Signals and Systems

Module-IV
Sampling of a Continuous-time Signal &
Reconstruction of the Signal from its
Samples
By
Dr Pranati Das
Prof., Electrical Engg. Department
IGIT Sarang

1
 Topics
1. Representation of a continuous-time signal by
its samples: The Sampling Theorem
(Section 7.0, 7.1.1 and 7.1.2 of Ch. 7 of Text
book)
2. Reconstruction of a Signal from its Samples
using interpolation(Section 7.2 of Ch. 7 of Text
book)
3. Tutorial examples

2
Properties of continuous-time unit impulse function
• Sampling property of unit impulse:
Multiplying x (t ) by a unit impulse, samples the value
of the signal at the point at which the impulse is
located, i.e. x (t ) (t  t )  xt  (t  t )
0 0 0
• Convolution with an impulse simply shifts a signal.
o In time domain
x(t )   (t  t 0 )  xt  t 0 .
o In frequency domain
X ( j )   (   0 )  X ( j (   0 )) .
• Convolution with an impulse train produces periodic
replicas of a signal at each impulse location.

3
Convolution of a Spectrum with a unit
impulse in frequency domain

Convolution with an impulse in frequency domain

simply shifts the spectrum 4
 Convolution of a Signal with a unit Impulse
Signal

Convolution of a Signal with a unit Impulse Signal in time domain

simply shifts the signal 5
 Sampling of a continuous-time signal
Sampling of a continuous-time signal x(t ) at regular
interval Ts can be done by multiplying x(t )with
a periodic unit impulse train p (t ) of period Ts to
obtain a discrete-time sequence xnTs   x p (nTs )
where x p (t )  xt  p t  ...............(1)

The sampling function, pt     t  nT s , (2)
n  
The sampling period: Ts ,
  2 .
The sampling frequency: s Ts
Using the property of impulse function x p (t ) is an
impulse train with amplitude of 
impulses equal tox(nTs )
i.e. x p t    xnTs  t  nTs  (3)
6
n  
 Sampling of a CT signal cont....
From the multiplication property of Fourier transform,
1 
X p ( j )  X ( j )  P( j )   X ( j ) P j    d ........(4)
1
2 2 
2 

and P j   T     k s  .
s k  
Since convolution with an impulse simply shifts the
signal [i.e. , X ( j )   (   0 )  X ( j (   0 )) ], it
follows that 1 
X P  j    X  j (  k s ) .........(5)
Ts k  
That is X P  j  is a periodic function of  consisting of
shifted copies of X  j  scaled by 1T .
s
7
Figure 7.2 of the Text book-I
Impulse-train sampling.

8
 Figure 7.3 of the Text Book-I
Effect in the frequency domain of sampling in the time domain:
(a) spectrum of original signal;(b) spectrum of sampling function;

9
Figure 7.3 of the Text Book-I Continued
(c) spectrum of sampled signal with  s  2  M ;
(d) spectrum of sampled signal with  s  2  M .

10
 Effect of sampling in the time domain on
spectrum in the frequency domain
• Spectrum of the sampled signal X P  j  is a periodic
function of  consisting of superposition of shifted
copies of X  j  scaled by 1 as shown in figure.
Ts
• In Figure (c),  M  ( s   M ) or equivalently,
 s  2  M , and thus the shifted and scaled copies of
the spectrum X  j  in X P  j  do not overlap.( It is
the case of oversampling.) Consequently the signal x(t )
can be reconstructed by passing the sample train
through an ideal low pass filter with gain Ts and cut-
off frequency  c greater than  M and less than  s   M .
• In Figure (d), with  M  ( s   M ) there is overlap. It is
the case of undersampling and the signal reconstruction is
11
not possible in this case because of aliasing.
 Sampling Theorem
Let x(t ) be a band limited signal having maximum
frequency  M with Fourier transform
X ( j )  0 for    M .
Then x(t ) is completely determined by its samples
x(nTs ), n  0,  1,  2,  3,. and x(t ) can be
2
reconstructed for  s  2  M where  s  . The
Ts
signal x(t ) can be reconstructed by passing this sample
train through an ideal low pass filter with gain Ts and
cut-off frequency  c greater than  M and less than
 s   M . The frequency 2  M which must be exceeded by
the sampling frequency  s is called the Nyquist rate.
12
 Sampling Theorem
A continuous time band-limited signal x(t ) having
maximum frequency  M can be represented by its
samples and can be reconstructed back using a low
pass filter when the sampling frequency  s is greater
than or equal to twice the maximum frequency
present in the signal x(t ).
That is  s  2  M
where  s  2 , and Ts is the sampling interval.
Ts
13
Interpolation

