Chapter

Linear Time Invariant

System 2
Introduction
Two most important attributes of systems are linearity and time-invariance. In this chapter we will develop
the fundamental input-output relationship for systems having these attributes. It will be shown that the input-output
relationship for LTI system is described in terms of a convolution operation. The importance of the convolution
operation in LTI systems steps from the fact that knowledge of the response of an LTI system to the unit impulse
input allows us to find its output to any input signals.

2.1 Response of Continuous Time LTI System and Convolution Integral

We assume that the system is continuous time and linear time invariant. That is the system follows superposition
and time invariance property. First of all let us try to write any arbitrary continuous time signal in terms of impulse
((t)) function.
Let us take any signal x(t) and practical delta function p(t),

x (t )
p(t)

1/

t t
–/2 /2

Fig. 2.1 : Arbitrary signal x(t) and practical delta function p(t)

Now, x(t) p(t) is x(0) p(t). Plot of x(0) p(t) is shown in Fig. 2.2(a) and plot of x(0) p(t) is shown in
Fig. 2.2(b). Similarly Fig. 2.2(c) and Fig.2.2 (d) show x(t) p(t – ) i.e. x() p(t – ) and x() p(t – ).

GATE MASTERS PUBLICATION

 64 Signals & Systems

x(0)p( t) x(0)p(t )

x(0)/ x(0)

t t
–/2 /2

(a) (b)

x()p( t – ) x()p( t – )

x( )/ x()

t t


(c) (d)

Fig. 2.2 : Multiplication of x(t) with practical impulse function

Now, we form a new signal x ( t ) :



x ( t ) =  [ x(k ) 
k  
p (t  k )]  ...(1)

we can see that x ( t ) is staircase approximation of x(t) of Fig. 2.1.

x~ (t )

t
5 3  0  3 5 7
  
2 2 2 2 2 2 2

Fig. 2.3 : Plot of x ( t )

Now as  0 the staircase approximation x ( t ) will become x(t),

Lim x ( t ) = x(t) ...(2)

0

Since,  0 will make input practical delta function into ideal delta function and summation of equation (1)
will change to integration. Thus,


x(t) =

 x ( ) ( t  ) d  ...(3)

The above equation show how any time domain signal can be written in terms of impulse function.
Remember: We can intuitively think that x(t ) as a sum of weighted shifted impulses, where the weight on the impulse
(t  ) is x()d 

GATE MASTERS PUBLICATION

 Linear Time Invariant System 65

Response of Continuous Time LTI System for any Arbtrary Input

Suppose any LTI system in given, we first of all find the impulse response of system. That is at input (t) is
applied and output is the impulse response h(t).

LTI
(t ) h( t)
system
Impulse Impulse
input response

Fig. 2.4 : Impulse response of system

Sy stem
( t )   h( t )
Since, the system is time invariant so.
Sy stem
( t  )   h( t )
Since system is linear so if we multiply input with a constant then output also get multiplied by same
constant.
Sy stem
 x( ) ( t  )   x( ) h( t  )
Since, integration is a linear operation applying this operation at input and output side.
 
Sy stem
 x( ) ( t  ) d  

  x ( ) h( t  ) d 



Sy stem
 x ( t )    x ( ) h( t  ) d 


So output, y(t) =  x ( ) h( t  ) d 

...(4)

Equation (4) is referred as the convolution integral or superposition integral and symbolically
represented as,
y ( t ) = x(t)  h(t) ...(5)
Thus, in case of any linear time invariant system we can find output for any arbitrary input x(t) by
convolution of x(t) with impulse response of system h(t).
Procedure that should be followed to find out convolution of two signals:


Suppose, y ( t ) = x ( t )  h( t )   x( )  h( t  ) d 


Signals x(t) and h(t) are given and we have to find y(t), or we have to compute the above integral. Here
the integration variable is ‘’ and ‘t’ is a constant.
S t e p - 1 : Change variable ‘t’ by ‘’ in x(t) to get x().
S t e p - 2 : From h(t) find h() by changing variable ‘t’ by ‘’ then plot h(–) by taking mirror image of h()
from amplitude axis. To get h(t – ) we have to do the required shifting in h(–).

S t e p - 3 : To find y(t) multiply x() and h(t – ) and find,  x ( )  h( t  ) d  .


Lets look an example, x ( t ) = e –t u(t),

h(t) = u(t)
and y ( t ) = x(t)  h(t)

GATE MASTERS PUBLICATION

66 Signals & Systems

First of all we plot x(t) and h(t),

x(t) h(t)

1.0

e –t
1.0

t t
0 0

Now to find x() and h(),we will simply change t by ,

x() h( )

1.0

e–
1.0

 
0 0

Now h(–) will be inversion of h()

h( –)

1.0


0

Now to find h(t – ) we have to shift h(–). For t > 0 the h(–) will shift right by t units.

h( t – )

1.0
(for t > 0)


t

For t < 0 the h(–) will shift left by t units.

h( t – )

1.0 (for t > 0)


t 0

GATE MASTERS PUBLICATION

 Linear Time Invariant System 67

Now to find y(t) we have to multiply x() and h(t – ),

Case-1: For t > 0 the plot of x() and h(t – ).

x( ) h(t –  )
1
–
–
1.0 e
e
× =

  
t 0 t

t

For t > 0, y(t) = e d
0

= (1 – e–t)
Case-2: For t < 0, plot of x() and h(t – ) are
x( ) h( t – )
1.0
–
e
× = 0

 
t

y(t) = 0 for t < 0

(1  e t ), t  0
so, y(t) = 
 0, t 0
= (1 – e–t) u(t)

Example 2.1

Find the convolution of two continuous time signals,

x ( t ) = e –t  and h(t) = e –t u(t – 1)
Solution 2.1
We have to find y(t) = x(t)  h(t). First let us drawn x(t) and h(t),
x (t )
h( t)
1.0

–t –t –t
e e e

t t
1.0

The plot of x() and h() is

x()
h( )
1.0

– –
e e e

 
1.0

GATE MASTERS PUBLICATION

68 Signals & Systems

The plot of h(–) is

h(–)


e


–1.0

Plot of h(t – ) will be shifted h(–) by ‘t’, if t is positive h(–) shift right by ‘t’ units and if t is negative h(–) shift
left by ‘t’ units. Another way is to use equation to visualise, h(t) = et u(t  1)

(t  ) e (t ) t    0 e(t ) t

Thus h(t )= e u (t  )   
 0 t    0  0 t

These two figures show two cases when t > 1 and t < 1.
h(t – ) h(t – )

e–(t – ) e–(t – )

 
( t – 1) ( t – 1)

SInce y(t) is convolution of x(t) and h(t),



y(t) =  x( ) h( t  ),

putting x() we get

0 

=  e  h(t  ) d    e  h(t  ) d 
 0

Analysing the plot of h(t – ) we can see that if t < 1 then h(t – ) = 0 for t > 0. So for t < 1
0

y(t) = e

h( t  ) d   0

For t < 1, h(t – ) = e–(t – ) for  < t – 1

t 1 t 1
  ( t  ) t 2
 y(t) =  e e d  e e d
 

e( t  2)
y(t) = for t  1
2
Now for t > 1, h(t – ) = e–(t – ) for  < t – 1
0 t 1

y(t) =  e  e( t  ) dt  e

e ( t  ) dt
 0

0 t 1
t 2 t e t
= e e dt  e d =  et ( t  1)
 0 2
 e t 
y(t) =   e t ( t  1)  for t  1
 2 

GATE MASTERS PUBLICATION

 Linear Time Invariant System 69

Example 2.2

Consider a LTI system with impulse response h(t) and input x(t), the output of system is y(t). The value of
output at t = 5 is
x (t ) h( t)

1.0
2.0

t t
–5 –4 4 5 0 10

Solution 2.2
Since y(t) is output,
y ( t ) = x(t)  h(t)

y(t) =  x ( ) h ( t  ) d 


Now, y( t ) t  5 =  x() h(5  ) d 



Now we plot x(t) and h(5 – ),

x() h(5 – )

1.0 2.0

 
–5 –4 4 5 –5 5


So, y(5) =  x( ) h(5  ) d 


= Area of x(t)  h(5 – t)

x () h(5 – )

2.0


–5 –4 4 5

So, y(5) = 18

Example 2.3

Find the convolution of a rectangular signal x(t) with itself.

x(t) = A rect(t/T)
Solution 2.3
We have to find y(t), y(t) = x(t)  x(t)

y(t) =  x ( ) x( t  ) d 


GATE MASTERS PUBLICATION

70 Signals & Systems

x( ) x( –)

 
–T/2 T/2 –T/2 T/2

The plot of x(t – ) will be, for t > 0 the signal x(–) will move right side by t units
So, when t > 0
 x(t – )
y(t) =  x ( ) x( t  ) d 


 T T
 A,     
we can seen that, x() =  2 2
t – /2 t + /2
 0 else wh ere

 T T
A, t     t 
and x(t – ) =  2 2
 0 else wh ere
T /2

 y(t) =  A2 d 
t T / 2

y ( t ) = A2 (T – t)
So, y(t) = A2 (T – t) when t > 0 and t < T
because overlap between x() and x(t – ) exist for t < T and for t > T the overlap of x() and x(t – ) will
be zero and their product will also be zero thus y(t) = 0 for t > T.
Now when t is negative then x(–) will move to left by t units, so plot of x(t – ) for t < 0 is
x(t – )


t – /2 t +  /2

 T T
 A,    
Now, x() =  2 2
 0 , else wh ere

 T T
A, t     t 
and x(t – ) =  2 2
 0 , else wh ere

 t T / 2

y(t) =  x (  ) x ( t  ) d  =  A2
 T / 2

y(t) = A2(t
+ T) for t < 0 and t > – T
because for t < –T overlap between x() and x(t – ) do not exist and y(t) = 0 for t < –T.

