Sampling

• In DSP, sampling is the reduction of a continuous signal to

a discrete signal.
• A common example is the conversion of a sound wave (a
continuous signal) to a sequence of samples (a discrete-time
signal).
• A sample refers to a value or set of values at a point in time
and/or space.
• A sampler is a subsystem or operation that extracts samples
from a continuous signal.
• A theoretical ideal sampler produces samples equivalent to
the instantaneous value of the continuous signal at the
desired points.
Periodic Sampling
 Periodic Sampling
Two
 Motivation
• Signals are time-continuous in nature;
• DSP: processing of digital signals
• How to convert analog signals to digital
ones?
 Motivation (contd.)
• The sampling process should not yield any loss of
the information.
• In other words, the original analog signal should be
reconstructed (restored) based on the time-discrete
sequence.

x(t ) x[n]  x(nTs )

sampling

Analog Discrete-time
signal sequence
 Motivation (contd.)
• The problem is how to choose the sampling
interval Ts so that the original analog signal
can be reconstructed.

x(t ) x[n]  x(nTs )

sampling

Analog Discrete-time
signal sequence
 Sampling
• The sampler takes a snapshot of the x(t) for
every Ts

x(t ) x[n]  x(nTs )

Analog Discrete-time
signal sequence
 Sampling of a sinusoid
• Given an analog sinusoid
x(t )  A cos(t   )
• The discrete sequence after sampling is

x[n]  x(nTs )  A cos(nTs   )  A cos(ˆ n   )

• ̂ is called the normalized radian frequency

f
ˆ  Ts  2
fs
 Sampling of a sinusoid
• What is the difference between x(t) and x[n]?
• Radian frequency has the unit of rad/sec;
• Normalized frequency has the unit of rad—
dimensionless quantity;
• X(t) is a continuous time function. x[n] is just a
number sequence carrying no information about the
sampling period.

x(t )  A cos(t   )

x[n]  x(nTs )  A cos(nTs   )  A cos(ˆ n   )

Sampling of a sinusoid
 Sampling of a sinusoid
• With different sampling frequency, sampling
of an analog signal will different discrete
sequence
• Sampling of different analog signals may
yield the same discrete sequence
• Sampling frequency must be employed in
order to reconstruct the original analog signal
 Shannon Sampling Theorem

A continuous-time signal x(t) with frequencies

no higher than fmax can be reconstructed
exactly from its samples x[n]=x(nTs), if the
samples are taken at a rate fs=1/Ts that is
greater than 2fmax, which is called Nyquist
sampling rate (or frequency)
 Aliasing distortion
• Aliasing refers to an effect that causes
different signals to become indistinguishable
(or aliases of one another) when sampled.
• It also refers to the distortion or artifact that
results when the signal reconstructed from
samples is different from the original
continuous signal.
 Aliasing and Folding
• Aliasing:
• when the sampling rate is lower than the
Nyquist rate, the reconstruction is not possible
as the solution is not unique.
 Aliasing and Folding
• Consider the following signals

x(t )  A cos(2f 0t   ) and

y(t )  A cos(2 ( f 0  lf s )t   ) and

w(t )  A cos(2 ( f 0  lf s )t   )
• Sampling of above signals at the rate of fs will yield
the same discrete sequence
 Aliasing and Folding

x[n]  x(nTs )  A cos(2f 0 nTs   )

Note: the frequency is f0

 Aliasing and Folding

y[n ]  y ( nTs )  A cos(2 ( f 0  lf s )nTs   )

 A cos(2f 0 nTs  2nl   )
 A cos(2f 0 nTs   )  x[n ]

The frequency is f 0  lf s
 Aliasing and Folding

w[n ]  w(nTs )  A cos(2 (  f 0  lf s )nTs   )

 A cos( 2f 0 nTs  2nl   )
 A cos(2f 0 nTs   )  x[n ]

The frequency is  f 0  lf
 Aliasing and Folding
• Hence x[n] might be the samples of the
following signals:
– A sinusoid with the frequency of f 0
– Sinusoids with frequencies f 0  lf
– Sinusoids with frequencies  f 0  lf
• However, only one of the above is the
original signal
 Aliasing and Folding
• Aliasing frequencies

f 0  lf  f 0  lf

Where l is a positive or negative integer

 aliasing

500Hz
 Aliasing and Folding
• The way for signal reconstruction is to take
the sinusoid with the frequency less than half
of the sampling rate.
fs
2
• Wrong decision may be made if the following frequencies are within
the above range.

