

sin(!0 )( ) d;

Z1
1

cos(!0 )( ) d;

cos x
dx;
x

Z1
1

Z2
p

u(t)et ej!t dt

cos x
dx;
x

sin(x2 )
(x 2) dx
ln(x=)

dx
;
x2

Z1
0

dx
;
x2

dx
p ;
x

exp[j sin2 (!0 t)] (t + )

2 sin(3t=2) (t )
2
dt
t3 tan(7t) + ln(sin t)
t + 2t 4 sin1 (t=)
35

Z1
0

dx
p
x

3. Prove that convolution is commutative:

f (t) ~ g(t) = g(t) ~ f (t):
4. Evaluate u(t) ~ u(t).
5. Show that (t) = (t) is an even function.
6. If f (t) is continuous at t = t0 , show that f (t)(t t0 ) = f (t0 )(t t0 ).

Relationship Between Impulse Response and Step Response

d
(t) = u(t)
dt

(t)

u(t)

u(t) =

h(t)

h(t) =

( ) d

s(t)
s(t) =

d
s(t)
dt

h( ) d

To see this, consider the general linear system input/output relationship.

x(t)

y(t)

h(t)

y(t) = x(t) ~ h(t) =

d
y(t) =
dt

Z1
1

x( )h(t ) d =

d
x( ) h(t ) d =
dt

Z1
1

h( )

h( )x(t ) d

d
x(t ) d
dt

= x(t) ~ h (t) = x (t) ~ h(t)

The output of these two systems is identical, since the order of dierentiation and passing
through the linear system having a given impulse response h(t) does not matter. More
concisely, the operators h(t)~ and d=dt commute.

x
e(t)

d
dt

h(t)

d
dt

h(t)

ye(t)

Time-Domain RC Circuit Analysis

t=0

V0 cos(!t)u(t) = vg (t)

v(t)

The capacitor is initially charged, such that v(t) = Vc0 prior to the closing of the switch
at time t = 0. Find an expression for the signal v(t) that is valid for all time t 0.
The constants R, C, !, V0 , and Vc0 are given and thus known.
The initial condition on the voltage v(t) that applies to our circuit for time t 0 is v(0+ )
and v(0+ ) = v(0 ) = Vc0 since the voltage across a capacitor must be a continuous function
of time. That is, if the capacitor current is bounded, because the voltage/current terminal
relationship of a capacitor is
d
iC (t) = C vC (t)
(1)
dt
where iC (t) and vC (t) are related by the passive sign convention. Inverting this, that is,
writing the voltage as a function of the current, we have
1
vC (t) =
C

Zt
1

1
iC ( ) d =
C

1
iC ( ) d +
C

1
iC ( ) d = vC (0) +
C

iC ( ) d: (2)

The Kirchho current law applied to the (single unknown) node of potential v(t), for time
t 0 is
dv(t) v(t) vg (t)
C
+
=0
(3)
dt
R
which is rearranged as
dv(t)
1
V0
+
v(t) =
cos(!t)u(t):
(4)
dt
RC
RC
Here we are already working in the time zone t 0, so with the stipulation that t 0, the
Heaviside unit step is probably unnecessary so we can write
dv(t)
1
V0
+
v(t) =
cos(!t):
dt
RC
RC

(5)

The homogeneous solution is the solution to

1
dvh (t)
+
vh (t) = 0
dt
RC
38

(6)

and it is
vh (t) = Ket=RC

(7)

where the constant K cannot be determined until we construct the complete solution. The
particular solution is some specic or particular function vp (t) that yields the right-hand
side (forcing term) when operated upon by the dierential operator

V0
d
1
+
vp (t) =
cos(!t):
(8)
dt RC
RC
There are no unknown constants (unknown by the dierential operator) in the particular
solution.3 Since the forcing function is cos(!t) and since cos(!t) is not a homogeneous
solution, then we see that the particular solution must be a linear combination of cos(!t)
and sin(!t), as in
(9)
vp (t) = A cos(!t) + B sin(!t):
The constants A and B come directly from the dierential equation (8) , which readily
surrenders them (to us): All we have to do is insert the form (9) into the dierential
equation (8). To do this, rst write the required derivative
dvp (t)
= !B cos(!t) !A sin(!t):
dt
Insertion of (9) and (10) into (8) yields

A
B
V0
+ !B cos(!t) +
!A sin(!t) =
cos(!t):
RC
RC
RC

(10)

(11)

The only way this can be true for all t 0 is if the coecients of the cos(!t) and the
sin(!t) terms are separately equal on both sides of the equation, so that
A
V0
+ !B =
RC
RC

(12)

B
!A = 0:
RC
The solution to this two-by-two system of simultaneous, linear equations is
A=

V0
1 + (!RC)2

and

!RCV0
;
1 + (!RC)2

(13)

(14)

and so our particular solution (9) is

vp (t) =

V0
cos(!t)
+
!RC
sin(!t)
:
1 + (!RC)2

3 In

(15)

other words, the particular solution comes only from the dierential equation, that is, the particular
solution is completely independent and ignorant of any initial conditions.
39

Again, note that vp (t) comes completely from the dierential equation alone. We now can
assemble the complete solution v(t) = vh (t) + vp (t) or
v(t) = Ket=RC +

V0
cos(!t)
+
!RC
sin(!t)
:
1 + (!RC)2

(16)

Application of the one initial condition v(0) = v(0+ ) = Vc0 (note that it is the plus side of
zero that applies to our dierential equation, because the dierential equation is valid for
time t 0) to (16) gives
Vc0 = K +

V0
1 + (!RC)2

K = Vc0

V0
:
1 + (!RC)2

(17)

Let's write out the complete solution in two forms. Firstly, observe that for t 0

V0
v(t) = Vc0
et=RC
1 + (!RC)2
|
{z
}
homogeneous solution

V0
cos(!t)
+
!RC
sin(!t)
1 + (!RC)2
|
{z
}

(18)

particular solution

is exactly in the form in which we constructed the solution, that is, the sum of the homogeneous plus particular solutions. If we instead group terms proportional to Vc0 and V0 ,
we have for t 0
v(t) =

t=RC

Vc0 e
| {z

zero-input response

h
i
V0
t=RC
cos(!t) + !RC sin(!t) e
:
1 + (!RC)2
|
{z
}

(19)

zero-state response

If there is no input signal (or generator voltage vg (t) or forcing term), then V0 = 0 and
the output signal v(t) is caused solely by the initial energy (or non-zero initial state of
the system) stored in the capacitor, and is proportional to the non-zero initial condition
voltage Vc0 . On the other hand, if the initial state of the system is zero, so that Vc0 = 0,
then the response due to the input vg (t) that is proportional to V0 is called the zero-state
response.

Step Response of an RC Circuit

..... ..... .....

V0 u(t) = vg (t)

+
v(t)

The step response is, by denition, a zero-state response so the capacitor C is initially
uncharged prior to the turning on of the input step function at time t = 0. Therefore, by
the continuity of capacitor voltage we have the initial condition
v(0+ ) = v(0 ) = 0:

(20)

Ultimately, the step response s(t) = y(t) should be the response due to a non dimensioned
input signal u(t) = x(t), but in doing circuit analysis I nd it better to explicitly express
all physical signals such as voltages and currents in appropriate mksC (SI) units. At the
end we can normalize by V0 , for example. In systems engineering (\black-box") notation,
we have the general form of a zero-state response y(t) to some arbitrary input x(t) like
this:

x(t)

y(t)

s(t)

For the step response, we have:

u(t)

Our ensuing circuit analysis regards the input as x(t) = vg (t) = V0 u(t) and our output
signal is y(t) = v(t). The Kirchho current law applied to the node of potential v(t) in
the circuit is for time t 0
C

dv(t) v(t) vg (t)

+
=0
dt
R

(21)

which is rearranged as
1
V0
dv(t)
+
v(t) =
u(t):
dt
RC
RC
41

(22)

Here we are already working in the time zone t 0, so with the stipulation that t 0, the
Heaviside unit step is probably unnecessary so we can write
1
V0
dv(t)
+
v(t) =
:
dt
RC
RC
The homogeneous solution is (still) of the form
vh (t) = Ket=RC

(23)

(24)

where K is some (new) unknown constant. But now the particular solution, due to a
constant forcing term, is itself a constant, call it
vp (t) = D:

(25)

This constant D is, again, completely determined (that is known by) the dierential equation (24). Insertion of (25) into (23) gives the rather trivial
V0
D
=
or
D = V0 :
RC
RC
The complete solution is now v(t) = vh (t) + vp (t) or
v(t) = Ket=RC + V0 ;
whereupon the initial condition (20) gives K = V0 so that

v(t) = V0 1 et=RC
(t 0):
A better (systems engineering style!) way to write this is

v(t) = V0 1 et=RC u(t):

(26)

(27)

(28)

(29)

The unit step not only eliminates the need for the disclaimer t 0, but its inclusion is
invaluable in emphasizing the causal nature of the response, and it is critically important
if (when!) we dierentiate the response.
Now we can normalize by V0 , as advertised, and obtain the true, non dimensional step
response

s(t) = 1 et=RC u(t):

(30)

The impulse response is the derivative of the step response

d
1 t=RC
h(t) = s(t) =
e
u(t) + 1 et=RC (t)
dt
RC
1 t=RC
e
u(t):
(31)
=
RC
Note that the units of h(t) are [()(F)]1 = (s)1 . In this particular instance, the chainrule dierentiation that also dierentiated the u(t) to get a (t) might appear superuous,
since the contribution

t=RC
t=RC
1e
(t) = 1 e
(t) = 0(t) = 0
(32)

t=0

is zero. But that will not happen in general.

Step Response of Another RC Circuit

+ vC (t)
...
...
..
..
..
..
.
....................................................................
.
...
..
.......................................................................
..
..
..
..
..
...
....
...
...
..
..
..
..
.....
......
...
...
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......
..... .........
.
.
.
.
.
.....
.
...
...
..
........
............
.....
..
........
...
...
.
..
....
.......
.......................
.............
...
.
...
....
...
..
...
...
...
..
...
...
...
.......................................................................................................................................................

V0 u(t) = vg (t)

v(t)

In this circuit, we take our output signal to be the voltage v(t) across the resistor, but the
capacitor voltage vC (t) must also be specically addressed because of the importance of
continuity of capacitor voltage for determining initial conditions. The zero-state response
demands that there is no initial energy storage in the capacitor so vC (0 ) = 0 prior to the
unit step input turning on at time t = 0. By the continuity of capacitor voltage
vC (0+ ) = vC (0 ) = 0:

(33)

The application of the Kirchho current law to the node of potential v(t) gives

d
v(t)
+C
v(t) vg (t) = 0
R
dt

(34)

dv(t)
1
dvg (t)
+
v(t) =
:
(35)
dt
RC
dt
The forcing term is now the derivative of the voltage generator, which puts us in the
unfamiliar realm of solving a dierential equation with a wild Dirac-delta forcing term
1
dv(t)
+
v(t) = V0 (t):
dt
RC

(36)

If we stay away from t = 0, then the delta function is zero and we have (note the time
restriction)
1
dv(t)
+
v(t) = 0
(t > 0):
(37)
dt
RC
It looks as though we have lost our excitation: The dierential equation is now homogeneous. Information about the excitation, embodied by the source strength V0 , is included
in the initial condition. Observe in the circuit that
vg (t) = vC (t) + v(t)
43

(38)

and at time t = 0+ this is

vg (0+ ) = vC (0+ ) + v(0+ ):

(39)

The initial condition (33) on the capacitor voltage together with vg (0+ ) = V0 reveals that
the non-zero initial condition on our output signal v(t) is v(0+ ) = V0 . No particular
solution is required for the complete solution to our homogeneous dierential equation
(37), which is readily seen to be
v(t) = V0 et=RC

(t > 0):

(40)

Inclusion of the Heaviside unit step function gives

v(t) = V0 et=RC u(t)

(41)

so that the time zone restriction t > 0 can be removed. The non dimensional step response
is therefore
(42)
s(t) = et=RC u(t)
and the corresponding impulse response is
h(t) =

since

1 t=RC
d
s(t) = et=RC (t)
e
u(t)
dt
RC
1 t=RC
e
= (t)
u(t)
RC

t=RC
e

= 1:

(43)

(44)

t=0

Observe that the proper use of chain-rule dierentiation and explicitly including the u(t)
in the step response s(t) is necessary to get the true impulse response h(t), which most
denitely contains a (t) term.

Relationship Between the Two RC Circuits

(t)

....
..
................................................................
...
..

ha (t)

(t)

hb (t)

In the lab, simply connect R and C in series and hang the series leg across the output
terminals of a (ideal?) voltage source. Then we see that the output of circuit (a) is the
voltage across C and the output of circuit (b) is the voltage across R. Kirchho's voltage
law clearly shows the simple relationship that is displayed by comparing (30) and (42) and
by comparing (31) and (43).
............................................................................
..
...
...
...
........
...
......
.
.
.
.
....
.
.
.......
...
........
..
...
........
.............
.....
.......
...
..
...
.......
..............
....
.
.
.
.
.
.
.
.
.
.
.. ........
...
...
......
...
..
...
..
...
.....
..
.
..
...
.
.
..
....
.
.
.
..
.
....................
...
...
..
....
.
..
...............................
...
...
...
..............................
.
....
...
...
..
...
..
..
..
..
...
............................................................................

vg (t)

vb (t)

vg (t) = va (t) + vb (t)

(t) = ha (t) + hb (t)

+
va (t)

u(t) = sa (t) + sb (t)

Homework Problems
1. Observe that the circuit on page 41 is a special case of the circuit on page 38, if Vc0 = 0
and ! = 0. Therefore, show for yourself that the response (29) is the corresponding special
case of the response (19).
2. Use convolution to nd the zero-state (or driven) response of the RC circuit on page 41
to a rectangular input pulse vg (t) = V0 [u(t)u(tT )]. Graph this response using MATLAB
for several numerical values of the important (non dimensional) parameter T =RC.
3. Go backwards, and obtain the two step responses (30) and (42) by integrating the
corresponding impulse responses (31) and (43), since
d
h(t) = s(t)
dt

()
45

s(t) =

Zt
1

h( ) d:

Step and Impulse Response of Series RLC Circuit

+ v (t)

i(t)

vg (t)

vR (t)

Fig. 1. Circuit layout.

vg (t) = x(t)

y(t) = Ri(t)

Fig. 2. System block diagram.

The system step response is a \zero-state" response: No energy is initially stored in the
system prior to the onset of the exciting Heaviside step function that turns on at t = 0.
For deniteness and to keep track of the physical units, let
vg (t) = V0 u(t)
46

(1)

with the (real) constant V0 being the amplitude of the exciting step function. Since the
instantaneous energy stored in the inductor and capacitor at time t are
wL (t) = 12 Li2 (t)

2
wC (t) = 12 CvC
(t);

and

(2)

respectively, zero initial (at time t = 0 ) energy storage requires

i(0 ) = 0

vC (0 ) = 0:

and

(3)

If the inductor voltage

di(t)
dt
is to be bounded, then its current must be a continuous function of time, and hence
vL (t) = L

i(0+ ) = i(0 ):

(4)

Similarly, if the capacitor current

dvC (t)
dt
is to be bounded, then its voltage must be a continuous function as in
iC (t) = C

vC (0+ ) = vC (0 ):

(5)

The equation-of-motion for the series circuit of Fig. 1 is Kirchho's voltage law
Zt

di(t)
1
+ Ri(t) +
L
dt
C

i( ) d = vg (t);

(6)

an integrodierential equation in the unknown series current i(t). With the forcing function
vg (t) known, the unique solution of equation (6), whose highest derivative term is of degree
one, requires the specication of one initial condition. The physics of the circuit dictates
i(0+ ) = 0;

(7)

from (3) and (4). If we dierentiate (6) once to eliminate the integral operator, the resulting
second-order dierential equation
di(t)
d2 i(t)
1
+R
(8)
+ i(t) = vg0 (t)
2
dt
dt
C
now requires two initial conditions for a complete (unique) solution. This second initial
condition was already contained in (6), before we dierentiated. Evaluation of (6) at time
t = 0+ is
Li0 (0+ ) + Ri(0+ ) + vC (0+ ) = vg (0+ ):
(9)
L

The capacitor voltage, from (3) and (5), starts o at zero

vC (0+ ) = 0;

(10)

so that together with (1) and (7), statement (9) gives

V0
:
L
Note that the dimensions are consistent in that (A/s) = (V/H). X
i0 (0+ ) =

(11)

The proper formulation of the above initial value problem, especially stating the correct
number (two) and type of initial conditions (on the current and its derivative), required
our detailed understanding of the electrophysics embodied in the circuit. In one view
the ensuing construction of the solution to the dierential equation may seem like a pure
exercise in mathematics, but that is not the view we will take. Instead, we want to let each
step in our solution be guided by the physics that the mathematics represents, so that our
mathematics will be ecient and keep us in touch with the important physics that is our
quest. If we do this right, then our symbols and formulas and solution will be in a form
that not only allows, but even begs for physical interpretation.
Let's restate our initial value problem:
L

1
di(t)
d2 i(t)
+ i(t) = vg0 (t)
+R
2
dt
dt
C
i(0+ ) = 0
i0 (0+ ) =

(12)
(13)

V0
:
L

(14)

Since our voltage generator is vg (t) = V0 u(t), the forcing function on the right-hand side
of (12) is a Dirac-delta function, i.e.
L

d2 i(t)
di(t)
1
+ i(t) = V0 (t):
+R
2
dt
dt
C

(15)

That looks troublesome, I believe, but if we stay away from t = 0, then the delta function
is zero and our equation is homogeneous
L

di(t)
1
d2 i(t)
+
R
+
i(t) = 0
dt2
dt
C

(t > 0):

(16)

It looks as though the homogeneous dierential equation (16) has lost the source, but note
that initial condition (14) has information about the original source in the circuit. If there
is no initial energy storage to drive the \dead" circuit, the excitation of a nonzero response
has to come from an explicit input. And our input (1) is proportional to the constant V0
that explicitly appears in the nonzero (\nonhomogeneous") initial condition (14).
Homogeneous constant-coecient linear ordinary dierential equations (LODE's) are particularly easy to solve, since the solutions are exponentials, say of the form i(t) = exp(rt)
where two such linearly independent forms (i.e. two dierent r values), are required in our
case of a second-order equation. Before proceeding, it is expedient to condense the notation
and in fact, combine the element values R, L, and C by in a manner that anticipates the
solution. Of course, no one could truly do this in advance if they had not already solved
the problem (or read someone else's solution). But that's not the point: We choose the
best path because it draws attention to what is important and because it is ecient and
48

because it works. Firstly, normalize (16) so that the coecient of the highest derivative is
unity
1
d2 i(t) R di(t)
+
i(t) = 0
+
dt2
L dt
LC
and then introduce constants
1
LC

(17)

d2 i(t)
di(t)
+
2
+ !02 i(t) = 0:
2
dt
dt

(18)

=
to write

R
2L

and

!02 =

Insertion of i(t) = exp(rt) into (18) gives

r + 2r + !02 ert = 0;
and since the exponential cannot equal zero, we arrive at the characteristic polynomial for
the two roots
r2 + 2r + !02 = 0:
(19)
The quadratic formula provides the two roots immediately
r1;2

q
= 2 !02 :

(20)

Two distinct possibilities exist: The case of an over-damped circuit having > !0 and an
under-damped circuit when < !0 . The case when = !0 is called critically-damped, and
can be obtained from either of the other two cases in the limit as ! !0 . Or, it can be
constructed directly.
Under-damped circuit. If < !0 , introduce yet another parameter (constant)
q
= + !02 2

(21)

so that the two roots (20) are a complex-conjugate pair

r1;2 = j:

(22)

The solution to the homogeneous dierential equation (18) is a linear combination

i(t) = K1 er1 t + K2 er2 t = K1 e(j)t + K2 e(+j)t = et K1 ejt + K2 e+jt

= et [A cos(t) + B sin(t)] :

(23)

It is critically important to recognize that a linear combination of exp(jx) is equivalent

to a (dierent) linear combination of cos x and sin x. Although either form is ne, the
49

trig form is better here because it facilitates the application of initial conditions. Initial
condition (13) gives
i(0) = A = 0
(24)
immediately! The derivative of the remaining form

i(t) = Bet sin(t)

(25)

i0 (t) = Bet cos(t) sin(t)

(26)

i0 (0) = B:

(27)

which has initial value

Initial condition (14) is now
V0
L
and the complete solution to the initial value problem is

(28)

B =

i(t) =

V0 t
e
sin(t)
L

(t > 0)

or, with attention drawn to its \turned-on" nature,

i(t) =

V0 t
e
sin(t)u(t):
L

(29)

Note that the dimensions of and are both (s1 ) and that the dimensions of the product
L are (H/s= ), so that the current is in (A). The signal that is regarded as the output
in the linear system diagram of Fig. 2 is the voltage vR (t) = Ri(t), so that when the input
vg (t) is regarded as
x(t) = V0 u(t);
(30)
the output is
y(t) =

RV0 t
e
sin(t)u(t):
L

(31)

If we normalize both signals by the amplitude V0 , then the nondimensional input/output

signals are
R t
x(t) = u(t)
and
y(t) = s(t) =
e
sin(t)u(t):
(32)
L
The response to the Heaviside unit step function is called the step response, and is denoted
here as s(t).
The derivative of the convolution
z(t) = f (t) ~ g(t) =

Z1
1

f ( )g(t ) d =
50

Z1
1

g( )f (t ) d

(33)

z 0 (t) = f 0 (t) ~ g(t) = f (t) ~ g 0 (t):

(34)

Therefore, the system impulse response is the derivative of its step response
h(t) = s0 (t):

(35)

The impulse response of the RLC circuit of Fig. 1 is therefore

R t

h(t) = e
cos(t) sin(t) :
L

(36)

Note that the units of h(t) are (s1 ), as are the units of (t).

1.00
0.75
0.50
L
s(t)
R

0.25
0.00
0:25
0:50
0:75

t
Fig 2. Normalized step response of underdamped RLC circuit.
Sold curve: = = 0:1; Dashed curve: = = 0:5

Over-damped circuit. If > !0 , then both roots of the characteristic quadratic polynomial (19) are real, so with
r1 = 1

q
with 1 = 2 !02

and
r2 = 2

with 2 = +

2 !02 ;

(37)

(38)

we see that 0 < 1 < 2 and the solution to the homogeneous dierential equation (16)
can be written as the linear combination
i(t) = k1 e1 t + k2 e2 t

(t > 0):

(39)

Initial condition (13) is i(0+ ) = 0 so that k2 = k1 and therefore

i(t) = k1 e1 t e2 t :

(40)

i0 (t) = k1 2 e2 t 1 e1 t

(41)

The required derivative is

and at time t = 0+ this must be

V0
i0 (0+ ) = k1 2 1 =
L
and therefore

(42)

V0
:
(2 1 )L

(43)

1 t

V0
e
e2 t u(t)
(2 1 )L

(44)

k1 =
The solution for the current is now
i(t) =

and the nondimensional step response is

s(t) =

1 t

R
e
e2 t u(t):
(2 1 )L

(45)

Homework. Given that a (scaled) rectangular pulse

1
x(t) =
T

t
T

behaves as the Dirac-delta function in the limit as T ! 0, how small does T have to be so
that the response y(t) of a system to the pulse x(t) is a good approximation to the true
system impulse response h(t) ? Consider the underdamped RLC circuit with parameters
and . (Note that specifying and is equivalent to, and in fact more illuminating than,
specifying the circuit element values R, L, and C.) Use MATLAB to compare graphs of
h(t) with y(t) for several values of the pulse-width T . The important parameter(s) of the
problem should be nondimensional.

Homework Exercises. Find the impulse response of these four circuits.

(t)

..... ..... .....

(t)

ha (t)

..
...
................................................................
....
.

hb (t)

........ .......... ......... ..........

hc (t)

(t)

What is the relationship between ha (t) and hb (t) ?

hd (t)

Between ha (t) and hd (t) ?

Homework. Find the step and impulse responses of the series RLC circuit that is critically
damped: when = !0 and the characteristic quadratic polynomaial has a single, repeated
root.

Evaluation of an Important Integral via the Laplace Transform

1.0
0.8
0.6
sin(x)
x

0.4
0.2
0.0
0:2
0:4

1
2n+1
sin(x)
1X
1
x3
x5
n x
sinc(x) ,
=
(1)
=
x
+
:::
x
x n=0
(2n + 1)!
x
3!
5!

x2
x4
= 1
+
:::
3!
5!
sin(x)
=1
x!0
x

sinc(0) = lim

sinc(x) = sinc(x)
We want to show that

is even

sin(x)
dx = :
x

(1)

(2)
(3)

(4)

Since sinc(x) is even, observe that

sin(x)
dx = 2
x

Z1
0

sin(x)
dx:
x

(5)

Our approach is to nd the Laplace transform of the function

Si(t) ,

sin(x)
dx;
x

(6)

called the \sine integral." Our desired integral (half of it!) is the nal value of Si(t), that
is Si(1). Perhaps it is helpful to write Si(t) using as the dummy variable of integration
Si(t) ,

sin( )
d:

(7)

We will use the following Laplace transform properties. If L[f (t)] = F (s), then

Z1
f (t)
L
= F () d
t

(8)

2 t
3
Z
F (s)
L 4 f ( ) d 5 =
s

(9)

lim f (t) = lim sF (s)

t!1

nal value theorem

s!0

(10)

The Laplace transform of

g(t) = sin(t)u(t)

(11)

1
:
+1

(12)

is
G(s) =

The Laplace transform of

h(t) =

g(t)
= sinc(t)u(t)
t

is then
H(s) =

d
:
+1

(13)

(14)

This integral can be found in a table (say in your calculus book, or you can use a symbolic
mathematical tool such as MAPLE, or you can use a web-based integrator4 ). Although
not important for our purposes, let's do it ourselves just for fun.
Consider y = tan(x) so that
dy
d sin(x)
=
= 1 + tan2 (x)
dx
dx cos(x)
4 If

you know one, tell the class. If you don't know one, nd one!
55

Z
Z1
s

dy
1 + y2

dy
= dx
1 + y2
Z
= dx = x = tan1 (y)

= tan1 (1) tan1 (s) = tan1 (s) = tan1 (1=s)

2
1+y
2

tan1 (s) + tan1 (1=s) =

Therefore the Laplace transform of h(t) = sinc(t)u(t) is

H(s) = tan1 (1=s):

(15)

The Laplace transform of

f (t) = Si(t) =

sinc( ) d

(16)

is now
F (s) =

tan1 (1=s)
s

(17)

and the nal value theorem gives

Si(1) = f (1) = lim sF (s) = lim tan1 (1=s) =
s!0

s!0

:
2

(18)

Proposition (Spectral form of the Dirac-delta). This Fourier integral satises the
denition of the Dirac-delta (distribution or generalized function):
1
2

ej!t d! = (t):

(1)

Given the integral

sin x
dx = ;
x

(2)

Dirichlet's discontinuous integral falls out

Z1
1

sin x
dx = sgn() =
x

( ? 0):

(3)

Integrate the left hand side of (1) from a to b:

Zb
a

1
2

j!t

1
=
2

1
d! dt =
2

Z1
1

Z1 Zb
1 a

sin b! sin a!