GATE MASTERS PUBLICATION

 Linear Time Invariant System 71

The plot of y(t) is

2
AT


–T T

Thus, A rect (t/T)  A rect (t/T) = A2T (t/T)

Study Notes
Generally in GATE exam whenever we have to find convolution of two signals we will not use the method shown above
because this is a lenghty method, it is better to use laplace transform to do the same. We will learn laplace transform in
Chapter 4. The above method will be helpful in those questions where x(t) is input to an LTI system with impulse response
h(t) and we have to find value of output at a particular time instant as shown in Example 2.2.

2.2 Properties of Convolution Integral

Continuous-time convolution satisfied the following properties:

2.2.1 Commutative Property

The convolution operation possesses the commutative property, or
x(t)  h(t) = h(t)  x(t) ...(6)
This property (see Fig. 2.5), implies that the role of the input signal and impulse response are interchangeable.
It is often used to simplify the evaluation or interpretation of the convolution integral.

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

Fig. 2.5
Proof: By definition,


x(t)  h(t) =  x ( ) h( t  ) d 


If we perform a change of variables by letting (t – ) = , then  = (t – ), d = –d,  as  –, and
 – as  . Therefore,
 

x(t)  h(t) =   x(t  ) h() d    h() x(t  ) d   h(t )  x(t )

 

Study Notes
Suppose, y(t) = x(t)  h(t),
to find the result we can perform,
 

x(t)  h(t) =  x ( ) h( t  ) d  or h(t)  x(t) =  h() x(t  ) d 

 

If we have signal x(t) simpler than h(t) then we perform h(t)  x(t) and if h(t) is simpler then we perform
x(t)  h(t).

GATE MASTERS PUBLICATION

 72 Signals & Systems

For example if h(t) = e–t u(t) and x(t) = u(t)

then to find y(t) it will be easier to perform,


h(t)  x(t) =  h() x(t  ) d 



and if x( t) = e –t u(t) and h(t )  u (t )

then to find y(t) it will be easier to perform,


x(t)  h(t) =  x ( ) h( t  ) d 


2.2.2 Associative Property

The continuous convolution is associative, i.e.,
x(t)  [h1(t)  h2(t)] = [x(t)  h1(t)]  h2(t) ...(7)
Associativity, as shown in Fig. 2.6 implies that a cascade combination of LTI systems can be replaced by a
single system whose impulse response is the convolution of the individual impulse responses.

x ( t) h1 ( t ) h2 ( t ) y (t ) x (t ) h1 ( t )  h2 ( t ) y (t )

Fig. 2.6

Since continuous convolution is associative, there is no need to indicate which convolution is to be

performed first, hence the square brackets can be removed from equation (7). Since continuous convolution is
also commutative, then
x(t)  h 1 (t)  h 2 (t) = x(t)  h 2 (t)  h 1 (t) = ... = h 2 (t)  h 1 (t)  x(t)

2.2.3 Distributive Property

The convolution operation possesses the distribution property, or
x(t)  [h 1 (t) + h 2 (t)] = x(t)  h 1 (t) + x(t)  h 2 (t) ...(8)
Distributivity, as shown in Fig. 2.7 states that a parallel combination of LTI systems is equivalent to a
single system whose impulse response is the sum of the individual impulse responses in the parallel configuration.

h1(t)

x(t)  y(t) x(t) h1(t) + h2(t) y(t)

h2(t)

Fig. 2.7

2.2.4 Shift Property

According to the shift property,
if x(t)  h(t) = y( t) ,
then, x(t)  h(t – t0) = y(t – t0) ...(9)
and x(t – t1)  h(t – t2) = y(t – t1 – t2) ...(10)

GATE MASTERS PUBLICATION

 Linear Time Invariant System 73

Proof: By definition,

x(t)  h(t) =  x( ) h( t  ) d   y( t )



x(t)  h (t – t0) =  x( ) h( t  t 0   ) d   y( t  t 0 )


Equation (10) and follows from equation (9).

2.2.5 Convolution with an Impulse

The convolution of a signal x(t) with a unit impulse function results in the signal x(t) itself
x(t)  (t) = x(t) ...(11)
Proof: By definition,

x(t)  (t) =  x( ) ( t  ) d 


From the sampling property of the impulse function, we have

x() (t – ) = x(t) (t – )
 

and therefore, x(t)  (t) =  x ( ) ( t  ) d   x ( t )  ( t  ) d   x ( t )

 

similarly we can prove that,

x(t)  (t – t0) = x(t – t0)
1 x( t )
x(t)  (at) = x ( t )  ( t ) 
a a

 b
x t  
1 b a
x(t)  (at + b) = x ( t )   ( t  )  
a a a

Study Notes
To perform the convolution of a signal with delta function, we should follow these steps
1
• First of all find the area and location of delta function and write delta function as ( t -location) .
area
• Now convolution of signal x(t) with given delta fuction is found by replacing variable ‘t’ in x (t) by (t - location) and
divide expression by area of delta function.

2.2.6 Width Property

The width of the non-zero extent (the interval of time between the first and last non-zero values) of the
continuous convolution of two functions equals the sum of the widths of the non-zero extents of the two functions.
In other words, if the durations (widths) of x(t) and h(t) are finite, given by Wx and Wh, respectively, then the
duration (width) of x(t)  h(t) is Wx + Wh.

GATE MASTERS PUBLICATION

 74 Signals & Systems

Study Notes
Suppose y(t) = x(t)  h(t), signal x(t) is non-zero for TxL < t < TxH and h(t) is non-zero for ThL < t < ThH then y(t) will be non-
zero only for ThL + TxL < t < TxH + ThH ,and width of x(t) and h(t) is Wx = TxHTxL and Wh = ThHThL.

2.2.7 Differentiation Property

According to the differentiation property,
If, x(t)  h(t) = y(t)

d  d  d
then,  x ( t )   h( t ) = x ( t )   h( t )   y( t ) ...(12)
 dt   dt  dt
Proof : By definition,


y(t) = x ( t )  h( t )   x( ) h( t  ) d 


By differentiating both sides of the above equation w.r.t. t, we get


d d  d 
dt
y( t ) =  x( )  dt h( t  )  d   x ( t )   dt h( t ) 


Similarly we can prove that

d2 d 2 x( t ) d 2 h( t ) dx ( t ) dh( t )
2
y( t ) =  h( t )  x ( t )   
2
dt dt dt 2 dt dt

2.2.8 Time-Scaling Property

According to the time-scaling property,
if, x(t)  h(t) = y(t),
1
then, x(at)  h(at) = y (at )
a
This property states that if both x(t) and h(t) are time-scaled by a, their convolution is also time-scaled by a
and multiplied by 1/[a].
Proof : Case-I: a> 0

By definition, x(t)  h(t) =  x( ) h( t  ) d   y( t )


 
x(at)  h(at) =  x ( a ) h( a( t  )) d    x( a) h( at  a) d 
 

A change of variables is performed by letting a = , which also yields,

1
d =   d ,    as  , and   as  
a

1 1
Therefore, x(at)  h(at) =  x (  ) h( at   d   y( at )
a  a

GATE MASTERS PUBLICATION

 Linear Time Invariant System 75

Case-II: a < 0

By definition, x(t)  h(t) =  x ( ) h( t   d   y( t )


 

x(–at)  h(–at) =  x ( a ) h(  a( t   d    x( a) h( at  a  d 

 

A change of variables is performed by letting –a =, which also yields,

1
d =    d ,    as  – and   – as  
a

1 1
Therefore, x(at)  h(at) =  x(  ) h( at   ) d  a y( t )
a 
From the above two cases it is evident that,
1
x(at)  h(at) = y( at )
a

REMEMBER • The convolution of an odd and even function is an odd function.

• The convolution of an even and even function is an even function.
• The convolution of an odd and odd function an even function.
• The convolution has the property that the area of convolution integral is equal to product of
area of two signals which are convolved. That is if y(t) = x(t)  h(t), then Ay = Ax  Ah
d
• x( t )   ( t )  x ( t )
dt
• u(t)  u(t) = r(t)
t

• x( t )  u ( t )   x( ) d 


Example 2.4

Determine and sketch the convolution of the following signals:

 1, 1  t  1
x(t) = 
 0 , else where
and h ( t ) = (t + 1) + 2(t + 2)
Solution 2.4
Using the result,
y(t) = x(t)  h(t) = x(t)  [(t + 1) + 2(t + 2)]
= [x(t)  (t + 1)] + [x(t)  2(t + 2)]
= x(t + 1) + 2x(t + 2)]
= x(t + 1) + 2x(t + 2)

GATE MASTERS PUBLICATION

76 Signals & Systems

The signals x(t), h(t), and y(t) are

y ( t)
3
h( t)

x (t ) 2 2

1
1 1

t t t
–1 0 1 –2 –1 0 –3 –2 –1 0

Example 2.5

Consider an LTI system with input and output related through the following equations:
t
 (t  )
y(t) = e x(  2) d 


What is the impulse response h(t) for this system?

