Sunteți pe pagina 1din 9

# LTI Discrete-

## Discrete- Time Systems in LTI Discrete-

Discrete- Time Systems in
the Transform Domain the Transform Domain
An LTI discrete-time system is completely Such transform-domain representations
characterized in the time-domain by its provide additional insight into the behavior
impulse response sequence {h[n]} of such systems
Thus, the transform-domain representation It is easier to design and implement these
systems in the transform-domain for certain
of a discrete-time signal can also be equally applications
applied to the transform-domain
We consider now the use of the DTFT and
representation of an LTI discrete-time the z-transform in developing the transform-
system domain representations of an LTI system
1 2

Finite-Dimensional LTI
Finite- Finite-Dimensional LTI
Finite-
Discrete-- Time Systems
Discrete
Discrete-- Time Systems
Discrete
Applying the DTFT to the difference
In this course we shall be concerned with equation and making use of the linearity and
LTI discrete-time systems characterized by the time-invariance properties of Table 3.2
linear constant coefficient difference we arrive at the input-output relation in the
equations of the form: transform-domain as
N M N M
j k
dk y[n k ] = pk x[n k ] dke Y (e j ) = pk e j k X (e j )
k=0 k=0 k =0 k =0
j j
whereY (e ) and X ( e ) are the DTFTs of
y[n] and x[n], respectively
3 4

Finite-Dimensional LTI
Finite- Finite-Dimensional LTI
Finite-
Discrete-- Time Systems
Discrete Discrete-- Time Systems
Discrete
Applying the z-transform to both sides of
In developing the transform-domain the difference equation and making use of
representation of the difference equation, it the linearity and the time-invariance
has been tacitly assumed that X (e j ) and properties of Table 3.9 we arrive at
Y ( e j ) exist N M
k k
The previous equation can be alternately dk z Y(z ) = pk z X (z )
written as k=0 k =0
where Y(z) and X(z) denote the z-transforms
N M
d k e j k Y ( e j ) = pk e j k X ( e j ) of y[n] and x[n] with associated ROCs ,
k =0 k =0 respectively
5 6

1
Finite--Dimensional LTI
Finite The Frequency Response
Discrete- Time Systems
Discrete- Most discrete-time signals encountered in
A more convenient form of the z-domain practice can be represented as a linear
representation of the difference equation is combination of a very large, maybe infinite,
given by number of sinusoidal discrete-time signals
N M of different angular frequencies
d k z k Y ( z ) = pk z k X ( z )
k=0 k=0 Thus, knowing the response of the LTI
system to a single sinusoidal signal, we can
determine its response to more complicated
signals by making use of the superposition
7 8 property

## The Frequency Response The Frequency Response

An important property of an LTI system is Consider the LTI discrete-time system with
that for certain types of input signals, called an impulse response {h[n]} shown below
eigen functions , the output signal is the
input signal multiplied by a complex x[n] h[n] y[n]

constant
Its input-output relationship in the time-
We consider here one such eigen function domain is given by the convolution sum
as the input
y[ n] = h[k ] x[ n k ]
k =
9 10

## The Frequency Response The Frequency Response

If the input is of the form Then we can write
x[ n] = e j n , < n < y[ n] = H ( e j ) e j n
then it follows that the output is given by Thus for a complex exponential input signal
e j n , the output of an LTI discrete-time
y[ n] = h[ k ] e j (n k ) = h[k ] e j k e j n system is also a complex exponential signal
k = k =
of the same frequency multiplied by a
Let
complex constant H (e j )
H (e j ) = h[ k ] e j k Thus e j n is an eigen function of the system
k =
11 12

2
The Frequency Response The Frequency Response
H (e j ) , in general, is a complex function
j of with a period 2
The quantity H (e ) is called the frequency
response of the LTI discrete-time system It can be expressed in terms of its real and
H (e j ) provides a frequency-domain imaginary parts
description of the system H (e j ) = H re (e j ) + j Him (e j )
H (e j ) is precisely the DTFT of the impulse or, in terms of its magnitude and phase,
response {h[n]} of the system H (e j ) = H (e j ) e j ()
where
() = arg H ( e j )
13 14

## The Frequency Response The Frequency Response

The function H (e j ) is called the In some cases, the magnitude function is
magnitude response and the function () specified in decibels as
is called the phase responseof the LTI G( ) = 20 log10 H (e j ) dB
discrete-time system where G() is called the gain function
Design specifications for the LTI discrete- The negative of the gain function
time system, in many applications, are
given in terms of the magnitude response or A ( ) = G()
the phase response or both is called the attenuation or loss function
15 16

## The Frequency Response The Frequency Response

Note: Magnitude and phase functions are Likewise, for a real impulse response h[n],
real functions of , whereas the frequency H re (e j ) is even and H im( e j ) is odd
response is a complex function of Example - Consider the M-point moving
If the impulse responseh[n] is real then it average filter with an impulse response
follows from Table 3.4 that the magnitude given by
function is an even function of : h[n] = 1 / M , 0 n M 1
0, otherwise
H (e j ) = H ( e j ) Its frequency response is then given by
and the phase function is an odd function of M 1
: H (e j ) = M1 e j n
17
() = () 18
n =0