!
!

j!t

1
dt d! =
2

ejb! eja!
d!
j!

1
d! =
[ sgn(b) sgn(a)] =
2

This is the sampling property of (t) :

1; a < 0 < b
0; o.w.

(4)

Fourier Transform

F (!) = F[f (t)] =

1
f (t) = F 1 [F (!)] =
2

f (t)ej!t dt

()

F (!)ej!t d!

Z1
Z1

f (t)ej!t dt =
f (t) dt
f (t)ej!t dt

Therefore a sucient (not necessary) condition for the existence of F (!) is the absolute
integrability of f (t)
Z1

f (t) dt < 1:
1

Note that the zero frequency value of the Fourier transform (or spectrum)

F (0) =

f (t) dt

is the DC or \average value" of the signal f (t).

We can easily demonstrate the legitimacy of the Fourier transform pair by taking the
inverse transform of the forward transform
2
3
Z1 Z1
n
o

1
4
F 1 F f (t) =
f ( )ej! d 5 ej!t d!
2
1
1
{z
}
|
F (!)

Z1
1

f ( )

1
ej!(t ) d! d
2
1
|
{z
}
( t)

f ( )( t) d = f (t):

Fourier Transform Exercises

Find an expression for the Fourier transform of the following time-domain signals, and
sketch BOTH f (t) and F (!):
0. f (t) = eat u(t) causal, decaying exponential (a > 0)
1. f (t) = A(t=T ) rectangular pulse
2. f (t) = B(t=T ) triangular pulse
3. f (t) = C constant
4. f (t) = (t t0 ) delayed Dirac impulse
2

5. f (t) = Aet Gaussian signal \and it is a trick"

6. f (t) = ejtj

cos
7. f (t) =
(!0 t)
sin
8. f (t) = u(t) Heaviside unit-step function (why is this one so hard?)
hint: write u(t) = lim eat u(t)

and note lim

a!0 a2

a!0

9. f (t) = sgn(t)

a
= (!)
+ !2

signum (or signature or sign) function

10. f (t) = m(t) cos(!0 t)

AM modulated signal
i.e. !0 =2 = 1250 (kHz) and m(t) = \Your Cheatin' Heart"
11. f (t) = sinc(t)
12. f (t) = (t=T ) ~ sgn(t)
13. f (t) = t(t=T )
14. f (t) = D combT (t) = D

1
P

n=1

15. f (t) =

1
P

k=1

16. f (t) = A

(t nT ) Woodward's comb function

g(t kT ) periodic replication where F[g(t)] = G(!)

1
P

k=1

[(t kT )= ] rectangular pulse train

Fourier Series
f (t) = a0 +

1
X

cn ejn!0 t

(!0 = 2=T )

n=1

1
a0 =
T

1
X

[an cos n!0 t + bn sin n!0 t] =

tZ
0 +T

f (t) dt

an
bn

2
=
T

tZ
0 +T

f (t)

cos
sin

(n!0 t) dt

1
cn =
T

tZ
0 +T

f (t)ejn!0 t dt

Fourier Transform
F (!) = F[f (t)] =

j!t

f (t)e

Properties
1. superposition
2. time delay

7. multiplication

f (t)
F (t)

10. integration

c1 F1 (!) + c2 F2 (!)

F (!)G(!)
1
F (!) ~ G(!)
2

f (t)g(t)

jtf (t)

F (!)ej!t d!

F (! !0 )

f (t) ~ g(t)

9. frequency dierentiation

2f (!)

f (t)ej!0 t

df (t)
dt

1
[F (!)] =
2

ej!t0 F (!)
1 !
F
jj

f (t t0 )

8. time dierentiation

Transform

c1 f1 (t) + c2 f2 (t)

4. duality

6. convolution

dt () f (t) = F

Signal

3. time scale

5. frequency translation

j!F (!)
d
F (!)
d!

f ( ) d

F (0)(!) +

1
F (!)
j!

Z1
1

1
jf (t)j dt =
2

u(t) ()
F

(t= ) ()

1
+ (!)
j!
sin(! =2)
= sinc(! =2)
! =2

sinc(t) ()
F

combT (t) ()

jF (!)j2 d!

(!=2)

2
comb2=T (!)
T

Z-Transform
Zfx[n]g = X(z) =
Properties
1. superposition
2. time delay

1
X

x[n]z n

n=1

Signal

Transform

c1 x1 [n] + c2 x2 [n]

z m X(z)

x[n m]

3. modulation

x[n] n

4. convolution

x[n] ~ y[n]

5. frequency dierentiation

nx[n]

6. accumulation

n
P

c1 X1 (z) + c2 X2 (z)

X(z=)
X(z)Y (z)
d
X(z)
dz
1
X(z)
1 z 1
z

x[k]

k=1

Lff (t)g = F (s) =

f (t)est dt

is the transform pair

f (t) , F (s)

property

t-domain

linearity

f (t) + g(t)

s-domain
F (s) + G(s)
est0 F (s)

f (t t0 )u(t t0 )

time delay
time scaling

1
F (s=)

f (t)

exponential modulation

et f (t)

time dierentiation

df (t)=dt

time integration
time integration
s-domain dierentiation

sF (s) f (0)

s2 F (s) sf (0) f 0 (0)

F (s)=s
F (s)=s +
dF (s)=ds
R1
F () d
s

tf (t)

division by t

f (t)=t

t-domain convolution

f (t) ~ g(t)

F (s)G(s)

Initial value theorem

lim f (t) = lim sF (s)

t!0

s!1

lim f (t) = lim sF (s)

Final value theorem

t!1

s!0

Short Table of Laplace Transforms (s = + j!)

(t)
u(t)
eat u(t)
teat u(t)
tn u(t)
cos(t)u(t)
sin(t)u(t)

( > 0)

F (s + )

d2 f (t)=dt2
Rt
f ( ) d
0
Rt
f ( ) d
1

reprise

(t0 0)

1
1
s
1
s+a
1
(s + a)2
n!
n+1
s
s
2
s + 2

2
s + 2

(8s)
( > 0)
( > Re[a])
( > Re[a])
( > 0)
( > 0)
( > 0)

1
s

f (t) dt

Fourier Transform of the Gaussian

then F (!) =
exp[! 2 =4].

If f (t) = exp[t2 ]

This one is a bit involved to show, especially without complex variable theory. But it's
so pervasive in signal analysis and communications that we use it almost daily, so here is
a classical derivation that uses nothing more than integration-by-parts and a simple rst
order dierential equation.
A preliminary integral

ex dx =

is required. The trick5 to derive this is to consider

I =

y 2

Z1 Z1

dy =

e(x

+y 2 )

dx dy:

1 1

Conversion to polar coordinates via

x = r cos

and

y = r sin

gives
2

I =

Z2Z1
0

1
=
r dr d = (2)
2

and so

Our desired Fourier integral is

F (!) =

t2 j!t

dt = 2

et cos !t dt

since f (t) = f (t) is even. Dierentiation w.r.t. ! gives

dF (!)
= 2
d!

tet sin !t dt:

Integration by parts

5 \And

u dv = uv

it is a trick!"
63

v du

p
:

with
u = sin !t

du = ! cos !t dt
2

et
v=

dv = 2te
gives

1
Z

! 1 t2
1 t2
dF (!)

sin !t

e
cos !t dt
= e
d!

0
t=0

dF (!)
!
= F (!)
d!
2
dF (!)
d! = !
F (!)
2
d
!
ln F (!) =
d!
2
2
!
+C
ln F (!) =
4
F (!) = eC e!
Z1

e = F (0) =

dt =

F (!) =

!2 =4
e

solid curves: = 1, dashed curves: = 2

f (t)
1.0
0.8
0.6
0.4
0.2
0.0

F (!)
2.0

1.5
1.0
0.5
0.0

8 6 4 2 0
!

t
64

Fourier Transform of the Heaviside Unit Step Function

F (!) = F[f (t)] =

j!t

f (t)e

()

f (t) = F

1
[F (!)] =
2

F (!)ej!t d!

Recall that a sucient (but not necessary) condition for the Fourier transform F (!) of a
signal f (t) to exist is that the signal be absolutely integrable, that is
Z1
1

jf (t)j dt < 1:

The Heaviside unit step function is not absolutely integrable

Z1
1

ju(t)j dt =

dt 1

and therefore we might anticipate that its Fourier transform will be both dicult to evaluate and exhibit singular behavior. Recall or note that the periodic signal cos(!0 t) is not
absolutely integrable, and its spectrum consists of Dirac-deltas
2 1
3
Z1 j!0 t
Z
Z1
j!0 t
e
+e
1
ej!t dt = 4
F[cos(!0 t)] =
ej(!!0 )t dt +
ej(!+!0 )t dt5
2
2
1
1
1

= (! !0 ) + (! + !0 ) :
Two preliminary formulae or ideas:
First:

lim

=
= (!)
+ !2

!0 2

(1)

This is the Cauchy or Lorentz Dirac-delta sequence (that we already studied back on page
7). Clearly
=
lim 2
= 0 if ! 6
= 0.
!0 + ! 2
The integral

Z1
1

d!
1
=
2
2
+!

Z1
1

dx
1
=2 (=2)
1

=
tan
(x)
=
=1

2
1+x

x=1

gives unity independently of , so the truth of (1) is veried.

Rewrite it as

j!
1
=
2
+!
j!

lim

Second:

(2)

!0 2

1
1
=
2
!0 j![1 + (=!) ]
j!
lim

which is clearly true if ! 6

= 0. The only other possibility is that our limit also has a
Dirac-delta at the origin of strength or weight K
1
1
=
+ K(!):
!0 j![1 + (=!)2 ]
j!
lim

Integrate both sides of this from ! = to ! = +

1
j

d!
![1 + (=!)2 ]

1
j

d!
!

(!) d!

and observe that the integral on the left side and the rst integral on the right side vanish
because of the odd symmetry of their integrands. Therefore the only possibility is K = 0
and the truth of (2) is established.
One of our rst examples (or exercises) of Fourier integral evaluations was the causal,
decaying exponential

1
( > 0):
F et u(t) =
+ j!
The condition that > 0 is necessary to ensure convergence of the integral, by making
the time-domain function decay suciently fast as t ! 1. Although the restriction > 0
enabled us to evaluate the Fourier transform of the decaying exponential, let's try to get
our desired Fourier transform of the plain Heaviside unit-step by taking the limit of the
above result as ! 0. That is

F[u(t)] = lim F et u(t) = lim

1
:
!0 + j!

Rewrite

1
j!

j!
=
= 2
+ 2
2
+ j!
( + j!)( j!)
+!
+ !2

so that
F[u(t)] = lim

!0 2

j!
1
+ lim 2
= (!) +
2
2
!0 + !
+!
j!

from (1) and (2) above.

(F)

Since
2u(t) = 1 + sgn(t)
we now have the Fourier transform of the signum function as

2
1
F[sgn(t)] = 2F[u(t)] F[1] = 2 (!) +
2(!) =
:
j!
j!
With
F[u(t)] = (!) +

1
j!

then by the time-dierentiation property of the Fourier transform

df (t)
F
= j!F[f (t)]
dt
where
(t) =

du(t)
;
dt

we have the correct transform

1
F[(t)] = j! (!) +
= 1:
j!
Using the convolution property of the Fourier transform
F[f (t) ~ g(t)] = F (!)G(!)
and the integral identity
Zt

f ( ) d = f (t) ~ u(t);

we have the Fourier transform of the integral as

1
F (!)
f ( ) d 5 = F (!) (!) +
= F (0)(!) +
:
j!
j!

Duality Property of the Fourier Transform

Recall our notation for the Fourier transform
F (!) = F[f (t)] =

j!t

f (t)e

dt () f (t) = F

1
[F (!)] =
2

F (!)ej!t d!:

For our duality property, we want the Fourier transform of some signal F (t):
F[F (t)] =

F (t)ej!t dt:

(1)

The fact that our time domain signal is a \big" F of t means that we interpret our big F
as related to some \little" f via a direct Fourier transform

F (x) =

f ()ejx d

(2)

and therefore the particular little f that must have spawned our big F must be
Z1

1
f () =
2

F (x)ejx dx:

(3)

According to (2), we can also write

F (t) =

f ()ejt d

(4)

and insertion of this into (1) yields

F[F (t)] =

8
Z1 < Z1
1

f ()ejt d

9
=
;

ej!t dt:

(5)

Reverse the orders of the and t integrations

F[F (t)] =

Z1
1

f () 4
|

ej(+!)t dt5 d
{z

recognize as 2(+!)

(6)

and so
F[F (t)] = 2

f ()( + !) d = 2f (!):

(7)

Our desired duality property of the Fourier transform is here F[F (t)] = 2f (!) where
F
f (t) , F (!) is the notation for the original Fourier transform pair F[f (t)] = F (!).
Example. Since we already know the Fourier transform pair

t F

, T sinc(!T =2);
T

(8)

!
F[T sinc(tT =2)] = 2
T

(9)

it follows from duality that

2 !

(10)
T
T
by the linearity of the Fourier transform operator, and invoking the even symmetry of the
pulse function . Now T is just a remnant parameter from the original pair (8), so let's
introduce a new (better!) constant = T =2 such that the above is
F[sinc( T2 t)] =

!
F[sinc(t)] =
:

(11)

Add this entry in your table of Fourier transforms!

Another View. Consider a function (x) related to a given function (u) through the
direct Fourier transform operator

(x) =

(u)ejux du:

(12)

Consider a second function (x) related to our given function (u) through the inverse
Fourier transform operator
Z1
1
(x) =
(u)ejux du:
(13)
2
1

Observe that (x) = 2(x). If we know the forward transform, then we also know the
reverse transform of any function, by inspection.

Half-Power Bandwidth or Pulse width

f1 (x)
...
....
.........................
.
.
.
.....
....
.....
....
....
..
....
...
...
..
...
.
...
.
.
.
..
.
....
.
.
.
.
.
.
...
.
.
.
..
...
.
.
.
.
....
.
.
.
.
....
..
.
.
.
.
.
....
.
.
.
.
.
....
.
..
.
.
.
.
.
.....
..
.
.
.
.
....
.
..
.
.
.....
.
.
.
.
.
.
......
.
.
.
.
.
.
..
.........
.
.
.
.
.
.
.
.
.
.
.
............................................................................................................................................................................................................................................................................................................................................................................................
...

f2 (x)
..
...
.........
..
...
.... ......
..
...
..
.
..
..
.
...
..
...
..
..
..
.
..
.
..
...
.
.
..
...
.
.
..
..
.
..
.
..
.
..
.
.
...
..
.
..
.
..
.
................................
......
..
.
...
.......
........ ................
.
..
...
.
.......
....
........
...
.. ...
.........................
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.. .......
.
... ..
.
...
.......
.
.......................
............................................................................................................................................................................................................................................................................................................................................................
....
...........................
.... .....
.. ...
....
.
.
.
.
.
.
.
.
......
..... ......
...............
...

f3 (x)
.....
....
..
... ....
...
........
....
.
.
.....
....
..
.. .... .... ....
.....
...
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
..
..
...
... ..
...
. ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.....
.. .. ..
..
..
.. ...
.. ..
.. ...
..
......
..
.. .. ...
..
.. ..
.. ... ...
.. ..
.. ..
..
.. .... ....
.. ....
.. ....
..
.. ....
.. .... ...
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.. .
.....
. .
. ...
. .. .
. ..
. . ...
....... .......
..
... ..
..
.. .. ... .. ... ... .. .. ... ..
.. ...
... ...........
.......
..
..
.......
... .. .. .... ... .... ... ... ..... ... .... ...
.......
.... .... ..
..
..
....
.. .... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. . . .. .
.. .
..
.. .
. .
. .
..
.
.............................................................................................................................................................................................................................................................................................................................................................
.....
.. .
..
.. ..
.. .. .. .. ..
.. ..
..
..
..
....
.. .. ... ... ..
.
... .
.
.
.
.. . . .. .
..
...
... .. .. .. ..
..
...
.... ... .....
.. ....
..
... ..
.
.
... ...
...

Consider the three functions (or signals) graphed above. The function f1 (x) has a clear
maximum and a \function width" or \pulse width" about the maximum where most of its
energy is concentrated. The same observation applies to the signal f2 (x), but its \pulse
width" is perhaps not as obvious. The oscillatory and somewhat wild signal f3 (x) possesses
no obvious maximum that we see in its graph and it is unclear how to dene any kind of
a pulse width for it.
Often our time-domain signals are one of the \linear" signals in an electric circuit, such as
a voltage v(t) or a current i(t). In a mechanical system, the displacements x(t) or velocities
v(t) are examples of one of the so-called \linear" signals that we might be interested in.
The instantaneous power associated with one of the linear signals f (t) is proportional to
jf (t)j2 . If the signal f (t) does possess a clear maximum and most of its energy or power 6
6 This

is an example of hazy (perhaps lazy, too) language. Power p(t) (in Watts) is not equal to energy
E(t) (in Joules) but we are approximately thinking that p(t) / E(t) / jf (t)j2 .
70

is concentrated around the maximum (such as the functions f1 and f2 above), then we
dene the half-power pulse width as
T1=2 = t2 t1
where

jf (t1;2 )j2
1
=
2
jfmax j
2

jf (t1;2 )j
1
=p :
jfmax j
2

When the signal magnitude is given or graphed in units of decibels (dB), the half-power
points t1;2 are also called the \3 (db)" points since
p
10 log10 (1=2) = 20 log10 (1= 2) 3:
The half-power pulse width is also then called the \3 (dB) pulse width." This measure of
\signal duration" applies equally well in the frequency domain, where the term bandwidth
is commonly used instead of pulse width. The symbol for bandwidth is typically BW in
communications work, but a simple B1=2 seems ne.
Example: Gaussian pulse. Note in this case that jf (t)j = f (t) and jF (!)j = F (!).
This will not be true in general.
r

F (!) =
exp(! 2 =4)
f (t) = exp(t2 )

..
..
... ... ... ...................... ... ... ...
.
max
.
. ... .....
.
.
... .... ....
.
.
.
.
...
..
....
.....
.....
max
.
... ........ ... ... ... ..... ... ... ............
.
....
.
.
. ..
..
.
.
.
.
.. .
...
... ....
...
.. .
.. ....
.. ...
..
.
... ....
.
.
.
..
.. ..
.
.. ...
.
.
..
...
.
..
..
..
.
..
.
.
.
...
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
.
...
..
.
.
.
.
.
.
...
.
.
..
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
....
..
.
.
.
.
.
.
.
......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....................................................................................................................................................................................................................................................
.
.
.
.
........................
.........................
1=2 .
.

..
..
... ... ... ... ... ... ... ... ...................................... ... ... ... ... ... ... ... ...
.
......
.
..
max
.
..
.
.
.....
.
...
.
..
....
..
....
.
.
.
.
.
.
.
...
...
....
.... ..
.. .....
.
max
.
... ... ............... ... ... ... ... ... ... ............ ... ... ... ... ... ................. ...
.
.
.
. ......
.
.
.
.. .
.
.
.
.
.
.
.
.. .
...
... ........
... ..
....
...
....
..
..
.....
....
..
.
.
.
.
....
.
.
.
.
.
..
.
.
.
.
...
.
.
.
.
.
..
....
.
.
.
.
.
.
.
.
....
..
.
.
.
.
.
.
.
....
.
.
.
.
.
.
.
....
.
.
.
..
.
.
.
.
.
.
.
.....
..
.
.
.
.
.
.
.
.....
.
.
.
.
.
.
.
.
.
......
.
.
.
.
.
...
..
..
....
..
..
..
.
.
.
..............................................................................................................................................................................................................................
.
..
... ..
...............................
1=2 ............................
.

jf j

jF j

jf j
p
2

T1=2

jF j
p
2

exp(t22 )
1
=p
1
2
r
ln 2
t2 =
2
r
2 ln 2
= t2 t1 = 2t2 =

exp(!22 =4)
1
=p
1
2
!2 =

p
2 ln 2

B1=2 = !2 !1 = 2!2 =

The \BT product" B1=2 T1=2 = 4 ln 2 is independent of .

p
8 ln 2

Example: The sinc function. Examine the two graphs of j sinc(x)j = j sin(x)=xj, the
one on the top using a linear ordinate scale and the one on the bottom using a logarithmic
(dB) ordinate scale. The abscissa is x, which can serve as a scaled frequency ! for the
spectrum of a time-domain rectangular pulse.
1.0
0.8
p
1= 2
0.6
jsinc(x)j

0.4
0.2

0.0
10
0
3
5
10
jsinc(x)j
(dB) 15
20
25

0
x

30
10
sin(x1 )
1
=p
x1
2

8
=)

6
x1 =

p
2 sin(x1 )

0
x

x1 = 1:391557378251510

If f (t) = (t=T ) is the standard rectangular pulse of pulse width T , then since the spectrum
is F (!) = T sinc(!T =2), we see that the half-power bandwidth is
B1=2 = 2!1

2 2 1:39
5:56

:
T
T
72

Homework

1. Find the pulse width, bandwidth, and BT product of these two signals:
f (t) = exp(jtj)

and

g(t) = exp(t)u(t):

Graph jf (t)j and its amplitude spectrum jF (!)j, and also jg(t)j and jG(!)j. Use at least
two values for the constant .

2. Compare the half-power pulse width T1=2 described above with the (statistical class)
of pulse width Dt that is used in the mathematical analysis of the Uncertainty Principle
that appears on page 74. If the signal is normalized to have unit energy
Z1
1

jf (t)j2 dt = 1

then the (square of) pulse width is

Dt2

Z1
1

t2 jf (t)j2 dt:

In other words, simply take Dt as dened here, to be yet another measure of a signal's
pulse width. For a specic signal, use the Gaussian pulse.

Uncertainty Principle for the Fourier Integral

To simplify our notation, assume that the signals under consideration are normalized to
have unit energy:
Z1
Z1
1
jf (t)j2 dt =
jF (!)j2 d! = 1:
2
1

If we further assume that both f (t) and its Fourier transform F (!) are centered about
t = 0 and ! = 0, respectively, then it is reasonable to dene the t and !-domain durations
by
Z1
Z1
Dt2 =
t2 jf (t)j2 dt
and
D!2 =
! 2 jF (!)j2 d!:
1

Our so-called \uncertainty principle" states:

lim

t!1

jtjf (t) = 0;

Then Dt D!

:
2

prf: Equality holds in Schwarz's inequality

2
b
Z

Zb
Zb

g1 g2 dt jg1 j2 dt jg2 j2 dt

only if g2 (t) = kg1 (t). Let g1 (t) = tf (t) and g2 (t) = f 0 (t):
1
2
Z

2
Z1
Z1

df
(t)
df
(t)
2

tf
(t)
dt
jtf
(t)j
dt

dt dt:

Integrate by parts

Z1
1

1
Z1
df
tf 2 (t)
1
1
tf dt =

jf j2 dt = :

dt
2
2
2
1
| {z 1}
| {z }
=0 by above
=1 by above

From the Fourier transform pair f 0 (t) () j!F (!) and Parseval's theorem,
Z1 2
Z1
df
1
dt =
j!F (!)j2 d!
dt
2

and so Schwarz's inequality gives

Z1
1

1
jtf (t)j2 dt
2
74

Z1
1

j!F (!)j2 d!

Dt2 D!2 :
2

Equality holds if
df (t)
= ktf (t):
dt
The solution of this dierential equation is
f (t) = Cekt

Let k = 2; this is the inescapable Gaussian function

f (t) =

1=4

() F (!) =

1=4 r

!2 =4
=
e

1=4

Theorem (Riemann-Lebesgue Lemma). Let f (x) 2 L1 (1; 1). That is, f (x) is
absolutely integrable
Z1
jf (x)j dx < 1:
1

Then the integrals

f (x) cos x dx;

f (x) sin x dx;

tend to zero as ! 1.
Proof. Consider the cosine integral. Let be a given positive number. Then we can choose
X so large that
X
Z1
Z
jf (x)j dx < ;
jf (x)j dx < :
1

Hence

1
Z

f (x) cos x dx < ;

X
Z

<
f
(x)
cos
x
dx

for all values of . Next, we can dene a function (x), absolutely continuous in the
interval (X; X), such that
ZX
jf (x) (x)j dx < :
X

Then

X
Z

ff (x) (x)g cos x dx <

for all values of . Finally

ZX
X

(X) sin X
(X) sin X
1
(x) cos x dx =
+

0 (x) sin x dx;

and (for a xed X) we can choose 0 so large that the modulus of this is less than for
> 0 . Then
1

f (x) cos x dx < 4

( > 0 ):

This proves the theorem for the cosine integral; a similar proof applies to the sine integral.

The above is the working-class proof of the Riemann-Lebesgue lemma, as given by E.C.
Titchmarsh in Introduction to the Theory of Fourier Integrals, 1937. Perhaps you will
prefer this one, from R.R. Goldberg, Fourier Transforms, Cambridge University Press,
1962.
Definition (the Class Lp ). Suppose 1 p < 1. The function f on (1; 1) is said
R1
to be of class Lp (written f 2 Lp ) if
jf (x)jp dx < 1. If f 2 Lp then kf kp is dened to
be

11=p

jf (x)jp dxA

The symbol kf kp is read as the Lp norm of f .