Solution 2.5
t
 (t   )
Given that, y(t) = e

x(   2) d 

A change of variables is performed by letting  = ( – 2), which also yields d = d,   – as   –,
and   (t – 2) as   t. Therefore,
t2 
y(t) =  e( t  2  ) x (  ) d   e
( t  2  )
u ( t  2   ) x(  ) d
 


Comparing with, y(t) =  h( t   ) x ( ) d


we get, h(t – ) = e–(t – 2 – ) u(t – 2 – )

Therefore, the impulse response of the system is
h ( t ) = e–(t – 2) u(t – 2)

Example 2.6

What is (t – 0.5)  (t – 0.5)?

Solution 2.6
Let, y ( t ) = (t – 0.5)  (t – 0.5)
If we assume that x ( t ) = (t – 0.5)
Thus y ( t ) = x(t)  (t – 0.5)
= x(t – 0.5)
so, y ( t ) = (t – 0.5 – 0.5)
= (t – 1)

GATE MASTERS PUBLICATION

 Linear Time Invariant System 77

Study Notes
The step response s(t) of a continuous-time LTI system is defined as the response of the system when the input is u(t).
In many applications, the step response s(t) is also a useful characterization of the system. The step response s(t) can be
easily determined by equation, that is
 t

s(t) = h(t)  u(t) =  h( ) u ( t  ) d  


 h( t ) d 


Thus, the step response s(t) can be obtained by integrating the impulse response h(t). Differentiating with respect to t, we
get
d s( t )
h(t) = s( t ) 
dt
Thus, the impulse response h(t) can be determined by differentiating the step response s(t).

2.3 Interconnection of LTI System

2.3.1 Series or Cascade Connection of LTI System

Two LTI systems are connected in cascade and they have impulse response h1(t) and h2(t).

x ( t) h1 ( t ) h2 ( t ) y (t )  x (t ) h( t ) y (t )

Fig. 2.8 : Series or Cascade connection

y(t) = x(t)  h1(t)  h2(t) ...(14)
y(t) = x(t)  h(t)
So, this cascade can be replaced by a single system with impulse response h(t),
h(t) = h1(t)  h2(t) = h2(t)  h1(t) ...(15)
That is systems can interchange location also.

2.3.2 Parallel Connection

The two systems having impulse response h1(t) and h2(t) are connected in parallel as shown in figure.
We can see that,
y(t) = x(t)  h1(t) ± x(t)  h2(t)
= x(t)  (h1(t) ± h2(t))
= x(t)  h(t) ...(16)

h1( t)
+
x ( t)  y (t )  x (t ) h1(t ) ± h2( t) y (t )
–
h2( t)

Fig. 2.9 : Parallel connection of two LTI system

So, parallel connection can be replaced by single system with impulse response,
h(t) = h1(t) ± h2(t)

GATE MASTERS PUBLICATION

 78 Signals & Systems

2.4 Response of Discrete Time LTI System and Convolution Sum

In this section we assume that system is discrete and linear and time invariant. Signal x[n] is any arbitrary
signal and we can write it as a sum of shifted and scaled unit impulses. For example:
 x[1], n  1
x[–1] [n + 1] = 
 0, n  1

 x[0], n  0
x[0] [n] = 
 0, n  0
 x[1], n  1
x[1] [n – 1] = 
 0, n  1
More generally, by including additional shifted, scaled impulses, we can write
x[n] = ... + x[–3] [n + 3] + x[–2] [n + 2] + x[–1] [n + 1] + x[0] [n]
+ x[1] [n – 1] + x[2] [n – 2] + x[3] [n – 3] + .... ...(17)
For any value of n, only one of the terms on the right-hand side of equation (17) is non-zero, and the scaling
associated with that term is precisely x[n]. Writing this summation in a more compact form, we have,

x[n] =  x[k ] [n  k ]
k  
...(18)

This corresponds to the representation of an arbitrary sequence as a linear combination of shifted unit
impulses [n – k], where the weights in this linear combination are x[k].

Response of Discrete-time LTI System for any Arbitrary Input:

Suppose a discrete-time LTI system is given and impulse response of system h[n] is also given (that is h[n] is
output when [n] is input)

LTI
[n ] Impulse response
system
h[n ]

Fig. 2.10 : Impulse response of system

Sy stem
[n ]   h[n ],
Since system is time invariant, so
Sy stem
[n  m ]   h[n  m ]
Since system is linear so if input is multiplied with a constant then output get multiplied by same constant.
Sy stem
x[m ] [n  m ]   x[m ] h[n  m ]
Since summation is a linear operation, applying summation on both side we get,
 
Sy stem

m  
x[m ] [n  m ]   
m  
x[m ] h[n  m ]


Sy stem
 x[n ]   
m  
x[m ] h[n  m ]

Thus, output is y[n] = 

m  
x[m ] h[n  m ] ...(19)

GATE MASTERS PUBLICATION

 Linear Time Invariant System 79

Equation (19) is referred as the convolution sum or superposition sum and symbolically represented as

y[n] = x[n]  h[n] = 
m  
x[m ] h[n  m ] ...(20)

Thus in case of any linear time invariant we can find output for any arbitrary input x[n] by convolution of
x[n] with impulse response of system h[n].
Procedure that should be followed to find out convolution of two signals:


Suppose, y[n] = x[n]  h[n] = 

m  
x[m ] h[n  m ]

Signal x[n] and h[n] are given and we have to find y[n] or we have to compute above summation. In this
summation variable is ‘m’ and ‘n’ is a constant. So we will follow this procedure.
S t e p - 1 : Change variable ‘n’ to ‘m’ in x[n] and h[n] to get x[m] and h[m].
S t e p - 2 : Now we get h[–m] by inversion of h[m]. To get h[n – m] we have to do shifting in h[–m].

S t e p - 3 : To find y[n] we multiply x[m] and h[n – m] and find 
m  
x[m ] h[n  m ] .

Let’s look at an example,

x[n] = (0.5)n u[n] and h[n] = u[n]
we have to find, y[n] = x[n]  h[n]
First of all we plot x[n] and h[n]
x [n ] h[n ]

1.0 1.0
0.5 .....
.....
n n
0 1 2

Now to find x[m] we simply change variable n by m.

(0.5) m , m  0
So, x[m] = (0.5) m u [m ]  
 0, m 0

 1, m  0
and for h[m] = u [ m ]  
0 , m  0
Now, h[–m] = u[–m]

1, n  m  0 1, m  n
Thus, h[n – m] = u[n – m] =  = 
0 , n  m  0 0, m  n
For n  0,
h[n – m]

1.0
..... ...
m
–4 –3 –2 –1 0 1 2 n

GATE MASTERS PUBLICATION

 80 Signals & Systems

and for n < 0

h[n – m ]

Now, y[n] =  x[m ] h[n  m ]
1.0
m  
.....
For n > 0, using definition of x[m], h[n – m] we get m
n
n
((0.5) n  1  1)
y[n] =  (0.5) m  1  (0.5)  1
0

y[n] = 2[1 – (0.5)n + 1] for n  0

For n < 0 we can see that y[n] = 0 because for n < 0 h[n – m] is zero for m > 0 and x[m] h[n – m] will be zero.
Thus, y[n] = 2(1 – (0.5)n + 1)

Study Note
The discrete-time LTI system which have impulse response h[n] non-zero only for finite value of ‘n’ is called finite impulse
response filter on FIR filter. If impulse h[n] is non-zero for infinite values of ‘n’ then it is called infinite impulse response
filter IIR filter.