f 0  lf  f 0  lf
 Aliasing and Folding
• If we consider the frequency component
within the following range

1
[0, f s )
2
• All aliasing components are outside the
range, and hence correct signal
reconstruction is possible
 Aliasing and Folding
However, if f s  2 f0 ( f s / 2)  f 0
then
 f 0  lf s

may also fall within the range [0,fs/2). So the signal restored will not
be the true original one.
 Aliasing and Folding
• The way for determining the occurring of aliasing to
see if the following aliasing frequencies

 f 0  lf s

fall within the range [0,fs/2). In other words if the

following conditions are met:

 ( f 0 / f s )  l  0.5
 Spectrum View of Sampling:
Over-sampling
• Example
x(t )  A cos(2 (100)t   )
sampled at rate of 1000Hz >>100Hz

x[n]  A cos(2 (0.1)n   )

the aliasing frequencies are

 100  1000l  0.1  l

Normalized frequency
Spectrum View of Sampling:
Over-sampling (fs>>f0) aliasing

500Hz
 Spectrum View of Sampling:
Over-sampling
• In the cases of over-sampling, the aliasing
frequencies are outside the range [0,0.5]
(normalized frequency) or [0,0.5fs].
• There is no overlap between the aliasing
frequencies and f0 . Hence it is possible to
reconstruct the original signal
 Spectrum View of Sampling:
Under-sampling (fs<f0)
• We still consider a sinusoid
x(t )  A cos(2 (100)t   )
sampled at rate of 80Hz

x[n]  A cos(2 (1.25)n   )

the aliasing frequencies are

 100  80l or
 1.25  l
Spectrum View of Sampling:
Under-sampling (fs<faliasing
0)

20Hz
 Spectrum View of Sampling:
Under-sampling (fs=f0)
• We still consider the same sinusoid
x(t )  A cos(2 (100)t   )
sampled at rate of 100Hz

x[n]  A cos(2n   )
the aliasing frequencies are

 100  100l or
1 l
Spectrum View of Sampling:
Under-sampling (fs=faliasing
0)
 Spectrum View of Sampling:
Under-sampling (f0 <fs<2f0)
• We still consider a sinusoid
x(t )  A cos(2 (100)t   )
sampled at rate of 125Hz

x[n]  A cos(2 (0.8)n   )

the aliasing frequencies are

 100  125l or
 0.8  l
Spectrum View of Sampling:
Under-sampling (fs<faliasing
0)

25Hz
 Signal Reconstruction
• In DSP, reconstruction usually means the
determination of an original continuous signal
from a sequence of equally spaced samples
(discrete time signal).
 Interpolation
• In the mathematical field of numerical
analysis, interpolation is a method of
constructing new data points within the range
of a discrete set of known data points.
• In DSP, the term interpolation refers to the
process of converting a sampled digital signal
(such as a sampled audio signal) to a higher
sampling rate (Upsampling) using various
digital filtering techniques (e.g., convolution
with a frequency-limited impulse signal)
Interpolation (contd.)
 Discrete-to-Continuous Conversion
y(n) y(t)

D-to-C

D-to-C conversion is implemented based on the principle of

interpolation:


y (t )   y[n] p(t  nT )
n  
s

p(t) is the characteristic pulse shape of the converter.

 Discrete-to-Continuous Conversion

y (t )   y[n] p(t  nT )
n  
s

• choose an interpolation pulse p(t);

• shift the pulse by nTs,
n=…-3,-2,-1,0,1,2,3,…
• modify the amplitude of the shifted pulse
by y[n]
• add the modified pulses together to yield
the output
 Discrete-to-Continuous Conversion

y (t )   y[n] p(t  nT )
n  
s

• The operation is equivalent to pass the discrete sequence through

a system with the impulse response of p(t)
 Discrete-to-Continuous Conversion

• Interpolation pulses
 Discrete-to-Continuous Conversion
• Zero-Order Hold Interpolation

 1 1
1  Ts  t  Ts
p (t )   2 2
0 otherwise

1 Rectangle pulse

t
1 1
 Ts Ts
2 2
Spectrum of Product of sinusoids
 Discrete-to-Continuous Conversion
• Linear Interpolation

 t
1  Ts  t  Ts
p(t )   Ts
 0 otherwise
Triangle pulse
1

t
 Ts Ts
Spectrum of Product of sinusoids
 Discrete-to-Continuous Conversion
• Parabolic Interpolation

Four parabolic segments (the

2nd order polynomial)

Duration 4Ts

Note: p( t )  0 for t  0,Ts ,2Ts ,...