Theorem (Riemann-Lebesgue). If f 2 L1 then
Z1

lim F (!) = lim

!!1

!!1
1

Proof. Since

F (!) =

ej!t f (t) dt = 0:

ej!t f (t) dt;

(1)

then
F (!) =

ej![t+(=!)] f (t) dt =

Z1
1

ej!t f (t =!) dt:

(2)

Subtracting (2) from (1) we obtain

2F (!) =

Z1
1

Hence
2jF (!)j
But since f 2 L1 ,
lim

!!1
1

ej!t [f (t) f (t =!)] dt:

jf (t) f (t =!)j dt:

(3)

jf (t) f (t =!)j dt = 0

(4)

by the continuity-in-the-mean theorem. The theorem follows from (3) and (4).
77

Riemann-Lebesgue Lemma
If f 0 (t) is absolutely integrable, then F (!) ! 0 as ! ! 1.
PROOF: Integrate by parts
F (!) =

Z1
1

j!t

f (t)e

1
Z1
f (t)ej!t
1
dt =
+
f 0 (t)ej!t dt

j!
j!
t=1
1

If f (1) is nite, then the boundary terms go to zero as ! ! 1. The integral is the
Fourier transform of the derivative of f (t)
Z1
1

f 0 (t)ej!t dt = F [f 0 (t)]

and this Fourier integral exists (has nite magnitude jF [f 0 (t)] j < 1) if the derivative is
absolutely integrable
Z1
jf 0 (t)j dt A < 1:
1

If jF [f 0 (t)] j < 1, then the integral term

F [f 0 (t)]
! 0:
j!
!!1
Exercise. Examine the high frequency limiting behavior (! ! 1) of the spectra of
these signal pairs and interpret in terms of the Riemann-Lebesgue Lemma:
F

u(t) ()

1
+ (!)
j!

(t=T ) () T sinc(!T =2)

F
sinc(t) () (!=2)

1
F
et u(t) ()
+ j!
F

(t) () 1
1 F
() sgn(!)
t
j

Convergence of the Fourier Integral Representation

F (!) = F[f (t)] =

j!t

f (t)e

()

f (t) = F

1
[F (!)] =
2

F (!)ej!t d!

Assume f 0 (t) is absolutely integrable so that the Riemann-Lebesgue lemma guarantees

lim F (!) = 0:

!!1

Denote the inverse Fourier transform of the Fourier transform as

1
fe(t) = lim
!1 2
= lim

Z Z1

f ( )ej! d ej!t d!

!1
1
Z1

1
f ( )
2

sin[( t)]
d
( t)

= lim

f ( )

= lim

f (x + t)

!1
1
Z1
!1
1

ej!(t ) d! d

sin x
dx:
x

If f is continuous at t, then clearly fe(t) = f (t) if we accept the Dirac-delta representation

sin x
= (x):
!1 x
lim

If f suers a jump discontinuity at t, in that

lim f (t + x) = f (t ) 6
= f (t+ ) = lim+ f (t + x);

x!0

then it is appropriate to break the integral above at x = 0 so that

fe(t) = lim

!1
1

= lim

!1
1
Z1

sin x
f (x + t)
dx + lim
!1
x

f (x + t)

sin x
dx
x

f (x + t) f (t )
sin x dx + lim f (t )
!1
x

sin x
dx
x

f (x + t) f (t+ )
sin x dx + lim f (t+ )
!1
x

+ lim

Z1
0

sin x
dx
x

The function

f (x + t) f (t )
x

is continuous at x = 0 , and is well-behaved everywhere else by our initial assumption.

(We are assuming that f is continuous everywhere except at the particular t of interest. If
f actually has a nite number of such jump discontinuities, then each one can be handled
in a similar fashion.) By the Riemann-Lebesgue lemma, the rst term on the right-hand
side of the above integral vanishes as ! 1. The same argument applies to the function
f (x + t) f (t+ )
x
and its Fourier integral. From the famous integral
Z1

sin u
du =
u

and its variation

sin u
du = sgn ;
u

then for > 0 we have

Z0
1

sin x
dx =
x

sin x
dx = 12 :
x

Therefore, we have shown

F 1 fF [f (t)]g = fe(t) =

1
2

f (t ) + f (t+ ) :

System Transfer Function

x(t)

h(t)

y(t)

The zero-state response (in the time domain) of our linear, time-invariant (LTI) system is
the convolution of the input signal with the system impulse response
y(t) = x(t) ~ h(t):

(1)

The direct Fourier transform of this equation is

Y (!) = X(!)H(!)

(2)

which exhibits the important role of the system transfer function

Ffh(t)g = H(!):

(3)

By transforming to the frequency domain, the nontrivial convolution operator has been
replaced by a simple multiplication! Besides algebraic simplicity, working in the frequency
domain also gives us a whole another conceptual view of the input/output dynamics of our
systems. Electrical engineers work so much in the frequency domain that we sometimes
loose touch with the original time domain!
For an arbitrary input signal x(t), the time-domain response y(t) now requires an inverse
Fourier transform
(4)
y(t) = F 1 fX(!)H(!)g:
However, the transfer function H(!) immediately gives the response of a LTI system to a
time-harmonic signal. The simplest time-harmonic input signal is the complex exponential
form
(5)
x(t) = ej!0 t :
(The only other choice would be a real-valued cosine or sine, but as we have seen repeatedly
this semester, the mathematics of signal analysis usually (always?) is much simpler in terms
of the complex exponential form. And in deriving important concepts and principles, we
always take the simpler path.) Note the use of !0 to denote some particular frequency, to
avoid confusion with the Fourier transform frequency variable plain !. The response of
our system to the excitation (5) is
y(t) = h(t) ~ x(t) =

= ej!0 t

Z1
1

h( )x(t ) d =

Z1
1

h( )ej!0 d = ej!0 t H(!0 ):

h( )ej!0 (t ) d

(6)

Observe that the response (steady-state since the excitation has been on forever) of a
LTI system to a time-harmonic signal is a time-harmonic signal of the same frequency as
the excitation. Only the amplitude and phase of the response have been modied by the
amplitude and phase of the system transfer function, evaluated (of course) at the excitation
frequency. The amplitude and phase are apparent in the polar representation
H(!) = jH(!)jej(!) :

(7)

Note that if the system impulse response h(t) is a real-valued function of time, then we
can recover the separate responses to the real and imaginary parts of a complex input by
inspection:

x(t) = xr (t) + jxi (t)

y(t) = xr (t) ~ h(t) + jxi (t) ~ h(t)

Refx(t)g = xr (t)

h(t) real

yr (t) = Refy(t)g

Imfx(t)g = xi (t)

yi (t) = Imfy(t)g

h(t)

h(t) real

F Exercise 1. Show that the time-harmonic response of a real LTI system to a cosine
excitation x(t) = A cos(!0 t + ) of arbitrary amplitude A and phase is
y(t) = AjH(!0 )j cos[!0 t + + (!0 )]
where the amplitude and phase of the transfer function are dened in (7).
Hint: write x(t) = RefAej ej!0 t g. In AC circuit analysis, we call Aej the complex phasor.
F Exercise 2. Specialize the above result to the case where x(t) = sin(!0 t):
F Exercise 3. Now consider the periodic excitation
1
X
x(t) =
cn ejn!0 t
n=1

applied to a LTI system having arbitrary (not necessarily real) impulse response h(t). The
transfer function is the generic H(!). What is the form of the response? Think about the
linearity of the system.
82

Superposition Interpretation of System Transfer Function

x(t) =

1
2

j!t

X(!)e

H(!)

1
2

H(!)X(!)ej!t d! = y(t)

.................................................
...
...
..
.. H(! )X(! )ej!1 t
...
1
1
...
.
.
.
.
.
.............................................................................................
.
...............................................
...
..
...
.
....
....
....
...
..
.
..............................................
...
...
...
...
.................................................
...
.
.
...
..
j!2 t
j!2 t ....
.
X(!2 )e
. H(!2 )X(!2 )e
....
....
.
...
..........................................
......................................................
.....
...
...
...
.....
...
..
...
.
.
.
.
.
...
...
.....
.
.
.
.....
..............................................
....
.....
...
.....
.....
..........
.....
....... ................
.................................................
......
..............
...
...
....
...
...
..
..
...
...
...
..
j!n t
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
....................................................................... ..
............................................ .
.........................................
...
...
....
...
...
..
..
.
.
.
......
...
...
.
...... ...
.....
................................................
....... ....................
.....
.......
.....
.
.
.
.
.
.
.....
...
....
...
..................................................
.....
..
..
..
.....
.
.
.
..
...
..
.
.
.
.
.
.
...
.
.
.
.
.....................................................
..
............................................
..
..
..
...
...
...
..
..
..
..
................................................
..
..
...
..
.................................................
...
.
.
.
j!n t ...
j!n t ...
..
.
.
X(!n )e
H(!
)X(!
)e
.
n
n
.
.
...
..............................................................................................
.....................................................
...
...
..
..
.
...
................................................

X(!1 )ej!1 t

H(!)

x(t) =

P
n

X(!n )e

H(!)

P
n

H(!n )X(!n )ej!n t = y(t)

Transfer Function for Linear Electric Circuits

Example.

x(t)

y(t)

Recall (look back in your notes; no one memorizes such specics) that the impulse response
of the given RL circuit is

R
R
h(t) = (t) exp t u(t):
(8)
L
L
Most of us would agree that the correct derivation of this time-domain result is a nontrivial
exercise. We rst found the step response s(t) and then dierentiated to get h(t) = s0 (t).
Fortunately, the Fourier transform allows us to circumvent the time-domain dierential
equations and work with purely algebraic equations. Temporarily set the input x(t) = vg (t)
and the output y(t) = vL (t) for obvious dimensional consistency and to use signal labels
for voltages that look like voltages. At least this helps my circuit analysis.
.
.
.
...... ...... ......
..
.
.
.......................................................................................... ..... ..... ..... ..... ..... ...............................................................................
....
....
....
..
..
..
..
..
.
.....
...
.
..
......................
.........
....
...
.
...
.
...
.
.
....
.......
...
..
..
L
.
.
...
...
......
.........
..................
...
....
.
.......
..
..
..
.
...
........................................................................................................................................................................................................

i(t)

vg (t)

v (t)

Kirchho's voltage law is

vg (t) = Ri(t) + L

di(t)
dt

(9)

and the direct Fourier transform of this is

Vg (!) = [R + j!L]I(!):

(10)

Note the natural occurrence of the familiar complex impedance. The ratio of output to
input voltage spectra
VL (!)
j!L
ZL
=
=
(11)
Vg (!)
R + j!L
ZR + ZL
could have been written by inspection as a simple voltage divider. Our transfer function is
Y (!)
j!L
j!L=R
j!=!b
H(!) =
=
=
=
(12)
X(!)
R + j!L
1 + j!L=R
1 + j!=!b
where the constant !b = R=L (s1 ) is called the break frequency or half-power frequency
or maybe even cut-o frequency of this simple high-pass lter.
F Exercise 4. Verify that the Fourier transform of impulse response (8) gives transfer
function (12).
84

0
5
10
15
jHj (dB)
20
25
30
35

40
102

101

100
!=!b

101

102

(a) Magnitude response.

3=8
arg H
=4

102

101

100
!=!b

101

(b) Phase response.

Fig. 1. Frequency behavior of single-pole high-pass lter.

102

The magnitude and phase response of the transfer function (12)

j!=!b
(13)
1 + j!=!b
are graphed versus a logarithmic and normalized frequency abscissa in Fig. 1. We also use
a logarithmic (unnatural or base 10) scale to express the magnitude in decibels according
to
jHj (dB) = 20 log10 jHj:
(14)
H(!) =

Note that the low and high frequency approximations

8
j!=!b ; ! !b
>
>
<
j!=!b
j
H(!) =
=
(15)
; ! = !b
1 + j!=!b >
1+j
>
:
1;
! !b
clearly show that this lter passes the high frequencies unscathed. In the low frequency
regime (! !b ) where this lter has the greatest eect between the input and output,
the frequency-domain relationship is
j!X(!)
Y (!)
:
(16)
!b
Inverse Fourier transformation of (16) to the time domain gives
1 dx(t)
(17)
!b dt
so we see that a high-pass lter acts as a dierentiator when it is having its biggest eect
on the input.
y(t)

F Exercise 5. Consider the simple circuit having the positions of R and L reversed
relative to the circuit above. Find its transfer function and show that this low-pass lter
acts as an integrator when it is having its greatest eect on the input signal.
Homework Problem: Rectangular Pulse Response of Ideal Low-Pass Filter.
Consider an ideal low-pass lter of bandwidth B and subsequent transfer function
!
H(!) =
:
2B
Find the transient response y(t) of this lter to a rectangular pulse of pulse-width T
x(t) = (t=T ):
Graphical results simplify with normalized time t=T as the independent variable, for various
values of the dimensionless parameter BT . This problem is most readily solved in terms
of a special function, the sine integral7
Zu
sin
Si(u) =
d
Si(u) = Si(u)
Si(1) = =2:

0
7 M.

Abramowitz and I.A. Stegun, Handbook of Mathematical Functions. National Bureau of Standards,
1970, pp. 231{232, Table 5.1. (\Big Red"or \Citizen's Handbook"). MATLAB has it built-in, too!
86

Parseval's Theorem for the Fourier Integral

The energy associated with the signal f (t) is
E=

Z1
1

jf (t)j2 dt =

Z1
1

f (t)f (t) dt
2

1
4 1
F (!)ej!t d! 5 4
F ()ejt d5 dt
2
2
1
1
1
2
32
3
1
1
Z
Z
Z1
1
4 1
=
F (!)ej!t d! 5 4
F ()ejt d5 dt
2
2
1
1
1
8
9
1
1
1
Z
Z
Z
<
=
1
1
=
F (!)
F ()
ej(!)t dt d d!
: 2
;
2
=

1
=
2
=

1
2

1
Z1
1
Z1
1

1
Z1

F (!)

1
F ()( !) d d! =
2

F (!)F (!) d!

jF (!)j2 d!

F Exercise 6. Derive the generalized Parseval theorem

f (t)g (t) dt =

F Exercise 7. Find a simple evaluation for the integrals

sin x
dx
x

sin x
x

dx:

We met the rst one back on page 54. The similarity (more than similar! ) in the numerical
values strikes me as a strange coincidence. What do you think? (You'll need the answers,
rst!)
87

Other Conventions for the Fourier Transform Pair

Note that our forward Fourier transform could be scaled by any constant K as long as we
divide by K in performing the inverse:
Z1
Z1
1
F
j!t
F (!) = K
f (t)e
dt
()
f (t) =
F (!)ej!t d!:
2K
1
1
p
The common choice K = 1= 2 gives a pleasingly symmetric pair
Z1
Z1
1
1
F
j!t
F (!) = p
f (t)e
dt
()
f (t) = p
F (!)ej!t d!:
2
2
1

One-sided (unilateral) trigonometric forms

Fc (!) =

Fs (!) =

0
Z1

f (t) cos(!t) dt

f (t) sin(!t) dt

()
F

()

2
f (t) =

f (t) =

Fc (!) cos(!t) d!

0
Z1

Fs (!) sin(!t) d!

Note that the sine transform is a natural choice if f (0) = 0, while the cosine transform ts
the case when f 0 (0) = 0.
The use of frequency f (Hz=cycle/s) instead of ! (1/s)
Introduce the frequency f (Hz) such that ! = 2f , and our usual Fourier transform pair
Z1
Z1
1
F
j!t
X(!) =
x(t)e
dt
()
x(t) =
X(!)ej!t d!
2
1

becomes
X(f ) =

Z1
1

x(t)ej2f t dt

()

x(t) =

Z1
1

X(f )ej2f t df:

This form has the advantage that both variables t and f appear symmetrically without
any multiplying factors outside the integrals. Its disadvantage (to me) is the appearance
of a 2 in every kernel function exp(j2f t). Its supporters are often experimentalists,
because of their natural preference to measure frequency f in (Hz), as opposed to the
natural frequency !. The use of f can also pave the way for a slightly smoother transition
to the discrete-time world. Either way you look at it, the Fourier transform pair has
to have 2's somewhere: The question is, do you place them as harmless scaling factors
outside the integrals, or do you turn them loose as pesky irritants inside the arguments
of the complex exponentials, so that every dierential and integral operation results in a
factor of 2 or (2)1 ? Seriously, the world is divided and you will inevitably need to
translate between the two forms, but not in our course, where, at least for now, ! rules!
88

ECE 370

QUIZ 7

9 October 1940

Name

Ig
or Reiru
of, III

Take-Home: due next class, Wednesday October 31 12:00 noon

Consider a real-valued function f (t) that possesses a denite Fourier transform F (!).
Assume that f (t) is absolutely integrable so that we are assured of the niteness of F (!)
for all frequencies.
1.) If f (t) = f (t) is even, what can you deduce about the symmetry (even/odd parity)
of F (!)? Is F (!) pure real, pure imaginary, or arbitrarily complex?
2.) If f (t) = f (t) is odd, what can you deduce about the symmetry (even/odd parity)
of F (!)? Is F (!) pure real, pure imaginary, or arbitrarily complex?
3.) What is the physical (geometric) signicance of F (0)?
Z1

F (!) =

f (t)ej!t dt

SOLUTION

F (!) =

Z1
1

f (t) cos(!t) dt j

Z1
1

f (t) sin(!t) dt = Fr (!) jFi (!)

If f (t) = f (t) is even, then Fi (!) = 0 and Fr (!) = 2

f (t) cos(!t) dt = Fr (!):

That is, the spectrum of an even, real signal is pure real and is an even function of !.

If f (t) = f (t) is odd, then Fr (!) = 0 and Fi (!) = 2

f (t) sin(!t) dt = Fi (!):

That is, the spectrum of an odd, real signal is pure imaginary and is an odd function of !.

F (0) =

f (t) dt is the integral of f (t); the DC value, total area under the curve.

Cascade of Two LTI Systems

x(t)

h (t)

h(t) = h1 (t) ~ h2 (t)

H(!) = H1 (!)H2 (!)

Linear Dispersionless Filter

x(t)

h(t) = (t t )

H(!) = ej!t0

y(t) = x(t t0 )

\linear phase"

y(t)

Homework Problem: Rectangular Pulse Response of Ideal Low-Pass Filter.

Consider an ideal low-pass lter of bandwidth B and subsequent transfer function
!
:
(1)
H(!) =
2B
Find the transient response y(t) of this lter to a rectangular pulse of pulse-width T
x(t) = (t=T ):

(2)

Graphical results simplify with normalized time t=T as the independent variable, for various
values of the dimensionless parameter BT . This problem is most readily solved in terms
of a special function, the sine integral8
Si(z) =

sin
d

Si(z) = Si(z)

Si(1) = =2:

(3)

x(t)

h(t)

y(t)

The lter impulse response is

h(t) = F 1 [H(!)] =

B
sinc(Bt)

(4)

which can be found from your table of Fourier transforms or by direct evaluation of the
inverse transform integral. Similarly, the spectrum of the input signal is
X(!) = F[x(t)] = T sinc(!T =2):

(5)

At least three dierent, but related, approaches to nding the output y(t) are worth
considering.
Time-Domain Convolution.
y(t) = x(t) ~ h(t) =

Z1
1

B
=

T =2
Z
T =2

x( )h(t ) d =

sin[B(t )]
B
d =
B(t )

T =2
Z

h(t ) d

(6)

T =2
T =2
Z
T =2

8 M.

sin[B( t)]
1
d =
B( t)

B(T
Z=2t)
B(T =2t)

sin
d

(7)

Abramowitz and I.A. Stegun, Handbook of Mathematical Functions. National Bureau of Standards,
1970, pp. 231{232, Table 5.1. (\Big Red"or \Citizen's Handbook"). MATLAB has it built-in, too!
91

1
Si[B(T =2 t)] Si[B(T =2 + t)] =
Si[B(t + T =2)] Si[B(t T =2)]

(8)

1
t
1
t
1
=
Si BT
+
Si BT

(9)

T
2
T
2

y(t) =

Note that y(t ! 1) = 0. Also note that time appears normalized to the incident pulse
width T as in t=T , and that the single (dimensionless!) parameter that characterizes the
lter response to the rectangular pulse is the product BT .

Frequency-Domain Multiplication.

y(t) = F
1
=

1
[X(!)H(!)] =
2

j!t

X(!)H(!)e

sin(!T =2) j!t

e d!
!

T sinc(!T =2)ej!t d!

only the even part of ej!t contributes

1
=

1
d! =
2

sin(!T =2)
2
cos(!t) d! =
!

sin(!T =2)
cos(!t) d!
!

(10)
(11)

(12)

Use the trig identity 2 sin X cos Y = sin(X Y ) + sin(X + Y ) so that

1
y(t) =

d!
:
sin[!(T =2 t)] + sin[!(T =2 + t)]
!

(13)

Let = !(T =2 t) so that

d!
sin[!(T =2 t)]
=
!

B(T
Z=2t)

sin
d = Si[B(T =2 t)]

(14)

whereupon (13) is

1
Si[B(T =2 t)] + Si[B(T =2 + t)]

1
=
Si[B(t + T =2)] Si[B(t T =2)]:

y(t) =

(15)

Linear Combination of Step-Responses. Recognize that our particular input can be

written as
x(t) = (t=T ) = u(t + T =2) u(t T =2)
(16)
so that the response is
y(t) = s(t + T =2) s(t T =2)

(17)

where the step response is

Zt
ZBt
B
1
sin
h( ) d =
sinc(B ) d =
s(t) =
d

1
1
1
2 0
3
Bt
Z
Z
i 1
sin
sin 5 1 h
14
1
=
d +
d =
+ Si(Bt) = + Si(Bt):

2
2
1

(18)

(19)

The linear combination (17) of shifted step responses now yields

y(t) =

1
Si[B(t + T =2)] Si[B(t T =2)]:

(20)

The MATLAB routine sinint.m calls the symbolic toolbox, built upon MAPLE, and does
an extremely slow and inecient numerical integration to compute the function Si(x).
It doesn't work at all on my computer, so here is an alternate computation that uses
MATLAB's routine expint.m for the exponential integral, a dierent but related special
function.
E1 (z) =

et
dt
t

Si(x) =

+ ImfE1 (jx)g
2

The routine for E1 (jx) will not work for (real) values of x 0, but that's ok since we know
that Si(0) = 0 and it is an odd function Si(x) = Si(x). Therefore Si(x) = Si(jxj) sgn(x).

function f=sinintRWS(x)
% f=sinintRWS(x) sine integral Si(x) for real x
f=zeros(size(x));
k=find(x=0);
f(k)=(pi/2+imag(expint(i*abs(x(k))))).*sign(x(k));

1.2

1.0

dashed curve: BT = 10
solid curve: BT = 50

0.8

0.6
y(t)
0.4

0.2

0.0

0:2
2:0

1:5

1:0

0:5

0.0

0.5

1.0

t=T

Fig. 1. Rectangular pulse response of ideal low-pass lter.

1.5

2.0

Ideal Band-Pass Filter

H(!)
....
..
..
..
...
...
...
...
...
...
..
..
..
..
...
...
...
...
..
..
..
.
..
..
.
.
.
.
.
.. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
............................................
.
.
.
.
.
.
.
.
.................................... .
.................................... ...
.................................
.... ...
..
..
..
....
...
..
..
..
..
...
...
...
...
..
..
.
..
..
.
..
..
.
.
.
.
.
.
.
...............................................................................................................................................................................................................................................................................................................................................................................
.
.

H(!) =

! !0
B

! + !0
B

f (t)

B
sinc
2
e

F (!)

Bt
2

!
B

F (! ! )

f (t)

B
sinc(Bt=2) ej!0 t + ej!0 t
2
B
= sinc(Bt=2) cos(!0 t)

h(t) =

h(t)
...
...
......
.
....
.......
....
........ ............ .........
...... .......... ......
.
.
0
.. ... ........ .. ...
.....
.....
...... .. ... ....... .. ... .....
...... .. .. ........ .. .. .....
.... .... ..... .... .............. ..... ..... ..... ....
. .. . .. ... .. . . ..
... .. ... ... ... ...... .. ... .. ... ...
...... .. .. ... ... ........ ... ... .. .. ......
....... ... .... .... .... ......... ... .... ... .... .........
. .. . . . . ... .. .. . . . ..
... ... .. ... .. ... .. ... .. .. .. ... ... .. ...
... .... ... .... .... ... ... .... .... ... .... ..... ... .... .... ..
.
.
.
.
.
.
.
... . .. .. ... .. ... .. .. .. .. ... .. ... .. ... .....
.....
...
...... ......
... ... .. ... ... ... .. .. .. .. .. .. .. ... .. ... .. ... ...
.. .. .. .....
... ... ... ... ......
... .... .... ..... ... .... ... .... ... .... ... ... .... ..... .... .... ..... ..... .....
........ .... ..... ..... ..... ....
..... ......... ...
.
.
....... ......
.
.
.
.
...... ..... ..... ..... ..... .... ..... ....
.
... . .. .. ... .. .. . . .. .. .. .. .. . .. .. ..
... .. .. .. ...
... ... .. ... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....... ... ... ... ... .......
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
.
.. ... .. .. .. .. .. .....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... .. ... .. .. .. .. .
...... ... .. .... ... .... ..
...... ..... ..... .......... .
...... .... .. ... .. ... .. .. .. .. .. .. .. .. .. .. ... ... ......
..... ... ... .......
...... ... ... ... ... ....
.
.
... ... .. .. .....
. .. .. .. . .. .. .. .. .. .. . ... ... .. .. ... .. ..
...
...
... ... .. .....
... ... .. .. .. .. .. . .. .. .. .. .. .. .. .. ...
...... ....... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
..
... ... ... .. .. .. .. .. .. .. .. .. ... .. .. ... ...
... ... ... .. ... .. ... .. .. .. .. ... ... ... .. ... ...
...... ... .. ... ... .. .. ... .. .. .. ... .. ... ......
.... .. ... ... ... .. .. .. .. .. ... ... ... ... ....
... .. .. .. .. . . .. .. .. .. . ..
... ... ... ... ... .. ... ... .. .... ... .... ...
.. ... ... ... ... ... ... .. ... ... .. ... ....
..... .. . ... .. .. .. .. .. .. ....
...... .. ... .. ... .. ... ... ... . .......
... ... . ... ... .. ... .... .. .. ...
.. ... .. ... ... .. .. .. ...
..... .. . . .. .. .....
..... .. ... ... ... . .....
...... ..... .. ..... ......
...
....
....... ... .....
...... .. ......
..... ... .....
. .. ..
..
...