14
 Interpolation
The process of reconstructing a continuous time
signal x(t) either approximately or exactly from its samples is
known as interpolation. In frequency domain, the
interpolation can be considered as a filtering process.
• Zero-order hold -The interpolating function is the rectangular
function as the impulse response of the LTI system. It
represents an approximation to the ideal low pass filter
required for exact interpolation
• Band-limited interpolation -The interpolating function for
recovery of a band-limited signal is the Sinc function as
impulse response of an ideal lowpass filter. Exact
reconstruction is possible if sampling frequency satisfies
Nyquist criterion .
• 1st-order hold -The adjacent sample points are connected by
straight lines. The interpolating function is the triangular
function as the impulse response of the LTI system. The
associated transfer function of the filter is
1  sin  Ts 2 
2

H  j   
Ts   / 2  15
16
 Figure 7. 13 of the text book-I
Linear interpolation (first-order hold) as impulse-train sampling followed
by convolution with a triangular impulse response:

Figure (a) system

for sampling and
reconstruction;
(b) impulse train of samples;
(c) impulse response
representing a first order hold;

17
18
Figure 7.13 Continued (d) firstorder
hold applied to the sampled signal;
(e) comparison of transfer function
of ideal interpolating filter and firstorder
hold.

19
 The Effect of Undersampling: Aliasing
• For reconstruction of a signal from its samples,
Nyquist criterion must be satisfied. If we sample at
below the Nyquist rate i.e.  s  2  M , then
reconstruction is impossible - this problem is
known as aliasing.
• Aliasing occurs when there is an overlap in the
shifted, perioidic copies of our original signal's FT,
i.e. spectrum.

20
 Figure 7. 15 of the text book-I
Effect in the frequency domain of oversampling and undersampling

Figure: (a) spectrum of original

sinusoidal signal;
(b), (c) spectrum of sampled signal with
  2
s o ;
(d),(e) spectrum of sampled signal with
  2 ;
s o
As we increase o in
moving from (b) through (d), the
impulses drawn with solid lines move to
the right, while the impulses drawn with
dashed lines move to the left.
In (d) and (e), these impulses have
moved sufficiently that there is a
change in the ones falling within the
pass band of the ideal lowpass filter.

21
 Zero-order hold
• Zero-order hold system samples x(t ) at a given instant
and holds that value until the next sampling instant.
• The output of the Zero-order hold can be generated by
impulse train sampling followed by an LTI system with
rectangular impulse response h0 (t ) . That is  T 
2 sin   
s
1 0  t  Ts T
 j s
 2
h0 (t )   H o  j   e 2
0 otherwise, 
• Reconstruction of x (t ) from the output of a Zero-order
hold can be carried out by low pass filtering. The
cascade combination of H 0 ( j ) and H R ( j ) is the
ideal low pass filter H ( j ), i.e. H ( j )  H 0 ( j ) H R ( j ).
22
 Figure 7.5 of the Text book-I
Sampling utilizing a zero-order hold.

23
 Figure 7.6 of the text book-I
Zero-order hold as impulse-train sampling followed by
an LTI system with a rectangular impulse response.

24
 Figure 7.7 of the text book-I
Cascade of the representation of a zero-order hold
(Figure 7.6) with a reconstruction filter.

25
 Zero-order hold (cont...)
H  j 
Then H r  j   ,
H 0  j 
e jTs / 2 H  j 
or H r  j  
2 sin Ts / 2
.

•So the reconstruction filter for Zero-order hold has no
constant gain in pass band like ideal low pass filter.
In practice H r ( j ) can not be exactly realized and
the output of the zero order hold x 0 (t ) is considered a
coarse approximation to the original signal x(t ) without
additional low pass filtering.
•The interpolating function for Zero-order hold is the
rectangular function h0 (t ) as the impulse response of the
LTI system. So H 0 ( j ) represents an approximation to
the ideal lowpass filter required for exact interpolation.26
Reconstruction of a Signal
from its Samples

27
 Reconstruction of a signal from its samples by
interpolation
The signal x (t ) can be exactly reconstructed from its
samples by using an ideal low pass filter of gain Ts if the
signal x (t ) is band-limited and the sampling frequency
satisfies Nyquist criterion (i.e.  s  2  M ). The transfer
function of the ideal low pass filter is H  j   TS ,    c

0,   c ,
where the cut-off frequency  c is greater than  M and
less than  s   M .
The interpolating function is the impulse response of the
ideal low pass filter (i.e. Sinc function)
 c TS sin  c t 
ht    (3)
 c t 28

It is referred to as band-limited interpolation.