Example 2.7

Find the convolution of two sequences:

 0, n < -5  0, n3
 n  n
x[n] =  1  and h[n] =  1 
≥   , n  3
  , n - 5
2
   3 

Solution 2.7
So we have to find, y[n] = x[n]  h[n]

= 
m  
x[m ] h[n  m ]

 0, m  5
 m
we can define, x[m] =  1 
  , m   5
 2 

 0, nm 3  0, m  n 3
 nm  nm
and h[n – m] =  1    1 
  , nm 3   , m n3
 3   3 
Now, using above definition we can say that, product x[m] h[n – m] will be non zero for –5  m  n – 3,
 n 3 m nm
1 1
y[n] =  x[m ] h[n  m ]      
2 3
m   m  5

and for this it is required that (n – 3)  –5 that is n  –2.

n 3 n m
1  3
 For n  –2, y[n] =   3   
 2
5

GATE MASTERS PUBLICATION

 Linear Time Invariant System 81

 3 n  3   3  n  3 
n 5
   1     1 
5 n
1 3   2    2   1   2  
=     =    
3 2  3 3  3 1
  1 2
2

5 n n 3
 2   1   3  
So, for n  –2, y[n] = 2        1
 3   3   2  
and for n  –2, y[n] = 0

Example 2.8

Find convolution of rectangular signal with itself.

 n 
x[ n] = rect  
2N 
Solution 2.8
That is we have to find out, y[n] = x[n]  x[n]
 n  1,  N  n  N
x[n] = rect  
 2 N  0 , else wh ere

Since, y[n] =  x[m ] x[n  m ]
m  

 m  1, N  m  N
Since, x[m] = rect  
 2 N  0 , else where

1, N  n  m  N
and x[n – m] = 
0 , else wh ere

1, N  n  m  N  n
= 
0 , else wh ere
when n  0 that is when n is positive then,
 N
y[n] =  x[m ] x[n  m ]   1  (2 N  n )
m   N  n

and for n > 2N, overlap between x[m] and x[n – m] is 0 so,
y[n] = 0 for n > 2N
Now for n < 0, using above definition we get,
 N n
y[n] =  x[m] x[n  m]   1  2N  n
m   N

and y[n] = 0 for n < –2N as overlap between x[m] and x[n – m] is zero.
So, y[n] = 2N – m 0  n  2N
y[n] = 2N + n –2N  n < 0

GATE MASTERS PUBLICATION

82 Signals & Systems

Example 2.9

Compute the convolution y(n) = x(n)  h(n) of the following pairs of signals:
( a ) x(n) = h(n) = u(n)
( b ) x(n) = (0.8) n u(n) and h(n) = (0.4) n u(n)
( c ) x(n) = h(n) = a n u(n)
( d ) x(n) = u(n – 1) and h(n) =  n u(n – 1)
( e ) x(n) = r(n) = nu(n) and h(n) = a –n u(n – 1), where a < 1
Solution 2.9
By definition:
(a) y ( n ) = x(n)  h(n)
 
=  x ( k ) h( n  k )   u( k ) u ( n  k )
k   k  

The lower limit on the convolution sum simplifies to k = 0 [because u(k) = 0, k < 0], the upper limit
to k = n [because u(n – k) = 0, k > n], and we get
n
y(n) =  1  ( n  1) u ( n )  r( n  1)
k 0

( b ) By definition, y ( n ) = x(n)  h(n)

 
=  x ( k ) h( n  k )   (0.8)k u ( k ) (0.4) n  k u ( n  k )
k   k  

The lower limit on the convolution sum simplifies to k = 0 [because u(k) = 0, k < 0], the upper limit
to k = n [because u(n – k) = 0, k < n], and we get
n n
y(n) =  (0.8)k (0.4) n  k  (0.4) n  ( 2)k
k 0 k 0

n
n
1  2 n 1
= (0.4)  2 k  (0.4) n  (0.4) n ( 2 n  1  1)
k 0 12

= [2(0.8) n – (0.4) n ] u(n)

( c ) By definition, y ( n ) = x(n)  h(n)
 
=  x ( k ) h( n  k )   a k u( k ) a n  k u( n  k )
k   k  

The lower limit on the convolution sum simplifies to k = 0 [because u(k) = 0, k < 0], the upper limit
to k = n [because u(n – k) = 0, k > n], and we get
n n
y(n) =  ak a n  k  an  1  ( n  1) a n u ( n )
k 0 k 0

1, n  1
( d ) Given that, x ( n ) = u ( n  1)  
0 , n  1

 n , n  1
and h ( n ) =  n u (n  1)  
 0, n  1

GATE MASTERS PUBLICATION

 Linear Time Invariant System 83


By definition, y(n) = x ( n )  h( n )   x ( m ) h( n  m )
m  

1, m  1
Here x ( m ) = u ( m  1)  
0 , m  1
 n  m , n  m  1  n  m , m  n  1
and h(n – m) =  n  m u ( n  m  1)   
 0 , n  m  1  0 , m  n 1
using the above definitions we can see that, x(m) h(n – m) will be zero for m < 1 and m > n1, and non
zero for m  n  1 and m  1 ,thus n has to be greater than equal to 2
n 1 n 1
and y(n) =  x( m ) h( n  m ) =  a n m
m 1 m 1

n 1 n 2 n 2
n   n
=   am  n  ( m 1)   n  1  ( m )  1
m 1 m 0 m 0

  n
 , n2
Therefore, y(n) =  1 
 0, n0


  n
or, more compactly, y(n) = u ( n  2)
1

n , n  0
( e ) Given that, x ( n ) = n u( n )  
0 , n  0

n  a  n , n  1
and h ( n ) = a u ( n  1)  
 0 , n  1

By definition, y ( n ) = x ( n )  h( n )   x ( k ) h( n  k )
k  

k , k  0
Since, x ( k ) = k u( k )  
0 , k  0

( n  k ) a ( n  k ) , n  k  1 a ( n k ) , k  n  1
and h(n – k) = a u ( n  k  1)   
 0 , n  k  1  0 , k  n 1
Thus product x(k) h(n – k) will be non zero for k  n  1 and k  0 , that is 0  k  n – 1 and n  1. Thus
n 1 n 1
y(n) =  x ( k ) h( n  k ) =  k a ( n k ) ;
k 0 k 0

n 1
n a n 1
= a  k ak 
(1  a )2
[1  n a n  1  ( n  1)a n ]
k 0

 a  n 1
 [1  n a n 1  ( n  1) a n ], n  1
Therefore, y ( n ) =  (1  a )2

 0; n 1

a n 1
or, more compactly, y(n) = 2
[1  n a n 1  ( n  1) a n ] u ( n  1)
(1  a )

GATE MASTERS PUBLICATION

84 Signals & Systems

Example 2.10

Consider a LTI system with impulse response h[n] = u[n] and input x[n] then, output y[n] will be _____.
Solution 2.10
Since, y[n] = x[n]  h[n]


= 
m  
x[m ] h[n  m ]

Since, h[n] = u[n]

1, n  m  0
 h[n – m] = u[n – m] = 
0 , n  m  0

1, m  0
h[n – m] = 
0 , m  0
n
Thus, y[n] = 
m  
x [m ]

Example 2.11

Fin d:
( a ) x[n]  [n – n 0 ]
( b ) x[n]  [an]
( c ) x[n]  [an + b]
Solution
n
(a) x[n]  [n – n0] = 
m  
x[m ] [n  m  n 0 ]

Since, x[n] = 
m  
x[m ] [n  m ]

So, we can say that,

x[n]  [n – n0] = x[n – n0]
( b ) x[n]  [an], since [an] = [n]
so, x[n]  [an] = x[n]
( c ) x[n]  [an + b], first of all we find [an + b]

 b
[an + b] =   n   if b/a is integer and
 a
if b/a is not integer then [an + b] = 0.

 b
So, if b/a is integer then x[n]  [an + b] = x  n  .
 a 
Otherwise, x[n]  [an + b] = 0.

GATE MASTERS PUBLICATION

 Linear Time Invariant System 85

2.5 Properties of Convolution Sum

The structure of the convolution sum is similar to that of the convolution integral. Moreover, the properties
of the convolution sum are similar to those of the convolution integral. We shall enumerate these properties
here without proofs. The proofs are similar to those for the convolution integral.

2.5.1 Commulative Property

The convolution sum possesses the commutative property:
x(n)  h(n) = h(n)  x(n) ...(21)

Study Notes
Suppose, y(n) = x(n)  h(n),
to find the result we can perform,

x(n)  h(n) =  x ( k ) h( n  k )
k  


or h(n)  x(n) =  h( k ) x ( n  k )
k  

If we have signal x(n) simpler than h(n) then we perform h(n)  x(n) and if h(n) is simpler then we perform
x(n)  h(n).
For example if, h(n) = an u(n) and x(n) = u(n)
then to find y(n) it will be easier to perform,

h(n)  x(n) =  h( k ) x ( n  k )
k  

and if, x(n) = an u(n) and h(n) = u(n)

then to find y(n) it will be easier to perform,

x(n)  h(n) =  x ( k ) h( n  k )
k  

2.5.2 Associative Property

The convolution sum possesses the associative property:
x(n)  [h1(n)  h2(n)] = [x(n)  h1(n)]  h2(n) ...(22)

2.5.3 Distributive Property

The convolution sum possess the distributive property:
x(n)  [h1(n) + h2(n)] = [x(n)  h1(n)] + [x(n)  h2(n)] ...(23)

2.5.4 Shifting Property

According to the shift property:
If, x(n)  h(n) = y(n) ...(24)
then, x(n)  h(n – n0) = x(n – n0)  h(n) = y(n – n0)
and x(n – n1)  h(n – n2) = y(n – n1 – n2) ...(25)

GATE MASTERS PUBLICATION

 86 Signals & Systems

2.5.5 Convolution with an Impulse

Convolution of a signal x(n) with a unit impulse function results in the signal x(n) itself:
x(n)  (n) = x(n) ...(26)
Similarly we can prove that,
x(n)  (n – n0) = x(n  no)

Study Notes
To perform the convolution of a signal with delta function, we should follow these steps
• First of all find the location of delta function if location is an integer then move to step 2 else answer is zero.
• Now convolution of signal x(n) with given delta fuction is found by replacing variable ‘n’ in x(n) by
(n – location), where location is location of the delta function.