! = 100, B = 20

Complex Phasors for Time-Harmonic Circuit Analysis. Consider the sinusoidalsteady-state response of the given RLC circuit, that is nd vo (t) if vi (t) = Vp cos(!0 t + ).
The time-domain integro-dierential equation (Kirchho's voltage law) that is appropriate
for this network is
Zt
di(t)
1
L
+
i( ) d + Ri(t) = vi (t)
dt
C
1

and the output voltage is simply a scaled version of the mesh current, that is vo (t) = Ri(t).
...
...
..
..
.................................................
.................................................................................
........................................................................ ... ... ... ...................................
.....
...
..
...
...
..
...
...
...
..
..
...
...
...
...
.
...
.. ..........
......... ..
..........
........ ................
.
.
.
.
.
.
.
.
.............
.. .. ...
...
........
.. ... ...
...
.... .. .... ... ....
......
.
.............
... .. ..
.
i
o
..
......
... ... ..
.
.
.
.... ......
.
......
...
.......
............
.....................
..
...
...
...
..
..
...
..
.
........................................................................................................................................................................................................................................................
......
......

i(t) +

v (t)

(1) Solve for vo (t) by representing the time-domain signals in terms of their Fourier
transforms.
(2) Show that the Fourier transform analysis is equivalent to, and is therefore the motivation for, the (simple and compact!) complex phasor analysis of time-harmonic
circuits such as this one. Recall that the time-harmonic signal i(t) is represented
in terms of the complex phasor I via
i(t) = RefIe+j!t g:
The Fourier transform of the input or generator voltage is

Vp j(!0 t+)
j(!0 t+)
Vi (!) = F[vi (t)] = F[Vp cos(!0 t + )] = F
+e
e
2

Vp
=
2ej (! !0 ) + 2ej (! !0 ) :
2
The Fourier transform of the entire (KVL) equation above (both the LHS and RHS) is

1 I(!)
j!LI(!) +
+ I(0)(!) + RI(!) = Vp ej (! !0 ) + ej (! + !0 ) :
C j!
Physically argue that I(0) = 0 here, since we expect zero steady-state DC current through
the capacitor, whereupon

1
R + j!L +
I(!) = Vp ej (! !0 ) + ej (! + !0 ) :
j!C
|
{z
}
Z(!)

Now solve this algebraic equation for the unknown spectrum

ej (! !0 ) + ej (! + !0 )
:
I(!) = Vp
Z(!)
Fourier inversion proceeds as
1
i(t) =
2

Z1
1

I(!)ej!t d!
Z1

Vp 4 j
(! !0 ) j!t
(! + !0 ) j!t 5
e
e d! + ej
e d!
2
Z(!)
Z(!)
1
1

j!0 t
j!0 t
e
Vp j e
=
e
+ ej
note Z(!0 ) = Z (+!0 )
2
Z(!0 )
Z(!0 )

1 Vp ej j!0 t Vp ej j!0 t
=
e
e
+
2 Z(!0 )
Z (!0 )

1 Vp ej j!0 t
=
e
+ c.c.
an expression plus its complex conjugate
2 Z(!0 )

Vp ej j!0 t
e
= Re
= Re Iej!0 t
Z(!0 )
=

where the complex phasor

Vp ej
Vi
=
Z(!0 )
Z(!0 )

introduced itself! If you still want it,

Vo = RI

and

vo (t) = Ri(t):

It looks like the two steps alluded to in the problem statement are more naturally done
together.

Homework Problem on Uncertainty Principle

Consider the signal
1
x(t) = p (t=T )
T
where T > 0 and the total energy in x(t) is

Ex =

Z1
1

jx(t)j2 dt = 1:

(The multiplying factor T 1=2 ensures that Ex is independent of T .) If x(t) is the input
voltage to the given electric circuit, with xed values of R and L, consider the energy in
the output voltage y(t)
Z1
jy(t)j2 dt:
Ey =
1

Again, note that changing T has no eect on the total input energy. If T is increased, will
the output energy Ey increase, decrease, or remain the same? Explain.
........................................................................
...
....
....
....
............................................................................
............................................
..
....
...
..
.....
......
....
........
.............
.......
.......
...
........
.............
.......
......
....
........
.............
...
..
...
....................................................................................................................................................................................................

x(t)

y(t)

Approximate Analysis of Dispersion

x(t)

y(t)

H(!)

Fig. 1. System Block Diagram

If the modulating signal or envelope of a high frequency sinusoid of frequency !0 is

the baseband signal a(t), having bandwidth considerably less than !0 (say 10%), then the
spectrum of the the modulated carrier
x(t) = a(t) cos(!0 t)

(1)

X(!) = 12 [A(! !0 ) + A(! + !0 )]:

(2)

is
An example spectrum is shown in Fig. 2.
jX(!)j

Fig. 2. Typical Narrow-Band Spectrum

For real signals x(t) and h(t), the spectra satisfy X(!) = X (!) and H(!) = H (!).
The response of the system is the inverse Fourier transform of X(!)H(!), with the positive
and negative frequency contributions written separately as

1
y(t) =
2

1
2

Z1
0

1
2

j!t

X(!)H(!)e

=
=

1
2

0
Z1

1
d! +
2

X(!)H(!)ej!t d! +

1
2

X(!)H(!)ej!t d!

1
Z1

1
2

X()H()ejt d

X(!)H(!)ej!t d! + c.c.

0
1
2 [y+ (t)

+ y+
(t)]

(3)

The signal y+ (t) is called the pre-envelope of y(t), and consists of positive frequencies only
Z1
1
y+ (t) =
[A(! !0 ) + A(! + !0 )]H(!)ej!t d!
2
0

Z1
1

1
A(! !0 )H(!)ej!t d! =
2

A()H( + !0 )ej(+!0 )t d

(4)

where A() is the original, baseband or low-pass signal.

The phase or argument of the transfer function H(!) in a neighborhood of the carrier
frequency !0 is of most interest. Write
H(!) = exp[j(!)]:

(5)

A Taylor series about !0 of the phase function

(!0 + ) = (!0 ) + 0 (!0 ) + 12 2 00 (!0 ) + : : :

(6)

gives the approximate transfer function

H( + !0 ) ej(!0 ) ej (!0 ) :

(7)

An approximation for the pre-envelope (4) is therefore

Z1
0
1 j[!0 t+(!0 )]
y+ (t)
e
A()ej[ (!0 )+t] d
2
1

= ej[!0 t+(!0 )] a[t + 0 (!0 )]

(8)

and so the nal approximation for the output of the dispersive lter (5) is
y(t) a[t + 0 (!0 )] cos[!0 t + (!0 )]:

(9)

The phase delay (of the carrier) is due to (!0 ) and the group delay (of the envelope) is
due to 0 (!0 ).
100

Numerical Example. The phase function of a length of waveguide transmission line is

p
(!) = ! 2 !c2 t0
where !c is the cuto frequency of the guide and t0 = d=c where d is the length and c is
the free-space phase velocity.
If !c = 0, then 0 (!0 ) = t0 .
If !c = 25, !0 = 30, t0 = 3, then
(30)(3)
0 (!0 ) = p
= 5:43
302 252
The following gures are the result of an inverse FFT approximation for the signal
2

x(t) = et cos(!0 t)
applied to the input of the above dispersive waveguide section. The 4096 point FFT uses
! = 0:3.

101

1.0

0.5

y(t)

0.0

0:5
1:0

.
. .. .
.. .... ..
.... ..... .....
... ...... ...... ...... .....
.. ... ... ... ..
..... ..... ..... ..... .....
. .... .... .... .... .... .
..... ...... ....... ....... ....... ....... ......
.. .. .. .. .. .. ..
c
0
0
..... ..... ..... ..... ..... ..... .....
. .... .... .... .... .... .... .... .
...... ....... ...... ...... ...... ...... ...... ...... ......
...... ..... .... ..... ..... ...... ..... ..... .....
.. .. ... .. .. .. .. .. ..
... ..... ..... .... ..... ..... ..... ..... ..... ..... ...
...... ....... ...... ....... ....... ....... ....... ....... ...... ....... .......
.
.
.
.
.
... .. ... ..... ..... ..... .... .... .... .... ... .... ...
..... ..... ..... ..... .... .... ...... ..... ..... ..... ..... .... .....
..... ....... ......... ....... ....... ........ ........ ........ ....... ....... ....... ....... ........ ....... .......
.
... ... .. .. .. . . .. . .. . . . .. . . ... .. .... ... .. ...
......................................................................................... ... .... ... .... ... ... .... .... ... .... ... .... ... ... ........ .. .......................................................................................................................................................................................
. .... ..... ..... .... ..... .... .... ..... ..... .. .. .... ..... .... .... .... ..... ..
.. .... ... .... .... .... .... ... ... .... .... ... .... .... .... .
..... .... ... .... ... .... .... .... ... ... ... ... .... ....
.... ... ... ... .. ... ... .... .... .... .... .... .
.... .... ... ... ... ... ... ... ... ... ... ....
.... ..... ..... ...... ...... ...... ...... ...... ...... ...... ..... ....
..... .... ..... ..... ..... ..... ..... .... .... ....
..... ..... .... .... .... ..... ..... ..... ..... ......
... ..... ..... ..... ..... ..... ..... ..... ..... ...
..... ...... ...... ...... ...... ..... ...... ......
..... ..... ..... ..... ..... ..... .... .....
... .... .... ..... .... .... .... ....
. ..... .... ..... ..... ..... .....
.... .... ..... .... .... .....
.... .... ..... ..... ..... .....
... ..... .... .... .... ..
.... ..... .... .....
... .. .. ..
.... .... ..... ...
. ..... ...... ..
... ...

! = 0, t = 3, ! = 30

t
1.0

0.5

y(t)

0.0

0:5
1:0

. .. ..
.. .... .. .
. ..... .... ..... ...
...... ....... ....... ....... ........ ..
. ... .. ... .. .. ..
.... .... ..... ..... ..... ..... .....
.... ..... ..... .... ...... .... ..... ...
c
0
..... ....... ........ ....... ........ ....... ....... ........ ......
.. .. .. .. .. .. .. .. .. .
. .... ..... ..... ..... ..... ..... ..... ..... ..... .....
....... ....... ...... ...... ........ ....... ....... ........ ...... ........ ...... ...
.. .. .. .. .. .. .. .. .. .. .. ..
... ..... ..... ..... ..... ..... ...... ..... ..... ...... ..... ..... .... .
..... .... ..... ...... ...... ..... ..... ..... ..... ..... .... ..... ..... .....
...... ....... ....... ........ ....... ....... ........ ....... ........ ....... ....... ....... ...... ....... ...... .....
.. ... ... .... ... ... ... ... ... ... ... ... ... ... ... .... .... ..
. ..... .... ..... .... .... ...... ..... ...... ..... ..... ..... ..... ..... ..... ..... .... ..... ..... ....
.... ........ ........ ........ ........ ........ ........ ........ ........ ........ ......... ........ ........ ........ ........ ........ ........ ........ ....... ........ ........ ........ ...... .... .... .... ... ... .
.
.
.
.
.
.
.
.
........................................................................................................................ ..... .... ........... ..... .... ..... ..... ...... ..... ... .. ... .. ... ... ... .. .. .. ... .. ... .. ... ... .. .. ... .. .... ..... .... ............ ...... ..... ...... ..........................
.... ... ... ... ... ... .... ... ... ... .... ... ... ... ... ... ... .... ... ..... .... .... .. .
...... .... .... .... .... .... .... ..... .... .... .... .... .... .... .... .... ... .... ... ..
.... ..... ..... ...... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..... ..
... ..... .... .... .... .... .... .... .... .... .... .... .... .... .... .... ..
.... ..... ..... ..... ...... ..... ...... ..... ..... ..... ..... ...... ..... ..... .....
.. ... ... .... ... ... ... .... .... ... ... ... ... ...
..... .... .... .... ..... ..... .... .... ..... ..... .... ...... .....
... .... .... .... ... .... .... .... .... ... .... ....
. .... .... .... ..... .... .... .... .... .... .... ....
..... .... ..... ..... .... .... ..... .... ..... .... ..
.. ..... .... .... ...... ...... .... ..... ..... ......
.... ..... ..... .... .... .... ..... ... ..
... ... ... .... ... ... .. ... .
... .... ... ... ... ... ... ....
.... ...... ...... ...... ...... ...... ....
... .... .... ... .... .....
. .... ..... .... ... ..
.... ..... ...... .....
.. .. ..
. ..

! = 25, t = 3, !0 = 30

t
Fig. 3. Dispersive Filter Output

102

Fourier Integral Solution of Potential Problem

Laplace's equation
2

r (x; y) =

@2
@2
+ 2
@x2
@y

boundary conditions
(x; 0) =

(x; y) = 0

1; jxj < 1
0; jxj > 1

separate variables
(x; y) = X(x)Y (y)
X 00 (x) Y 00 (y)
+
=0
X(x)
Y (y)
| {z } | {z }
=k2

=+k2

solution form
(x; y) = cos(kx)ekjyj
(x; y) =

A(k) cos(kx)ekjyj dk

(x; 0) =

A(k) cos(kx) dk

inverse Fourier transform

2
A(k) =

2
(x; 0) cos(kx) dx =

cos(kx) dx =

2 sin k
k

2
(x; y) =

sin k cos kx kjyj

e
dk
k

sin[k(x + 1)] sin[k(x 1)] kjyj

e
dx
k
0

1
x+1
x1
1
1
=
tan
tan

jyj
jyj

using G&R 3.941(1).

103

0
3

Fig. 1. Equipotentials

104

Fourier Integral Solution of Potential Problem

M.J. Ablowitz and A.S. Fokas, Complex Variables, Cambridge, 1997, page 60 prob 9.
Laplace's equation
r2 (x; y) = 0
boundary conditions

(x; 0) = sgn(x) = 1 (x ? 0)

separate variables
(x; y) = X(x)Y (y)
X 00 (x) Y 00 (y)
+
=0
X(x)
Y (y)
| {z } | {z }
=k2

=+k2

solution form
(x; y) = sin(kx)ekjyj
(x; y) =

B(k) sin(kx)ekjyj dk

(x; 0) =

B(k) sin(kx) dk

inverse Fourier transform

2
B(k) =

2
(x; 0) sin(kx) dx =

sin(kx) dx =

a bit tricky! recall ECE 370 and sgn(x) = u(x) u(x)

2
(x; y) =

sin kx kjyj
e
dk
k
0

2
2
x
jyj
1
1
= tan
= 1 tan

jyj

using G&R 3.941(1).

105

2
k

The Fourier Transform of combT (t) is

2
T

comb2=T (!)

Recall our notation F (!) for the Fourier transform

F (!) = F[f (t)] =

j!t

f (t)e

dt () f (t) = F

1
[F (!)] =
2

F (!)ej!t d!:

(1)

The Fourier transform of Woodward's comb function, or impulse train

f (t) = combT (t) =

1
X

n=1

is therefore
F[combT (t)] =

1
X

(t nT )

ejn!T = F (!) = F (! + 2=T );

(2)

(3)

n=1

periodic in ! with period 2=T . Let's integrate F (!) over the range a < ! < b, where
=T < a < b < =T . The minimum and maximum restrictions cover one period, centered
about the origin (! = 0). We will make use of the discontinuous sum
1
X
sin(nx)
jxj
=
sgn(x)
n
2
n=1

(2 < x < 2)

(4)

where, in this context, we dene

8
>
< 1; x < 0
sgn(x) =
0; x = 0
>
:
+1; x > 0

(5)

in order to ensure the vanishing of the sum for x = 0. The integral of interest is
Zb X
1

a n=1

1 Z
X

jn!T

d! = b a + 2

cos(n!T ) d!

n=1 a
1
X

2
1
[sin(nbT ) sin(naT )]
T n=1 n

=ba+
jbj sgn(b)
jaj sgn(a)
T
T

2=T if
=T < a < 0 < b < =T
=
:
0
if a < b < 0 or 0 < a < b
=ba+

(6)

Therefore, we see that F[combT (t)] = 2

T comb2=T (!):
Now we are in a position to derive the Fourier series from the Fourier integral. And recall
that we essentially derived the Fourier integral from the \spectral representation of the
Dirac-delta" function. We are in excellent shape to now do some real damage.
106

()

:::

2T T

2
comb 2 (!)
T
T

combT (t)

:::
t

:::

2
T

4
T

Fourier Transform of a Periodic Signal

f (t) = f (t + T )
........
........
........
........
.. ..
.. ..
.. ...
.. ..
... ....
... ....
... ....
... ....
.....
.....
.....
.
.
.
.
.
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
..
..
..
...
...
...
...
.. .
... ...........
... ........... ....
... ........... ....
... ........... ....
....
....
....
....
........ ....
.. ...
.. ...
.. ...
.. ...
..
..
..
....
....
....
....
..
......
......
......
......
..
..
..
..
...
...
...
...
.
.
.
.
.
.
.
.
.
.
.
...
...
...
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
..
..
..
.
.
.
.
.
.
.
.
.
.
.
.
...
...
...
...
.
.
.
.
.. ......
.. ......
.. ......
.. ......
.. ...
.. ...
.. ...
.. ...
....................................................................................................................................................................................................................................................................................................................................................................................................
.
.
.
.

:::

g(t)
...
... ...
... ...
...
... ....
.
... ...
..
..
... .......... ....
...
.
....
.
..
.
.
.
.
.
.......
..
...
...
...
...
...
.
.
...
..
.
.
.
..
.
.
..
..
.
.
..............................................................................................................................................................................................................................................................................................................................................................................................
.
.
.
.

Consider the arbitrary periodic signal f (t) = f (t + T ), and let's focus our attention on the
particular period 0 t T . Any period would work, but since all periods are equivalent
we might as well work with the selected one. (The function f does not have to go to zero
at t = nT ; it was simply easier to draw one like this. And of course the function f does
not have to be positive, or even real-valued.) Introduce the new aperiodic function g(t),
also dened for all t according to

f (t); 0 t T
g(t) =
(1)
0;
t < 0 or t > T:
We can easily reproduce the entire periodic signal f (t) by replicating g(t) all up and down
the t-axis. That is, we write f (t) as the periodic extension of g(t) in the form
f (t) =

1
X

n=1

107

g(t nT ):

(2)

Recognize that (2) can be written as

f (t) = g(t) ~

1
X

n=1

(t nT ) = g(t) ~ combT (t):

(3)

The Fourier transform of (3) is

2
comb 2 (!)
T
T
1
X
2
G(!)
=
(! n!0 )
T
n=1

F (!) = G(!)

where !0 ,

2
T

1
2 X
=
G(n!0 )(! n!0 ):
T n=1

(4)

The spectrum F (!) of the periodic signal f (t) consists of discrete spectral lines at integer
multiples n!0 of the fundamental frequency !0 = 2=T . The nth multiple is called the
nth harmonic. Note that negative n are just as important as positive n.

108

!
!0

The inverse Fourier transform of F (!) should give us our original periodic f (t). It does.
In fact, it gives us a whole new representation for a periodic signal
1
f (t) =
2

Z1
1

j!t

F (!)e

Z1
1
1
X
X
G(n!0 )
G(n!0 ) jn!0 t
d! =
(! n!0 )ej!t d! =
e
T
T
n=1
n=1
1

(5)

where

G(n!0 ) = G(!)!=n!0 =

jn!0 t

g(t)e

dt =

jn!0 t

g(t)e

dt =

f (t)ejn!0 t dt:

(6)

We usually call the Fourier coecients cn instead of G(n!0 )=T and write the whole
package as
ZT
1
X
1
jn!0 t
f (t) =
cn e
cn =
f (t)ejn!0 t dt:
(7)
T
n=1
0

Of course any period will work for the integration limits, since the total integrand is
periodic with period T :
tZ
0 +T
1
f (t)ejn!0 t dt:
(8)
cn =
T
t0

You now have the complete and uncluttered derivation of the Fourier series representation
of a periodic signal, starting from the Fourier integral. The two key components are the
correspondingly appropriate spectral representations for the aperiodic Dirac-delta function
and for its periodic version, Woodward's comb function. All that was required on our part
was a reasonable uency with the Fourier transform properties and a few fundamental
signals.
F Exercise 8. Find an interested (or at least willing) fellow student or more who are at
or above your present level with respect to Fourier analysis, and explain to them, in detail,
the above derivation starting from the top of three pages back. If you explain it correctly
and with some degree of coherence, and your entire audience has not deserted you, buy the
survivor(s) a Coke. (They can split it to t any nancial boundary conditions, applied, of
course, to your total solution.) You deserve one, too. Otherwise go back and do any of the
homework exercises (including enough of the 16 Fourier transform pairs on page 59) that
strike your fancy. If none t that description, see me for more.

109

Periodic Signals and Fourier Series

The period of a periodic signal is the smallest positive number T such that f (t) = f (t + T )
for all t. It will be frequently convenient to dene !0 , 2=T to be the fundamental
frequency of the periodic signal, with n!0 the nth harmonic frequency for integer n =
0; 1; 2; : : : . An important property of a periodic signal is that the integral
tZ
0 +T
t0

f (t) dt =

T =2
Z

f (t) dt =

T =2

3T
Z =4

f (t) dt = : : :

T =4

is independent of the time t0 . That is, integration over any complete period yields the same
result. Another useful observation is that if f (t) = f (t + T ), then the product f (t)ejn!0 t
is also periodic with period T , so that the integral
tZ
0 +T

f (t)ejn!0 t dt

is also independent of the time t0 . F Be sure you can show/verify this important property
for yourself.
DEFN: Orthogonality. Two functions (generally complex) are said to be orthogonal on
the domain a t b if the inner product
b Zb

hf (t); g(t)i , f (t)g (t) dt = 0:

In any given situation, let's assume the interval of interest a t b is known and xed, so
that we can condense the notation on the inner product angle brackets. The inner product
of a signal and itself is sometimes called the energy in the signal
Zb

hf (t); f (t)i ,

jf (t)j2 dt = E

and the square-root of this necessarily non negative quantity is called the norm of the
signal
2 b
31=2
Z
p
kf (t)k = + hf (t); f (t)i = 4 jf (t)j2 dt5 :
a

When dealing with periodic signals so that the interval of interest is any period t0 t
t0 + T , the eective or root-mean-square value is
2 t +T
31=2
Z0
1
frms = 4
jf (t)j2 dt5 :
T
t0

110

Kronecker delta. This simple function is a function of two integers m and n and is
dened as

1; m = n
m;n =
0; m 6
= n:
It is unrelated to the Dirac-delta function (t) that is explicitly written as a function of
the continuous time variable t. The operational advantage of introducing this notational
convenience will be clear as we progress.
Orthogonality of the Fourier basis.
hejn!0 t ; ejm!0 t iT =

tZ
0 +T

hcos(n!0 t); cos(m!0 t)iT =

hsin(n!0 t); sin(m!0 t)iT =

ej(nm)!0 t dt = T n;m

tZ
0 +T

(n; m = 0; 1; 2; : : : )

cos(n!0 t) cos(m!0 t) dt =

T
n;m
2

(n; m = 1; 2; : : : )

sin(n!0 t) sin(m!0 t) dt =

T
n;m
2

(n; m = 1; 2; : : : )

tZ
0 +T
t0

hcos(n!0 t); sin(m!0 t)iT =

tZ
0 +T

cos(n!0 t) sin(m!0 t) dt = 0

(n; m = 0; 1; 2; : : : )

When n = 0 the nth cosine is unity, and we have

h1; cos(m!0 t)iT =

tZ
0 +T

cos(m!0 t) dt = T 0;m

(m = 0; 1; 2; : : : )

h1; sin(m!0 t)iT =

tZ
0 +T

sin(m!0 t) dt = 0

(m = 1; 2; : : : )

Note that sin(m!0 t) is trivial, and therefore not used, when m = 0.

F Homework: Verify/derive each of the above orthogonality relations. You will need to
use trig identities to simplify the products in the integrands.
111

Derivation of Fourier Coecients by Minimizing the Mean Square Error

The concept of the Fourier series arose naturally back on page 108 when we wrote a
periodic signal as the inverse Fourier integral of its Fourier transform. In doing so, the
Fourier coecients appeared naturally. Now let's start over by assuming that the periodic
signal f (t) = f (t + T ) can be represented by a Fourier series, call it
fe(t) =

1
X

cn ejn!0 t

n=1

to temporarily avoid any premature claims that fe(t) = f (t). One approach to derive the
expression for the Fourier coecients cn is to ask the question: What values of the cn
minimize the mean square error between the original signal f (t) and its proposed Fourier
series representation fe(t) ?
Firstly, write the mean square error
1
E =
T
2

1
=
T

tZ
0 +T
t0

1
jf (t) fe(t)j2 dt =
T

tZ
0 +T

f (t) fe(t) f (t) fe (t) dt

tZ
0 +T
t0

jf (t)j2 + jfe(t)j2 f (t)fe (t) f (t)fe(t) dt:

We need the following component integrals.

