72 Discrete-Time Signals and Systems Chap.

2.3 ANALYSIS OF DISCRETE-TIME LINEAR TIME-INVARIANT

SYSTEMS

In S ectio n 2.2 we classified system s in a c c o rd a n c e w ith a n u m b e r of c h aracteristic

p ro p e rtie s o r categ o ries, nam ely: lin e a rity , cau sality , stab ility , a n d tim e in variance.
H av in g d o n e so. we now tu rn o u r a tte n tio n to th e analysis o f th e im p o rta n t class
o f lin ear, tim e -in v a ria n t (L T I) system s. In p a rtic u la r, we shall d e m o n s tra te th a t
such system s are c h a ra c te riz e d in th e tim e d o m ain sim ply bv th e ir resp o n se to a
u n it sam p le se q u en ce. W e shall also d e m o n s tra te th a t any a rb itra ry in p u t signal
can b e d eco m p o sed an d r e p re s e n te d as a w eig h ted sum of u n it sa m p le seq u en ces.
A s a c o n s eq u en ce o f th e lin e a rity a n d tim e-in v arian ce p ro p e rtie s of the system ,
th e resp o n se o f th e system to any a rb itra ry in p u t signal can be e x p ressed in term s
of th e u n it sam p le re sp o n se of th e system . T h e g en eral form o f the ex p ressio n
th a t re la te s th e u n it sam ple re sp o n se o f the system an d th e a rb itra ry in p u t signal
to th e o u tp u t signal, called th e c o n v o lu tio n sum o r th e c o n v o lu tio n fo rm u la, is also
d eriv ed . T h u s we are ab le to d e te rm in e the o u tp u t o f any lin e a r, tim e-in v arian t
system to any a rb itra ry in p u t signal.

2.3.1 Techniques for the Analysis of Linear Systems

T h e re are tw o basic m e th o d s for an aly zin g th e b e h a v io r o r re sp o n s e of a lin ear

system to a given in p u t signal. O n e m e th o d is b a se d on th e d ire c t so lu tio n o f the
in p u t-o u tp u t e q u a tio n for th e system , w hich, in g e n e ra l, has th e form

v(ji) - F [v ( n - 1), v (?7 — 2 ) ........ y(n - N) , x ( n ) . x ( n — 1).......... x ( n - M)]

w h ere F[-] d e n o te s so m e fu n ctio n o f th e q u a n titie s in b ra c k e ts. Specifically, fo r

an L T I system , w e shall see la te r th a t th e g e n e ra l form o f th e in p u t-o u tp u t re la ­
tio n sh ip is

N M
(2.3.1)

w h ere an d {b*,.} are c o n sta n t p a ra m e te rs th a t specify th e sy stem an d a re in ­

d e p e n d e n t o f x( n) a n d y( n) . T h e in p u t-o u tp u t re la tio n sh ip in (2.3.1) is called
a d ifferen ce e q u a tio n a n d re p re se n ts o n e w ay to c h a ra c te riz e th e b e h a v io r of a
d isc re te -tim e L T I system . T h e so lu tio n o f (2.3.1) is th e su b je ct o f S ection 2.4.
T h e seco n d m e th o d fo r analyzing th e b e h a v io r o f a lin e a r sy stem to a given
in p u t signal is first to d e c o m p o se o r reso lv e th e in p u t signal in to a sum o f e le ­
m e n ta ry signals. T h e e le m e n ta ry signals a re se le c te d so th a t th e re sp o n se o f the
system to each signal c o m p o n e n t is easily d e te rm in e d . T h e n , u sin g th e lin earity
p ro p e rty o f th e sy stem , th e re sp o n se s o f th e sy stem to th e e le m e n ta ry signals are
ad d e d to o b ta in th e to ta l re s p o n s e of th e sy stem to th e given in p u t signal. T h is
seco n d m e th o d is th e o n e d e sc rib e d in th is se ctio n .
Sec. 2.3 Analysis o f Discrete-Time Linear Tim e-Invariant Systems 73

T o e la b o ra te , su p p o se th a t th e in p u t signal x (n ) is reso lv ed into a w eig h ted

sum o f elem en tary ' signal c o m p o n e n ts {*/. («)} so th a t

w h ere th e {ct} a re th e set of a m p litu d e s (w eighting coefficients) in the d e c o m ­

p o sitio n o f th e signal x( n) . N ow su p p o se th a t th e re sp o n se o f th e system to the
e le m e n ta ry signal c o m p o n e n t x k{n) is y k(n). T hus.

y*(n) = T [j.t (n)] (2.3.3)

assu m in g th a t th e sy stem is re la x e d an d th a t th e re sp o n se to ckx k(n) is cky k(n). as

a c o n s e q u e n c e o f th e scaling p ro p e rty o f th e lin e a r system .
F in ally , th e to ta l re sp o n se to th e in p u t x( n ) is

(2.3.4)

In (2.3.4) we u se d th e ad d itiv ity p ro p e rty o f th e lin e a r system .