2.5.6 Width Property

Suppose two signals x[n] and h[n] are convolved to get y[n] then, if the location of first non-zero sequence
value (Left edge) of x[n] and h[n] is NLx and NLh and location of last non-zero sequence value (Right edges) of x[n]
and h[n] is NHx and NHh. Then
Number of samples of x[n], length of x[n] = Lx = NHx – NLx + 1
Number of samples of h[n], length of h[n] = Lh = NHh – NLh + 1
Then length of y[n], Ly = Lx + Lh – 1
Also, first non-zero sequence value (left edge) of y[n] is NLx + NLh and right edge of yh[n] is NHx + NHh. Thus
length of convolution of two signals is equal to sum of length of two signals being convolved.

REMEMBER If h[n] is impulse response of a LTI system then step response will be h[n]  u[n]
n
[n] = h[n ]  u [n ]   h[ m ]

and from [n] we can get h[n] by
h[n] = [n] – [n – 1]

2.5.7 Sum Property

The sum of the impulses in a convolution sum of two discrete-time sequence is the product of the sums of the
impulses in the two individual sequences. Let the sum of all the impulses in the signals y(n), x(n) and h(n) be Sy, Sx,
and Sh, respectively. Then
S y = Sx Sh ...(25)
Proof: By definition,

y(n) = x ( n )  h( n )   x( k ) h( n  k )
k  

  
Sy =  y( n )    x( k ) h( n  k )
n   n   k  

Interchanging the order of the summation,

 
Sy =  x( k ) n h( n  k )  S x S h
 
k  
  
 

Sx Sk

GATE MASTERS PUBLICATION

 Linear Time Invariant System 87

2.6 Relation between LTI System Properties and the Impulse System
The impulse response completely characterizes the input-output behaviour of an LTI system. Hence, the
properties of the system, such as memory, causality, and stability, are related to the system’s impulse response.

2.6.1 LTI System With and Without Memory

The present output of a memoryless LTI system depends only on the present input. Exploiting the commutative
property of convolution, we may express the output of discrete-time LTI system as,
y(n) = h(n)  x(n)

=  h( k ) x( n  k )
k  

= ... + h(–2) x(n + 2) + h(–1) x(n + 1) + h(0) x(n) + h(1) x(n – 1) + h(2) x(n – 2) + ....
For the system to be memoryless, y(n) must depend only on x(n) and therefore cannot depend on x(n – k) for
k  0. Hence, every term in the above equation must be zero, except h(0) x(n). The condition implies that h(k) = 0 for
k  0, thus a discrete-time LTI system is memoryless if and only if
h(n) = K (n) ...(26)
where K = h(0) is a constant and the convolution sum reduces to the relation,
y(n) = K x(n)
If a discrete-time LTI system has an impulse response h(n) that is not identically zero for n  0, then the
system has memory.
From equation (4), we can deduce similar properties for continuous-time LTI systems with an without
memory. In particular, a continuous-time LTI system is memoryless if h(t) = 0 for t  0, and such a memoryless
LTI system has the form,
y ( t ) = K x(t)
for some constant K and has the impulse response,
h ( t ) = K (t) ...(27)

2.6.2 Causality for LTI Systems

The output of a causal system depends only on the present and past values of the input to the system.
Again, we write the convolution sum as,
  1    
y(n)=  h( k ) x( n  k )    h( k ) x( n  k )    h( k ) x( n  k )
k    k    k  0 
y ( n ) =[...+ h(–2) x(n + 2) + h(–1) x(n + 1)] + [h(0) x(n) + h(1) x(n – 1) + h(2) x(n – 2) +....]
We see that past and present values of the input x(n), x(n – 1), x(n – 2), .... are associated with indices
k  0 in the impulse response h(k), while future values of the input, x(n + 1), x(n + 2),... are associated with
indices k < 0. In order, then, for y(n) to depend only on past and present values of the input we require that
h(k) = 0 for k < 0. Hence, for a discrete-time causal LTI system,
h ( n ) = 0, for n < 0 ...(28)
and the discrete-time convolution takes the new form

y(n) =  h( k ) x( n  k )
k 0

GATE MASTERS PUBLICATION

 88 Signals & Systems

The alternative equivalent form is


y(n) =  x( k ) h( n  k )
k  

Similarly, a continuous-time LTI system is causal if,

h(t) = 0, for t < 0 ...(29)
and in this case the continuous convolution is given by
 t

y(t) =  h( ) x ( t  ) d  
0
 x ( ) h( t  ) d 


2.6.3 Stability for LTI System

A system is stable if every bounded input produces a bounded output. We shall now derive conditions on
h(t) that guarantee stability of the system by bounding the convolution sum. The magnitude of the output is
given by,
y( n ) = h( n )  x( n )

 
=  h( k ) x( n  k )  
k   k  
h( k ) x ( n  k )

If we assume that the input is bounded, or x( n )  Bx   , then x(n  k )  Bx , and it follows that,

y( n )  Bx 
k  
h( k )

From the above equation, we can conclude that if the impulse response is absolutely summable, i.e. if



k  
h( k ) < 

they y(n) is bounded in magnitude, and hence, the system is stable.

In continuous time, we obtain an analogous characterization of stability in terms of the impulse response of
an LTI system. Specifically, if x ( t )  Bx for all t, then

  
y( t ) =  h (  ) x ( t  ) d    h() x (t   d   Bx  h ( ) d 
  

Therefore, the system is stable if the impulse response is absolutely integrable, i.e., if


 h( ) d  < 


2.6.4 Invertibility for LTI Systems

Consider a continuous-time LTI system with impulse response h(t). This system is invertible only if an
inverse system exists that, when connected in series with the original system, produces an output equal to the
input to the first system. Further more, if an LTI system is invertible, then it has an LTI inverse.

GATE MASTERS PUBLICATION

 Linear Time Invariant System 89

y ( t)
x (t ) h( t) hI( t) x (t )

y ( n)
x (n ) h( n) hI( n) x (n )

Fig. 2.11 : Concept of a general invertible system

The relationship between the impulse response, h(t), and that of the corresponding inverse system hI(t), is
easily derived. The impulse response of the cascade connection in Fig. 2.11 is the convolution of h(t) and hI(t). We
require the output of the cascade to equal the input, or
x(t)  [h(t)  hI(t)] = x(t)
This requirement implies that,
h(t)  hI(t) = (t) ...(32)
Similarly, the impulse response of a discrete-time LTI system, hI(n), must satisfy
h(n)  hI(n) = (n) ...(33)

Remember: Generally all LTI systems are invertible.

Property Continuous-time System Discrete-time System

Memoryless h(t ) = K (t ) h(n) = K (n )
Causality h(t ) = 0, for t < 0 h(n) = 0, for n < 0
 
Stability  h(  ) dt   
n  
h( n )  


Invertibility h(t )  hI(t ) = ( t ) h(n)  hI(n ) = (n)

Table-2.1 : Properties of the impulse response representation for LTI system

REMEMBER In discrete-time LTI system also we can have series/cascade connection and parallel connection
as that of continuous time LTI system.

h[t ]
x [n] h1[n ] h2[n ] y [n ]  x [n] = h1[n ]  h2 [n]
y [n ]

h1[n]
+ h[n ]
x [t ]  y [n ]  x [n ] = h1[n ]  h2[n ]
y [n ]
–
h2[n]

2.7 Correlation of Signal

A mathematical operation that closely resembles convolution is correlation. Just as in the case of
convolution two signals are involved in correlation as well. In contrast to convolution, however, our objective in
computing the correlation between the two signals is to measure the degree to which the two signals are similar.

GATE MASTERS PUBLICATION

 90 Signals & Systems

2.7.1 Cross-correlation Function of Energy Signals

Suppose that we have two signals x(t) and y(t) each of which has finite energy. The cross-correlation of x(t)
and y(t) is a signal Rxy(t), which is defined as
 

Rxy() =  x ( t ) y  ( t  ) dt  
 x( t  ) y ( t ) dt ...(34)
 

or, if both signals x(t) and y(t) are real, then

 
Rxy() =  x ( t ) y( t  ) dt   x( t  ) y( t ) dt ...(36)
 

The index  is the (time) shift (or lag) parameter and the subscripts xy in the cross-correlation function Rxy()
indicate the signals being correlated. The order of the subscripts, with x proceeding y, indicates the direction in which
one signal is shifted relative to the other.
If we reverse the roles of x(t) and y(t) in equation (36) and therefore reverse the order of the indices xy, we
obtain the cross-correlation function,
 

Ryx() =  y( t ) x ( t  ) dt   y( t  ) x( t ) dt ...(37)
 

By comparing equation (36) with equation (37), we conclude that,

Rxy() = Ryx(–) ...(38)

Study Note
Therefore, Rxy(t) is simply the folded version of Ryx(t), where the folding is done w.r.t. t = 0. Hence, Rxy(t) provides the same
information as Ryx(t) w.r.t. the similarly of x(t) to y(t).