1
T

tZ
0 +T
t0

1
jfe(t)j2 dt =
T

tZ
0 +T
t0

1
X

cn e

n=1

fe(t)

1
X

1
=
cn
cm
T
n=1
m=1
=
=

n=1
1
X

1
X

jn!0 t

m=1

cn cn

n=1

112

m=1

} |

tZ
0 +T
t0

cm n;m

cm ejm!0 t dt
{z

fe (t)

ej(nm)!0 t dt

1
T

tZ
0 +T
t0

1
T

tZ
0 +T

f (t)fe(t) dt =

1
X

1
cn
T
n=1
1
X

f (t)f (t) dt =

n=1

1
T

tZ
0 +T

f (t)ejn!0 t dt

t0
tZ
0 +T

f (t)ejn!0 t dt

Now the mean square error is

1
E =
T
2

tZ
0 +T
t0

jf (t)j dt +

1
X

cn cn

n=1

1
X

tZ
0 +T

1
X

tZ
0 +T

cn
T
n=1

n=1

1
T

f (t)ejn!0 t dt

To minimize E with respect to the Fourier coecients cn , set to zero the partial derivative
with respect to each of the ck (for all integer k)
@ 2
1
E = ck
@ck
T
or
ck

1
=
T

tZ
0 +T

f (t)ejk!0 t dt = 0

tZ
0 +T

f (t)ejk!0 t dt

the conjugate of which gives the required Fourier coecient formula

1
ck =
T

tZ
0 +T

f (t)ejk!0 t dt

where k can now be replaced by n

1
cn =
T

tZ
0 +T

f (t)ejn!0 t dt:

113

Fourier Coecients by a Direct Inner Product

Write

1
X

f (t) =

cn ejn!0 t

n=1

and then take the inner product of both sides with ejm!0 t
tZ
0 +T

1
X

f (t)ejm!0 t dt =

n=1

1
X

tZ
0 +T

ej(nm)!0 t dt

cn T n;m = T cm

n=1

and therefore
cm

1
=
T

tZ
0 +T

f (t)ejm!0 t dt:

Trigonometric Form of the Fourier Series

1
X

f (t) = a0 +
an cos(n!0 t) + bn sin(n!0 t)
n=1

To nd the trig coecients an and bn (n = 0; 1; 2; : : : ) in terms of the complex exponential

coecients cn (n = : : : ; 1; 0; 1; 2; : : : ) set
1
X

cn e

n=1

c0 +

1 h
X

c0 +

jn!0 t

cn e

n=1

jn!0 t

1
X

= a0 +
an cos(n!0 t) + bn sin(n!0 t)
n=1

jn!0 t

+ cn e

1
X

= a0 +
an cos(n!0 t) + bn sin(n!0 t)
n=1

1 h
X

i
cn cos(n!0 t) + j sin(n!0 t) + cn cos(n!0 t) j sin(n!0 t)

n=1
1
X

= a0 +

n=1

an cos(n!0 t) + bn sin(n!0 t)

1 h
1
i
X
X

c0 +
(cn +cn ) cos(n!0 t)+j(cn cn ) sin(n!0 t) = a0 +
an cos(n!0 t)+bn sin(n!0 t)
n=1

n=1

114

Therefore
1
a0 = c0 =
T

tZ
0 +T

f (t) dt

an = cn + cn

2
=
T

tZ
0 +T

f (t) cos(n!0 t) dt

(n = 1; 2; : : : )

2
bn = j(cn cn ) =
T

tZ
0 +T

f (t) sin(n!0 t) dt

(n = 1; 2; : : : )

F Homework. Derive the trig Fourier coecients by taking the direct inner product of
1
X

f (t) = a0 +
an cos(n!0 t) + bn sin(n!0 t)
n=1

with each of the trig Fourier basis functions cos(m!0 t), sin(m!0 t), and unity.

115

Two Standard Notations for the Trigonometric Fourier Series

f (t) = a0 +

1
X

an cos(n!0 t) + bn sin(n!0 t)

n=1

1
a0 =
T

tZ
0 +T

f (t) dt

2
an =
T

tZ
0 +T

f (t) cos(n!0 t) dt

(n = 1; 2; : : : )

f (t) sin(n!0 t) dt

(n = 1; 2; : : : )

2
bn =
T

tZ
0 +T
t0

a0 X
f (t) =
+
an cos(n!0 t) + bn sin(n!0 t)
2
n=1

2
an =
T

tZ
0 +T

f (t) cos(n!0 t) dt

(n = 0; 1; 2; : : : )

2
bn =
T

tZ
0 +T

f (t) sin(n!0 t) dt

(n = 1; 2; : : : )

The second form uses the general form for an to also calculate the DC term a0 , avoiding
the need for a separate formula for a0 .
It is silly, but observe that this captures the spirit:
1

| X
f (t) =
+
an cos(n!0 t) + bn sin(n!0 t)
7 n=1
7
|=
T

tZ
0 +T

f (t) dt

F Homework. In some applications, the Fourier series is written in the form

1
X
f (t) = A0 +
An cos(n!0 t + n ):
n=1

Find an expression for An and n in terms of an and bn .

116

Riemann-Lebesgue Lemma - Fourier Coecients

f (t) = f (t + T ) =

1
X

cn ejn!0 t ;

!0 = 2=T

n=1

T cn =

tZ
0 +T

f (t)ejn!0 t dt

t0 +T
tZ
0 +T
f (t)ejn!0 t
1
=
+
f 0 (t)ejn!0 t dt

jn!0
jn!0
t=t0

by integration-by-parts. The boundary terms are zero by the periodicity of the integrand.
The remaining integral has

t +T
tZ
0 +T

0
jn!
t
0

f (t)e
dt
jf 0 (t)j dt K < 1

if f 0 (t) is absolutlely integrable. Therefore

jcn j

K
:
2n

If f 0 (t) is absolutlely integrable, then cn ! 0 as n ! 1.

Corollary.
If f (k) (t) is absolutely integrable, but f (k+1) (t) is not, then cn = O(nk ) as n ! 1.

117

Gibb's Phenomenon - Example Square wave

S5 (t)

1.0

0.5

f (t)

0.0

0:5
1:0

0.0

0.5

1.0

t=T
Fourier series of the above square wave is
f (t) =

1
4 X sin[(2k 1)!0 t]
;

2k 1

!0 = 2=T

k=1

Truncated Fourier series

N
4 X sin[(2k 1)!0 t]
SN (t) =

2k 1
k=1

Dierentiate to nd its extrema

0
SN
(t) =

N
8X
cos[(2k 1)!0 t]
T
k=1

write
cos[(2k 1)!0 t] =

sin(2k!0 t) sin[(2k 1)!0 t]

2 sin !0 t

All terms cancel except

0
SN
(t) =

4 sin(2N !0 t)
T sin !0 t
118

Smallest zero is

2N !0
h
i
h
i
(2k1)
N sin (2k1)
N
sin
X
X
2N
2N
4
4

SN (t0 ) =
=
(2k1)

2k 1

2N
t0 =

k=1

Riemann sum

k=1

N
X
sin 2k1
x
sin x
2
dx
x
2k1
x
2 x
k=1

where N x = a. Here x = =N so a = .
2
SN (t0 )

sin x
2
dx = Si() 1:18
x

1:18 1
100% 9%
total jump of 2

overshoot is

Pointwise Convergence of the Fourier Series

The Dirichlet kernel is
DN (x) =

N
X

jnx

jN x

2N
X

ejnx = ejN x

n=0

n=N

sin[(N + 1=2)x]
1 ej(2N +1)x
=
;
jx
1e
sin(x=2)

an even function DN (x) = DN (x). Consider a periodic function f (t) = f (t + T ) having

a single jump discontinuity at the point T =2 < t < T =2. Dene the right and left-hand
limits
and
f (t ) = lim+ f (t ):
f (t+ ) = lim+ f (t + )
!0

With the Fourier coecients

1
cn =
T

T =2
Z

f ( )ej2n =T d;

T =2

the truncated Fourier series

feN (t) =

N
X

n=N

119

cn ej2nt=T

is
feN (t) =

T =2
Z

T =2
Z
N
1 X j2n(t )=T
1
f ( )
e
d =
f ( ) DN [2(t )=T ] d
T
T
n=N

T =2

1
f ( ) DN [2( t)=T ] d +
T

T =2

I =

T =2

T =2
Z
1
f ( ) DN [2( t)=T ] d
T
t
{z
}
|
I+

1
[f ( ) f (t )] DN [2( t)=T ] d + f (t )
T

T =2

1
DN [2( t)=T ] d
T

T =2

Introduce the change-of-integration variable

x=
such that

2
(N + 1=2)( t)
T

1
DN [2( t)=T ] d =
T

sin x
dx

x
(2N + 1)
sin
2N + 1

= T =2 =) x = (2N + 1)(1=2 + t=T ) = X

= t =) x = 0
T
t=
(2N + 1)

1
I =

Z0
f
X

f (t )
+

Z0
X

x + t f (t )
(2N + 1)

sin x

(2N + 1) sin

x
2N + 1

Since X < x < 0 , as N ! 1,

sin x

(2N + 1) sin

T
f
x + t ! f (t )
(2N + 1)
N !1
120

x
2N + 1

so that the rst integral above vanishes. The denominator becomes

(2N + 1) sin

x
2N + 1

! x
N !1

and with X ! 1, the second integral gives

f (t )
I !

N !1

sin x
f (t )
dx =
:
x
2

The same philosophy yields

f (t+ )
I+ !

N !1
so that

N
X

j2nt=T

cn e

n=N

sin x
f (t+ )
dx =
x
2

f (t+ ) + f (t )
!
:
2
N !1

F Homework Consider the simpler case where f (t) is continuous everywhere, and therefore demonstrate that
N
X
cn ej2nt=T ! f (t):
N !1

n=N

121

ECE 370

16 January 1978

QUIZ 3

NAME

Take Home - Due Wednesday January 18

f (t)
.

:::

T =2

3T =2

1.) Find a Fourier series representation of the signal f (t).

2.) Use MATLAB to graph the N th partial sum of this Fourier series. For example, if
1

a0 X
f (t) =
+
an cos(n!0 t);
2
n=1
then let
N

a0 X
fN (t) =
+
an cos(n!0 t):
2
n=1
Pick several N to demonstrate (pictorially) the convergence of this Fourier series.

There are no sines (bn 0) in this even function, and the DC or average value is
a0
1
=
2
T

f (t) dt =

1
;
2

(1)

as the graph shows (perhaps after a little thought). The cosine coecients (for n = 1; 2; : : : )
are
2
an =
T

tZ
0 +T
t0

2
f (t) cos(n!0 t) dt =
T

T =2
Z

2
f (t) cos(n!0 t) dt = 2
T

T =2
Z
f (t) cos(n!0 t) dt (2)
0

T =2

122

by the even symmetry of the product of f (t) = f (t) and the nth cosine. The last integral
above requires an expression for f (t) in the range 0 t T =2, which is the line of slope
2=T through the origin: f (t) = 2t=T . Required now is
2
an = 2
T

T =2
Z

2t
8
cos(n!0 t) dt = 2
T
T

T =2
Z
t cos(n!0 t) dt:

(3)

In terms of dierent variables, we need the integral

x cos(x) dx:

(4)

It is perfectly acceptable to use a table of integrals (perhaps in your calculus book, or

a table such as the CRC). But here are two dierent techniques to evaluate it. The
rst method is based on the observation that the parameter is independent from the
integration variable x, so that we can write

d
x cos(x) dx =
d

sin(x) dx =

d 1 cos(d)
d

d sin(d) 1 cos(d)
=
:

(5)

The second method is integration by parts:

d(uv) = u dv + v du

Zb
a

The choice
u=x
dv = cos(x) dx

b Zb

u dv = uv v du:
a

du = dx
1
v = sin(x)

gives

Zd
0

Zd
d
x
1

x cos(x) dx = sin(x)

sin(x) dx

x=0
0

d
d
1

= sin(d) + 2 cos(x)

x=0
d
cos(d) 1
= sin(d) +
;

2
123

(6)

in agreement with the rst result (5). The needed integral in (3) has upper limit d = T =2
and frequency (the factor multiplying the integration variable t in the argument of the
cosine) = n!0 , whereupon d = n!0 T =2 = n (recall !0 T = 2). Note that the values
of the trig functions are thus sin(n) = 0 and cos(n) = (1)n . The Fourier coecients
(3) are now
8 2
1
2
<
n
8 1 (1)
2

; n odd
n
an = 2
=
[1 (1) ] =
(7)

n2
2
2
:
T
(n!0 )
(n)
0;
n even:
Together with the DC value (1), the entire trigonometric Fourier series for our even, periodic function f (t) is
2 X
1
2
1
1
f (t) =
cos(2nt=T ):
(8)
2

n2
n=1;3;5;:::
The sum is reasonably clear (at least to me), but it would more commonly be written in
the (probably nicer looking) completely equivalent form
2 X
1
1
1
2
cos[(2n 1)2t=T ]:
(9)
f (t) =
2
n=1 (2n 1)2
And it would be perfectly acceptable to write it in terms of the fundamental frequency as
2 X
1
1
2
1
f (t) =
cos[(2n 1)!0 t];
2
n=1 (2n 1)2
as long as it is perfectly clear that we have dened !0 = 2=T .
This series converges nicely, since jan j n2 as n ! 1 (in fact, these an are 1=n2 for all
odd n. In fact, Fourier series representations of known functions are used to sum certain
series. In this case, note that f (T =2) = 1 and so (9) gives
2 X
1
2
1
1
cos[(2n 1)]
1=
2
2

(2n

1)
n=1
or

1
X

1
2
=
:
2
(2n

1)
8
n=1
Note that the sum over both odd and even integers is
1
1
1
X
X
X
1
1
1
=
+
2
2
n
(2n 1)
(2n)2
n=1
n=1
n=1
1
X

1
1
1X 1
=
+
2
(2n

1)
4 n=1 n2
n=1

124

or
1
1
X
3X 1
1
=
2
4 n=1 n
(2n 1)2
n=1

=
and therefore

2
8

1
X
1
2
=
:
n2
6
n=1

This is the Riemann zeta function

1
X
1
(x) =
nx
n=1

of argument x = 2.

125

On the Convergence and Summation of Some Series

The basic form to consider is
S=

1
X

an = a1 + a2 + a3 + : : :

n=1

where the indexing can just as easily start at n = 0

1
X

an = a0 + a1 + a2 + : : :

n=0

and negative indices are also commonly encountered

1
X

an = a0 +

n=1

If the series

1
P

n=1

1
X

an +

n=1

jan j converges, then we say that

1
X

n=1
1
P

an :

an is absolutely convergent. Any series

n=1

that is absolutely convergent is convergent, since

1
X X

an
jan j:

n=1

A convergent series that is not absolutely convergent is called conditionally convergent.

Here is an important class of series: With p real

1
X
1
np
n=1

converges i p > 1.

This can be shown by using the integral argument (Do it!) on the next page.

126

If all of the terms of the original series are positive (true for exp(z) if z > 0), then a related
integral can provide error bounds by the following reasoning.
.....
..
..
..
...
...
...
...
...
..
...
..
...
...
...
..
...
...
..
...
...
...
...
...
..
...
..
...
...
...
1
..
...
1
...
...
...
...
..
...
..
...
...
...
..
...
...
...
n=N +1
..
...
N +1
...
...
...
........................................
..
......
...
..
.
.
...
.... ......
...
...
..
.. ......
...
.....
..
...
.
.....
...
...
..... .....
..
..
...... ..
..
.
...........................................
....
...
..........
...
..
..
... ..........
.
..
...
........ ....
...
......... .
...
...
...
...............................................
..
.
...
... ................ ...
.
..
.
................................................
...
...
....
...
... ..................... ..
...
...
..
..........................................................
..
...
.................................................................
...
...
..
...
...
...
...........................................
...
.
.
..
.
...
...
...
..
...
..
.
.
.
.
...............................................................................................................................................................................................................................................................................................................................................................................................
...
..
...
.

g(x)

N +1 N +2 N +3

g(n) >

g(x) dx

:::

...
..
..
..
...
...
..
...
...
...
..
...
...
...
...
...
..
...
..
...
...
...
..
...
...
...
...
...
..
...
..
1
...
1
...
...
..
...
...
...
..
...
...
...
...
...
..
...
..
n=N +1
...
...
N
...
..
.............................................
...
...
... ...
..
..
..
... .....
..
...
... .....
.....
...
..
...
.....
..
..
...
.....
...
...
...
.....
...
..
...
..
..
..
.................................................
..
.
... ........
...
...
........
....
...
...
..
.......
.
..
...
....
...
........................................................
...
..
.
.
... ....................
...
...
...
...
.............................................................
...
...
...
...
...
...
....................................................................
...
..
.
...
.......................................................................................................
...
.
.
..
.
.
.
...
...
...
......... . .......................
...
...
...
....
...
.
.
.
.
.
.
.
.
...........................................................................................................................................................................................................................................................................................................................................................................................
..
...
..
..

g(x)

N +1 N +2 N +3

g(n) <

g(x) dx

:::

Therefore, we have both a lower and an upper bound, so that the error or \tail" sum is
Z1
N +1

g(x) dx <

1
X

n=N +1

127

g(n) <

Z1
N

g(x) dx:

If the series

1
P

n=1

jan j converges, then we must have lim an = 0: The converse is not true,
n!1

as we demonstrate by example, since the series

1
X
1
np
n=1

is divergent for p = 1=2 but

1
p ! 0:
n n!1

Comparison Test. Let

1
P

an and

1
P

bn be two
n=1
bn for all n >
1
P

n=1

exists an index N such that an

1
P
bn implies the convergence of the series
n=1

implies the divergence of

1
P

n=1

series with nonnegative terms. If there

N , then the convergence of the series

1
P
an , and the divergence of the series
an
n=1

bn .

n=1

Weierstrass M-Test for Absolute Convergence. Let

1
P

n=1

an and

1
P

bn be series.

n=1

Suppose there exists an index N such that jan j bn for all n > N . Then a sucient
1
1
P
P
an is that the series
bn converges.
condition for absolute convergence of the series
n=1

n=1

an+1
= exists for the
Ratio (d'Alembert's) Test. Suppose that the limit lim
n!1
an
1
P
series
an .
n=1

(1) if < 1 then the series

1
P

an converges absolutely

1
P

an diverges

n=1

(2) if > 1 then the series

n=1

(3) there exist both absolutely convergent and divergent series for which = 1.

128

Acceleration of series convergence by subtracting out the asymptotic form

(Kummer transform):
Consider the series

1
P

f (n) where the large n form of f (n) is

n=1

f (n) ! fasy (n):

n!1

If the original terms f (n) decay so slowly with n as to be numerically disagreeable, but
the asymptotic form can be summed by some alternate method, then write
1
X

f (n) =

n=1

1
X

n=1

[f (n) fasy (n)] +

1
X

fasy (n):

n=1

The terms of the dierence f (n) fasy (n) will approach zero faster than the original f (n),
so that truncation of the dierence sum can be made at a smaller n to achieve a given
level of accuracy.
Example. One series that can be summed exactly using the Poisson sum formula (on
pages 150-151) is
1
X
1

= coth b:
2
2
n +b
b
n=1
We choose an example series where we know the exact value of the sum simply to be able
1
P
to check our result. Let's get this series into the form
f (n), where f (n) = f (n) is an
n=1

even function of the summation index,

1
X

1
X
1
1
1
=
2
+ 2
2
2
2
2
n +b
n +b
b
n=1
n=1

and therefore
S=

1
X

n=1

b coth b 1
1
=
:
2
+b
2b2

Now let's compare the brute-force summation of S as it stands, with a simple Kummer
transform. Having the exact value of S allows us to gauge our accuracy. The asymptotic
form of the summand f (n) is fasy (n) = n2 so that the Kummer transform is
X
1
1
X
1
1
1
S=

+
:
n2 + b 2
n2
n2
n=1
n=1
The asymptotic or \tail" sum is the Riemann zeta function of argument 2 (a well-known,
actually famous sum) that we encountered back on page 124
1
X
1
2
(2) =
=
:
2
n
6
n=1

129

The dierence terms

1
n2 (n2 + b2 )
b2
1

=
=
= O(n4 )
n2 + b2
n2
n2 (n2 + b2 )
n2 (n2 + b2 )
decay much faster than the original terms of O(n2 ). The Kummer transform of the
original sum S is now
1
X
1
2
S = b2
+
:
2 (n2 + b2 )
n
6
n=1
To do the comparison, label the truncated brute-force sum
S0 (N; b) =

N
X

n=1

1
+ b2

and the truncated Kummer-accelerated sum

SK (N; b) = b

N
X

1
2
+
:
n2 (n2 + b2 )
6
n=1

With the exact value

b coth b 1
2b2
let's compare the numerical performance of
S(b) =

S0 (N; b)
S(b)

with

SK (N; b)
S(b)

as a function of the truncation index N for one or more values of the parameter b.

S0 (N; 10)
S(10)

SK (N; 10)
S(10)

101
102
103
104

0:499726021056655
0:934787298579118
0:993428003029169
0:999342482848786

1:125492496191760
1:000214644029801
1:000000218841745
1:000000000219156

Exercise. Using (4) =

1
P

n4 = 4 =90, perform a second Kummer transform on the

n=1

series in the example and compare its convergence to the data above.

130

Homework Problem: Convergence of Fourier Series. Consider the three periodic

functions x(t), y(t), and z(t) as graphed below. Find appropriate Fourier series expansions
for each of the three functions. Study and discuss the convergence of each series, including
graphical display.
For example, say the odd function y(t) is most conveniently represented by the Fourier
sine series
1
X
y(t) =
bn sin(2nt=T )
n=1

where the coecients bn are known. Truncation of this Fourier series yields the approximation
N
X
bn sin(2nt=T ):
yN (t) =
n=1

Compare these truncated series yN (t) with the original function y(t) for several (you decide
which range of values has signicance to your convergence study) values of the nal or
truncation index N . Certainly, the rate of decay of jbn j as n ! 1 is the important
feature.
Note: How are x(t), y(t), and z(t) related? How then are their Fourier series related?
+1

x(t)

T =2

y(t)

z(t)

H (1)

131

Fourier Series of Full-Wave Rectied Sine Wave

sin( 12 !0 t) = sin(t=T )

f (t) = j sin( 12 !0 t)j = j sin(t=T )j

The even function f (t) = f (t) is expanded in a Fourier series

a0 X
f (t) =
+
an cos(n!0 t)
2
n=1

(1)

where its fundamental frequency is !0 = 2=T , in terms of its period T . Note that the
period of the sine wave prior to rectication is 2T . Equivalently, the frequency of the
un-rectied sine wave is half the frequency of the rectied signal f (t).
The DC or average value of f (t) is
a0
1
=
2
T

1
f (t) dt =
T

ZT
0

!
! 0
2

0
0
sin
t dt =
cos
t
2
!0 T
2

t=T

2
2
[1 cos ] = : (2)
2

For n = 1; 2; : : :
2
an =
T

tZ
0 +T
t0

2
f (t) cos(n!0 t) dt = 2
T

T =2
Z
!
0
sin
t cos(n!0 t) dt
2
0

132

(3)

since

!
!

0
0
t = sin
t
f (t) = sin
2
2
Use of the trig identity

for 0 t T =2:

(4)

2 sin A cos B = sin(A B) + sin(A + B)

yields
2
an =
T

T =2

1
1
sin (n + 2 )!0 t sin (n 2 )!0 t dt:

(5)

Let's economize the algebra and do both forms at once by considering

0
T =2
Z

2
2
1

1
1
sin (n 2 )!0 t dt =
cos (n 2 )!0 t
1

T
T (n 2 )!0
t=T =2
0

1
1 cos (n 2 )!0 T =2
1 cos (n 12 )
1
=
=
=
:
1
1
(n 2 )!0 T =2
(n 2 )
(n 12 ) (6)
Insertion of (6) into (5) gives

1
1
1
1 (n 12 ) (n + 12 )
1
an =

=
=
:
1
1
1
1
n+ 2
(n + 2 )(n 2 )
n 2
(n2 14 )

(7)

Together with the DC term (2), the full Fourier series for f (t) is now
1
2
1X
1
f (t) =
2
n=1 n

1
4

cos(2nt=T );

(8)

and its truncated version (N th partial sum) is

fN (t) =

N
1X
1
2

n=1 n2

1
4

cos(2nt=T ):

(9)

solid curve: N = 3
dashed curve: N = 20

1
133

t=T

Parseval's Theorem for the Fourier Series

Consider a periodic signal f (t) = f (t + T ) and its Fourier series representation
f (t) =

1
X

cn ejn!0 t

(!0 = 2=T ):

n=1

The average power (per period) \in" or \of" f (t) is

Pav

1
=
T

tZ
0 +T
t0

1
=
T

tZ
0 +T

jf (t)j2 dt
1
f (t)f (t) dt =
T

1
X

n=1

1
X

n=1

tZ
0 +T

m=1

1
T

tZ
0 +T

jn!0 t

cn e

n=1

t0
1
X

1
X

j(nm)!0 t

1
X

cm ejm!0 t dt

m=1

dt =

1
X

n=1

1
X

m=1

cm mn

1
X

cn cn

n=1

jcn j2

F Homework: Find the comparable expression in terms of the trigonometric Fourier

series coecients.