A lth o u g h to a larg e e x te n t, th e ch o ice o f th e e le m e n ta ry signals a p p e a rs to
b e a rb itra ry , o u r se lectio n is heavily d e p e n d e n t on th e class of input signals that
we w ish to co n sid er. If w e place n o re stric tio n on th e c h aracteristics of th e input
signals, its re so lu tio n in to a w eig h ted sum of u n it sam ple (im p u lse) se q u en ces
p ro v e s to b e m a th e m a tic a lly c o n v e n ie n t an d c o m p letely g en eral. O n th e o th e r
h a n d , if w e re stric t o u r a tte n tio n to a subclass o f in p u t signals, th e re m ay be
a n o th e r set o f e le m e n ta ry signals th a t is m o re c o n v e n ie n t m a th e m a tic a lly in the
d e te rm in a tio n o f th e o u tp u t. F o r ex am p le, if th e in p u t signal x( n) is p erio d ic
w ith p e rio d N , w e have a lre a d y o b se rv e d in S ectio n 1.3.5 th a t a m a th em atically
c o n v e n ie n t set o f elem en tary 7 signals is th e set o f e x p o n en tials

x k (n) = eJluin k = 0. l , . . . , i V - l (2.3.5)

w h e re th e fre q u e n c ie s {cok } a re h arm o n ic a lly re la te d , th a t is.

k = 0. 1.........N - 1 (2.3.6)

T h e fre q u e n c y 2 n / N is called th e fu n d a m e n ta l fre q u e n c y , an d all h ig h er-freq u en cy

c o m p o n e n ts a re m u ltip le s o f th e fu n d a m e n ta l fre q u e n c y c o m p o n e n t. T h is subclass
o f in p u t sig n als is c o n s id e re d in m o re d e ta il later.
F o r th e re so lu tio n o f th e in p u t signal in to a w eig h ted su m o f u n it sam ple
se q u en ces, w e m u st first d e te rm in e th e re sp o n se o f th e system to a u n it sa m ­
p le se q u e n c e a n d th e n use th e scaling a n d m u ltip lic ativ e p ro p e rtie s of th e lin ear
74 Discrete-Time Signals and Systems Chap. 2

sy stem to d e te rm in e th e fo rm u la fo r th e o u tp u t given any a rb itra ry input, T his

d e v e lo p m e n t is d escrib ed in d etail as follow s.

2.3.2 Resolution of a Discrete-Time Signal into Impulses

S u p p o se w e h av e an a rb itra ry signal x ( n ) th a t we w ish to reso lv e in to a sum of unit

sa m p le seq u en ces. T o utilize th e n o ta tio n e sta b lish e d in th e p re c e d in g se ctio n , we
select th e e le m e n ta ry signals x k (n) to be

x k(n) = 8{n - k) (2.3.7)

w h ere k re p re se n ts th e d elay o f th e u n it sam p le se q u e n c e . T o h a n d le an a rb itra ry

signal x ( n ) th a t m ay h ave n o n z e ro v alu es o v er an infinite d u ra tio n , th e set of unit
im p u lses m u st also b e infinite, to en co m p ass th e infinite n u m b e r of delays.
N o w su p p o se th a t we m u ltip ly th e tw o se q u e n c e s x( n) a n d <5(n - k). Since
8{n — k) is z e ro e v ery w h ere ex cep t a t n = k , w h e re its v alu e is u nity, the result
o f th is m u ltip lic atio n is a n o th e r se q u e n c e th a t is z e ro e v ery w h ere e x c e p t at n — k ,
w h ere its v alu e is x ( k) , as illu stra te d in Fig. 2,22. T h u s

x( n) 8( n — k) = x ( k) 8( n — k) (2.3.8)

