Topic 3

LTI System Analysis

Introduction
Linear Time Invariant System
An LTI system is linear and time invariant for both the input and
output. The system obeys superposition principle. For such system,
N

The input is x(t ) a1 x1 (t )  a 2 x 2 (t )  a3 x3 (t )  ........... ......  a N x N (t )  ai xi (t )

i 1
The corresponding output is
N
y (t ) a1 y1 (t )  a 2 y 2 (t )  a3 y 3 (t )  ........... ......  a N y N (t )  ai y i (t )
i 1

From the properties of impulse function, we know
that any signal can be represented as the weighted x(t )  x( )(t   )d

summation of shifted impulse function
Since the
T  (t   )  h(t   )
Let T  (t )  h(t )  Impulse Response system is TI
  
We have y (t ) T  x(t )  T  x( )(t   )d   x( ) T  (t    d
  

y (t )  x( ) h(t   )d x(t ) * h(t)  Convolution Integral / Convolution

Convolution

y (t )  x( ) h(t   )d


The operations involved in convolution are Folding,

Shifting, Multiplication and Integration.

Condition for Convolution

 Both x(t) and h(t) must be absolutely integrable
over  , 0
 Both x(t) and h(t) must be absolutely integrable
over 0, 
 Either x(t) or h(t )or both must be absolutely
integrable over  , 
 
y (t )  x( ) h(t   )d
Convolution Properties 

Commutative: x(t )  h(t ) h(t )  x(t )

Associative: series systems
x(t )  h1 (t )  h2 (t )  x(t )  h1 (t )  h2 (t )  x(t )  h1 (t )  h2 (t )
Distributive: parallel systems
x(t )  h1 (t )  h2 (t )  x(t )  h1 (t )  x(t )  h2 (t )
t
x(t )  (t )  x(t ) x(t )  u (t )  x( )d x(t)  h(t)  y(t)=h(t)  x(t)  y(t)

 
y (t )  x( ) h(t   )d
Convolution Example 

Let x(t ) e  bt u (t ) h(t ) e  at u (t ) a b

 
y (t )  x( ) h(t   )d  e  b u ( ) e  a (t   ) u (t   )d
 

1 0   t
Now u ( )u (t   ) 
0 otherwise
t t
y (t ) e b e  a (t   ) d e  at e ( a  b ) d
0 0

e  at ( a  b ) t e  at ( a  b )t
So, 
a b
 e  
0 a b
 e  1

1
  e  bt  e  at 
a b
Analysis of DT LTI System
N M
ak, bk  system coefficients y (n)   a k y (n  k )   bk x(n  k )
M, N  system order k 1 k 0

Also termed as autoregressive moving average (ARMA) system. It

has two subclasses as:

If bk = 0 for all k, the system is autoregressive (AR)

If ak = 0 for all k, the system is moving average (MA)

Two methods:
1. Direct solution of I/O relation:
y(n) = F[y(n-1), y(n-2), . . ., y(n-N), x(n), x(n-1), . . ., x(n-M)]
2. Impulse resolution method

We will use second method first

Impulse Resolution Method
Steps:
o Resolve or decompose the input signal into a sum of
elementary signals. The elementary signals are selected in such a
way that the response of the system to each elementary signal is
easily determined
o Find the response of the system to the elementary inputs
o Judiciously sum all the responses to find the total response
x ( n )  c k x k ( n ) yk (n) T [ xk (n)]
y ( n )  c k y k ( n )
k

k
Mostly used method of resolving input into impulses

x(n)   x(k ) (n  k )
k 
Impulse Resolution Method
Example: Resolve into impulses x(n) {2 4 0 3}
2 

x(n)   x(k ) (n  k ) 2 (n  1)  4 (n)  3 (n  2)

k  1

Since x(n) can be resolved into weighted sum of impulses, we can

find the response of the impulse function as
T [ (n)] h(n)  T [ (n  k )] h(n  k )
Using superposition summation
  
y (n) T [ x(n)] T  x(k ) (n  k )   x(k )T [ (n  k )]   x(k )h(n  k )
k  k  k 

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

Convolution: y(n) = x(n) * h(n)=h(n)*x(n)

Convolution

Operations y ( n)   x ( k ) h( n  k )
Folding: Find h(-k) from h(k) k 
Shifting: Find h(n-k) from h(-k)
right shift for +ve n and left shift for –ve n
Multiplication
Summation

Properties
Commutative  x(n) * h(n) = h(n) * x(n)
Associative  [x(n) * h1(n)]* h2(n) = x(n) * [h1(n) * h2(n)]
Distributive  x(n) * [h1(n)+ h2(n)]
= x(n) * h1(n)+ x(n) * h2(n)
 