134

Homework Problem: Power Supply Performance.

sin( 12 !0 t) = sin(t=T )

x(t) = j sin( 12 !0 t)j = j sin(t=T )j

x(t) =

1
2
1X
1

n=1 n2

1
4

cos(2nt=T )

Consider a real power supply, consisting of a full-wave rectier followed by a simple lowpass lter. Find an expression for, and graph, the time-domain output y(t) of the power
supply for several values of the important parameter !b =!0 . Dene the percent ripple as
the ratio of average power in the unwanted harmonics to the desired DC power. Graph
percent ripple versus !b =!0 . Recall Parseval's theorem for Fourier series.

x(t)

y(t)

135

Solution. The transfer function of the low-pass lter is

1
1
1
j!C
H(!) =
=
=
1
1 + j!RC
1 + j!=!b
R+
j!C

with !b ,

1
:
RC

In polar form
H(!) = jH(!)jej(!)
the magnitude of H(!) is
1
1 + (!=!b )2

jH(!)j = p
and the phase of H(!) is

(!) = tan1 (!=!b ):

Note that the lter impulse response is necessarily real, since the lter is an actual, physical
electric circuit. Also note that H(0) = 1 is real; that is, (0) = 0. With the lter input
written as the particular Fourier series
x(t) =

1
X

an cos(n!0 t);

n=0

the response of the lter is

y(t) =

1
X

n=0

an jH(n!0 )j cos[n!0 t + (n!0 )]:

Note that the DC term is now called a0 , instead of a0 =2. Recall that !0 , 2=T is the
fundamental frequency of the rectied sine, which is twice the frequency of the pure sine
wave that is the initial input to the full-wave rectier. Specically, the above expression is
y(t) =

1
1X
2
1

n=1 n2

1
4

1
p
cos n!0 t tan1 (n!0 =!b ) :
1 + (n!0 =!b )2

Note that the single (dimensionless!) numerical parameter required to characterize the
complete power supply output is !0 =!b . Time t is always multiplied by !0 = 2=T , so
that in terms of normalized time t=T , we have

1
t
1
2
1X
y
=
T
n=1 n2

1
4

1
t
1
p
cos 2n tan (n!0 =!b ) :
T
1 + (n!0 =!b )2

The form of the Fourier series for y(t) is, with An = an jH(n!0 )j and n = (n!0 ),
y(t) =

1
X

An cos(n!0 t + n )

n=0

136

(1)

and the total average power in the output y(t) is

Ptot

1
=
T

tZ
0 +T
t0

1
=
T

jy(t)j2 dt

tZ
0 +T " 1
X

n=0

t0
1
1 X
X

An cos(n!0 t + n )

Am cos(m!0 t + m )

m=0

An Am

n=0 m=0

1
X

1
T

tZ
0 +T

cos(n!0 t + n ) cos(m!0 t + m ) dt:

The DC terms (n = 0 and/or m = 0) are a little dierent, but I included them in the series
(starting at n = 0 instead of starting at n = 1) for notational convenience. Recall or note
that 0 0. The inner products of interest are:
1
T

tZ
0 +T
t0

1
T

tZ
0 +T
t0

1
T

tZ
0 +T

1 1 dt = 1

(n = m = 0)

1 cos(m!0 t + m ) dt = 0

cos(n!0 t + n ) cos(m!0 t + m ) dt =

(n = 0; m = 1; 2; : : : )

1
nm
2

(n; m = 1; 2; : : : ):

Therefore, the total average power in the signal y(t) is

Ptot = jA0 j2 +

1
1X
jAn j2 ;
2 n=1

which is the form of Parseval's theorem that applies for our particular representation of
y(t). The DC (n = 0) term is our desired power supply output, while all of the higher
harmonics (n = 1; 2; : : : ) give rise to the unwanted \ripple." Therefore, one denition of
percent ripple is
unwanted power in the ripples
100%
r,
total power output
or
1
P
1
jAn j2
2
n=1
r=
100%:
(2)
1
P
1
2
2
jA0 j + 2
jAn j
n=1

137

However, the problem statement evidently dened percent ripple as

1
2

1
P

n=1

jAn j2

jA0 j2

100%:

These two r expressions will be nearly identical when the majority of the output power is
in the DC component, but for a very poor power supply I suppose some marketing/sales
people would prefer to use (2). The second form might be greater than 100%!
1.2
!i =!b = 0:1....................................................

1.0

0.8

0.6
y(t)
0.4
0.2
0.0
0:2

input frequency of unrectied sine is !i = !0 =2

0:75

0:50

0:25

0.00

0.25

0.50

0.75

t=T
20
18
16
14
12
r (%) 10
8
6
4
2
0

!i =!b = freq of unrectied sine/lter break

138

Periodic Excitation of a Simple GLC Electric Circuit

v(t)

ig (t)

If the source signal ig (t) = ig (t + T ) is a periodic signal, then the steady-state solution
for all of the linear signals (common voltage v(t), all three branch currents) will also be
periodic with the same fundamental frequency !0 = 2=T of the excitation. If we know
the waveform ig (t), then we can nd its Fourier series representation
ig (t) =

1
X

cn ejn!0 t :

(1)

n=1

Here is a case where it is more convenient and natural to use the complex exponential
form of the Fourier series: The trigonometric form would suce in principle, but carrying
around two sets of coecients an and bn is more awkward. Plus dierentiating exp(jn!0 t)
is much cleaner than dierentiating cos(n!0 t) and sin(n!0 t).
Kirchho's current law supplies the equation-of-motion for the unknown voltage signal
v(t) in our circuit
Zt
1
dv(t)
+
v( ) d = ig (t):
(2)
Gv(t) + C
dt
L
1

This integrodierential equation is converted to a pure dierential equation by dierentiating once to eliminate the integral operator in the inductive current
C

d2 v(t)
dv(t)
1
+G
+ v(t) = i0g (t):
2
dt
dt
L

(3)

The order of the terms was rearranged to put the dierential equation in standard form,
and the prime on the generator current denotes dierentiation with respect to its argument,
which is our independent variable t. The periodic voltage signal is written as a Fourier
series
1
X
v(t) =
n ejn!0 t
(4)
n=1

139

where the Fourier coecients n are unknown, so far. Insertion of both series (1) and (4)
into the dierential equation (3) gives

1
1
X
X
1 jn!0 t
2
n (jn!0 ) C + (jn!0 )G +
e
=
(jn!0 )cn ejn!0 t :
(5)
L
n=1
n=1
Since the above two Fourier series in (5) are equal for all t and they are written in terms
of the same Fourier basis functions exp(jn!0 t), the corresponding coecients are equal

1
2
= (jn!0 )cn :
n (jn!0 ) C + (jn!0 )G +
(6)
L
Therefore the n are given explicitly in terms of the known cn
jn!0
jn!0 L
n =
cn ;
cn =
1
1 (n!0 )2 LC + jn!0 LG
(jn!0 )2 C + (jn!0 )G +
L
and our particular solution (4) for the voltage is
v(t) =

1
X

jn!0 L
cn ejn!0 t :
2
1 (n!0 ) LC + jn!0 LG
n=1

(7)

(8)

Note that the (SI or mksC) units of the numerator are those of !0 L, which are (H/s) or
(). All three terms in the denominator are dimensionless (check for yourself), and the
Fourier coecients cn dened in (1) are in (A). The units of the n are therefore (V), as
required.
Problems
1. Find an expression for the average power dissipated as heat in this circuit.
2. Give a physical explanation for the absence (zero value) of the n = 0 term in the voltage.
1
3. If the lossy element is removed from the circuit (G
p = 0 or G = R = 1), what is the
physical signicance of a harmonic having n!0 = 1= LC ?

4. Write down an approximation for the higher-order terms in the series for v(t), that
is as n ! +1. (The terms with negative indices behave similarly.) Interpret this high
frequency approximation in terms of the circuit.
5. If f (t) is a real function of time, show that the coecients cn in its complex exponential
Fourier series
1
X
cn ejn!0 t
f (t) =
n=1

must satisfy the requirement cn =

cn .

One way to see this is to examine f (t) = f (t).

6. If ig (t) is a real-valued source current, show that the expression (8) is also real.
7. Graph the signal v(t) when the current source is a square wave, say of amplitude I0 .
Show the eect of varying the circuit element values. Use normalized parameters, as much
as possible, to simplify the results.
140

Simple Low-Pass and Hi-Pass Filter Response to a Periodic Input

Problem: Find expressions for and graph the signals ya (t) and yb (t) for several values of
the important (nondimensional) parameters that appear naturally.
x(t)

...... ...... ......

x(t)

...
..
...............................................................
....
.

y (t)

x(t)

a
........................
..
..
...
..
..
..
......................................................................................................................................................................

yb (t)

Note that the excitation x(t) has both even and odd parts, but the even part is simply
A=2. The graph of x
e(t) = x(t) A=2 is clearly an antisymmetric (odd) function, so that its
Fourier series is most logically written in the trigonometric form with only sine functions.
x
e(t) = x(t) A=2

A=2

x
e(t) = e
x(t)
x
e(t) =
T =2
Z

2
bn =
T

T =2

4A
=
T

1
X

x
e(t) = A
bn sin(n!0 t)

1
t

T
2

(0 < t < T =2)

(!0 , 2=T )

n=1

4
x
e(t) sin(n!0 t) dt =
T

T =2
Z

t
1

T
2

sin(n!0 t) dt

141

T =2
Z
x
e(t) sin(n!0 t) dt
0

T =2
0
Z
cos(n!0 t)
1 cos(n!0 T =2)
1 cos(n)
1 (1)n
sin(n!0 t) dt =
=
=
=
n!0 t=T =2
n!0
n!0
n!0
0
8
< 2
n odd
n!0
=
(but this form is not as useful as it might appear)
:
0
n even

Z
u dv = uv v du

u=t

du = dt

dv = sin(n!0 t) dt

cos(n!0 t)
n!0

T =2
T =2
0
Z
Z
t cos(n!0 t)
1
t sin(n!0 t) dt =
+
cos(n!0 t) dt

n!0
n!0
t=T =2
0

T =2
sin(n!0 t)
T
sin(n)
T
=
cos(n!0 T =2) +
=

cos(n)
+
2n!0
(n!0 )2 t=0
2n!0
(n!0 )2
T (1)n
=
2n!0

Now go back and assemble the required Fourier coecient

4A (1)n
1 (1)n
A
4A
=
:
bn =

=
T
2n!0
2n!0
2n!0 T
n
The Fourier series representation of the input signal is therefore
1
A A X sin(n!0 t)
:
x(t) =
2
n=1
n

The transfer functions of our two lters, from a simple frequency domain voltage divider,
are
1
1
j!C
Ha (!) =
=
= jHa (!)jeja (!)
1
1 + j!RC
R+
j!C
Hb (!) =

R
1
R+
j!C

j!RC
= jHb (!)jejb (!) :
1 + j!RC

Let's not write out the polar representations explicitly: MATLAB will compute those for
us. That is, assume the magnitude function jHa (!)j and a (!) are known (they are!), and
obviously the same for the b circuit. Note that Ha (0) = 1 and Hb (0) = 0.
142

The Fourier series for the two output signals are therefore
1

A A X jHa (n!0 )j
sin n!0 t + a (n!0 )
ya (t) =
2
n=1
n
1

A X jHb (n!0 )j
yb (t) =
sin n!0 t + b (n!0 )
n=1
n

Let's use normalized time t=T and normalize by the amplitude A of the input to write
1

1
1 X jHa (n!0 )j
ya (t=T )
=
sin 2nt=T + a (n!0 )
A
2 n=1
n
1

yb (t=T )
1 X jHb (n!0 )j
=
sin 2nt=T + b (n!0 )
A
n=1
n

We need
Ha (n!0 ) =

1
!b
1
=
!
1 + jn!0 RC
1 + jn!0 =!b n!1 jn!0

Hb (n!0 ) =

jn!0 RC
jn!0 =!b
=
! 1
1 + jn!0 RC
1 + jn!0 =!b n!1

where
!b ,

1
RC

is the break frequency (or cut-o frequency or half-power frequency) of (both of) the lters.
Therefore, the important nondimensional parameter of our problem is !0 =!b = !0 RC.
Recall that !0 = 2=T is the fundamental frequency of the periodic excitation.
As n ! 1, the terms in the series for the low-pass lter output ya (t) vary like n2 while
the higher-order terms in the high-pass lter output yb (t) vary like n1 . Therefore, the
numerical convergence of the signal ya (t) is much faster. That is, truncation of the ya (t)
series at some nite value of n will give a much better approximation for the innite series
than will the same truncation of the series for yb (t). Note that the outputs ya (t) and yb (t)
are intimately related since
Ha (!) + Hb (!) = 1

ya (t) + yb (t) = x(t):

Therefore, we can numerically evaluate the series for ya (t) (because its series converges
faster), and then avoid computing the poorly convergent series for yb (t) simply by using
the known relationship
yb (t) = x(t) ya (t):

143

!0 =!b = 0:1
y (t)

.......
...........
a
............ ..
. .......... ..
................. .....
.................. .....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
........
..
........
.............
............
...
...
..............
...............
....
...
.................
................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
.
.
.
.
...
...
.
.
.
.
.
.
.
.................
.
....
.
.
.
.....
.........
.
.
.
.
.
.
.
........
....
.....
.
.
.
.
.
.
.
......
.
.....
.
.
......
.
.
.
.
...........
....
.
......
.
.
.
.
. ....
....
.....
.................
.
.
.
.
.
....
.
.
.....
.........
.
.
.
.
.
.
.
.
....
.....
...
.
.
.
.
.
.
.
.
.
.
.
.
....
.........
......
.
.
.
.
.
.
........
.
......
.
.. ...
.
.
.
.
.
.......
.. ...
.. ...
..................
.
.. ...
.
.
.
.
.. ...
.........
.
.
.
.
.
.. ...
.
.......
.. ...
.
.
.
.
.
.
.
.........
.
.. ...
.. ...
.
.
.
.
.
...
.
.
.
.
.
.
.. ....
.
.
.
... .....
.
.
......
....
.. ...
.. ..
.................
...............
.
.
.
.
.
.
.
.
.. ...
......
.
.................
.
.
.
.
.
.
.
.
.
.
.
.
... ....
.
.
.
.. ...
............
..............
.. ....
.. ...
............
.............
.. .............................
.. ............................
.. ......
.. ... .....
............................................................................................................................................................................................................................................................................................................................................................................

t=T

y (t)

.....
......
b
.... ..
.... ...
..... ..
.
.....
.
.
.
.
.
.
..
.
..
...
.....
..
....
..
....
.....
..
..
.....
....
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.....
...
..
...
.
.
.
.
..
..
.....
..
.
..
.
.
.
.
.
..
.
..
.
.
.
.
.
..
.
..
..
.
.
.
..
..
.
.
.
.
.
..
..
..
..
.....
..
.
.
.
.
.
..
.
..
.
..
.
.
..
.
..
.
.
.
.
..
.
.
..
.
.
.
.
..
..
..
.
.
.
.
..
..
.....
..
.
.
.
..
...
...
..
.
.
..
.
.
.....
.
.
..
.
..
..
....
.
.
.
..
.....
..
...
.
.
.
.
.
.
.
..
..
....
.....
..
..
.....
....
.. ......
.. ........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.............................................................................................................................................................................................................................................................................................................................................................................................................................
.
. .
.... ....
.... ....
.. ...
.. ...
... ..
... ..
.. ..
.. ..
... ..
... ..
.. ...
.. ...
... ..
... ..
.
.
..
.... ...
.... ...
.. ..
.. ..
... .
... .
.. ....
.. ....
... ...
... ...
....
....
.....
.....
....
....
.....
.....
....
....
.....
.....
.....
.....
.....
.....
....
....
.....
.....
.
.
..
.......
.......
..
...
...
...
..
..
...
...
..
..

144

t=T

!0 =!b = 1
y (t)

.....
......
a
.... ..
. ... ...
..... ..
.
.....
.
.
.
.
.
.
..
..
...
...
.
....
....
.....
.....
...........
...........
....
.....
........ ........
........ .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. ...
.
....
.. ....
........
........
.. ...
.....
........
........
.. ...
....
.. ...
........
........
.....
.......
........
. .....
.
.
.
.
.. ....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
...
...
.. ...
.
.
.
.
.
.
.
.. ....
.
.
.
.
.
.
.
.
.
.
...
.
...
.
.
.
.
...
......
.
.
.
..
.
..
.
.
.
.
...
...
.
.
.
.
...
.
.
.
.
.
..
.
.
...
..
.
..
.
.
.
.
.
...
.
.
.
.
.
.
.
.
.
...
..
..
...
....
.
.
.
.
.
.
.
.
.
.
.
.
.
....
..
...
....
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
....
....
.
...
.
.
..
.
.
.
.
.
.
.
.
.
.....
.
.
.
.....
.
.
....
.
.
.
.
.
.
.
.
.
..
.....
.
.
.....
.
.
.
...
...
.
.
.
.
.
......
......
.
.
..
.
.
.
.
.
.
..
....
.
.
.
........
........
.
.
.
.
.
.
.
.
.
.
.
.
..
.
.
. ..................
...........
...........
.
.
.
.
.
.
.
....................................................
..........
..
..
.
.
.
.
.
...
.
.
..
..
.
.
.
.
.
.
.
.....
.
.
.
.
..
.
.
.
.
..
.
.
.
.
.
.
.
.
..
.
.
.
..
.....
..
....
..
.....
.....
..
... .......
....
.. ........
. ... ...
...........................................................................................................................................................................................................................................................................................................................................................................

t=T

y (t)

.....
......
b
.... ..
.... ...
..... ..
.
.....
.
.
.
.
.
.
..
.
..
...
.....
..
....
..
....
.....
..
..
.....
....
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.....
...
..
...
.
.
.
.
..
..
.....
..
.
..
.
.
.
.
.
..
.
..
.
.
.
.
.
..
.
..
..
.
.
.
..
..
.
.
.
.
.
..
..
..
..
.....
..
.
.
.
.
.
..
.
..
.
..
.
.
..
.
..
.
.
.
.
..
.
.
..
.
.
.
.
..
..
..
.
.
.
.
..
..
.....
..
.
.
.
..
..
...
..
.
.
..
.
....
.
.
.
..
.
.....
.
..
.
.
.
.....
..........................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....................................................................................
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
...
...........
.
...
..........................
...
...
.....
....
..................
... ........
... ........
............
.
..........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
................................................................................................................................................................................................................................................................................................................................................................................
..
.
....
.......
.
...
...
......
.....
.
.....
.....
.
.
...
....
.
.....
..
.
.
.
.
..
.
.
.
.
...
....
....
....
...
..
..
....
...
...
...
...
....
...
.
..
.
.
.
.
.
...
....
...
...
..
..
...
...
...
...
...
...
.
.
..
.
.
.
.
...
....
...
...
..
..
...
...
...
...
...
...
.
.
..
.
.
.
... .....
.... .....
.. ...
.. ...
... .
... .
.. ...
.. ...
... ...
... ...
.. ....
.. ....
... ..
... ..
....
....
....
....
....
....

145

t=T

!0 =!b = 10
y (t)

.....
......
a
.... ..
. ... ...
..... ..
.
.....
.
.
.
.
.
.
..
..
...
...
..
....
....
..
.....
.....
..
..
....
.....
.
.
.
.
.
.
.
..
.
.
.
.
.
....
...
.
..
..
.
.
.
.
..
..
.
.
.
.
..
..
..
.....
..
.
..
.
.
.
.
.
..
.
..
.
.
.
.
.
.
.
.............................
...............................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.................................
..................................
..
.........
..........
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.......................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.........................................
.
.
.
.................................................................................................................................................................................
..
..
.
..
..
.....
.
.
.
..
.
.
....
..
..
.....
..
..
....
.....
..
..
.
.
.
..
.
.
.
..
.
..
...
.
.
..
..
.
.
.....
.
.
..
.
.
.
.
....
.
.
.
.
.
.
.
..
.
.
.
..
.....
....
..
...
.....
.....
..
....
. .......
.. .. ......
.. .....
.............................................................................................................................................................................................................................................................................................................................................................................

t=T

y (t)

.....
......
b
.... ..
.... ...
..... ..
.
.....
.
.
.
.
.
.
..
.
..
...
.....
..
....
..
....
.....
..
..
.....
....
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.....
...
..
...
.
.
.
.
..
..
.....
..
.
..
.
.
.
.
.
..
.
..
.
.
.
.
.
..
.
..
..
.
.
.
..
..
.
.
.
.
.
..
..
..
..
.....
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.... ...
..... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.
.
...
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
...
...
..
..........
..........
.....
...
...
..........
..........
....
..........
...
...
..........
.....
..........
.........
.
.
...
.
.
...
.
.
.
.
.
.
.
..........
.
....
...
.
.
...
.
...
...
.
.
.
.
.
.
.
.
...
.
.
.....
.
.
.
.
...
.
...
.
.
.
.
.
..
....
.
....
.
.
.
.
...
.
.
..
.
.
.
.
...
.....
.
...
.
.
.
...
.
.
.
.
.
.
.
.
.
.
.
...
...
...
.........
....
...
....
.........
.. ........
... ........
........
... ...
........
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.................................................................................................................................................................................................................................................................................................................................................................................
.
....
....
........
........
...
...
.......
.
.
.
.
...
...
.
.
...
.
.
.
.
...
...
.
.
...
.
.
.
.
.
...
...
.
..
.
.
.
.
.
.
.
...
...
...
.
.
.
.
.
.
...
.
.
.
.......
.....
.....
.
.
....
......
..
..
......
......
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
...
.......
..
.......
......
......
...
....
......
......
..
..
.......
.......
.
.
.
.
.
.
.
...
.
.
.
.
... .......
.. ...........
.
.... ..........
... .....
.....
.....

146

t=T

The Vibrating String

This is one of the most celebrated problems of mathematical physics. And since the early
1950's, the unchallenged lead instrument in the sound track of Western culture has been
constructed around a set of six such steel strings. Let's see what our Fourier analysis can
tell us about this simple and important wave equation.
y(x; t0 )
..
...
...................
.....
....
..
....
...
...
....
.
...
.
..
.
..
.
...
..
.
.
.
.
...
.
...
.
..
.
..
...
.
.
.
...
.
...
...
.
..
...
.
.
.
.
...
.
...
........
.
.
.
...
.
...
.... ......
.
.
.
...
..
.
...
...
.
.
.. ........
...
.
...
.
..
.. ...
...
.........................................................................................................................................................................................................................................................................................................................................................
.
...
...
.
...
.
... ...
.
.
.
.
...
.
.
...
...
.....
...
...
.....
.... ......
...
...........
...................
...
.....
.
.
.
..
...
....
...
...
...
...
...
.
.
..
...
...
...
..
...
..
.
...
.
..... .......
...

Fig. 1. Snapshot of string displacement at time t = t0 .

Fig. 1 shows the prole (at some particular time t0 ) of the vertical deection of a string, of
total length a, about its equilibrium position y = 0. The string is xed at each end; hence
y(0; t) = y(a; t) = 0

(1)

are the boundary conditions imposed on our unknown function y(x; t), for all t.
.............
...............
.......... .
...........
.
.
.
.
.
0
.
.
.
.
..
...........
............................................................................

........................................................
..
......
......
......
.
.
.
.
.
.
......
. ........
..........
............

x + x

Fig. 2. Free-body diagram of elemental string length.

147

Consider the free-body diagram (Fig. 2) of a dierential length of string. If the entire
string is subject to a tensile force T (N), then the sum of the forces in the y-direction,
according to Newton's second law, must be
T sin 0 T sin = x

@2y
@t2

(2)

where is the lineal mass density (kg/m) of the string. For small displacements about
equilibrium
sin 0 tan 0 =

@
y(x + x)
@x

and

sin tan =

@
y(x)
@x

(3)

resulting in
@
@
y(x + x; t)
y(x; t)
@2
@x
@x
=
y(x; t):
x
T @t2

(4)

In the limit as x ! 0, we have the wave equation

where c =

1 @2
@2
2 2 y(x; t) = 0
@x2
c @t

(5)

p
T = will be the wave speed in (m/s).
(x)
...
..
....................
..
...
...
..
...
...
...
...
.
..
...
.
.
.
...
...
.
.
.
...
...
.
.
.
..
.
...
.
.
..
.
...
.
.
...
.
...
.
..
....
.
...
.
.
.
.
.........................................................................................................................................................................................................................................................
...
...

Fig. 3. Initial displacement of string.

Let's excite our string by holding it in the initial rest position

y(x; 0) = (x)
@
y(x; 0) = 0
@t

(one possibility shown in Fig. 3)

(initially at rest)

(6)
(7)

and releasing it. We can solve our partial dierential equation (5) by expanding y(x; t)
into a Fourier series in x, which is equivalent to periodically extending the function y(x)
beyond our original physical domain 0 x a. Since y vanishes at the end points x = 0
148

and x = a, we can use a sine series, which is the odd periodic extension of period 2a. So
the form of our sought-after solution is
1
X

y(x; t) =

Bn (t) sin(nx=a);

(8)

n=1

where the functions Bn (t) (Fourier coecients for the x-expansion) are unknown at this
point. We need the appropriate Fourier series representation of our initial displacement
(6); write it as
1
X
(x) =
bn sin(nx=a)
(9)
n=1

where the bn are known, in principle and in reality, since we know the initial displacement
(x):
Za
2
(x) sin(nx=a) dx:
(10)
bn =
a
0

Insertion of our solution form (8) into the wave equation (5) yields

1
X
1 d2
n 2
2 2 Bn (t) sin(nx=a) = 0:

a
c dt
n=1

(11)

This is itself a Fourier series representation of zero, so each coecient must be zero; that
is
2
nc 2
d
+
(12)
Bn (t) = 0:
dt2
a
The solution of this homogeneous dierential equation is a linear combination of

cos
sin

nct
a

We select the cosine (discard the sine) in accordance with the initial condition (7). Now
we have Bn (t) = n cos(nct=a) in terms of (unknown) constants n , which gives
y(x; t) =

1
X

n cos(nct=a) sin(nx=a):

(13)

n=1

Evaluation at t = 0, according to the initial condition (6), provides n = bn , and so

y(x; t) =

1
X

bn cos(nct=a) sin(nx=a):

n=1

149

(14)

The physical interpretation of our solution(!) (this is it!!!) is much clearer upon rewriting
it (via a trig identity) as
1
i
h n
io
1 X n h n
bn sin
(x ct) + sin
(x + ct)
y(x; t) =
2 n=1
a
a

or, nally,
y(x; t) =

i
1 h~
~ + ct)
(x ct) + (x
2

(15)

(16)

~ means the odd periodic extension (Fig. 4) of the initial displacement (x) as
where ()
given in the sine series (9):
~
(x)
=

1
X

n=1

(x 2na)

1
X

n=1

[x (2n + 1)a]:

(17)

In (17) we are taking () to be zero outside of the original domain 0 a. Also note
~ ct) is a wave propagating with speed c in the x-direction.
that (x

~
(x)

Fig. 4. Odd periodic extension of initial displacement of string.

Problems
1.) Verify that any functions f (x + ct) and g(x ct) satisfy the wave equation (5).
Certainly, then, (16) is a solution. Such forms are called the d'Alembert solution.
2.) Verify that (16) satises the initial conditions (6) and (7), and the homogeneous
boundary conditions (1).
3.) Why do we say that f (x + ct) and g(x ct) are propagating in the x-directions,
respectively, with speed c?