Jt(n)

T TT 111 i ’ l l i , ‘ 1 I , Tt I
i [ -2-10113 i i 1
(a) JC(Jc)
6(/i-Jt)

(b)

*(*) 6(n —k )

k
0 n

Figure 2.22 M ultiplication of a signal x i n ) with a shifted unit sam ple sequence.
Sec. 2.3 Analysis of Discrete-Time Linear Time-Invariant Systems 75

is a se q u e n c e th a t is z e ro e v e ry w h e re ex cep t at n = k , w h e re its v a lu e is x( k ) . If we
w ere to re p e a t th e m u ltip lic a tio n of x ( n ) w ith <5(a? — m ), w h ere m is a n o th e r d elay
(im =6 k), th e re su lt will b e a se q u en ce th a t is z e ro e v e ry w h e re e x cep t at n = m,
w h ere its v alu e is x ( m ) . H e n c e
x( n ) 5 ( n — m) = x ( m) 8( n — m) (2.3.9)

In o th e r w o rd s, each m u ltip lic a tio n o f th e signal x( n ) by a u n it im p u lse at som e

d elay k, [i.e., <5(n — it)], in essen ce picks o u t th e single v alu e x ( k ) o f th e signal jc(n)
at th e d e la y w h e re th e u n it im pulse is n o n z e ro . C o n s e q u e n tly , if w e re p e a t this
m u ltip lic a tio n o v e r all p o ssib le delays, - o o < k < oo, a n d su m all th e p ro d u c t
se q u e n c e s, th e re su lt will be a se q u e n c e e q u a l to th e se q u e n c e x ( n ) , th a t is,
PC-
x( n) = ^ x ( k) 8( n — k) (2.3.10)
k=—oc
W e e m p h a s iz e th a t th e rig h t-h an d side of (2.3.10) is th e su m m a tio n of an
infinite n u m b e r o f u n it sa m p le se q u en ces w h ere th e u n it sa m p le se q u e n c e 6(n - k)
h as an a m p litu d e value o f x( k ) . T h u s th e rig h t-h a n d sid e o f (2.3.10) gives th e
re so lu tio n o f o r d e c o m p o s itio n o f any a rb itra ry signal jc(n) in to a w e ig h te d (scaled)
sum o f sh ifted u n it sam p le seq u en ces.
Exam ple 2.3.1
Consider the special case of a finite-duration sequence given as
jc(«) = (2, 4, 0,3)
T

Resolve the sequence x(n) into a sum of weighted impulse sequences.

Solution Since the sequence x(n) is nonzero for the time instants n = —1, 0. 2, we
need three impulses at delays k = —1. 0, 2, Following (2.3.10) we find that
x<n) = 2(5(n + 1) + 4<5(n) + 3<5(n —2)

2.3.3 Response of LTI Systems to Arbitrary Inputs: The

Convolution Sum

H av in g re so lv e d an a rb itra ry in p u t signal x ( n ) in to a w eig h te d su m o f im pulses,

w e a re no w re a d y to d e te rm in e th e re sp o n se of an y re la x e d lin e a r sy stem to any
in p u t signal. F irs t, w e d e n o te th e re sp o n se v(n, k) o f th e sy stem to th e in p u t unit
sa m p le se q u e n c e a t n = it by th e special sym bol h(n, k), —oo < k < oo. T h a t is,

y( n, k) = h(n. k ) = T[ S( n — £)] (2.3.11)

In (2.3.11) w e n o te th a t n is th e tim e index a n d k is a p a r a m e te r sh ow ing th e

lo c a tio n o f th e in p u t im p u lse. If th e im pulse at th e in p u t is sc aled by an a m o u n t
ct = jc(it), th e re sp o n s e of th e system is th e c o rre sp o n d in g ly sc aled o u tp u t, th a t is,

Ckh(n, k) = x(k)h(n, k) (2.3.12)

76 Discrete-Time Signals and Systems Chap. 2

F in ally , if th e in p u t is th e a rb itra ry signal x(/t) th a t is e x p re ss e d as a sum of

w eig h ted im p u lses, th a t is.

