of 66
7/30/2019 DSP Chapter4
1/66
Chapter 4. Sampling of
Continuous-Time Signals
. n ro uc on
4.1 Periodic Sampling4.2 Frequency Representation of Sampling
. -
4.4 Changing the Sampling Rate
4.5 Digital Processing of Analog Signals
DSP 4-1
7/30/2019 DSP Chapter4
2/66
4.0 Introductionues on o answer: ow o approx ma e a con nuous
(analog) linear system by a digital system?
Signals: continuous-time discrete-time
o a ons:
time-domain
frequency-domain
)(txc ][nxj
eX
Systems: continuous-time discrete-time
-
c
frequency-domain
tc
)( jH
n
jeH
DSP 4-2
7/30/2019 DSP Chapter4
3/66
4.1 Periodic Sampling-
continuous-time signal is through periodic sampling.
T is the sampling period
.-, = nnxnx c
is the sampling frequency (radians per second)
An ideal continuous-to-discrete-time C/D converte
s=
Ts /2=
DSP 4-3
7/30/2019 DSP Chapter4
4/66
Mathematical Representation of Sampling
Conversion
from impulse
)(ts
train to discrete-
time sequence
c c)(txs
train)impulseperiodic(the)()(
=
=n
nTtts
n)(modulatio)()()()()(
==ccs nTttxtstxtx
property)(sifting)()()(
= cs nTtnTxtx
DSP 4-4
=n
7/30/2019 DSP Chapter4
5/66
Periodic Sampling Examples
DSP 4-5
7/30/2019 DSP Chapter4
6/66
4.2 Frequency-Domain Representation
of Sampling: Time-Domain
continuous-time signals, obtaining
= cs tstxtx )()()(
= c nTttx )()(
==n
c
DSP 4-6
7/30/2019 DSP Chapter4
7/66
Frequency-Domain Representation
)2(where)(2)()()( kjSnTtts ss
===
Since
kn ==
)(*)(21)()()()( == jSjXjXtstxtx cscs
= kXX 2*1
=
=
kscs
kX
T
1
2
DSP 4-7
=kT
7/30/2019 DSP Chapter4
8/66
Observations of Frequency-Domain
Representation of Sampling
s equa on prov es e rea ons p e ween e ourer
transform of continuous-time signal and discrete-time signal
= scs kjXT
jX ))((1
)(
consists of eriodicall re eated and scaled co ies of the
=
XFourier transform of .
The copies of are shifted by integer multiples of the
s
)(i.e.,),( jXtx cc)( jXc
sampling frequency .
All copies of replicated spectrums are superimposed to produce
s
DSP 4-8
.
7/30/2019 DSP Chapter4
9/66
DSP 4-9
7/30/2019 DSP Chapter4
10/66
Sampling Rate and Bandwidth
-
NjX >= ,0)(There is no overlap between replicated spectrums, whenwe have the sam lin rate as followin
That means we CAN reconstruct the continuous-time signal withNs > 2
an ideal low-pass filter.
There will be aliasing distortion, or aliasing when
That means we CANNOT reconstruct the continuous-time signalfrom its samples.
Ns
DSP 4-10
7/30/2019 DSP Chapter4
11/66
How to Reconstruct a Signal?
Ideal low-pass Filtering
DSP 4-11
7/30/2019 DSP Chapter4
12/66
How to Reconstruct a Signal? (Cont'd)
sampling
-
Ideal low-passfiltering
Ideal low- ass Filterin Reconstructed Si nal
DSP 4-12
7/30/2019 DSP Chapter4
13/66
Sampling and Reconstruction Example
ven a s gna
What is the Fourier transform of the given signal?
ttxc 0cos=
se e u er equa on, we now a
tjtjeettx 00
1cos +==
According to continuous Fourier transform, we know2
Therefore, the Fourier transform of the given signal is
)(2)()( 00 == jXetx tj
++= X
DSP 4-13
7/30/2019 DSP Chapter4
14/66
Sampling and Reconstruction Example
(No Aliasing)
Original Signal
Reconstructed Signal
teetxtjtj
cos1
00 =+=
DSP 4-14
2
7/30/2019 DSP Chapter4
15/66
Sampling and Reconstruction Example
(With Aliasing)
Original Signal
Reconstructed Signal
teetxtjtj ss cos
1 )()( 00 =+=
DSP 4-15
2
7/30/2019 DSP Chapter4
16/66
Nyquist Sampling Theorem
uppose a s an - m e o a requency
interval , i.e.,
)()( aa Xtx
[ ]NN ,
Then x t can be exactl reconstructed from e uidistant
NX = for0)(
samples , if)/2()(][ sasad nxnTxnx == ,22 Ns
sT
=
,
sampling frequency (radians per second), is referred
ssT = /2 s
N
to as the Nyquist frequency, and is called the Nyquist
rate.