The similarities between the computation of the cross-correlation of two signals and the convolution integral
of two signals is apparent. For energy signals, there is a simple mathematical relationship between correlation and
convolution integral.
Rxy() = x()  y(–) ...(39)
Proof:
Rxy() = x()  y(–)
 

=  x ( t ) y( (   t )) dt   x ( t ) y( t  ) dt
 

Remember: For real signals cross correlation is defined as Rxy() = x()  y(–) and for non real signals Rxy() = x()  y*(–).
Similarly for real signals Ryx() = y()  x(–) and for non real signals Ryx() = y()  x*(–).

2.7.2 Cross-correlation Function of Power Signals

The cross-correlation function between two power signals x(t) and y(t) is mathematically defined by,
T /2 T /2
1 1
Rxy() = Lim  x ( t ) y  ( t  ) dt  Lim  x ( t  ) y  ( t ) dt ...(40)
T  T T / 2
T  T T / 2

GATE MASTERS PUBLICATION

 Linear Time Invariant System 91

If x(t) and y(t) are both real, then

T /2 T /2
1 1
Rxy() = Lim  x ( t ) y( t  ) dt  Lim  x ( t  ) y( t ) dt ...(41)
T  T T / 2
T  T T / 2

An important special case of correlation of power signals is the correlation between two periodic signals
whose fundamental periods are such that the product of the two signals is also periodic. This will happen any
time the fundamental periods of the two periodic signals have an LCM.
For two signals whose product has a period T, the general for of the correlation function (for real power
signals) is given by,
T /2
1
R xy () = Lim  x ( t ) y( t  ) dt
T  T T / 2

T /2
1
can be replaced by R xy () =  x ( t ) y( t  ) dt ...(42)
T T / 2

2.7.3 Auto-correlation Function of Continuous-time Signals

A very important special case of the correlation function is the correlation of a function with itself. This
type of correlation function is called the autocorrelation function. If x(t) is an energy signal, its autocorrelation
is
 
R xy () =  x ( t ) x ( t  ) dt   x ( t  ) x( t ) dt ...(43)
 

Remember: for real signal auto correlation is Rxx() = x()  x(–) and for non real signal Rxx() = x()  x*(–)
If x(t) is a power signal, its autocorrelation is
T /2 T /2
1 1
R xx () = Lim   x ( t ) x ( t  ) dt  Lim   x ( t  ) x ( t ) dt ...(44)
T  T T / 2 T  T
T / 2

2.7.4 Properties of Cross-correlation and Autocorrelation Functions

Following are the properties of cross-correlation and autocorrelation functions:
1. We have already demonstrated that the cross-correlation function satisfies the property
R xy () = R yx (–)
With y(t) = x(t), this relation results in the following important property for the autocorrelation
function:
R xx () = R xx (–) ...(45)
Note : Hence, the autocorrelation function is an even function.
2. Relation to signal energy and signal power:
Case-1: If x(t) is an energy signal, then

2
R xx ( 0 ) = Ex  x ( t ) dt ...(46)


Proof: R xx () =  x ( t ) x( t  ) dt


GATE MASTERS PUBLICATION

92 Signals & Systems


R x x ( )   0 =  x( t ) x( t ) dt



2
Rxx(0) = x

( t ) dt  Ex

Case-2: If x(t) is a power signal, then

T /2
1
Rxx(0) = Px = Lim  x 2 ( t ) dt ...(47)
T  T T / 2

T /2
1
Proof: Rxx() = Lim  x ( t ) x ( t  ) dt
T  T T / 2

T /2
R x x ( )   0 = Lim 1 x ( t ) x ( t ) dt
T  T 
T / 2

T /2
1
Rxx(0) = Lim  x 2 ( t ) dt  Px
T  T T / 2

3. The cross-correlation satisfies the following condition:

R x y ( )  R x x (0) R y y (0)  Ex Ey ...(48)

The autocorrelation satisfies the following condition:

Rxx ()  Rxx(0) = Ex ...(49)

4. The autocorrelation of a periodic signal with period T is periodic with the same period:
If, x(t) = x(t ± T)
then, Rxx() = Rxx( ± T) ...(50)
Proof: Let x(t) be a periodic signal with period T. Then,
x(t) = x(t + T)
x(t – ) = x(t –  + T)
T /2
1
Now, Rxx() =  x ( t ) x ( t  ) dt
T T / 2

T /2
1
Rxx( – T) =  x ( t ) x ( t  (   T )) dt
T T / 2

T /2 T /2
1 1
=  x ( t ) x ( t    T ) dt   x ( t ) x ( t  ) dt  R x x ( )
T T / 2 T T / 2

5. If Rxy(0) = 0, then x(t) and y(t) are orthogonal.

Proof: By definition,


R xy () =  x( t ) y( t  ) dt



R x y ( ) = R x y (0)   x( t ) y( t ) dt  0
0


GATE MASTERS PUBLICATION

 Linear Time Invariant System 93

2.7.5 Cross-correlation Sequence of Discrete-time Energy Signals

Suppose that we have two signals x(n) and y(n) each of which has finite energy. The cross-correlation of x(n)
and y(n) is a signal Rxy(m), which is defined as,
 
 
Rxy(m) =  x( n ) y
n  
(n  m)   x( n  m ) y ( n )
n  
...(51)

or, if both signals x(n) and y(n) are real, then

 
Rxy(m) =  x ( n ) y( n  m )   x ( n  m ) y( n )
n   n  
...(52)

The index m is the (time) shift (or lag) parameter and the subscripts xy on the cross-correlation function
Rxy(m) indicate the signals being correlated. The order of the subscripts with x preceeding y, indicates the
direction in which one signal is shifted relative to the other.
If we reverse the roles of x(n) and y(n) equation (52) and therefore reverse the order of the indices xy,
we obtain the cross-correlation function,
 
R yx( m ) =  y( n ) x ( n  m )   y( n  m ) x ( n )
n   n  
...(53)

By comparing equation (52) with equation (53), we conclude that,

R x y ( m ) = R yx (– m) ...(54)

Study Note
Therefore, Rxy(m) is simply the folded version of Ryx(m), where the folding is done w.r.t. m = 0. Hence, Rxy(m)
provides the same information as Ryx(m) w.r.t. the similarity of x(n) to y(n).
The similarities between the computation of the cross-correlation of two signals and the convolution
sum of two sequences is apparent. For energy signals, there is a simple mathematical relationship between
correlation and convolution sum:
R x y ( m ) = x(m)  y(–m) ...(55)

Remember: For real signals the crosscorrelation is Rxy(m) = x(m)  y(–m) and for non real signals crosscorrelation
is Rxy(m) = x(m)  y*(–m). Similarly for real signals the crosscorrelation is Ryx(m) = y(m)  x(–m) and for non real
signals crosscorrelation is Ryx(m) = y(m)  x*(–m).

Proof:
R x y ( m ) = x(m)  y(–m)
 
=  x( n ) y( ( m  n ))   x( n ) y( n  m )
n   n  

2.7.6 Cross-correlation Sequence of Power Signals

The cross-correlation function between two power signals x(n) and y(n) is mathematically defined by

1 
R x y ( m ) = Lim
N  2N 1
 x( n ) y ( n  m )
n  N
...(56)


1 
= Lim
N  2N  1
 x( n  m ) y ( n )
n  N

GATE MASTERS PUBLICATION

 94 Signals & Systems

If x(n) and y(n) are both real,


1
Rxy(m) = Lim
N  2N 1
 x ( n ) y( n  m )
n  N
...(57)


1
= Lim  x ( n  m ) y( n )
N  2N  1 n  N

An important special case of correlation of power signals is the correlation between two periodic signals
whose fundamental periods are such that the product of the two signals is also periodic. This will happen any
time the fundamental periods of the two periodic signals have an LCM.
For two signals whose product has a period N, the general form of the correlation function (for real
power signals).
N
1
R x y ( m ) = Lim
N 
 x (n) y (n  m)
2N  1 nN
can be replaced by,
N
1
R xy( m ) =  x ( n ) y ( n  m)
2 N  1 n  N
...(58)

2.7.7 Autocorrelation Sequence of Discrete-Time Signals

In the special case where y(n) = x(n), we have the autocorrelation sequence of x(n). If x(n) is an
energy signal, its autocorrelation is
 
Rxx( m ) = 
n  
x( n ) x( n  m )   x ( n  m ) x( n )
n  
...(59)

If x(n) is a power signal, its autocorrelation is

N
1
R x x ( m ) = Lim
N  2N  1
 x( n ) x( n  m )
n  N
...(60)

N
1
= Lim
N  2N  1
 x( n  m ) x( n )
n  N

Remember: For real signal autocorrelation is Rxx(m) =x(m)  x(–m) and for non real signal autocorrelation will
be R xx(m) =x(m)  x*(–m)

2.7.8 Properties of Crosscorrelation and Autocorrelation Sequences

Following are the properties of crosscorrelation and autocorrelation sequences:
1. We have already demonstrated that the crosscorrelation sequences satisfies the property,
R x y ( m ) = R yx (– m)
With y(n) = x(n), this relation results in the following important property for the autocorrelation
sequence,
R x x ( m ) = R xx (– m) ...(61)
Hence, the autocorrelation sequences is an even function.