(2.3.13)

th e n th e resp o n se o f the system to x(/i) is the c o rre sp o n d in g sum o f w eig h ted

o u tp u ts, th a t is,

y(rt) = 7"[.r(/;)] = T ^ x( k ) S( n — k)

= ^ x ( k ) T [ 5 (m - k)] (2.3.14)

ii= —oc
C learly , (2.3.14) follow s from th e su p e rp o sitio n p ro p e rty of lin e a r system s, and is
k n o w n as th e superposit i on s u mma t i o n .
W e n o te th a t (2.3.14) is an e x p ressio n for th e resp o n se o f a lin e a r system to
any a rb itra ry in p u t se q u e n c e x( n) . T his ex p ressio n is a fu n ctio n of b o th .v(») and
th e resp o n ses h(n. k) of the system to th e unit im pulses Sin — k) fo r —oc < k < oc.
In d eriv in g (2.3.14) w e used th e lin earity p ro p e rty o f th e system but not its tim e-
in v arian ce p ro p e rty . T h u s th e ex p ressio n in (2.3.14) ap p lies to any relax ed lin ear
(tim e -v a ria n t) system .
If. in a d d itio n , th e system is tim e in v a ria n t, th e fo rm u la in (2.3.14) sim plifies
c o n sid erab ly . In fact, if the resp o n se o f th e L T I system to th e u n it sa m p le seq u en ce
<5(rc) is d e n o te d as h(n). th a t is.
h(n) = T [ b( n ) \ (2.3.15)
th en by th e tim e-in v arian ce p ro p e rty , th e resp o n se o f the system to the delay ed
u n it sa m p le se q u e n c e <5(n - k) is
h(n — k) = T [ S( n — A')] (2,3.16)
C o n seq u en tly , th e fo rm u la in (2.3.14) re d u c e s to

(2.3.17)
k=-oc
N ow we o b serv e th a t th e relax ed L T I system is co m p letely c h a ra c te riz e d by a
single fu n ctio n h(n), n am ely , its resp o n se to th e u n it sam p le se q u e n c e In
c o n tra st, th e g en eral c h a ra c te riz a tio n of th e o u tp u t o f a tim e -v a ria n t, lin e a r sys­
tem re q u ire s an in fin ite n u m b e r o f u n it sa m p le re sp o n s e fu n ctio n s, h{n, k), o n e for
ea c h p o ssib le d elay .
T h e fo rm u la in (2.3.17) th a t gives th e re sp o n se y( n) of th e L T I system as a
fu n c tio n o f th e in p u t signal x ( n ) a n d th e u n it sa m p le (im p u lse) re sp o n se h(n) is
called a convol ut i on s um. W e say th a t th e in p u t jt(n ) is c o n v o lv ed w ith th e im p u lse
Sec. 2.3 Analysis of Discrete-Time Linear Tim e-Invariant Systems 77

response h(n) to yield the output y in ). W e shall now explain the procedure for
com puting the resp onse y (n ). both m athem atically and graphically, given the input
x ( n ) and the im pulse response h(n) o f the system .
Suppose that we wish to com pute the output of the system at som e time
instant, say n = n 0. A ccordin g to (2.3.17), the resp onse at n = no is given as
OC

y (n 0) = ^ 2 x ( k ) h ( n 0 - k) (2.3.18)
Jc=-oc

O ur first observation is that the index in the sum m ation is k , and h en ce both the
input signal x ( k ) and the im pulse resp onse h(no - k) are fun ction s o f k. Second,
w e ob serve that the sequ en ces x ( k ) and h(nQ — k) are m ultiplied togeth er to form
a product seq u en ce. T h e output >(«o) is sim ply the sum over all valu es o f the
product sequ en ce. T he seq u en ce h ( n 0 — k) is ob tain ed from h ( k ) by, first, folding
h (k) about k = 0 (the tim e origin), w hich results in the seq u en ce h ( —k). The
folded seq u en ce is then shifted by no to yield h(no — k). T o sum m arize, the process
o f com p utin g the con volu tion b etw een x ( k ) and h (k) in volves the follow in g four
steps.

1. Folding. F old h(k) about k = 0 to obtain h ( ~ k ) .

2. Shifting, Shift h ( —k) by n 0 to the right (left) if n o is p ositive (n egative), to
obtain h(no — £).
3. Multiplication. M ultiply by h (no — k) to obtain the product sequ en ce
v„ Jk) = x ( k ) h ( n 0 - k).
4. S u m m a t i o n . Sum all the values o f the product seq u en ce vnt)( k ) to obtain the
value o f the output at tim e n = n 0.