150

Poisson Sum Formula

Let f (t) be any function that is Fourier transformable. Then its periodic extension can be
represented by a Fourier series:
1
X

g(t) =

f (t + nT ) =

n=1

1
X

ck ej2kt=T

k=1

with Fourier coecients

1
ck =
T

g(t)ej2kt=T dt

0
1 Z
1 X
=
f (t + nT )ej2kt=T dt
T n=1
T

with = t + nT
(n+1)T
Z

1
1 X
=
T n=1

1
=
T

f ( )ej2k =T ej2kn d

nT
j2k =T

f ( )e

1
d = F
T

2k
T

The resulting Poisson sum formula

1
1 X
2k
f (t + nT ) =
F
ej2kt=T
T
T
n=1
1
X

k=1

is often written in the common form, with t = 0 and T = 1 such that

1
X

f (n) =

n=1

1
X

F (2k):

k=1

F Homework. Sum the slowly converging series

1
X

n=1

= coth b:
2
+b
b

151

Solution. Recall that the Fourier transform of g(t) = exp(jtj) with > 0 is
G(!) =

e(j!)t dt +

e(+j!)t dt =

1
2
1
+
= 2
:
j! + j!
+ !2

Duality says that the Fourier transform pair

g(t) () G(!)

G(t) () 2g(!):

also gives

so that the Fourier transform of

f (t) =
is
F (!) =

1
+ 2

j!j
e
:

We will also use the geometric series

1
X

zk =

k=1

The Poisson sum formula

1
X

z
1z

f (n) =

n=1

(jzj < 1):

1
X

F (2k)

k=1

therefore gives (assume b > 0 and note that the desired sum is even in b)
"
#

1
1
1
X
X
1
X 2jkjb
2e2b

2bk
1+
=
e
=
1+2
e
=
2 + b2
n
b
b
b
1 e2b
n=1
k=1

k=1

1+e
e +e

=
= coth(b):
2b
b
b
b 1e
b e e
b

It is interesting to look at the limiting case of very small b. If jbj 1, then

1
X

1
X
1
1
1
1
1
2
:
!
+2
= 2 + 2(2) = 2 +
n2 + b2 b!0 b2
n2
b
b
3
n=1
n=1

Formula 4.5.67 of Big Red shows that, for jzj < ,

coth z =
so that

1 z
z3
+
+ :::
z
3 45

b (b)3
1
2
1
coth(b) !
+

+::: = 2 +
+ O(b2 ):
b
b
b
3
45
b
3
b!0

How can you see the behavior of the sum for very large b, that is for b ! 1 ?
152

Sampling and Reconstruction

..........................................................
...
....
....
...
........ ........
..
........... .............
...
s
...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
................................................
......................................... ........... ..............................................................
..
..
...
... ...... ...... ...
...
......
.......
...
.
..
...............
...
.
.....
.........................................................
.
.........
...
..
...

x (t)

x(t)..

h(t)

y(t)

combT (t)
Consider a continuous-time (analog) signal x(t) that is uniformly sampled at the times
tn = nT (n = : : : ; 2; 1; 0; 1; 2; 3; : : : ) so that the sampling frequency is

2
(1)
T
in terms of the sampling period or interval T . Once x(t) has been sampled, all we have is
a table (list) of numbers x(nT ). But a useful representation of the sampled signal xs (t) in
continuous-time notation is
1
X
xs (t) = x(t) combT (t) =
x(nT )(t nT ):
(2)
!s =

n=1

From our table of Fourier transform properties we need

1
F (!) ~ G(!):
F f (t)g(t) =
2

(3)

The easiest way to get this is to take the inverse Fourier transform of the frequency-domain
convolution
Z1 Z1

1
1
F
F ()G(! ) d ej!t d!
F (!) ~ G(!) =
2
1 1

1
F ()
2

Z1
1

G(! )ej!t d! d:

The inner integral (in !) is

1
2

Z1
1

1
G(! )ej!t d! =
2

G()ej(+)t d = ejt g(t)

and therefore

F 1 F (!) ~ G(!) = g(t)

F ()ejt d = 2f (t)g(t)

which is identical to (3).

Typeset by AMS-TEX
153

The Fourier fransform of the time-domain comb function is

2
comb 2 (!)
F combT (t) =
T
T

(4)

and thus the spectrum of the sampled signal xs (t) is

Xs (!) =

1
X
1
1
2
(! k!s )
X(!) ~
comb 2 (!) = X(!) ~
2
T
T
T
k=1

1
T

1
X

k=1

X(! k!s ):

(5)

If x(t) is a \low-pass" signal, then

X(!) = 0 for j!j > !h :

(6)

This means that the spectrum of x(t) has a highest frequency !h . An example low-pass
spectrum appears in the rst row of the gure below.
X(!)

Xs (!) when !s =2 > !h

Xs (!) when !s =2 < !h

!s =2

2!s

!s =2

2!s

The middle curve shows Xs (!) for the case !s > 2!h and the bottom curve is for !s < 2!h .
If we pass xs (t) through an ideal low-pass lter having transfer function
H(!) = T
154

!
!s

(7)

then we see that, for all !,

Y (!) = X(!) if !s > 2!h :

(8)

In this case, y(t) = x(t) and we have exactly recovered the low-pass signal x(t) from its
sample values! The minimum sampling frequency is called the Nyquist rate or Nyquist
frequency 2!h . If we sample at too low a frequency, as shown in the bottom curve of the
previous gure, then
Y (!) 6
= X(!) if !s < 2!h
(9)

and of course then y(t) 6

= x(t). The error or distortion that results from the overlap in the
adjacent shifted transforms X(! k!s ) is called aliasing. The whole concept is usually
called:
Shannon's sampling theorem. If the low-pass signal x(t) having spectrum X(!) = 0
for j!j > !h is sampled uniformly at the rate !s > 2!h (the Nyquist rate), then x(t) is
exactly recoverable by passing the sampled signal through the ideal reconstruction lter
having transfer function H(!) = T (!=!s ).
The frequency domain renders this derivation clear and simple. Now let's examine the
output y(t) in the time-domain to get another perspective. The impulse response of the
ideal reconstruction lter is

h(t) = F 1 T (!=!s ) = sinc(!s t=2) = sinc(t=T );

(10)

and this ideal low-pass lter is also called the sinc interpolation lter. Independent of x(t)
being a low-pass signal or not, and for an arbitrary sampling frequency (that is, ignoring
both Nyquist and Shannon), the output of the lter in response to the sampled signal (2)
is always
y(t) = xs (t) ~ h(t) =

1
X

n=1

x(nT )(t nT ) ~ h(t) =

t
=
x(nT ) sinc
n :
T
n=1

1
X

n=1

x(nT )h(t nT )

1
X

(11)

If k is an integer, observe that

sin(k)
sinc(k) =
=
k

1; k = 0
= k0 ;
0; k =
60

(12)

in terms of the Kronecker-delta. Therefore, we see that at the sample times t = mT , the
output is always equal to the input in that
y(mT ) =

1
X

n=1

x(nT ) sinc [ (m n)] =

155

1
X

n=1

x(nT )mn = x(mT ):

(13)

Dissection of the Time-Domain Output of the sinc Interpolation Filter: Consider

a small number (four) of samples x[n] = x(nT ) for n = 0; 1; 2; 3. If all other (n = 4; 5; : : :
and n = 1; 2; : : : ) of the x[n] are zero (or absent), then the output (11) of the sinc
interpolation lter is

t
t
+ x[1] sinc
1
y(t) = x[0] sinc
T
T

t
t
+ x[2] sinc
2 + x[3] sinc
3 :
T
T
Each of these individual terms are graphed below for the example data
x[0] = 0:5

x[1] = 1:0

x[2] = 0:8

x[3] = 0:6

along with the sum of all four terms. Do you see which curve is which?

1.0

0.8

0.6

........... .
........ ................
.... .
. .......
.... ....
... ....
.
.
.. .
.. .....
....
... ...
.
.
...
.....
.. .
.
.
..
.....
.. ..
.
.
.....
. ..
.
.
.
...
......
. .
.
.
.
......
.
.
.
.
.
.
.
...
.. ..
.........
.
.
..
.
.. ...............
.
.
.
.
. ........
.
.
.
.
...
.
.......
.
..
.
.
.
..
.
......
.
.
.....
...
...
.
.
.
.
.
.
...
.
....
.
.
.
.
.
.
.
...
. .
.
.
.
.
.
.
...
.
... ..
.
.
.
.
.
.
.. .
...
.
.
.
.
.
..
.
.
.
.
.
.
.
....
.
...........
.
.
.
.
.
... ...........
.
.
.
.
.
.
.
.
.
.
...
.. ...
.
.
..
.
.
.
... ..
.
.
....
...
... ...
. ...
. ...
...
... ..
. ..
......
.
.
.
.
.
.
.
.
.
.
... ...
..
... .... ..... .
.
.
.
.
... ..
.
.. ..
. ..
.
.
.
.
.
.
... ...
.. .
...
. ...
.
.
.
.
.
....
... ..
..
. ...
.
.
.
.
.
.
... ...
.
. ..
..
.
.
.
.
.
..
... ..
..
. ..
.. ...
.
.
.
.
.
... ...
.
.
.
.
.
..
.
.
.
.
...
... ..
.
..
. ..
.
.
..
.
.
.
.
... ...
..
..
..
.
.
.
.
.
.
.
.
.
... ..
.
.
.
..
.
.
.
.
.
.
.
.
.
.
... ...
...
.
.
.
.
.
.
.
.
.
..
.
... ..
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
... ...
.
.
.
.
.
.
.
.
.
.
.
.
... ..
.
.
..
.
.
..
.
.
.
.
... ..
.
.
.
.
.
.
.
..
.
.
.
.
.
.
... ..
.
.
..
...
..
.
..
.
.
.
... ...
.
.
..
..
.
.
.
.
.
.
..
.. ..
.
.
.
..
.
..
.
.
.
.
... ..
.
.
.
.
.
.
...
.
.
.. ..
..
. ....
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. ..... .........
.
. ..
.
.
... .....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ....
.
.
..
.
.
...
..........
. ....... .....
.
.
.
.
.
.
.
.
.
.
.
.
...
..
. .
.
.
............
...
...
. .......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
.
.
.
.
.
.
.
... .. . ... ... ... ...
.... . .
..
.
.......
.
.
....
. ....
.... .. .
..... .. ..
...... ... .. ... ... ....
.. .. .....
... ... ......
... ......................... .... .............
.
.
.
.
.
.
.
.
... ... . ..
.
.
.
.
..... . ..... ..
...... .
.... . ...
..
..... . .........
.
..
..
.
.
.
.
.
.
.
.
.
.
.
. .
.
.
.
.
.
.
.
.
.
.
.
.
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................
.................. ..... . ....
..
.....
. .
. ........
... . ........
.........
.. .. .......
........ ...........
.... .... ..
............
.
.
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ...
..........
... ....
.. ................... . .
..
.. .....
. . .
. .......
.
.. ..
... .... .. ...... ....
.. ......... .........................
....... ..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
....
.
.
.
.
.
.
...
... ............... ....
...
........
.
.
.. .
.
.
..... ... ......
..
...
..
.
........... ... .......
.
.
..
..
...... ..........
.
.
...
.............
...
..
...
..
....
.
..
.
..
...
.
.
.
.
.
.
.
.
.
.
.
... ...
.
.... ....
.
.
...
..
.
.
.
.
.
..
...
.
..
.
.
.
.... ....
........ .....
.....

0.4

0.2

0.0

0:2
0:4

2
t=T

156

DISCRETE-TIME SIGNALS, SYSTEMS, AND TRANSFORMS

A common source for a discrete-time signal is to uniformly sample a continuous-time signal,
as in
x[n] = x(nT )
n = : : : ; 2; 1; 0; 1; 2; 3; : : :
The discrete-time signal x[n] is also called a sequence; it is a function of the integers
n. Usually we are not concerned with any underlying sample period T that was used to
obtain our sequence from some original x(t). Once we're in the discrete-time world, the
independent time variable is simply the integers n. Our sequence x[n] is simply a list or
table of ordered numbers. Note that the sequence x[n] can take on real or complex values.

Many of our familiar continuous-time signals have discrete-time analogues or cousins, but
be aware that their behavior is often (usually? always?) quite dierent. For starters, there
is no such thing as continuity in discrete-time. The continuous-time calculus operators of
dierentiation and integration have analogous operators of dierencing and accumulation
in discrete-time. Sometimes the discrete-time version of a common signal, for example
the Heaviside unit-step function u[n], goes by the same name and similar symbol as its
continuous-time counterpart. In other cases, dierent names are used. I suppose the
Heaviside unit-step sequence is a better name that draws attention to its dierence with
the continuous-time signal u(t).
Heaviside unit-step sequence.

0; n = 1; 2; : : :
u[n] ,
1; n = 0; 1; 2; : : :

u[n]

......

:::

:::
n

Unlike u(t) that is not well-dened at t = 0, the unit-step sequence clearly takes on the
value u[0] = 1. Note, however that u[0:5] is complete notational nonsense.
Kronecker delta sequence. (also called unit impulse sequence)
[n] .....
....
:::
:::
..

...
..
0; n = 1; 2; 3; : : :
.
..............
................................
................................
...............................................................
...............................
................................
...............................
.................. n
[n] ,
3 2 1
1; n = 0
0
1
2
3
4
Unlike the continuous-time Dirac-delta function (t) that is illegal at t = 0, the discretetime Kronecker-delta sequence clearly takes on the value [0] = 1. The Kronecker-delta
possesses the sampling property
x[n] =

1
X

k=1

x[k][k n]:

Relationship.
[n] = u[n] u[n 1]

u[n] =

n
X

k=1

157

[k]

Complex exponential. First, recall that for any positive real number ! (called the
frequency), the continuous-time complex exponential
x(t) = ej!t
is always periodic in time t with period T = 2=!, since
x(t + T ) = ej!(t+T ) = ej!T ej!t = ej!t = x(t)
whenever !T = 2, since e2j = 1.

The discrete-time complex exponential is substantially dierent. Consider the discretetime sequence
x[n] = ejn
where the frequency is any real number: Take it to be positive for starters. Is x[n]
periodic in the independent time variable n ? Let's denote its candidate period by the
specic integer N , and examine
x[n + N ] = ej(n+N ) = ejN ejn :
If x[n] is periodic with the period N , then x[n + N ] = x[n] which requires that
N = 2k
where k is any integer. This only happens if
=

k
2 = a rational multiple of 2:
N

However, the discrete-time complex exponential is a periodic function of the frequency

variable , since
ejn(+2) = e2nj ejn = ejn :
Therefore, in discrete-time work, we only have to consider frequencies in the primary
range < , for example. The highest frequency possible is = (or equivalently,
), and the lowest frequency is = 0 (which is identical with = 2).

158

N ......

Exercises.

N 1

x[n]

1. Write an expression, in terms of n, N ,

and u[], for the given sequence x[n].

:::

: : : N 1 N

N +1

:::
n

2. Graph the exponential (or geometric) sequence x[n] = an u[n].

Look at the following specic cases:
a = 1=2;

a = 1=2;

a = 2;

a = rej

where r = 1=2 and = 1

Discrete-Time Systems

x[n]

y[n]

The equation-of-motion, that describes the relationship between the input x[n] and the
output y[n], for a discrete-time system is often a dierence equation. An example of a
second-order, constant-coecient dierence equation is
c0 y[n] + c1 y[n 1] + c2 y[n 2] = b0 x[n] + b1 x[n 1]:
When accompanied by two (the dierence equation is second order) initial conditions of
the form
y[1] = y1 and y[2] = y2 ;
we have a discrete-time initial value problem. Classical solution techniques are analogous
to the classical methods used to solve ordinary dierential equations in continuous-time.
Dierence equations can sometimes be solved recursively:
y[n] =

c2
b0
b1
c1
y[n 1] y[n 2] + x[n] + x[n 1]:
c0
c0
c0
c0

The discrete-time version of the Laplace transform (or Fourier transform) is called the
Z-transform, and it, too, provides an attractive path to solve discrete-time dierence
equations by converting them into purely algebraic equations in the transform (z) domain.
The inverse Z-transform gets our solution back to the time [n] domain.
159

Linear Time-Invariant Discrete-Time Systems

x[n]

y[n] = T fx[n]g

Assume that the zero-state response sequence y[n] to the exciting sequence x[n] is given
by the operator or system transformation rule y[n] = T fx[n]g. If the operator T satises
the superposition property
T fc1 x1 [n] + c2 x2 [n]g = c1 T fx1 [n]g + c2 T fx2 [n]g
for any values of the constants c1 and c2 and for arbitrary sequences x1 [n] and x2 [n], then
the discrete-time system is linear.
Given that the zero-state response to the input sequence x[n] is y[n], if the response to the
delayed input x[n n0 ] is the delayed response y[n n0 ], then the system is time-invariant.

x[n]

x[n n0 ]

y[n] = T fx[n]g

y[n n0 ] = T fx[n n0 ]g

Input/Output relationship. Using the sampling property of the Kronecker-delta sequence, write the input sequence as
x[n] =

1
X

k=1

x[k][n k]:

The system operator T operates only on functions of the running time variable n, treating
all other variables as constants, so that the T operator can be moved inside the summation
in
( 1
)
1
X
X
y[n] = T fx[n]g = T
x[k][n k] =
x[k]T f[n k]g:
k=1

k=1

The response to a Kronecker-delta or unit impulse sequence is the system impulse response
h[n] = T f[n]g. By the (assumed or dened) time-invariance of the linear system, the
160

response due to a delayed unit impulse [n k] is the delayed h[n k] = T f[n k]g, and
therefore the zero-state response is
y[n] =

1
X

k=1

x[k]h[n k] = x[n] ~ h[n]:

This is the discrete-time convolution sum between the two signals x[n] and h[n].

x[n]

h[n]

y[n] = x[n] ~ h[n]

Example of a LTI Discrete-Time System. Let's look at one particular discrete-time

system in detail and study its input/output dynamics from several perspectives. One
common way to specify a discrete-time system is via a system block diagram. We need to
rst introduce several basic components of such system block diagrams.

summer

f [n]

.....................
....
...
...
...................................
.
.
.
.
.
.
...................................
....
..
...
...
.
.....
...................
...
.
......
..
...
..
...
...
.

f [n] + g[n]

g[n]

amplier/attenuator

unit delay

f [n]

..........
... ..........
........
.
.
........................................
.
.
.
.
.....................................
..
........
..... .............
........

f [n]

...............................
...
...
..
.

f [n] ............................................. z 1 ...........................................f [n 1]

...
.
.............................

The reason that the unit delay is represented by the symbol z 1 will be clear when we
study the Z-transform.

161

x[n]

y[n]

The output of the summer is

y[n] = x[n] + y[n 1]
and so the discrete-time system depicted by the block diagram above is governed by an
equation-of-motion that is a linear, constant coecient, rst-order dierence equation.
The impulse response h[n] is the response y[n] to the input sequence x[n] = [n], and being
a zero-state response, its initial condition is h[1] = 0. Sometimes dierence equations
can be solved recursively, and this is one
h[n] = h[n 1] + [n]

h[1] = 0

h[0] = h[1] + 1 = 1
h[1] = h[0] + 0 =
h[2] = h[1] = 2
h[3] = h[2] = 3
..
.
h[n] = n u[n]
With the unit impulse response h[n] in hand, the zero-state response due to any input
sequence is readily computed via the convolution sum. For example, the unit step response
is
s[n] = u[n] ~ h[n] =

1
X

k=1

h[k]u[n k] =

n
X

k=1

h[k] =

n
X

k=1

n+1

1
1

u[n]:

The required nite geometric series is detailed two pages ahead.

162

u[k] =

n
X

k=0

The Z-Transform
In order to derive the Z-transform from the Laplace transform, we need to represent our
sequence x[n] in continuous-time notation: Take a step backwards, so to speak, and write
a sampled (discretized) continuous-time signal as
.........
.. ....
1
1
.. .....
.... .... ....
.. ...
.
... ..
...
d
T
.
.. ...
... ...
...
..
.. ...
n=1
n=1
...
.. ...
.
...
...
...
..
...
...
..
...
.
...
...
...
..
...
..
..
..
...
.......
..
......
..
...
.. ...
.. ..
..
... ...
..
.
...
.. ...
........
.. ...
..
....
... ..
..
.....
.
.
...
.
.. ...
.. ...
......
..
... ...
..
.... ..
... ..
...
...
..
... ...
.
..
..
...
.
..
....
..
..
...
....
...
...
..
..
..
...
......
...........................................
..........................
.........................
..
..
..
.... ....
...
...............
..............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
..
..
.
.
.
.
...............
..
.
.
.
.....
.
.
.
.
.
.
....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...........................................................................................................................................................................................................................................................................................................................................................................................................
.
.
...
.
.
...
...
...
...
...
.
...
..
..
..
.
.. ........
... ...
...
...
...
... .....
... ...
..
..
..
.. ....
...
...
... ...
...
.....
.
.
..
..
... ..
.
..
... .......
...
...
... ...
.. ......
..
..
.....
......
...
...
....
..
..
...
.....
.
.
.
.
.
.
...
... ....
.... ....
.
..... ..
............
........
................

x (t) = x(t) comb (t) =

x(nT )(t nT ) =

x[n](t nT ):

The bilateral Laplace transform of a continuous-time signal f (t) is dened as the integral
Lff (t)g = F (s) =

f (t)est dt

where s = +j! is a complex frequency variable. The Laplace-transform of the discretized

signal xd (t) is therefore
Xd (s) =

1
X

n=1

x[n]

(t nT )e

dt =

1
X

x[n]ensT :

n=1

Introduce the transformed variable

z = esT
so that the above Laplace-transform is now called the Z-transform
Zfx[n]g = X(z) =

1
X

x[n]z n :

n=1

The Z-transform converts a function of the integers n (a sequence) to a function of a

complex variable z.
163

Geometric series. Firstly, consider the nite geometric series

N
X

n=0

where a is an arbitrary complex number. There is no question about the convergence

of this series, since it is simply the sum of a nite number of terms. We can obtain a
closed-form expression for the sum S by this neat trick:
S = 1+a + a2 + a3 + + aN 1 + aN

a + a2 + a3 + + aN 1 + aN + aN +1

aS =

S aS = 1aN +1

N +1

S(1 a) = 1 a

N
X

an =

n=0

1 aN +1
1a

Note that when a = 1, we have S = N + 1, which is the limit as a ! 1 of the above result.
Similarly, the expression for the innite geometric series
1
X

an =

n=0

1
1a

(jaj < 1)

explicitly displays its region-of-convergence jaj < 1.

Z-transform example: causal exponential sequence. Consider the sequence
x[n] = an u[n]
where a is an arbitrary complex number. Then the Z-transform of x[n] is
X(z) =

1 n
X
a

n=0

z
a = za
1
z

(jzj > jaj):

The region-of-convergence in the complex z-plane for X(z) to be a valid transform of x[n]
is the exterior of the circle of radius jzj = jaj.

164

Fibonacci Sequence - Dierence Equation and Z-Transform

f [n] = f [n 1] + f [n 2]

initial conditions: f [2] = 0;

Zff [n]g = F (z) =

Zff [n 1]g = f [1] + z 1 F (z);

1
X

f [1] = 1

f [n]z n

n=0

Zff [n 2]g = f [2] + z 1 f [1] + z 2 F (z)

F (z) = 1 + z 1 F (z) + z 1 + z 2 F (z)

F (z) =

z(z + 1)
1 + z 1
= 2
1
2
1z z
z z1

F (z)
z+1
z+1
= 2
=
p
p
z
z z1
z 1+2 5 z 12 5

1
1
p3
p3
1
+
1

2
2
5
5
p
p
=
+
via partial fraction expansion
z 1+2 5
z 12 5
1
f [n] =
2

p !n
p !n

3
1
3
1+ 5
1 5
1+ p
1 p
+
2
2
2
5
5

for n = 2; 1; 0; 1; 2; : : :
p
f [n + 1]
1+ 5
!
n!1
f [n]
2

f = 0; 1; 1; 2; 3; 5; 8; 13; 21; 34; 55; 89; 144; : : :

165

Convolution Theorem of the Z-Transform

Zff [n] ~ g[n]g =
=

1
X

n=1
1
X

1
X

k=1

f [k]

k=1

(let m = n k)

1
X

k=1
1
X

f [k]z

f [k]g[n k] z n

1
X

n=1
1
X

g[n k]z n

g[m]z (m+k)

m=1
1
X
k

g[m]z m

m=1

k=1

= F (z)G(z)

Z-Transform of a Delayed Sequence.