GATE MASTERS PUBLICATION

 Linear Time Invariant System 95

Proof: By definition,

Rxx(m) =  x( n ) x ( n  m )
n  

A change of variables is performed by letting n – m = k, which also yields k   as n and k  

as n  –. Therefore,
  
Rxx(m) =  x ( k  m) x ( k )   x ( k ) x ( k  m)   x(k ) x(k  ( m))
k  k  k 

= Rxx(–m)
2. Relation to signal energy and signal power.
Case-1: If x(n) is an energy signal, then

2
Rxx(0) = Ex  x
n  
( n) ...(62)

Proof: By definition,

Rxx(m) =  x( n ) x ( n  m )
n  

 
2
R x x ( m ) m  0  R x x (0) = 
n  
x( n ) x( n )  x
n  
( n )  Ex

Case-2: If x(n) is a power signal, then

N
1 2
Rxx(0) = Px = Lim
N  2N  1
x
n  N
(n) ...(63)

Proof: By definition,
N
1
Rxx(m) = Lim
N  2N  1
 x( n ) x( n  m )
n  N

N
1
R x x ( m ) m  0  R x x (0) = Lim
N  2N  1
 x ( n ) x( n )
n  N

N
1
= Lim  x 2 ( n )  Px
N  2N  1 n  N

3. The cross-correlation sequence satisfies the following condition:

Rxy ( m )  R x x (0) R y y (0)  Ex Ey ...(64)

The autocorrelation sequence satisfies the following condition:
R x x ( m )  Rxx(0) = Ex ...(65)

Example 2.12

Let h(t) be the triangular pulse shown in Fig. 2.12 (a) and let x(t) be the unit impulse train Fig. 2.12 (b)
expressed as:

x( t )   T ( t )   ( t  nT )
n  

GATE MASTERS PUBLICATION

96 Signals & Systems

Determine and sketch y(t) = h(t)  x(t) for the following values of T:
(a) T = 3, (b) T = 2, (c) T = 1.5.
h( t) T (t )

1 1

t t
–1 0 1 –2T –T 0 T 2T
(a) (b)

Fig. 2.12

Solution 2.12
y ( t ) = h(t)  (t)
   
= h( t )    ( t  nT )    h( t )  ( t  nT )
 n    n  


=  h ( t  nT )
n  

( a ) For T = 3, y(t) =  h(t  3n )

n  

which is sketched in Fig. 2.13 (a).

( b ) For T = 2, becomes

y(t) =  h(t  2n)
n  

which is sketched in Fig. 2.13 (b).

y(t)
T=3
1.0

t
–7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7
(a)
y(t)
T=2
1.0

t
–7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7
(b)

y(t)
T = 1.5

t
–7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7
(c)

GATE MASTERS PUBLICATION

 Linear Time Invariant System 97

(c) For T = 1.5, equations becomes,


y(t) =  h ( t  1.5 n )
n  

which is sketched in Fig. 2.13 (c). Note that when T < 2, the triangular pulses are no longer separated
and they overlap.

Example 2.13

The system shown in figure formed by connecting two systems in cascade. The impulse
responses of the systems are given by h 1 (t) and h 2 (t) respectively and
h 1 ( t ) = e –2t u(t);
h 2 ( t ) = 2e –t u(t)
( a ) Find the impulse response h(t) of the overall system shown in figure.
w( t )
x (t ) h1 ( t ) h2( t) y ( t)

(a)

x (t ) h( t ) y ( t)

(b)

( b ) Determine if the overall system is BIBO stable.

Solution 2.13
( a ) Let w(t) be the output of the first system,
w ( t ) = x(t)  h 1 (t)
Then we have, y ( t ) = w(t)  h 2 (t)
= [x(t)  h 1 (t)]  h 2(t)
But by the associativity property of convolution above equations can be rewritten as,
y ( t ) = x(t)  [h 1 (t)  h 2 (t)] = x(t)  h(t)
Therefore, the impulse response of the overall system is given by
h ( t ) = h 1 (t)  h 2 (t)
Thus, with the given h1(t) and h2(t), we have

h(t) =  h1 ( ) h2 ( t  ) d 


2 
= e u (  ) 2 e ( t  ) u ( t   ) d 

  t  
= 2 e t  e 
u (  ) u ( t   ) d   2 e t
  e d  u ( t )
  0 
= 2(e –t – e –2t ) u(t)
( b ) Using the above h(t), we have
 

 h(  ) d  = 2  (e  e 2  ) d 
 

   1
= 2   e d    e2  d   2 1    1  
 0 0   2

GATE MASTERS PUBLICATION

98 Signals & Systems

Thus, the system is BIBO stable.

Example 2.14

Consider a continuous-time LTI system whose step response is given by

s ( t ) = e –t u(t)
Determine and sketch the output of this system to the input x(t) shown in figure (a).
Solution 2.14
From figure (a) the input x(t) can be expressed as,
x ( t ) = u(t – 1) –u(t – 3)
Since the system is linear and time-invariant, the output y(t) is given by
y ( t ) = s(t – 1) –s(t – 3)
= e–(t – 1) u(t – 1) – e–(t – 3) u(t – 3)
which is sketched in figure (b).
y ( t)
x (t )

1
1
t
0 1 2 3
t
0 1 2 3 4 –1
(a) (b)

Example 2.15

Consider a continuous-time LTI system described by,

t T / 2
1
y ( t ) = T { x( t )}   x(  ) d 
T t T / 2

( a ) Find and sketch the impulse response h(t) of the system.

( b ) Is this system causal?
Solution 2.15
( a ) Equation can be rewritten as,
tT / 2 t T / 2
1 1
y(t) =  x( ) d    x ( ) d 
T 
T 
This equation can be expressed as,
1  T 1  T
y(t) = x( t )  u  t    x( t )  u  t  
T  2  T  2
1  T  T 
= x( t )   u  t    u  t    = x(t)  h(t)
T  2  2 
1   T   T 
Thus, we obtain h(t) = u  t  2   u t  
T     2 
1 T T
  t 
= T 2 2
 0 oth erwise
h(t) is sketched in figure.

GATE MASTERS PUBLICATION

 Linear Time Invariant System 99

(b) From figure, we see that, h(t)  0 for t < 0. Hence, the system is not causal.
h( t )

1/T

t
–T/2 0 T/2

Example 2.16

Compute y[n] = x[n]  h[n], where

( a ) x[n] =  n u[n], h[n] =  n u[n]
( b ) x[n] =  n u[n], h[n] =  –n u[–n], 0 <  < 1
Solution 2.16
( a ) We have,
 
y[n] =  x[k ] h[n  k ]   k u[k ]  n  k u[n  k ]
k   k  


=  k  n  k u[k ] u [n  k ]
k  

1 0  k  n
Since, u[k] u[n – k] = 
0 oth erwise
n n k

we have, y[n] =  k n  k  n  ,n0
k 0 k 0  

 n 1  (  / ) n  1
 u [n ]   
Thus , y[n] =  1  (  / )
 n
  ( n  1) u[n ] 

 1
 (  n  1   n  1 ) u[n ]   
or, y[n] =    
  n ( n  1) u[n ] 

 
(b) y[n] =  x k ] h[n  k ]   k u[k ] ( n k ) u [( n  k )]
k   k  


=    n 2 k u[k ] u[k  n ]
k  

For n  0, we have
1 nk
u[k] u[k – n] = 
 0 oth erwise

n 2k

n
Thus, we have y[n] =     ,n  0
k 0 1  2
For n greater than zero we have
1 nk
u[k] u[k – n] = 
 0 oth erwise

GATE MASTERS PUBLICATION

100 Signals & Systems

n 2k n

2 n
y[n] =     
k n 1   2
n
= ; n 0
1  2
n

we obtain, y[n] = ; ( All n )
1  2
which is sketched in figure.
y [n]

n
–2 –1 0 1 2 3

Example 2.17

Consider a discrete-time LTI system whose input x[n] and output y[n] are related by,

y[n] =  2 k  n x[k + 1]
k= 
Is the system causal?
Solution: 2.17
By the definition the impulse response of system is given by
n n n
h[n] =  2 k  n [k  1]   2 ( n  1) [k  1] = 2 ( n 1)  [k  1]
k   k   k  

By changing the variable k + 1 = m, we obtain

n 1
h[n] = 2 ( n 1)  [ m ]  2 ( n 1) u[n  1]
m  

From equation we have h[–1] = u[0] = 1  0. Thus, the system is not causal.

Example 2.18

Consider a discrete-time LTI system with impulse response h[n] given by,
h[n] =  n u[n]
( a ) Is this system causal?
( b ) Is this system BIBO stable?
Solution: 2.18
( a ) Since h[n] = 0 for n < 0, the system is causal.
( b ) For BIBO stability,
 
 h[k ] =   k u[n ]
k   k  


k 1
=   
1 
(   1)
k  

Therefore, the system is BIBO stable if   1 and unstable if   1.