W e note that this p rocedure results in the resp onse o f the system at a sin ­
gle tim e instant, say n = n 0. In gen eral, we are interested in evaluating the
response o f the system over all tim e instants - o o < n < oo. C onsequently,
steps 2 through 4 in the sum m ary m ust be rep eated , for all p ossible tim e shifts
—oo < n < oo.
In order to gain a b etter understanding o f the procedure for evaluating the
convolution sum , w e shall dem onstrate the p rocess graphically. T he graphs will
aid us in explaining the four steps in volved in the com p utation o f the convolution
sum.

Example 2.3.2
The impulse response of a linear time-invariant system is
/!(«) = [1 .2 ,1 ,-1 } (2.3.19)
T

Determine the response of the system to the input signal

x(n) = {1,2.3,1} (2.3.20)
t
 Discrete-Time Signais and Systems Chap. 2

Solution We shall com pute the convolution according to the formula (2.3.17). but
we shall use graphs of the sequences to aid us in the com putation. In Fig. 2.23a we
illustrate the input signal sequence x(k) and the impulse response h{k) of the system,
using k as the time index in order to be consistent with (2.3.17),
The first step in the com putation of the convolution sum is to fold h(k). The
folded sequence h(~k) is illustrated in Fig. 2.23b. Now we can compute the output
at n = 0. according to (2.3.17), which is

v(0) = (2.3.21)
*=-cx

Since the shift n = 0, we use h( —k) directly without shifting it. The product sequence

= x(k)h(-k) (2.3.22)

h(k) x(k
3 4i

* -i
T •
1
-1 0 ! j 10 1 2 3

h(-k) L'n(k 1
2
. . -2 T ’t . . . .
-1 0 1 2
(b)

Shift Product
vAk)
h(\-k)

111
To
(c)

,,. Product
7
Br
T

L’ |{t)
sequence
T 2
1■
. ~3 T !
j-2 -1 0 1 k 0 12 *
(d)

Figure 2.23 G raphical com putation of convolution.

Sec. 2.3 Analysis of Discrete-Time Linear Time-Invariant Systems 79

is also shown in Fig. 2.23b. Finally, the sum of all the terms in the product sequence
yields

■(°) = £ vott) = 4

We continue the com putation by evaluating the response of the system at n = 1.

According to (2.3.17),

(2.3.23)

The sequence h(\ —k) is simply the folded sequence h( —k) shifted to the right by one
unit in time. This sequence is illustrated in Fig. 2.23c. The product sequence
V] (k} = x(k)h{l — k) (2.3.24)

is also illustrated in Fig. 2.23c. Finally, the sum of all the values in the product
sequence yields

y(l) = £ ui(*) = 8

In a similar m anner, we obtain y(2) by shifting h ( - k ) two units to the right,

forming the product sequence ih(A) = x(k)h(2 — k) and then summing all the terms
in the product sequence obtaining y(2) = 8. By shifting h(—k) farther to the right,
multiplying the corresponding sequence, and summing over all the values of the re­
sulting product sequences, we obtain v(3) = 3. v(4) = - 2 , y(5) = - 1 . For u > 5, we
find that v(n) = 0 because the product sequences contain all zeros. Thus we have
obtained the response y(n) for n > 0.
Next we wish to evaluate v(n) for n < 0. We begin with n = Then

(2.3.25)

Now the sequence h (—1 —k) is simply the folded sequence h ( —k ) shifted one time
unit to the left. The resulting sequence is illustrated in Fig. 2.23d. The corresponding
product sequence is also shown in Fig. 2.23d. Finally, summing over the values of the
product sequence, we obtain

V (-1) = 1
From observation of the graphs of Fig. 2.23, it is clear that any further shifts of
h (—1 - k) to the left always results in an all-zero product sequence, and hence
y(n) = 0 for n 5 —2
Now we have the entire response of the system for —oc < n < oc. which we
summarize below as
y(n) = 0 ,0,1, 4. 8, 8. 3, - 2 , - 1 , 0 . 0 . . . .) (2.3.26)
t
80 Discrete-Time Signals and Systems Chap. 2

In E x am p le 2.3.2 w e illu stra te d th e c o m p u ta tio n o f th e co n v o lu tio n sum .