 Cont....
The reconstructed signal at the output of the lowpass
filter is x r (t )  x p t   ht  ...............(1)
The sampled signal is x p t    xnTs  t  nTs ,  (3)
 n  
So x r t   (  xnTs  t  nTs  ) * ht  (3)
n  


Or x r t    xnTs ht  nTs  (2)

n  
Equation (2) describes how exact interpolation can be
carried out between the sample points.

 cTs sin  c t  nTs 
x r t    xnTs   (4)
n     c t  nTs 
It is interpolation by superposition of shifted Sinc
functions each weighted by corresponding sample value.
29
 Sampling and Reconstruction by using
Ideal Lowpass Filter

pt     t  nT  s
n  

H  j 
Ts

x(t ) x p t  x r (t )
Sampler  c 0 c
 Ideal Lowpass Filter
x p t  
 xnT  t  nT  s s
n  
1 
X P  j    X  j (  k s )
Ts k  

 cTs sin  c t  nTs 
x r t    xnTs 
n     c t  nTs  30
 Representative Spectrums
X  j 
a) Spectrum for x(t ) 1

b) Spectrum for x p t   M 0
X p  j 
M 
1
Ts
s  2 M
Guard
band
 s s  M M s s
 0 
2
H  j  2 s  M
c) Ideal Lowpass Filter to
recover X r  j  from X p  j 
Ts  M   c  ( s   M )
 c 0 c 
d) Spectrum for x r t  X r  j 
1

 M 0 M 
31
 Figure 7.4 of the text book-1
Exact recovery of a continuous-time signal from its
samples using an ideal low pass filter:

Figure (a) system for sampling

and reconstruction;
(b) representative spectrum
for x (t );
(c) corresponding spectrum

for x p t ;
(d) ideal low pass filter to
recover X  j  from X p  j  ;
(e) spectrum of x r (t ).

32
 Example problem
In the figure, we have a sampler, followed by an
ideal low pass filter, for reconstruction of x(t ) from
its samples x p t  . From the sampling theorem, we
know that if  2 is greater than twice the
s
highest frequencyTs  M present in x(t ) and  c  ,
s

2
then the reconstructed signal x r (t ) will exactly equal
.x(t ). If this condition on the bandwidth of x(t ) is
violated, then x r (t ) will not equal x(t ) . We seek to
s
show in this problem that if  c  , then for any
2
choice of Ts , x r (t ) and x(t ) will always be equal
at the sampling instants; That is
x r (t )  x(nT ), n  0,  1,  2,  3,. 33
Example problem on Reconstruction cont..
We know that reconstructed signal at the output
of the ideal lowpass filter is

 c Ts sin  c t  nTs 
x r t    xnTs 
n     c t  nTs 

With  c  s , this becomes
2 
sin t  nTs 

Ts 
1 
x r t    xnT    xnTs Sinc t  nTs 
s
  Ts
n  
t  nTs  n   
Ts

So x r (t )  x(nT ) for any integer value of n.

34
Reconstruction of Band-limited Signal from
its Samples using Ideal Lowpass Filter

1 
s  t  nTs 
 Ts 

x p (t ) x r (t ) s
c 
2

35
 Cont...
• Band-limited Interpolation is based on the
impulse response of the ideal low pass filter
(i.e. Sinc function) as the interpolating function.
• Interpolation using Sinc function consists of
superposition of shifted Sinc functions each
weighted by corresponding sample value. The
Sinc functions attain zero value at integer
s
multiple of sample period Ts for  c  .
2
• The interpolated value at any time in between
the sample points is the sum of contributions
from infinitely many weighted Sinc functions. 36
Additional slides for reference

37
 Duality (Example 4.4 and 4.5)
Consider the Fourier transform pair in Example 4.4
t 1 t  T  2 sin T1 
x(t )  rect (
2T
) 1
 X  j   
1 0 otherwise. 
Consider the Fourier transform pair in Example 4.5

sin Wt       1,   W

xt    X  j   rect   0,   W .
t  2W  
Because of the symmetry between Fourier transform and
inverse Fourier transform relations, for any Fourier transform
pair, there exists a dual pair with time and frequency variables
interchanged.

38
 Fourier Transform pair (Example 4.4 )
x(t )  rect (
t
) 
1 t  T 
  2 sin  T 

1
  X j  1
12T 0 otherwise.

Increase in T1
broadens x(t )
but makes X  j 
Narrower.

39
 Fourier Transform pair (Example 4.5 )

  
sin Wt  X  j   rect    1,   W
xt    W  0,   W .
t
As W increases
X  j  broadens,
while the main
peak of x(t ) at
t  0 becomes
higher and the
width of 1st lobe of
x(t ) becomes
narrower. x(t )
converges to an
impulse as W   40
 Duality of CTFT(Example 4.4 and 4.5)
t  2 sin T1 
x(t )  rect ( 
) 1 t  T1  X  j  
2T1 0 otherwise. 
sin Wt      1,   W
xt    X  j   rect    0,   W .
t  2W  


41
Multiplication Property and its effect in
the frequency domain(Example 4.21)
s (t ). p (t )  1 S ( j ) * P ( j )


2
•Multiplication of a signals (t )
with a periodic signal
results in translation of the
spectrum S ( j )
•Convolution of a spectrum
with an impulse in
frequency domain simply
shifts the spectrum.

42
 References
1. Signals and Systems by O&W- Chapter 7
Ch 7.0, 7.1.1 and 7.1.2 and 7.2 of Ch. 7
of Text book)

43
Thank you.

44