3
The Frequency Response The Frequency Response
Or, H (e j ) = 1 e j n e j n

Thus, the magnitude responseof the M-point
M
n=0 n= M moving average filter is given by
jM
( )
= 1 e j n 1 e jM = 1
M
1 e
M 1 e j H (e j ) = 1
sin(M / 2 )
n =0 M sin( / 2)

## 1 sin( M / 2) j ( M 1 ) / 2 and the phase response is given by

= e
M sin( / 2 ) ( M 1) M/2 2 k
() = + ( M )
2 k =0
19 20

## Frequency Response Frequency Response

Computation Using MATLAB
Computation Using MATLAB
The function freqz(h,w) can be used to
Example - Program 4_1 can be used to
determine the values of the frequency
generate the magnitude and gain responses
response vectorh at a set of given
of an M-point moving average filter as
frequency points w
shown below
Fromh, the real and imaginary parts can be 1
M=5
100

## computed using the functions real and 0.8

M=14 50

Phase, degrees
0
imag, and the magnitude and phase
Magnitude

0.6
-50

## functions using the functions abs and 0.4

-100
0.2 M=5
angle -150 M=14
0 -200
21 22 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
2001, S. K. Mitra

## Frequency Response Frequency Response

Computation Using MATLAB Computation Using MATLAB
The phase response of a discrete-time To this end the function unwrap can be
system when determined by a computer used, provided the computed phase is in
may exhibit jumps by an amount 2 caused radians
by the way the arctangent function is The jumps by the amount of 2 should not
computed
be confused with the jumps caused by the
The phase response can be made a zeros of the frequency response as indicated
continuous function of by unwrapping the
phase response across the jumps in the phase response of the moving average
filter
23 24

4
Note that the frequency response also
determines the steady-state response of an We can express the input x[n] as
LTI discrete-time system to a sinusoidal x[ n] = g[ n] + g * [n ]
input where
Example - Determine the steady-state g [n ] = 1 Ae je jon
2
output y[n] of a real coefficient LTI
Now the output of the system for an input
discrete-time system with a frequency
response H (e j) for an input e jon is simply
H (e jo )e jon
x[ n] = A cos(on + ), < n <
25 26

Because of linearity, the responsev[n] to an Combining the last two equations we get
input g[n] is given by y[ n] = v[n ] + v * [ n]
v [n] = 1 Ae jH (e j o )e jo n = 1 Ae j H ( e jo )e jon + 1 Ae j H (e jo )e jon
2 2 2

## Likewise, the output v*[n] to the input g*[n] = 1 A H (e j o ) {e j(o )e je j o n +e j(o )e j e j on }

2
is
= 1 A H ( e jo ) cos(o n + ( o ) + )
v * [ n] = 1 Ae j H (e jo )e jon 2
2
27 28

Response to a Causal
The expression for the steady-state response
Thus, the output y[n] has the same sinusoidal developed earlier assumes that the system is
waveform as the input with two differences: initially relaxed before the application of
(1) the amplitude is multiplied by H ( e jo ), the input x[n]
the value of the magnitude function at = o In practice, excitation x[n] to a discrete-time
(2) the output has a phase lag relative to the system is usually a right-sided sequence
input by an amount ( o ), the value of the applied at some sample index n = no
phase function at = o We develop the expression for the output
for such an input
29 30

5
Response to a Causal Response to a Causal
Exponential Sequence Exponential Sequence
Without any loss of generality, assume x[ n] = 0
n jn
for n < 0 Or, y[ n] = h[k ] e jk e [n ]
From the input-output relation k=0
y[ n] =
k = h[ k ] x[ n k ] The output forn < 0 is y[n] = 0
we observe that for an input The output for n 0 is given by
x[ n] = e jn[n ] n
y[ n] = h[k ] e jk e jn
the output is given by k =0
n jn jn
y[ n] = h[ k ] e j( n k ) [ n] = h[k ] e j k e h[k ] e jk e
31 k=0 Copyright 2001, S. K. Mitra 32 k =0 k = n+1

## Response to a Causal Response to a Causal

Exponential Sequence Exponential Sequence
Or, The second term on the RHS is called the
jn
y[ n] = H (e j ) e jn h[k ] e jk e transient response:
k = n+1
ytr [n ] = h[ k ] e jk e jn
The first term on the RHS is the same as k = n+1
that obtained when the input is applied at To determine the effect of the above term
n = 0 to an initially relaxed system and is on the total output response, we observe
j jn
ytr [n ] = h[k ] e j( k n ) h[ k ] h[ k ]
y sr[ n] = H ( e )e k =n +1 k = n+1 k =0
33 34