using g rap h s o f th e se q u en ces to aid us in visualizing th e ste p s in volved in th e
c o m p u ta tio n p ro c e d u re .
B e fo re w o rk in g o u t a n o th e r ex a m p le , w e w ish to show th a t th e co n v o lu ­
tio n o p e ra tio n is c o m m u ta tiv e in th e se n se th a t it is irre le v a n t w hich of th e tw o
se q u e n c e s is fo ld ed a n d shifted. In d e e d , if w e b eg in w ith (2.3.17) a n d m ak e a
ch an g e in th e v ariab le o f th e su m m a tio n , fro m k to m , by defin in g a new index
m — n — k, th e n k = n — m an d (2.3.17) b e co m es
CC

y(n) = ^2 x( n — m ) h { m ) (2.3.27)
m—
—
Since m is a d u m m y in dex, w e m ay sim ply re p la c e m by k so th a t

y{n) = x(n-k)h(k) (2.3.28)

T h e ex p ressio n in (2.3.28) involves leav in g th e im p u lse re sp o n s e h( k) u n a lte re d ,

w hile th e in p u t se q u en ce is fo ld ed a n d sh ifted . A lth o u g h th e o u tp u t v(n) in (2.3.28)
is id en tical to (2.3.17), th e p ro d u c t se q u e n c e s in th e tw o fo rm s o f th e co n v o lu tio n
fo rm u la are not id en tical. In fact, if w e define th e tw o p ro d u c t se q u e n c e s as

v„(k) = x ( k ) h ( n — k)

w n(k) = x( n — k) h(k)

it can b e easily show n th at

un(£) = w n (n — k)

an d th e re fo re ,
CC CC

v(n) = ^ ^ ~ k)
k——oc oc

since b o th se q u en ces co n tain th e sam e sa m p le v alu es in a d iffe re n t a rra n g e m e n t.

Example 2.3.3
Determ ine the output y(n) of a relaxed linear tim e-invariant system with impulse
response

h(ri) = a"u(n), \a\ < 1

when the input is a unit step sequence, that is,

x (n) = u(n)

Solution In this case both /j(n) and jc(n) are infinite-duration sequences. We use
the form of the convolution formula given by (2.3.28) in which x (k) is folded. The
Sec. 2.3 Analysts of Discrete-Time Linear Tim e-Invariant Systems 81

h{k\ x(k)

TT i f 1 1 2 3 4 *
(b)

x(- k) I'oCA')
l
* 1

-3 -2 -1 0 -1 1 - A

(c)
A(1 -k)
i'i(i-)
> a

1
- 1 0 1 k

v( 2—k) 1':(!>

II

- 2 - 1 0 1 2 3 4 5 k

Figure 2.24 Graphical computation of convolution in Example 2.3.3.

sequences h(k), x(k). and x{—k) are shown in Fig. 2.24. The product sequences vo(k).
v\(k), and v2(k) corresponding to x ( —k)h(k), x (l —k)h(k), and x(2 - k)h(k) are illus­
trated in Fig. 2.24c, d. and e. respectively. Thus we obtain the outputs

v(0) = 1

y(l) = 1 + a
y( 2) = 1 + a + a 1
82 Discrete-Time Signals and Systems Chap. 2

Clearly, for n > 0, the output is

y(n) = 1 + a 4- a2 + ■• ■+ a”
1 _ an+1 (2.3,29)
= 1-a
On the other hand, for n < 0, the product sequences consist of all zeros. Hence
v(n) = 0 n < 0
A graph of the output y(n) is illustrated in Fig. 2.24f for the case 0 < a < 1.
Note the exponential rise in the output as a function of n. Since |a| < 1, the final
value of the output as n approaches infinity is

v(oo) = lim v(n) = ------- (2.3.30)

n-*oc ' 1 —a

T o su m m arize, th e co n v o lu tio n fo rm u la p ro v id e s us w ith a m ean s fo r co m ­

