IEEE TRANSACTIONS ON EDUCATION, VOL. 45, NO.

1, FEBRUARY 2002 65

A Systematic Method for the Analytical

Evaluation of Convolution Integrals
Irwin S. Goldberg, Michael G. Block, and Roland E. Rojas

Abstract—Analytic methods are presented for the systematic methods for the evaluation of the integrals to be difficult
evaluation of convolution integrals that contain piecewise-con- and error prone, especially when discontinuous functions
tinuous or piecewise-smooth functions. These methods involve are involved. In contrast, a systematic analytical method will
breaking the convolution integral into a sum of integrals and
expressing each integral in a standard form that can be easily be presented for the time-domain evaluation of convolution
evaluated. Mathematical insight is gained by looking at the result integrals containing piecewise-continuous functions and/or
as a superposition of terms, where each term is nonzero or active piecewise-smooth functions. This analytical method involves
during a specific time interval. Applications to both causal and the following steps:
noncausal linear systems are considered. Additionally, an example
is included that provides students with insight into the construc- Step 1) expressing each discontinuous function, and/or
tion and the behavior of the impulse function. These methods have , as sums and differences of terms containing
been successfully taught to undergraduate engineering students to unit-step functions;
enhance their understanding of convolution and to facilitate their Step 2) substituting the dummy integration variable for the
ability to evaluate convolution integrals. independent variable in one function, either
Index Terms—Convolution, convolution integral, impulse func- or . Also, substituting for the independent
tion, impulse response. variable in the other function;
Step 3) writing the integral of the product
I. INTRODUCTION [or of the product ] as a sum of inte-
grals—each integral containing a unit-step function
LECTRICAL engineering students, during their junior
E year at St. Mary’s University, San Antonio, TX, are
required to take a Linear Signals and Systems Analysis course.
or a product of two unit-step functions;
Step 4) evaluating each integral in Step 3) in terms of stan-
dard integral forms given in (4b), (23), (24b), and
In this course, the students are introduced to continuous-time (25). The result is expressed in terms of sums of de-
convolution integrals. layed unit-step functions.
For a continuous, linear, time-invariant system with an im- The following three properties are useful as checks of the
pulse response , the response function, , to an input convolution result.
function, , is determined by a convolution integral. This
1) When both of the two convoluted functions are bounded
convolution relation, denoted as , is
(i.e., when impulse functions are not included), the result
of the convolution integral is a continuous function
(1a) of . Particularly, the result should be continuous at
points where the functional form of the solution abruptly
where is a dummy integration variable. Alternatively, because changes.
can also be evaluated as 2) The area of the convolution result is equal to the product
of the areas of the two functions that are involved in the
(1b) convolution operation [1]–[3].
3) When the two convoluted functions are of finite time du-
ration (that is, when both functions are of bounded sup-
The students are first taught to calculate continuous-time convo- port), the total duration of the result is equal to the sum of
lution results by evaluation of the above superposition integral the durations of the two functions involved in the convo-
in the time domain. Later, the students are taught to use trans- lution operation [3], [4].
form methods to calculate continuous-time convolution results. The expressions to be evaluated may contain the unit-step
Semigraphical methods are commonly presented in text- function and the ramp function , which is defined as
books for the time-domain evaluation of the convolution
integral [1]–[3]. Students often find these semigraphical for
(2)
for
Manuscript received March 9, 2001; revised October 9, 2001.
I. S. Goldberg and M. G. Block are with the Engineering Department, The method is facilitated by the realization that the ramp
St. Mary’s University, San Antonio, TX 78228-8534 USA (e-mail: engrirwin@ function can be expressed in terms of the unit-step function as
stmarytx.edu).
R. E. Rojas is with SBC Communications Inc., Austin, TX 78252 USA.
Publisher Item Identifier S 0018-9359(02)01298-0. (3)
0018–9359/02$17.00 © 2002 IEEE
66 IEEE TRANSACTIONS ON EDUCATION, VOL. 45, NO. 1, FEBRUARY 2002

II. CONVOLUTION INVOLVING CAUSAL SYSTEMS WITH INPUTS

APPLIED FOR
A. The First Standard Form
To obtain convolution results for causal systems with inputs
applied for , the convolution integrals can be put into a
standard form. This standard form is given by

(a)

if
(4a)
if

where represents a generic function of the integration

variable and the nonintegrated time variable . The relation
(4a) can be expressed as

(4b)
(b)
where and can be constants or they can be functions of .
Fig. 1. Graphs illustrating example problem 1. (a) Graph of x (t) or h (t)
The term is equal to zero if . The relation given verses t, as specified by (5). In example problem 1, x (t) is equal to h (t). (b)
by (4b) will be referred to as the first standard form. Graph of the convolution solution, y (t) verses t, given by (13). The arrows
To derive (4a), one must first consider the case where . It beneath the graph denote the times when different components of the signal
should be noted that is equal to zero when , and y (t) are active.
is equal to zero when . Therefore, when ,
the limits of integration can be changed from to After one substitutes (3) into (6), and takes delays into account,
. the result is
Next, the case where must be considered. The term
equals zero when and the term equals
zero when . Thus, the integrand in (4a) is equal to zero (7)
for all values of for . Therefore, when ,
the integral is equal to zero. Equation (4b) and its derivation are which simplifies to
similar to results given by Zuhao [5].
The first standard form, defined here by (4b), is the common (8)
standard form that is used when causal systems are considered
with . When the system is noncausal, or if inputs are In step 2), is substituted for in , and is substituted
nonzero when , three additional standard forms may need for in , with the resulting convolution integral
to be considered. These other standard forms are given in Sec-
tion III, where noncausal systems are discussed.