Zff [n n0 ]g =

1
X

n=1
n0

f [n n0 ]z n =

1
X

f [m]z (m+n0 ) = z n0

m=1

1
X

f [m]z m

m=1

F (z)

System Transfer Function.

y[n] = x[n] ~ h[n]

()
H(z) =

Y (z) = X(z)H(z)

Y (z)
X(z)

Recall the system block diagram for the example four pages back. The Z-transform of the
governing dierence equation
y[n] y[n 1] = x[n]
is
(1 z 1 )Y (z) = X(z)
and therefore the system transfer function is
Zfh[n]g = H(z) =

1
z
=
:
1 z 1
z

166

Discrete-Time Fourier Transform. With z written in polar form as z = rej , if we

restrict z to lie on the unit-circle (jzj = r = 1), then we have the discrete-time Fourier
transform
1

j
= X(e ) =
x[n]ejn :
X(z)
jzj=1

n=1

The discrete-time Fourier transform (DTFT) is a mapping from a function of the integers
n to a periodic function of the continuous real variable , with period 2. Observe that
the DTFT is essentially a Fourier series representation of the periodic function X(ej ),
where x[n] are the Fourier coecients! Therefore, the inverse DTFT is simply obtained
by applying the operator
Z
1
d ejm
2

to the DTFT above

1
2

X(e )e

Z
1
d =
x[n]
ej(mn) d = x[m]:
2
n=1

|
{z
}
1
X

The discrete-time Fourier transform pair is therefore

DTFTfx[n]g = X(e ) =

1
X

x[n]ejn

n=1

DTFT

1
fX(e )g = x[n] =
2
j

X(ej )ejn d:

Discrete Fourier Transform. Consider a nite or partial sequence x[n], where our
attention is centered on the indices n = 0; 1; 2; : : : ; N 1. Its DTFT is
j

X(e ) =

N
1
X

x[n]ejn :

n=0

If we now sample the DTFT at N values of

k =

2k
N

(k = 0; 1; 2; : : : ; N 1)

that are uniformly distributed around the unit circle, we have

X(e

N
1
X

x[n]ej2nk=N :

n=0

167

This is a mapping from a function x[n] of the integers n = 0; 1; 2; : : : ; N 1 to another

function of the same integers k = 0; 1; 2; : : : ; N 1. Rather than write it as the cumbersome
X(ejk ), we focus our attention on the discrete frequency parameter k and write it as the
discrete Fourier transform or DFT
X[k] =

N
1
X

x[n]ej2nk=N

(k = 0; 1; 2; : : : ; N 1):

n=0

The inverse DFT is most easily derived by rst looking at the comparable discrete-time
spectral representation of the Kronecker-delta sequence
[n] =

N 1
1 X j2nk=N
e
:
N
k=0

The truth of this last equation can be established in a number of ways. Observe that
the sum is a nite geometric series (page 163): The result follows at once. Therefore,
application of the operator
N 1
1 X j2mk=N
e
N
k=0

(m = 0; 1; 2; : : : ; N 1)

to the DFT above gives

N 1
N
1
N 1
X
1 X
1 X j2(mn)k=N
j2mk=N
X[k]e
=
x[n]
e
= x[m]:
N
N
n=0
k=0
k=0
|
{z
}
mn

Hence, the DFT transform pair is

DFTfx[n]g = X[k] =

N
1
X

x[n]ej2nk=N

n=0

DFT

N 1
1 X
fX[k]g = x[n] =
X[k]ej2nk=N
N
k=0

(k = 0; 1; 2; : : : ; N 1)
(n = 0; 1; 2; : : : ; N 1):

The DFT (FFT is the fast algorithm that eciently computes the DFT) is therefore a
transformation that converts N numbers into N other numbers. But each of the sets of
N numbers, x[n] and X[k], are a single period of periodic sequences, each with period N .
To see this, observe that
x[n + N ] =

N 1
N 1
1 X
1 X
X[k]ej2(n+N )k=N =
X[k]ej2nk=N ei2k
N
N
k=0

1
N

N
1
X

k=0

X[k]ej2nk=N = x[n]

k=0

168

and
X[k + N ] =

N
1
X

x[n]e

N
1
X

x[n]ej2nk=N = X[k]:

j2(k+N )n=N

n=0

N
1
X

x[n]ej2nk=N ej2n

n=0

Exercises.
1. Find the Z-transform and its region-of-convergence for the nite, rectangular pulse
sequence x[n] = u[n] u[n N ], where N > 0.
2. Given the sequence x[n] = [n] (n = 0; 1; 2; : : : ; N 1), nd its DFT X[k].
3. Given the sequence x[n] = 1 (n = 0; 1; 2; : : : ; N 1), nd its DFT X[k].
4. The Bessel function of order n and argument t is dened as the integral
1
Jn (t) =
2

ej(nt sin ) d:

Approximate this integral using the DFT. Use the fft algorithm in MATLAB to generate
values of Jn (1) and Jn (50) for n = 0; 1; 2; : : : ; 20. Compare with Table 9.4 of Big Red:
M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions with Formulas,
Graphs, and Mathematical Tables. National Bureau of Standards (also reprinted by Dover).
5. Recall from page 165 that the Z-domain transfer function of the system having block
diagram on page 161 is
1
:
H(z) =
1 z 1

Graph the magnitude and phase response H(ej ) as a function of for these values of the
constant :
(a) = 1 and = 1 (graph on the same axes)
(b) = 1=2 and = 1=2 (also graph on common axes)

Graph 20 log10 fjH(j)jg (dB) for the magnitude characteristic. Explain the frequency
behavior with respect to the sign of .

169

DFT APPROXIMATION OF THE CONTINUOUS-TIME

FOURIER TRANSFORM

F (!) =

j!t

f (t)e

1
X

N=2
j!n t

f (n t)e

n=1

t t

f (n t)ejn! t ;

n=N=2+1

subject to appropriate restrictions. Evaluation at specic frequencies gives

k2
N t

N=2

f (n t)ej2nk=N :

n=N=2+1

Dene the periodic sequence

x[n] =

f (n t);
n = 0; 1; : : : ; N=2
f ((n N )t); n = N=2 + 1; : : : ; N 1

where x[n + N ] = x[n]. Also note that exp(j2nk=N ) is periodic in both n and k with
period N . The approximate Fourier transform above is now
F

k2
N t

N
1
X
n=0

x[n]ej2nk=N t X[k]

in terms of the DFT X[k] of the sequence x[n]. If F (!k ) is desired for negative k, employ
the periodicity of X[k] = X[k + N ]. Also note that the range of available frequencies is

1
!
:
N
t

170

...
.
..................................
...
.

Numerical Example of Using the FFT to Evaluate the Fourier Integral:

Time-Domain Response of an RC Filter to a Gaussian Pulse
....
..
..
...
.........................................................................................................................
..
..
...
...
..
......
......
.
........
.
.
.
.
.
.
........
........
......
...
........
............
........
......
...
.......
.............
....
..
...
....................................................................................................................................................................................................

....
..
..
.
................................................................
...
..
..
..

x(t)

y(t)

y(t) = x(t) ~ h(t)

()

Y (!) = X(!)H(!)

(1)

The time-domain input to the simple RC lter is a Gaussian pulse

x(t) = et

(2)

(3)

having spectrum
X(!) =

!2 =4
:
e

The circuit transfer function is

H(!) =

Y (!)
=
X(!)

R
1
R+
j!C

j!RC
j!=!b
=
1 + j!RC
1 + j!=!b

(4)

where the lter break frequency is dened to be

!b =

1
:
RC

(5)

The time-domain response is the inverse Fourier transform

1
y(t) =
2

j!t

H(!)X(!)e

1
d! =
2

Z1
1

j!=!b
1 + j!=!b

!2 =4 j!t
e
e d!:

(6)

Let = 2 so that (6) becomes a little clearer

1
y(t) = p
2

Z1
1

1
t
!b
2
d!:
exp 4 (! ) exp j!
j!

1+
!b
171

(7)

Now introduce the normalized variables

z = !;

zb = !b ;

(8)

so that the circuit input and output signals are

x(v) = ev

1
y(v) = p
2

(9)
Z1
1

jz=zb z2 =4 jzv
e
e dz:
1 + jz=zb

(10)

Normalized time is v = t= and the single parameter requiring a numerical value is

zb = !b =

1
=p
:
RC
RC

(11)

This parameter is essentially the ratio of the input \pulse width" to the circuit timeconstant. Note that as the cut-o frequency of the high-pass lter approaces zero, the
output approaches the input
1
y(v) ! p
zb !0 2

1
x(v) = p
2
Let

z 2 =4 jzv

=4 jzv

dz = x(v):

(12)

1
dz p
2

N=2

e(nz)

=4 jnz v

n=N=2+1

2k
(k = N=2 + 1; : : : ; N=2)
N z

N=2
X
2
2k
z
p
x
e(nz) =4 ej2nk=N
N z
2
vk =

n=N=2+1

x[k] =

N z
p FFT1 fX[n]g
2

with

x[n] = e(nz)

172

1.00

x(t)

0.75

0.50
y(t)
0.25

0.00

0:25

0:50

0:1

=RC = 10

Fig. 1. Response of RC High-Pass Filter to a Gaussian Pulse x(t) = exp[(t= )2 ]

Cases: =RC = 0:1; 1; 10

173

Example: DFT Approximation of a Fourier Sine Integral

8 0;
>
>
>
2t
>
>
< ;
T
f (t) =
2
>
>
>
(T t);
>
...........
.
.
.
.
.
.
>
T
........ .......
:
............ ................
.
.
.
.........
....
.........
.......
0;
..........
.........
..
........

f (t)

t<0
0 t < T =2

T =2

T =2 t < T
tT

N t

exact Fourier sine transform for comparison (never start with one you don't know!)

Fs (!) =

2T
2
sin(!T
=2)

sin(!T
)
(!T )2

f (t) sin(!t) dt =

Fs (!)

2k
N t

N
1
X

f (nt) sin(n!t)t

n=0

N
1
X

f (nt) sin(2nk=N )

n=0

N 1

jt X
f (nt) ej2nk=N ej2nk=N
=
2 n=0

discrete Fourier transform (DFT) of a sequence x[n] is the sequence

X[k] =

N
1
X

x[n]ej2nk=N

n=0

(k = 0; 1; 2; : : : ; N 1)

both sequences are periodic with period N , that is X[k + N ] = X[k]

let x[n] = f (nt) for n = 0; 1; 2; : : : ; N 1

o
2k
jt n
Fs

X[k] X[N k]
N t
2
hopefully good for k < N=2 or slightly less

174

In this particular run, T = 1, t = 0:02, and N = 1024. In the range 0 < ! < 20
the approximate and exact curves are indistinguishable to the naked eye. Therefore, the
percentage error
Fap (!) Fex (!)
100%
E(!) =
Fex (!)
is also graphed.

0.4
0.3
Fs (!)
0.2
0.1
0.0
0:1

1.4
1.2
1.0
0.8
E(!)

0.6
0.4
0.2
0.0

10
!

175

10
!

% DFTsineFT.m DFT for a sine Fourier Transform

% 9 November 2008 R.W.Scharstein
% Practice for the Army 3-dim slotted cylinder problem
% and a good example for ECE 370.
%
% Original f(t) is causal, and an isosoles triangular waveform of total
% pulse width T.
T=1; N=1024;dt=0.02;
n=0:(N-1); t=n*dt;
x=(2*t/T).*(t<=(T/2))+(2*(T-t)/T).*(((T/2)<t)&(t<=T))+0*(t>T);
%plot(t,x)
X=fft(x); Y=[X(2:end) X(1)]; XX=fliplr(Y);
%[n' X.' XX.']
Fs=(i*dt/2)*(X-XX); w=2*pi*n/(N*dt);
% Fexact has a divide by zero at w=0, but so what!
Fexact=2*T*(2*sin(w*T/2)-sin(w*T))./(w*T).^2;
% look at the first N/4 data points
m=1:N/4; wm=w(1:N/4); Fsm=Fs(1:N/4);
Fexactm=[0 Fexact(2:N/4)];
[m' wm' Fsm.' Fexactm.']
figure
plot(wm,Fexactm,wm,real(Fsm))
axis([0 20 -0.1 0.5])
error=100*(Fsm-Fexactm)./Fexactm;
figure
plot(wm,error)
axis([0 20 0 2])

176

Discrete-Time Approximation for the Step Response

of an Underdamped RLC Series Circuit
R

V0 u(t)

i(t)

+
vC (t)

Exact continuous-time solution for verication:

Kirchho voltage law gives the equation of motion as the integro-dierential equation
di(t)
1
L
+ Ri(t) +
dt
C

i( ) d = V0 u(t):

The inductor forces the initial condition i(0) = 0. For t > 0 application of L1 d=dt to
eliminate the integral yields the homogeneous second-order dierential equation
d2 i(t) R di(t)
1
+
i(t) = 0
+
dt2
L dt
LC

(t > 0)

A second initial condition i0 (0) = V0 =L comes from the rst integro-dierential equation
evaluated at t = 0+ with the capacitor initially uncharged. A nice, clean statement of the
pertinent initial value problem is now
di(t)
d2 i(t)
+ 2
+ !02 i(t) = 0
2
dt
dt
with initial conditions
i(0) = 0

and

i0 (0) = V0 =L;

where the introduction of

R
2L

and

!02 ,

1
LC

simplies the mathematics by clarifying the physically-important combinations of the original R, L, and C circuit values. The exact solution to this well-posed initial value problem
is
V0 t
i(t) =
e
sin t
L
where
q
, !02 2 :
p
This form is most natural for the subject underdamped case where R < 2 L=C.
177

Discrete-time approximation:
The discretization of the derivative derives from the dening limit as
f (nT ) f ((n 1)T )
f [n] f [n 1]
df (t)
f (t) f (t t)
, lim
=)
=)
:
t!0
dt
t
T
T
A successive application of this gives the second-derivative
d df (t)
=)
dt dt

f [n] f [n 1] f [n 1] f [n 2]

f [n] 2f [n 1] + f [n 2]
T
T
=
:
T
T2

The discretized forms of the initial conditions i(0) = 0 and i0 (0) = V0 =L are thus
i[0] = 0

i[1] = V0 T =L:

and

The dierence equation that approximates the continuous-time dierential equation is

similarly

1 + 2T + (!0 T )2 i[n] 2(1 + T )i[n 1] + i[n 2] = 0:

Using the two initial conditions at n = 1 and n = 0, a recursive solution can be constructed for n = 1; 2; : : : according to
i[n] =

2(1 + T )
1
i[n 1]
i[n 2]:
2
1 + 2T + (!0 T )
1 + 2T + (!0 T )2

Numerical Results:
The circuit values
R = 200 ()

L = 20 (H)

C = 40 (F)

give
= 5 (s1 )

and

= 35 (s1 ):

The MATLAB code discreteRLC.m compares the normalized exact solution

L
i(t) = et sin t
V0
to the recursive solution of the discrete-time problem.

178

Normalized Current
1.0
solid curve: exact solution
dashed curve: sampling period T = 0:010 (s)

0.5

0.0

0:5

0.0

0.1

0.2

0.3

0.4

0.5

t (s)

Normalized Current
1.0
solid curve: exact solution
dashed curve: sampling period T = 0:001 (s)

0.5

0.0

0:5

0.0

0.1

0.2

0.3
t (s)

179

0.4

0.5

Sujûd As-Sahw (Corrected)
No ratings yet
Sujûd As-Sahw (Corrected)
4 pages
Applied Mathematics in Integrated Navigation Systems, Second Edition
No ratings yet
Applied Mathematics in Integrated Navigation Systems, Second Edition
3 pages
Tuhfatul Banat
No ratings yet
Tuhfatul Banat
98 pages
The Multiple Feedback
No ratings yet
The Multiple Feedback
10 pages
A Course in Robust Control Theory - A Convex Approach
100% (1)
A Course in Robust Control Theory - A Convex Approach
396 pages
Linear Equations Rubric
No ratings yet
Linear Equations Rubric
2 pages
Chapter 7 Notes
No ratings yet
Chapter 7 Notes
9 pages
Feedback Control Systems - HOzbay
100% (2)
Feedback Control Systems - HOzbay
227 pages
Great Women of Islam - Islamic Network
No ratings yet
Great Women of Islam - Islamic Network
5 pages
Positioning by Global Positioning System
100% (1)
Positioning by Global Positioning System
56 pages
Partial Differencial Equations Book
No ratings yet
Partial Differencial Equations Book
790 pages
Walker - Intro To Abstract Algebra
No ratings yet
Walker - Intro To Abstract Algebra
323 pages
Fazail e Amaal Ki Haqeeqat
No ratings yet
Fazail e Amaal Ki Haqeeqat
14 pages
Analysis Tools With Applications - Bruce Driver
No ratings yet
Analysis Tools With Applications - Bruce Driver
790 pages
Proteus Schematic ISIS 7.10 Help Converted by Atop CHM To PDF
No ratings yet
Proteus Schematic ISIS 7.10 Help Converted by Atop CHM To PDF
223 pages
Optimum Array Processing Detection Estimation and Modulation Theory Part IV 1st Edition Harry L. Van Trees - Quickly download the ebook to never miss any content
100% (2)
Optimum Array Processing Detection Estimation and Modulation Theory Part IV 1st Edition Harry L. Van Trees - Quickly download the ebook to never miss any content
83 pages
Introduction To Machine Learning
No ratings yet
Introduction To Machine Learning
209 pages
The Variational Approach To Optimal Control
100% (1)
The Variational Approach To Optimal Control
48 pages
Mathematical Analysis I - Paolo Guiotto +7
No ratings yet
Mathematical Analysis I - Paolo Guiotto +7
228 pages
Fixed-Point Signal Processing
No ratings yet
Fixed-Point Signal Processing
133 pages
Quantum Computing - Wikipedia
100% (1)
Quantum Computing - Wikipedia
196 pages
(Modern Approaches in Geophysics 15) Frank Scherbaum (Auth.) - of Poles and Zeros - Fundamentals of Digital Seismology (2001, Springer Netherlands) PDF
No ratings yet
(Modern Approaches in Geophysics 15) Frank Scherbaum (Auth.) - of Poles and Zeros - Fundamentals of Digital Seismology (2001, Springer Netherlands) PDF
281 pages
The Islamic Ruling On The Permissibility of Self-Sacrificial Operations Suicide or Martyrdom
No ratings yet
The Islamic Ruling On The Permissibility of Self-Sacrificial Operations Suicide or Martyrdom
60 pages
LinearTI Dynamic Systems PDF
No ratings yet
LinearTI Dynamic Systems PDF
439 pages
Introduction To Quantum Mechanics
100% (1)
Introduction To Quantum Mechanics
26 pages
IDL Programming Techniques 2nd Edition
No ratings yet
IDL Programming Techniques 2nd Edition
465 pages
CM Merged
No ratings yet
CM Merged
230 pages
QM Lecture Notes
100% (1)
QM Lecture Notes
636 pages
Ordinary Differential Equations and Stability Theory: An Introduction
From Everand
Ordinary Differential Equations and Stability Theory: An Introduction
David A. Sanchez
No ratings yet
Television Broadcasting Equpment Systems Ennes
No ratings yet
Television Broadcasting Equpment Systems Ennes
655 pages
Syllabus Quantum Mechanics
100% (2)
Syllabus Quantum Mechanics
7 pages
Bridging The Gap Between System Dynamics PDF
No ratings yet
Bridging The Gap Between System Dynamics PDF
11 pages
Energy System Modeling Lecture-Vii
No ratings yet
Energy System Modeling Lecture-Vii
12 pages
Book 1
No ratings yet
Book 1
389 pages
Physics
100% (1)
Physics
18 pages
The Grand Scheme of God, in The Light of Quran
No ratings yet
The Grand Scheme of God, in The Light of Quran
109 pages
Next-Generation Sequencing Data Analysis 2nd Edition
No ratings yet
Next-Generation Sequencing Data Analysis 2nd Edition
86 pages
Digital Communication over Fading Channels
From Everand
Digital Communication over Fading Channels
Marvin K. Simon
3/5 (8)
Dr. Joseph Pedlosky (Auth.) - Waves in The Ocean and Atmosphere - Introduction To Wave Dynamics-Springer-Verlag Berlin Heidelberg (2003) PDF
No ratings yet
Dr. Joseph Pedlosky (Auth.) - Waves in The Ocean and Atmosphere - Introduction To Wave Dynamics-Springer-Verlag Berlin Heidelberg (2003) PDF
258 pages
Vector and Matrix Calculus
100% (1)
Vector and Matrix Calculus
383 pages
Spiral Vector Theory of Brushless DC Motor
No ratings yet
Spiral Vector Theory of Brushless DC Motor
8 pages
Kalman Filter FPGA
No ratings yet
Kalman Filter FPGA
100 pages
The Least-Squares Finite Element Method Theory and Applications in Computational Fluid Dynamics and Electromagnetics
No ratings yet
The Least-Squares Finite Element Method Theory and Applications in Computational Fluid Dynamics and Electromagnetics
424 pages
Classical and Quantum Mechanics of The Damped Harmonic Oscillator - Dekker
No ratings yet
Classical and Quantum Mechanics of The Damped Harmonic Oscillator - Dekker
110 pages
Differential and Integral Calculus 1902 PDF
100% (1)
Differential and Integral Calculus 1902 PDF
348 pages
Introduction To Communication Systems
100% (1)
Introduction To Communication Systems
469 pages
Engineering Thermodynamics 2nd Edition M. Achuthan all chapter instant download
100% (6)
Engineering Thermodynamics 2nd Edition M. Achuthan all chapter instant download
85 pages
LightTools - 2024.03 Release Notes
No ratings yet
LightTools - 2024.03 Release Notes
54 pages
(Applied Mathematical Sciences 167) Otmar Scherzer, Markus Grasmair, Harald Grossauer, Markus Haltmeier, Frank Lenzen (Auth.) - Variational Methods in imaging-Springer-Verlag New York (2009)
No ratings yet
(Applied Mathematical Sciences 167) Otmar Scherzer, Markus Grasmair, Harald Grossauer, Markus Haltmeier, Frank Lenzen (Auth.) - Variational Methods in imaging-Springer-Verlag New York (2009)
323 pages
An Islamic Model For Stabilization and Growth
100% (1)
An Islamic Model For Stabilization and Growth
261 pages
Bulk and Surface Acoustic Waves - Fundamentals, Devices, and Applications (Guigen Zhang) (Z-Library)
No ratings yet
Bulk and Surface Acoustic Waves - Fundamentals, Devices, and Applications (Guigen Zhang) (Z-Library)
361 pages
The Abbe Principle Revisited by Bryan 1979
No ratings yet
The Abbe Principle Revisited by Bryan 1979
4 pages
20m All Mode Receiver Kit HR20
No ratings yet
20m All Mode Receiver Kit HR20
28 pages
Pre-Phd Syllabus ME
100% (1)
Pre-Phd Syllabus ME
24 pages
Linear Circuits Decarlo Solution Manual
100% (1)
Linear Circuits Decarlo Solution Manual
4 pages
Classical Electrodynamics - W. Greiner
100% (2)
Classical Electrodynamics - W. Greiner
569 pages
Anti-Lock Braking System: Presented By, 15/EE/68 Nit Durgapur
No ratings yet
Anti-Lock Braking System: Presented By, 15/EE/68 Nit Durgapur
25 pages
(Synthesis Lectures On Communications) Jerry D. Gibson - Analog Communications - Introduction To Communication Systems-Springer (2023)
No ratings yet
(Synthesis Lectures On Communications) Jerry D. Gibson - Analog Communications - Introduction To Communication Systems-Springer (2023)
102 pages
Refinement Calculus
No ratings yet
Refinement Calculus
513 pages
Graphs and Tables of the Mathieu Functions and Their First Derivatives
From Everand
Graphs and Tables of the Mathieu Functions and Their First Derivatives
James C. Wiltse
No ratings yet
Vineet PDF
No ratings yet
Vineet PDF
160 pages
Nonlinear Systems: Analysis, Stability, and Control
No ratings yet
Nonlinear Systems: Analysis, Stability, and Control
9 pages
General Mathematics Mock Board Exam Formula and Concept Use in 1Q 2015-2016 Algebra
No ratings yet
General Mathematics Mock Board Exam Formula and Concept Use in 1Q 2015-2016 Algebra
2 pages
CT2 - Maths II - KAS203T
No ratings yet
CT2 - Maths II - KAS203T
2 pages
VTAMPS 16 SS Set 3 (Corrected Solution Manual) PDF
No ratings yet
VTAMPS 16 SS Set 3 (Corrected Solution Manual) PDF
15 pages
6.2 Trigonometric Integrals and Substitutions
No ratings yet
6.2 Trigonometric Integrals and Substitutions
37 pages
FST-3C (ICSE-G-2) Math+Bio (C-10) 02-08-2024 - SOL
No ratings yet
FST-3C (ICSE-G-2) Math+Bio (C-10) 02-08-2024 - SOL
7 pages
B.Tech. & Integrated Dual Degree (B.Tech.+M.Tech.) Examination
No ratings yet
B.Tech. & Integrated Dual Degree (B.Tech.+M.Tech.) Examination
3 pages
Advanced Engineering Mathematics
No ratings yet
Advanced Engineering Mathematics
1 page
Application of Residue Theorem To Solve Several Re
No ratings yet
Application of Residue Theorem To Solve Several Re
7 pages
EE364a Homework 7 Solutions
No ratings yet
EE364a Homework 7 Solutions
16 pages
Bellman
100% (1)
Bellman
8 pages
Magnetic Field Around Finite Solenoid
No ratings yet
Magnetic Field Around Finite Solenoid
42 pages
SFPN Project Pseudospectra
No ratings yet
SFPN Project Pseudospectra
22 pages
P1 Chp4 GraphsAndTransformations PDF
No ratings yet
P1 Chp4 GraphsAndTransformations PDF
35 pages
Ch-05-07
No ratings yet
Ch-05-07
12 pages
Mathematical Modeling of Biological Processes: Avner Friedman Chiu-Yen Kao
100% (1)
Mathematical Modeling of Biological Processes: Avner Friedman Chiu-Yen Kao
152 pages
SSFCH 1
No ratings yet
SSFCH 1
13 pages
Distance On A Hexagonal Grid
No ratings yet
Distance On A Hexagonal Grid
2 pages
CBSE Class 10 Maths Standard Marking Scheme and Answer Key Term 1 2021 22
No ratings yet
CBSE Class 10 Maths Standard Marking Scheme and Answer Key Term 1 2021 22
5 pages
Spherical Harmonics: "Ylm" Redirects Here. For Other Uses, See
No ratings yet
Spherical Harmonics: "Ylm" Redirects Here. For Other Uses, See
2 pages
Teoria Algebraica de Numeros
No ratings yet
Teoria Algebraica de Numeros
152 pages
2 Dimensional Manifolds PDF
No ratings yet
2 Dimensional Manifolds PDF
6 pages
Quadratic Formula Explained (Article) - Khan Academy
No ratings yet
Quadratic Formula Explained (Article) - Khan Academy
13 pages
ADVANCE
No ratings yet
ADVANCE
136 pages
2.7 Related Rates Problems
No ratings yet
2.7 Related Rates Problems
4 pages
Isolated Point
No ratings yet
Isolated Point
3 pages
RANK REFINER SHEET: Quadratic Equation: Connect With Om Sir
No ratings yet
RANK REFINER SHEET: Quadratic Equation: Connect With Om Sir
9 pages
Divisibility Rules
No ratings yet
Divisibility Rules
4 pages
Some Simple Results On Cevian Quotients
No ratings yet
Some Simple Results On Cevian Quotients
5 pages