p u tin g th e re sp o n s e o f a re lax ed , lin ear tim e -in v a ria n t system to an y a rb itra ry in p u t
signal x( n). It ta k e s o n e o f tw o e q u iv a le n t form s, e ith e r (2.3.17) o r (2,3.28), w h ere
jt(n ) is th e in p u t sig n al to th e system , h (n ) is th e im p u lse re s p o n s e of th e system ,
an d y (n ) is th e out p u t o f th e system in re sp o n s e to th e in p u t signal x (n ). T he
e v a lu a tio n o f th e co n v o lu tio n fo rm u la involves fo u r o p e ra tio n s , n am ely: f ol di ng
e ith e r th e im p u lse re sp o n s e as specified by (2.3.17) o r th e in p u t se q u e n c e as sp ec­
ified by (2.3.28) to yield e ith e r h ( —k) o r x { —k). resp ec tiv ely , shifting th e folded
se q u e n c e by n u n its in tim e to yield e ith e r h{n — k ) o r x { n — k ). mul t i pl yi ng the
tw o se q u e n c e s to yield th e p ro d u c t se q u en ce, e ith e r x { k) h{ n — k) o r x ( n - k ) h ( k ) ,
a n d finally s u m m i n g all th e v alu es in th e p ro d u c t se q u e n c e to y ield th e o u tp u t v (n )
o f th e sy stem a t tim e n. T h e folding o p e ra tio n is d o n e only o n ce. H o w ev er, th e
o th e r th re e o p e ra tio n s a re re p e a te d fo r all p o ssib le shifts —oc < n < oo in o rd e r
to o b ta in y (n ) fo r —oo < n < oc.

2.3.4 Properties of Convolution and the Interconnection

of LTI Systems

In th is se ctio n w e in v estig ate so m e im p o rta n t p ro p e rtie s o f co n v o lu tio n an d in ­

te r p re t th e se p r o p e rtie s in te rm s o f in te rc o n n e c tin g lin ear tim e -in v a ria n t system s.
W e sh o u ld stress th a t th ese p ro p e rtie s h o ld fo r e v e ry in p u t signal.
It is c o n v e n ie n t to sim plify th e n o ta tio n by using an a s te ris k to d e n o te the
c o n v o lu tio n o p e ra tio n . T h u s
OC

y( n) = x{n) * h(n) = ^ x ( k ) h ( n — k) (2.3.31)

Jt = - O C

In th is n o ta tio n th e se q u e n c e follow ing th e aste risk [i.e., th e im p u lse re sp o n se /i(«)]

is fo ld e d an d sh ifted . T h e in p u t to th e sy stem is ;c(n). O n th e o th e r h a n d , we also
sh o w ed th a t
OC

>>(n) = h{n) * x( n) = ^ h ( k ) x ( n - k) (2.3.32)

k=-oc
86 Discrete-Time Signals and Systems Chap. 2

2.3.5 Causal Linear Time-Invariant Systems

In S ectio n 2.2.3 w e d efin ed a causal system as o n e w hose o u tp u t at tim e n d e p en d s

o n ly on p re s e n t an d p ast in p u ts b u t d o es n o t d e p e n d on f u tu re in p u ts. In o th e r
w o rd s, th e o u tp u t o f the. sy stem at so m e tim e in s ta n t n, say n = no, d e p e n d s only
on v alu es o f jc(«) fo r n < n 0-
In th e case o f a lin e a r tim e -in v a ria n t system , cau sality can b e tra n sla te d
to a c o n d itio n o n th e im p u lse resp o n se. T o d e te rm in e this re la tio n sh ip , le t us
co n sid e r a lin ear tim e -in v a ria n t system having an o u tp u t a t tim e n = no given by
th e c o n v o lu tio n fo rm u la
OC
v (« o ) = ^2 h ( k ) x (n o ~ k )
k —- o c

S u p p o se th a t w e su b d iv id e th e sum in to tw o sets o f term s, o n e se t involving p re se n t

a n d p a st v alu es o f th e in p u t [i.e.. x{n) for n < n 0] a n d o n e se t involving fu tu re
valu es o f th e in p u t [i.e., n > no]. T h u s we o b ta in
OC -I

y (n u) — ^ h ( k ) x ( n o - k) + ^ h( k) x( nu - k)
i=0 k ——oc

= [/ ?(0)x (n ()) + h( \ ) x( r tu - 1) + h ( 2 ) x ( n 0 - 2 ) + ■■•]

+ [/?( —1 )x(«(] + 1) + h ( - 2 ) x ( n o + 2) + ■■ ■]