## Response to a Causal Response to a Causal

Exponential Sequence Exponential Sequence
For a causal, stable LTI IIR discrete-time For a causal FIR LTI discrete-time system
system, h[n] is absolutely summable with an impulse response h[n] of length
As a result, the transient response ytr [n ] is a N + 1, h[n] = 0 forn > N
bounded sequence Hence, ytr [n ] = 0 for n > N 1
Moreover, as n , Here the output reaches the steady-state
k =n +1 h[ k] 0 value ysr[ n] = H ( e j )e jn at n = N
and hence, the transient response decays to
zero as n gets very large
35 36

6
The Concept of Filtering The Concept of Filtering
One application of an LTI discrete-time The key to the filtering process is
system is to pass certain frequency
components in an input sequence without x[ n] = 21 X (e j ) e j n d

any distortion (if possible) and to block
other frequency components It expresses an arbitrary input as a linear
weighted sum of an infinite number of
Such systems are called digital filters and exponential sequences, or equivalently, as a
one of the main subjects of discussion in linear weighted sum of sinusoidal sequences
this course
37 38

## The Concept of Filtering The Concept of Filtering

Thus, by appropriately choosing the values To understand the mechanism behind the
of the magnitude function H ( e j ) of the design of frequency-selective filters,
LTI digital filter at frequencies consider a real-coefficient LTI discrete-time
corresponding to the frequencies of the system characterized by a magnitude
sinusoidal components of the input, some of function
these components can be selectively heavily c
H (e j )
1,
attenuated or filtered with respect to the 0, c <
others
39 40

## The Concept of Filtering The Concept of Filtering

We apply an input As
x[n ] = A cos 1n + B cos 2 n, 0 < 1 < c < 2 < H (e j1 ) 1, H (e j 2 ) 0
to this system the output reduces to
Because of linearity, the output of this y[ n] A H ( e j1 ) cos (1n + (1 ))
system is of the form Thus, the system acts like a lowpass filter
y[ n] = A H (e j1 ) cos (1n + ( 1) ) In the following example, we consider the
design of a very simple digital filter
+ B H (e j2 ) cos (2 n + ( 2 ))
41 42

7
The Concept of Filtering The Concept of Filtering
Example - The input consists of a sum of two The convolution sum description of this
sinusoidal sequences of angular frequencies filter is then given by
0.1 rad/sample and 0.4 rad/sample y[ n] = h x[n ] + h x[n 1] + h[ 2] x[n 2]
We need to design a highpass filter that will = x[ n] + x[ n 1] + x[n 2]
pass the high-frequency component of the
input but block the low-frequency component y[n] and x[n] are, respectively, the output
and the input sequences
For simplicity, assume the filter to be an FIR
filter of length 3 with an impulse response: Design Objective : Choose suitable values
h = h = , h = of and so that the output is a sinusoidal
43 44
sequence with a frequency 0.4 rad/sample

## The Concept of Filtering The Concept of Filtering

The magnitude and phase functions are
Now, the frequency response of the FIR H (e j ) = 2 cos +
filter is given by ( ) =
H (e j) = h[ 0] + h e j + h[ 2] e j 2 In order to block the low-frequency
= (1 + e j2 ) + e j component, the magnitude function at
e j + e j j = 0.1 should be equal to zero
= 2 e + e j
2 Likewise, to pass the high-frequency
= ( 2 cos + )e j component, the magnitude function at
= 0.4 should be equal to one
45 46

## The Concept of Filtering The Concept of Filtering

Thus, the two conditions that must be Thus the output-input relation of the FIR
satisfied are filter is given by
H (e j0. 1) = 2 cos(0. 1) + = 0 y[ n] = 6.76195 ( x[ n] + x[n 2]) + 13. 456335 x[n 1]
H (e j 0. 4 ) = 2 cos( 0. 4) + = 1 where the input is
Solving the above two equations we get x[ n] = {cos(0.1n) + cos(0. 4n)}[ n]
= 6.76195 Program 4_2 can be used to verify the
= 13.456335 filtering action of the above system
47 48

8
The Concept of Filtering The Concept of Filtering
Figure below shows the plots generated by The first seven samples of the output are
running this program shown below
4
y[n]
x 2[n]
3
x 1[n]
Amplitude

-1
0 20 40 60 80 100
49 Time index n 50

## The Concept of Filtering The Concept of Filtering

From this table, it can be seen that,
neglecting the least significant digit, Therefore, first two output samples
y[ n] = cos(0.4( n 1)) for n 2 constitute the transient part of the output
Computation of the present value of the Since the impulse response is of length 3,
output requires the knowledge of the the steady-state is reached at n = N = 2
present and two previous input samples Note also that the output is delayed version
Hence, the first two output samples, y of the high-frequency component cos(0.4n)
and y, are the result of assumed zero of the input, and the delay is one sample
input sample values at n = 1 and n = 2 period
51 52