 Discrete-to-Continuous Conversion
• Parabolic Interpolation
 Discrete-to-Continuous Conversion

• None of the interpolation pulses give the

perfect reconstruction;
• The error can be reduced if small Ts is used
• Hence over-sampling helps to reconstruct the
signal
Discrete-to-Continuous Conversion
 Discrete-to-Continuous Conversion
• Ideal band-limited interpolation: the following pulse shape will result
in perfect reconstruction:
 Discrete-to-Continuous Conversion
• Ideal band-limited interpolation: the following pulse shape will result
in perfect reconstruction:


sin t
 Ts
p(t )  sin c( t )  for    t  
Ts 
t
Ts
Discrete-to-Continuous Conversion
 Discrete-to-Continuous Conversion

y (t )   y[n] p(t  nT )
n  
s

• The operation is equivalent to pass the discrete sequence through

a system with the impulse response of p(t)
• system with sinc impulse response behaves as an ideal band-limited
filter
 Discrete-to-Continuous Conversion
• 
sin t
Ts
p(t )  for    t  

t
Ts
Fourier transform

1
fs
2
 Ideal C-to-D Converter

• Mathematical Model for A-to-D

x[n]  x(nTs )
FOURIER
TRANSFORM
of xs(t) ???
 Periodic Impulse Train

 
2
p (t )   (t  nTs )   ak e jk s t
s 
Ts
n  k  
Ts / 2
1  jk s t 1 Fourier Series
ak 
Ts   ( t )e dt 
Ts
Ts / 2
 FT of Impulse Train
 
2
p(t )   (t  nTs )  P( j )    (  k s )
n  k  Ts

2
s 
Ts
 Impulse Train Sampling

 
xs (t)  x(t)   (t  nTs )   x(t) (t  nTs )
n n


xs (t)   x(nTs ) (t  nTs )
n
 Illustration of Sampling
x(t)

t

xs ( t )   x(nTs ) (t  nTs )
n  
x[n]  x(nTs )

n
 Sampling: Freq. Domain


  ak e jk s t EXPECT
FREQUENCY
k   SHIFTING !!!
 
p (t )   (t  nTs )   ak e jk s t

n  k  
 Frequency-Domain Analysis
 
xs (t)  x(t)   (t  nTs )   x(nTs ) (t  nTs )
n n
 1 jk st 1  jk st
xs (t)  x(t)  e   x(t)e
Ts
k Ts k

1 
Xs ( j )   X( j(  k s ))
Ts
k 2
s 
Ts
Frequency-Domain Representation of
Sampling
“Typical”
bandlimited signal

1 
Xs ( j )   X( j(  k s ))
Ts
k
 Aliasing Distortion

“Typical”
bandlimited signal

• If s < 2b , the copies of X(j) overlap, and we

have aliasing distortion.
 Reconstruction of x(t)


xs (t)   x(nTs ) (t  nTs )
n
1 
Xs ( j )   X( j(  k s ))
Ts
k
Xr ( j )  Hr ( j )Xs ( j )
 Reconstruction: Frequency-Domain

If  s  2b , the copies of

H r ( j ) X ( j ) do not overlap, so
X r ( j )  H r ( j ) X s ( j )
 Ideal Reconstruction Filter

Ts 

 Ts
Hr ( j )   
0 
 Ts


sin T t hr (0)  1
hr (t)  
s

Ts
t hr (nTs )  0, n  1, 2,
 Signal Reconstruction

xr (t)  hr (t)  xs (t)  hr (t)  x(nTs ) (t  nTs )
n

xr (t)   x(nTs )hr (t  nTs )
n


 sin T (t  nTs )
xr (t)   x(nTs ) 
s

n Ts
(t  nTs )
Ideal bandlimited interpolation formula
 Shannon Sampling Theorem
• “SINC” Interpolation is the ideal
– PERFECT RECONSTRUCTION
– of BANDLIMITED SIGNALS
Reconstruction in Time-Domain
 Ideal C-to-D and D-to-C

 sin T (t  nTs )
xr (t)   x[n] 
s
x[n]  x(nTs ) n Ts
(t  nTs )
Ideal Sampler Ideal bandlimited interpolator

1  Xr ( j )  Hr ( j )Xs ( j )
Xs ( j )   X( j(  k s ))
Ts
k
 Summary and Exercise
Summary
• Sampling
• Reconstruction
Next
• System Analysis
Exercise
Solve examples # 4.2, 4.3

Digital Signal Processing, Lecture 9, Spring

2013