W e o b se rv e th a t th e term s in th e first sum involve jr(no), x( no — 1).........w hich are

th e p re s e n t a n d p ast values of th e in p u t signal. O n th e o th e r h a n d , th e te rm s in
th e se co n d sum in volve th e in p u t signal c o m p o n e n ts ;c(no + l) , x { n o -f 2 ) .........N ow ,
if th e o u tp u t a t tim e n = n 0 is to d e p e n d only on th e p re s e n t a n d p a st in p u ts, th en ,
clearly , th e im p u lse re sp o n se o f th e system m u st satisfy th e c o n d itio n
ft(n) = 0 n < 0 (2.3.38)
Since h{n) is th e re sp o n se o f th e re lax ed lin e a r tim e -in v a ria n t sy stem to a u nit
im p u lse a p p lied a t n = 0, it follow s th a t h(n) = 0 fo r n < 0 is b o th a necessary
an d a sufficient c o n d itio n fo r causality. H e n c e an L T I syst em is causal i f a n d onl y
i f its i mpul se respons e is z e r o f o r negative values o f n.
Since fo r a cau sal sy stem , h(n) = 0 fo r n < 0, th e lim its on th e su m m a tio n of
th e co n v o lu tio n fo rm u la m ay be m odified to reflect th is re stric tio n . T h u s w e h ave
th e tw o e q u iv a le n t form s
OC

y ( n ) = ^ h ( k ) x ( n - k) (2.3.39)
Jt=0
n
= ^ x{k) h{n — k) (2.3.40)
k=~oc
A s in d icated p rev io u sly , cau sality is re q u ire d in an y re a l-tim e signal p ro c e ss­
ing a p p licatio n , since at a n y given tim e n w e have n o access to f u tu re v alu es o f th e
 Sec. 2.3 Analysis of Discrete-Time Linear Time-Invariant Systems 87

in p u t signal. O n ly th e p re s e n t a n d p ast v alu es o f the in p u t signal are av ailab le in

c o m p u tin g th e p re s e n t o u tp u t.
It is so m e tim e s c o n v e n ie n t to call a se q u en ce th a t is z e ro fo r n < 0, a causa!
s e q u e n c e , an d o n e th a t is n o n z e ro fo r n < 0 a n d n > 0. a n on c aus al sequence. T his
te rm in o lo g y m e a n s th a t su c h a se q u e n c e could be th e u n it sa m p le re sp o n se of a
causal o r a n o n c a u sa l system , resp ectiv ely .
If th e in p u t to a causal lin e a r tim e -in v a ria n t system is a causal se q u e n c e [i.e.,
if jr(n) = 0 fo r n < 0]. th e lim its on th e co n v o lu tio n fo rm u la a re fu rth e r restricted .
In th is case th e tw o e q u iv a le n t form s o f th e c o n v o lu tio n fo rm u la b eco m e
n
y ( n ) = ^ h( k ) x ( n — k) (2.3.41)
*=o
n
= -k) (2.3.42)
*■=<)
W e o b se rv e th a t in th is case, th e lim its on the su m m a tio n s fo r the tw o a lte rn a tiv e
fo rm s are id en tical, a n d th e u p p e r lim it is grow ing w ith tim e. C learly , th e resp o n se
o f a cau sal sy stem to a causal in p u t se q u e n c e is causal, since y( n) — 0 fo r n < 0.
Example 2.3.5
Determ ine the unit step response of the linear time-invariant system with impulse
response
h{n) = a " u( n ) \a\ < 1

Solution Since the input signal is a unit step, which is a causal signal, and the system
is also causal, we can use one of the special forms of the convolution formula, either
(2.3.41) or (2.3.42). Since x(n) = 1 for n > 0. (2.3.41) is simpler to use Because of the
simplicity of this problem, one can skip the steps involved with sketching the folded
and shifted sequences. Instead, we use direct substitution of the signals sequences in
(2.3.41) and obtain

y(n) = y ~ v
*=(I
1 - a"*1
1 -ci
and y(n) = 0 for n < 0. We note that this result is identical to that obtained in Ex­
ample 2.3.3. In this simple case, however, we computed the convolution algebraically
without resorting to the detailed procedure outlined previously.

2.3.6 Stability of Linear Time-Invariant Systems

A s in d ic a te d p rev io u sly , sta b ility is an im p o rta n t p ro p e rty th a t m u st be co n sid e re d

in an y p ra c tic a l im p le m e n ta tio n o f a system . W e d efin ed an a rb itra ry relax ed
system as B IB O sta b le if a n d only if its o u tp u t se q u e n c e y («) is b o u n d e d fo r every
b o u n d e d in p u t x ( n) .