B. Example Problem 1
(9)
As a first example, involving a causal system with inputs
applied at , the authors will evaluate the convolution
The above result is equivalent to
, where

for
(5)
otherwise

as shown in Fig. 1(a).

The first step requires expressing and in terms
of sums of (delayed) step functions. As a preliminary result,
these functions will be expressed in terms of delayed unit-step
functions and ramp functions as
(10)
(6)
GOLDBERG et al.: ANALYTICAL EVALUATION OF CONVOLUTION INTEGRALS 67

Recognizing that each of these integrals is of the first standard

form, applying (4b) to (10), one gets

(11)

Evaluating the integral expressions and simplifying, the result

is

(12a)
Fig. 2. The solution to example problem 2. Graph of the convolution result
y (t) versus t, given by (22). The arrows beneath the graph denote the times
can be expressed as a sum of three terms as when different components of the signal y (t) are active.

(12b)
These expressions can be written as
where , , and represent terms that are multi- (16)
plied by step functions with specific delays. The result repre-
sents the superposition of three terms, where each term is active and
or nonzero during a specific time interval. The terms in (12b) (17)
can be recognized as follows. The term is active when
; the term is active when ; the term is Substituting for in (16) and substituting for in (17),
active when . Thus, when the convolution integral becomes
for when
when . Because
when . A graph-
ical representation of these three terms is presented in Fig. 1(b), (18)
where the time when each of the three signal components is
active (nonzero) is represented by a bold arrow. With the su- Equation (18) is equivalent to
perposition of all active terms, the result given by (12a) can be
expressed as

for
for (19)
(13)
for
for Each of the above integrals is in the first standard form; substi-
tuting (4b) into (19), the result is
as plotted in Fig. 1(b). As expected, is a continuous func-
tion of .
As a supplemental exercise, the result for example problem 1,
given by (13), could be obtained using Laplace transforms. This (20)
could be easily accomplished starting with the expressions for After these integrals have been evaluated, the result is
and given by (7), using known properties of Laplace
transforms of time-delayed functions, and taking the inverse
transform of the product of the transform of with the trans- (21)
form of .
This result represents the superposition of two terms. The ar-
C. Example Problem 2 rows in Fig. 2 show when each of these two terms is active. The
terms in (21) can be recognized as follows. The term multiplied
As a second example involving a causal system, the convolu-
by (denoted by in Fig. 2) is active when ; the term
tion will be evaluated, where
multiplied by (denoted by in Fig. 2) is active when
. With superposition of all active terms, the result, ,
for
(14) can be constructed as
otherwise
and for
for for (22)
(15)
for . for .
68 IEEE TRANSACTIONS ON EDUCATION, VOL. 45, NO. 1, FEBRUARY 2002

A graph of is shown in Fig. 2. Note that is a con-

tinuous function of .

III. NON-CAUSAL SYSTEMS

A. Additional Standard Forms
For evaluation of convolution integrals involving noncausal
systems or for inputs applied when , three standard integral
forms are needed in addition to the first standard form, which
was given as (4b). (a)
The second standard form is given by

(23)

The third standard form is given by

if
(24a)
if
(b)
Expression (24a) can be written as
Fig. 3. Graphs illustrating example problem 3. (a) Graphs of h (t) versus t, as
specified by (26). The solid curve corresponds to k = 10 and the dashed curve
corresponds to k = 2. (b) Graphs of the convolution solutions y (t) versus t,
given by (34). The solid curve corresponds to k = 10 and the dashed curve
corresponds to k = 2. The arrows beneath the graphs denote the times when
(24b) different components of the signal y (t) are active.

Similarly, the fourth standard form is given by the value of the parameter ). For increasing values of , the
functions resemble a sequence of functions that could be
used to construct the impulse function.
After one substitutes for in (28) and substitutes for
in (29), the convolution integral becomes
(25)

B. Example Problem 3
A third example involves a noncausal system. The convolu- (30)
tion will be evaluated, where
The result in (30) is equivalent to
for
(26)
for
with a real-positive parameter, and where is a square
pulse given by
for
(27)
otherwise.
can be expressed as

(28) (31)