GATE MASTERS PUBLICATION

 Linear Time Invariant System 101

Example 2.19

Find the impulse response h[n] for each of the causal LTI discrete-time systems satisfying the following
difference equations and indicate whether each system is an FIR or an IIR system.
(a) y[n] = x[n] – 2x[n – 2] + x[n – 3]
( b ) y[n] + 2y[n – 1] = x[n] + x[n – 1]
1
( c ) y[ n]  y [ n  2]  2 x[ n ]  x[ n  2]
2
Solution 2.19
( a ) By definition,
h[n] = [n] – 2[n – 2] + [n – 3]
or, h[n] = {1, 0, –2, 1}
Since, h[n] has only four terms, the system is an FIR system.
(b) h[n] = –2h[n – 1] + [n] + [n – 1]
Since, the system is causal, h[–1] = 0.
Then, h[0] = –2h[–1] + [0] + [–1]
= [0] = 1
h[1] = –2h[0] + [1] + [0]
= –2 + 1 = –1
h[2] = –2h[1] + [2] + [1]
= –2(2) = –2 2

h[n] = –2h[n – 1] + [n] + [n – 1] = (–1)n 2n – 1
Hence, h[n] = [n] + (–1)n 2n – 1 u[n – 1]
Since, h[n] has infinite terms, the system is an IIR system.
1
(c) h[n] = h[n  2]  2 [n ]  [n  2]
2
Since, the system is causal,
h[–2] = h[–1] = 0, Then
1
h[0] = h[2]  2 [0]  [2]  2 [0]  2
2
1
h[1] = h[1]  2 [1]  [1]  0
2
1 1
h[2] = h[0]  2 [2]  [0]  (2)  1  0
2 2
1
h[3] = h[1]  2 [3]  [1]  0
2

Hence, h[n] = 2[n]
Since, h[n] has only one term, the system is an FIR system.



GATE MASTERS PUBLICATION

 102 Signals & Systems


5 . Let, y(t) = e t u ( t )   ( t  3 k )
k  

Subjective Practice Problems Show that, y(t) = Ae–t for 0  t < 3,

and determine the value of A.

1 . Consider the signal, 6 . Which of the following impulse response

n 1 correspond(s) to stable LTI systems?
1 (a) h 1(t) = e –(1 – 2j)t u(t)
h[n ]    {u[n  3]  u[n  10]}
2 (b) h2(t) = e–t cos(2t) u(t)
Express A and B in terms of n so that the following
7 . Which of the following impulse response
equation holds:
correspond(s) to stable LTI systems?
 1  n  k  1 
 ; Ak B (a) h1[n ]  n cos  n  u [n ]
h[n  k ]   2  4 

 0; elsewh ere (b) h2[n] = 3n u[–n + 10]

8 . For each of the following statements, determine

1 ; 0  n  9 whether it is true or false:
2 . Let, x[n] = 
0 ; elsewh ere (a) If x[n] = 0 for n < N1 and h[n] = 0 for n < N2,
then x[n]  h[n] = 0 for n < N1 + N2.
1 ; 0  n  N
and h[n] =  (b) If y[n] = x[n]  h[n], then
0 ; elsewh ere y[n – 1] = x[n – 1]  h[n – 1].
where, N  9 is an integer. Determine the value of (c) If y(t) = x(t)  h(t), then y(–t) = x(–t)  h(–t).
N, given that, (d) If x(t) = 0 for t > T1 and h(t) = 0 for t > T2,
y[n] = x[n]  h[n] then x(t)  h(t) = 0 for t > T1 + T2.
y[4] = 5, y[14] = 0
9 . Consider the cascade of the following two systems S1
3 . A linear system S has the relationship, and S2, as depicted in figure.
 w[n]
x [n] S1 S2 y [n ]
y[n] =  x[k ] g[n  2 k ]
k  
S1 : Causal LTI,
between its input x[n] and its output y[n],
1
where g[n] = u[n] –u[n – 4]. w[n] = w[n  1]  x[n ]
2
(a) Determine y[n] when x[n] = [n – 1]. S2 : Causal LTI,
(b) Determine y[n] when x[n] = [n – 2]. y[n] = y[n – 1] + w[n]
(c) Is S LTI? The difference equation relating x[n] and y[n] is,
(d) Determine y[n] when x[n] = u[n].
1 3
y[n] = 
y[n  2]  y[n  1]  x[n ]
4 . Suppose that, 8 4
(a) Determine  and .
1 ; 0  t  1
x( t) =  (b) Determine the impulse response of the cascade
0 ; elsewh ere connection of S1 and S2.

t  1 0 . Consider the cascade interconnection of three causal

and h(t) = x   where, 0 <  1
 LTI systems, illustrated in figure (a). The impulse
(a) Determine and sketch y(t) = x(t)  h(t). response h2[n] is
(b) If dy(t)/dt contains only three discontinuities, h2[n] = u[n] – u[n – 2]
what is the value of ? and the overall impulse response is as shown in
figure (b).

GATE MASTERS PUBLICATION

 Linear Time Invariant System 103

t
x [n ] h1[n] h2[n ] h2[n] y [n ] ( t  )
y(t) = e x (   2) d 
(a) 

11
What is the impulse response h(t) for this system?
10 (b) Determine the response of the system when the
8 input x(t) is as shown in figure.
5
4 x ( t)
1 1
n
–1 0 1 2 3 4 5 6 7 1
(b)
t
(a) Find the impulse response h1[n]. –1 2
(b) Find the response of the overall system to the
input, 1 5 . Suppose that the signal,
x[n] = [n] – [n – 1] x( t) = u(t + 0.5) – u(t – 0.5)
is convolved with the signal,
1 1 . The following are the impulse responses of discrete-
time LTI systems. Determine whether each system h(t) = e j0 t
is causal and/or stable. (a) Determine a value of 0 which ensures that,
n
y(0) = 0
1
(a) h [n ]    u[n ] where, y(t) = x(t)  h(t)
5
(b) Is your answer to the previous part unique?
(b) h[n] = (0.8)n u[n + 2]
n
1 6 . Determine whether each of the following statements
1 concerning LTI systems is true or false. Justify your
(c) h [n ]    u [ n ]
2 answers.
(d) h[n] = (5)n u[3 – n] (a) If h(t) is the impulse response of an LTI system
 1
n and h(t) is periodic and non-zero, the system is
(e) h [n ]     u [n ]  (1.01) n u [n  1] unstable.
 2
(b) The inverse of a causal LTI system is always
n
 1 causal.
(f) h [n ]     u [n ]  (1.01) n u [1  n ]
 2
(c) If h[n ]  K for each n, where K is a given
n
1
(g) h [n ]  n   u[n  1] number, then the LTI system with h[n] as its
3 impulse response is stable.
(d) If a discrete-time LTI system has an impulse
1 3 . The following are the impulse response of continuous
response h[n] of finite duration, the system is
time LTI systems. Determine whether each system
stable.
is causal and/or stable.
(e) If an LTI system is causal, it is stable.
(a) h(t) = e–4t u(t – 2)
(b) h(t) = e–6t u(3 – t) (f) The cascade of a non-causal LTI system with a
(c) h(t) = e–2t u(t + 50) causal one is necessarily non-causal.
(d) h(t) = e2t u(–1 – t) (g) A continuous-time LTI system is stable if and
6 t
only if its step response s(t) is absolutely
(e) h( t )  e integrable – that is, if and only if
(f) h(t) = t e –t u(t) 
(g) h(t) = (2e–t – e(t – 100)/100) u(t)  s( t ) dt  

1 4 . (a) Consider an LTI system with input and output
(h) A discrete-time LTI system is causal if and only
related through the equation,
if its step response s[n] is zero for n < 0.

GATE MASTERS PUBLICATION

 104 Signals & Systems

1 7 . Consider the cascade of two systems shown in figure. Show that the commutativity property does not
The first system, A is known to be LTI. The second hold for these two systems by computing the
system, B is known to be the inverse of system A. Let impulse response of the cascade combination in
y1(t) denote the response of system A to x1(t), and figure (a) and (b), respectively.
let y2(t) denote the response of system A to x2(t).
System System
x [n] y [n ]
LTI y ( t) System A B
x ( t) x (t )
system A B
(a)
(a) What is the response of system B to the input
ay1(t) + by2(t), where a and b are constants? System System
x [n] y [n ]
B A
(b) What is the response of system B to the input
y1(t – )? (b)

1 8 . In the text, we saw that the overall input-output (b) Suppose that we replace system B in each of the
relationship of the cascade of two LTI systems does interconnected systems of figure by the system
not depend on the order in which they are cascaded. with the following relationship between its input
This fact, known as the commutativity property, w[n] and output z[n];
depends on both the linearity and the time z[n] = w[n] + 2
invariance of both systems. In this problem, we Repeat the calculations of part (a) in this case.
illustrate the point.
(a) Consider two discrete-time systems A and B, 
where system A is an LTI system with unit sample
response h[n] = (1/2)n u[n]. System B, on the
other hand, is linear but time varying.
Specifically, if the input to system B is w[n], its
output is
z[n] = n w[n]

GATE MASTERS PUBLICATION