Convolution: Problem y ( n)   x ( k ) h( n  k )
k 

Find the convolution between

h(n) = {1 2 1 -1} and x(n) ={1 2 3 1} (Ex. 2.3.2 P 77)
 

Answer: y(n) = { . . 1 4 8 8 3 -2 -1 . . }


Solution: Given h(k) = {1 2 1 -1} and x(k) ={1 2 3 1}

 
h(-k)= {-1 1 2 1} h(1-k) = {-1 1 2 1} and so on
 


n=0 y ( 0)   x ( k ) h (  k ) = 12+2 1 = 4
k 

n = 1 y (1)   x(k ) h(1  k ) = 1 1+2 2+3 1 = 8
k 

and find y(2), y(3), y(4), y(5) and y(-1)

Why? Total will be M+N-1, arrow depends on original signals

Recursive Solution
N
For a practical causal sequence 
y ( n)  h( k ) x ( n  k )
If N is finite the system is FIR (finite impulse response) k 0

If N is infinity, the system is IIR (infinite impulse response)

How can we solve IIR system? Consider cumulative average

1 n
y ( n)  
n  1 k 0
x(k ) n 0,1, 2, .......
n 1
y ( n)  y (n  1)  x ( n)
n 1 n 1
General equation of recursive system is y (n) a y (n  1)  x(n)
Putting n = 0, 1, 2, …., n
n
y ( n) a n 1
y ( 1)   a k x(n  k )
k 0
Recursive Solution
n
y (n) a n 1 y ( 1)   a k x(n  k )
k 0

y(-1) is called initial condition. If the system n

is initially relaxed at n = 0, then y(-1) = 0 y zs ( n )  

k 0
a k
x(n  k )
and hence the response will depend on x(n)
only. This output is termed as zero-state
This is actually the
response or forced response, for all n ≥ 0
convolution sum of a causal
sequence with h(n) = an u(n)
If the system is initially non relaxed and
x(n) = 0 for all n, the output is called zero- y zi (n) a n 1 y ( 1)
input response or natural response

The complete response is the summation of two

Example
Find the homogeneous solution of the difference equation
y(n) – 3y(n-1) – 4y(n-2) = x(n) +x(n-1)

For homogeneous solution, all the terms containing x(n) will be 0 and
hence let
y(n) = n

1 n 16
yzi (n)  [ y ( 1)  4 y ( 2)]( 1)  [ y ( 1)  y ( 2)](4) n n 0
5 5
LTI System Stability

Let x(t )  B for all t. We have,

  
y (t )  h( )x(t   )d  h( ) x(t   ) d  B h( ) d
  

The output will be bounded provided that


S h  h(t ) dt  


So, an LTI system will be BIBO stable if its impulse

response is absolutely integrable.
LTI System Example
 h(t ) e  2t u (t )  e3t u ( t )  (t  1)
The system is anti-causal since h(t ) 0 for t<0
It is dynamic since h(t ) k(t )
  0 
11
S h  h(t ) dt e  2t
dt  e
3t
dt  (t  1)dt  6 Stable
 0  

 h(t ) 3e 2t u (t )
The system is causal, dynamic and unstable

S h  3e 2t dt 
0

 h(t ) 5(t  5)
The system is anti-causal, dynamic and stable

S h  5(t  5)dt 5


5t
 h(t ) 10 sin
t
The system is non-causal, dynamic and stable

5t
S h  10 sin dt 20

t
Stability of LTI System

BIBO Stability: y(n) is bounded for every bounded
x(n). If x(n) is bounded, there exists a constant M x
k 

y ( n)  h( k ) x ( n  k )
such that |x(n)| ≤ Mx <  for all n. Similarly, |y(n)| ≤
My <  for all n. 

Taking the absolute values y ( n)   h( k ) x ( n  k )

k 
Absolute value of the summation of terms is 
always less than or equal to the summation y ( n)   h( k ) x ( n  k )
of the absolute values of terms k 

For finite x(n), |x(n)| ≤ Mx. So, y(n) is bounded 

S h   h( k )  
provided that the summation of the absolute k 
values of the impulse response is finite

LTI system is stable if its impulse response is absolutely summable

Stability Example
Example: Determine the range of values of a and b for which an LTI
system with impulse response h(n) = an u(n) + bn u(-n-1) is stable

Answer: Stable if both |a| < 1 and |b| > 1

Realization of LTI System
y (n)  a1 y (n  1)  b0 x(n)  b1 x(n  1)