Also, can be expressed as In the result given by (31), the first two integrals are of the
first standard integral form [ (4b) with ], and the last two
(29)
integrals are of the fourth standard form [ (25) with ].
Therefore, using (4b) and (25) with (31), one gets
Graphs of for and are shown in Fig. 3(a).
The total integrated area of is equal to 1.00 (independent of
GOLDBERG et al.: ANALYTICAL EVALUATION OF CONVOLUTION INTEGRALS 69

techniques commonly presented in textbooks [1]–[3]. During

the past fourteen years, one of the authors (Goldberg) has taught
the derivation and application of this method as part of the Sig-
nals and Systems course for electrical engineering students in
their junior year at St. Mary’s University. The method presented
in this paper has been found to be easier for the students to
use than the conventional graphical method. The students com-
(32) monly report that, using the analytic method presented in this
paper, they have less difficulty evaluating convolution integrals
Evaluating these integral expressions, one gets of discontinuous functions. The students gain mathematical in-
sight by observing the convolution result as a superposition of
terms, where each term is nonzero (or is active) during a specific
time interval. Derivations of (4b), (23), (24b), and (25) are good
mathematical exercises to provide the students with insight into
(33) the behavior of integrals containing step functions. Addition-
ally, example problem 3 provides the students with insight into
The result, given by (33), represents the superposition of four the construction of the impulse function and into the behavior
terms. This result is shown in Fig. 3(b). The arrows in Fig. 3(b) of convolution integrals involving impulse functions.
show the times when each of these four terms is active. The
terms in (33) can be recognized as follows. The term multiplied REFERENCES
by [denoted by the arrow in Fig. 3(b)] is active for
[1] C. D. McGillem and G. R. Cooper, Continuous and Discrete Signal and
; the term multiplied by [denoted by the arrow System Analysis, 3rd ed. Philadelphia, PA: Saunders , 1991.
in Fig. 3(b)] is active for ; the term multiplied by [2] S. S. Soliman and M. D. Srinath, Continuous and Discrete Signals and
[denoted by the arrow in Fig. 3(b)] is active for ; and Systems, 2nd ed. Upper Saddle River, NJ: Prentice-Hall, 1998.
[3] R. E. Ziemer, W. H. Tranter, and D. R. Fannin, Signals and Systems, 4th
the term multiplied by [denoted by the arrow in ed. Upper Saddle River, NJ: Prentice-Hall, 1998.
Fig. 3(b)] is active for . The result, from the superposition [4] H. Kwakernaak and R. Sivan, Modern Signals and Systems. Engle-
of all active terms in (33), can be expressed as wood Cliffs, NJ: Prentice-Hall, 1991.
[5] Z. Zuhao, “The square matrix rule of the convolution integral,” IEEE
Trans. Educ., vol. 33, pp. 369–372, Nov. 1990.
[6] R. N. Bracewell, The Fourier Transform and its Application, 3rd
ed. New York: McGraw-Hill, 2000.
for [7] A. V. Oppenheim and A. S. Willsky, Signals and Systems, 2nd
(34) ed. Upper Saddle River, NJ: Prentice-Hall, 1997.
for
for .

It should be noted that for finite values of , is a continuous

Irwin S. Goldberg received the B.S. degree in electrical engineering from
function of . Comparison graphs of for and Drexel University, Philadelphia, PA, in 1963 and the M.S. and Ph.D. degrees in
are shown in Fig. 3(b). As the parameter becomes large, physics from Lehigh University, Bethlehem, PA, in 1965 and 1970, respectively.
closely approximates the square pulse, given by ; this He has been on the faculty at St. Mary’s University, San Antonio, TX, since
1980. He is currently a Professor of Engineering. His research interests include
approximation improves as increases. This result is expected elastic wave propagation, fluid dynamics, electrodynamics, and mathematical
because, for large values of , approximates the impulse modeling.
function 1 [2], [6], and [2], [6], [7]

(35)
Michael G. Block received the B.S. degree in visual science and the O.D. from
Illinois College of Optometry, Chicago, IL, in 1975 and 1977, respectively, and
This example illustrates the manner in which the impulse func- the M.S. degree in physiological optics from the University of Alabama, Birm-
tion can be constructed from the sequence of functions as ingham, 1988. He is currently pursuing a graduate degree in electrical engi-
. Also, this example demonstrates the behavior of con- neering at St. Mary’s University, San Antonio, TX.
From 1979 to 1999, he served in the Air Force. He is currently a contracted
volution integrals containing impulse functions, as expressed by research employee at Air Force Research Laboratory, Brooks Air Force Base,
(35). San Antonio, TX.

IV. SUMMARY
An analytic method is presented for the evaluation of con- Roland E. Rojas received the B.S. degree in electrical engineering from St.
volution integrals. This method does not rely on the graphical Mary’s University, San Antonio, TX, in 1995 and the M.E.E. degree from Rice
University, Houston, TX, in 1998.
1More correctly, for large values of k , the evaluation of integrals with h (t) Since 1997, he has been with SBC Communications, Inc., Austin, TX, and
included in the integrand approximates the evaluation of these integral expres- is currently a Technical Sales Specialist II. He served as Co-Chair for Student
sions with the generalized function  (t) replacing h (t). Activities for IEEE GlobeCom 2001.