N M
y (n)   a k y (n  k )   bk x(n  k )
k 1 k 0

Steps of converting
direct form I (a) to
direct form II
realization (c)
Realization of LTI System
N M
y (n)   a k y (n  k )   bk x(n  k )
k 1 k 0
M
v(n)  bk x(n  k )
k 0
N
y ( n)   a k y ( n  k )  v ( n)
k 1

Delay units: M+N

Adder: M+N

Direct form I of generalized

LTI system
Realization of LTI System

Delay units: Max{M, N}

Adder: M+N

Direct form II (also

known as canonic form
of generalized LTI
system
Realization of FIR System
M
y (n)  bk x(n  k )
k 0
1 M
y ( n)  
M  1 k 0
x(n  k )

1 M 1
y ( n)  
M  1 k 0
x ( n  1  k ) 
M 1
[ x(n)  x(n  1  M )] Non recursive realization of
FIR moving average system
1
 y (n  1)  [ x(n)  x(n  1  M )]
M 1

Recursive realization of FIR

moving average system
Correlation
Measurement of the degree of similarity between two signals, cross-
correlation and auto-correlation
 
rxy (l )   x(n) y (n  l )  x(n  l ) y (n) l 0, 1, 2, ...
n  n 
 
ryx (l )   x(n  l ) y (n)  x(n) y (n  l )
n  n 
   

rxx (l )   x(n) x(n  l )  x(n  l ) x(n) rxx (0)   x(n) x(n)  x(n) 2  E x
n  n  n  n 

N k1 N k1

rxy (l )   x ( n) y ( n  l ) rxx (l )   x ( n) x ( n  l )
n i
n i

i = l, k = 0 for l ≥ 0, and i = 0, k = l for l < 0

Correlation Properties
Symmetry  rxy(l) = ryx(-l) So, rxx(l) = rxx(-l)  even function
Maximum correlation: Let x(n) and y(n) have finite energy. If the time shift is l,
we have a composite signal as, s = ax(n) + by(n-l). The energy of the signal is,
   
E s   [ax(n)  by (n  l )] a
2 2
 [ x(n)] 2
b 2
 [ y(n  l )] 2
 2ab  x(n) y (n  l )
n  n  n  n 

a 2 rxx (0)  b 2 ryy (0)  2abrxy (l )

Since Es ≥ 0 a 2 rxx (0)  b 2 ryy (0)  2abrxy (l ) 0
2
This is a quadratic equation of the  a a
rxx (0)    2rxy (l )  ryy (0) 0
function a/b. Now, if a quadratic  b b
equation has nonnegative value, its
discriminant must non-positive. So, [rxy (l )] 2 rxx (0)ryy (0)  rxy (l )  E x E y
Autocorrelation has maximum value at 0 lag. To scale-
down the correlation function, normally is normalized rxx (l ) rxx (0)
between -1 to 1. The normalized correlations rare, rxy (l )
xx (l )
 xx (l )   xy (l ) 
rxx (0) rxx (0)ryy (0)
Correlation of Periodic Sequences
N1 N1
1 1
rxy (l ) 
N
 x ( n) y ( n  l )
n 0
rxx (l ) 
N
 x ( n) x ( n  l )
n 0

Application: Let y(n) = x(n) + w(n). Suppose we observe M samples

of y(n), 0 ≤ n ≤ M-1, where, M>>N. We can assume for practical
purpose, y(n) = 0 for n<0 and n ≥ M. Using normalized factor 1/M,
M1 M1
1 1
ryy (l ) 
M

n 0
y ( n) y ( n  l ) 
M
 [ x(n)  w(n)][ x(n  l )  w(n  l )]
n 0
M1 M1 M1 M 1
1 1 1 1

M

n 0
x ( n) x ( n  l ) 
M

n 0
x(n) w(n  l ) 
M

n 0
w(n) x(n  l ) 
M
 w(n)w(n  l )
n 0

rxx (l )  rxw (l )  rwx (l )  rww (l )

Detection of periodic signal buried in noise

Input-Output Correlation
For an LTI system, we can calculate the output using the convolution
of input and impulse response
ryx(l) = y(l) * x(-l) = h(l) * [x(-l) * x(-l)] = h(l) * r xx(l)
So, ryx(l) is the output of LTI system when input is rxx(l)

Similarly, rxy(l) = h(-l) * rxx(l)

Using autocorrelation,
ryy(l) = y(l) * y(-l) = [h(l) * x(l)] * [h(-l) * x(-l)]
= [h(l) * h(-l)] * [x(l) * x(l)] = rhh(l) * rxx(l)
rhh(l) exists if the system is stable
Thank you