N2
DSP 4-16
7/30/2019 DSP Chapter4
17/66
How to obtain discrete-time Fourier
transform (DTFT)?
ven e sampe s gna as
=
Since we have the following continuous-time Fourier
=ncs
transform (CTFT) pair ( ) nTjenTt
Thus we have the continuous-time Fourier transform of the
sampled signal as
=
=n
Tnj
cs enTxjX )()(
DSP 4-17
7/30/2019 DSP Chapter4
18/66
How to obtain discrete-time Fourier
transform (DTFT)? (Cont'd)
nce we now e rea ons p e ween e sampe
signal and the discrete-time sequence)(nTxc ][nx
We also have the DTFT of is defined as
nxnx c=
][nx
= njj enxeX ][)(
By comparing with
=
= Tnjcs enTxjX )()(
DSP 4-18
How to obtain discrete time Fourier
7/30/2019 DSP Chapter4
19/66
How to obtain discrete-time Fourier
transform (DTFT)? (Cont'd)
As we compare the following two equations
=
=n
enxe ][=
=n
cs enx
).()()(T
s eXeXjX ===
21 k 1
=
= TTTe
k
c s
k
csT
==
)( T=
DTFT representation of sampling !
is simply a frequency-scaled version of with the frequencyscaling specified by . This scaling is a normalization of the
)( jXsT=
)( j
eX
DSP 4-19
for .s
s
2= )( jeX
7/30/2019 DSP Chapter4
20/66
Example 4.1 (Without Aliasing)-
with sampling period T=1/6000.
Continuous-time Fourier transform
cosxc =
)( jXs Discrete-time Fourier transform
Problem Analysis)( jeX
Fourier transform of the original signal
)4000()4000()( ++=jXc.40000
=
Sampling frequency
Fourier transforms of the sampled signal
.12000/2 == Ts
= scs kjXjX ))((1
)(
= cj
kjXeX ))2
((1
)(
DSP 4-20
= =
7/30/2019 DSP Chapter4
21/66
Example 4.1 (Cont'd)
)()/( TT =)( T=
DSP 4-21
7/30/2019 DSP Chapter4
22/66
Example 4.2 (With Aliasing)-
with sampling period T=1/6000.
Continuous-time Fourier transform
cosxc =
)( jXs Discrete-time Fourier transform
Problem Analysis)( jeX
Fourier transform of the original signal)16000()16000()( ++=jXc.160000
=
Sampling frequency
Fourier transforms of the sampled signal are exactly same as the
.12000/2 == Ts
prevous one, w y
===
DSP 4-22
7/30/2019 DSP Chapter4
23/66
k=1 k=-1 k=0k=2k=-2
=
)()/( TT =)( T=
160000 =
DSP 4-23
7/30/2019 DSP Chapter4
24/66
Example 4.3 (with Aliasing)-
with sampling period T=1/1500.
Continuous-time Fourier transform
cosxc =
X Discrete-time Fourier transform
Problem Analysis)( jeX
Fourier transform of the original signal
)4000()4000()( ++=jXc
Sampling frequency
The discrete-time Fourier transform is the same as previous one.
.3000/2 == Ts
y
)3/2cos()1500/10002cos()1500/4000cos( nnnn =+=
DSP 4-24
7/30/2019 DSP Chapter4
25/66
k=1 k=-1 k=0
=
k=2k=-2
)( T= )()/( TT =
DSP 4-25
7/30/2019 DSP Chapter4
26/66
4.3 Reconstruction of a Band-limited
Signal from Its Samples,
modulated impulse train is filtered by an appropriate low-pass
filter, then the Fourier transform of the filter output will beidentical to the Fourier transform of the original signal.
Given a sequence of samples x[n], we form the impulse train
==
n
s nTtnxtx )(][)(
If the impulse train is the input to an ideal low-pass continuous-
time filter with impulse response)(
thr
=
=
=
==n
rr
n
rsr nTthnxthnTtnxthtxtx )(][)(*)(][)(*)()(
DSP 4-26
7/30/2019 DSP Chapter4
27/66
4.3.1 Ideal Reconstruction System
DSP 4-27
7/30/2019 DSP Chapter4
28/66
Ideal Reconstruction System (Cont'd)
==TTt
thr
sinc/
)(
DSP 4-28
7/30/2019 DSP Chapter4
29/66
Ideal Reconstruction System (Cont'd)
)(][)(nTthnxtx
rr=
n =
==tTt
thr
sinc)/sin(
)(
=
TnTtnxtx
]/)(sin[
If and then
= n TnTt /)(
nTxnx = =we have
c
txtx =
c
DSP 4-29
7/30/2019 DSP Chapter4
30/66
Ideal Band-limited Interpolatione ea ow-pass er n erpo a es e ween e mpuses
of to construct a continuous-time signal][nx
=r
TnTtnxtx
]/)(sin[][)(
If there is no aliasing, the ideal low-pass filter interpolates
=
correct reconstruction between the samples.
However, the ideal low-pass filter has infinite length whichis not realizable in practice. Finite length low-pass
filtering wil l result in some reconstruction error.
DSP 4-30
7/30/2019 DSP Chapter4
31/66
Ideal Band-limited Interpolation (Cont'd)
-
Sampled signalReconstructed signal
DSP 4-31
7/30/2019 DSP Chapter4
32/66
Ideal D/C Converter
DSP 4-32
7/30/2019 DSP Chapter4
33/66
Ideal D/C Converter (Cont'd)
seen in the frequency-domain.
TnTt ]/)(sin[ntnxtx rn
r == =
=n
rTnTt
nxx/)(
= Tnj
Linearity of continuous-timeFourier transform
=nrr
Time shifting leads to anexponential factor in theFourier transform
).()()( Tjrr eXjHjX=
Discrete-time Fourier
DSP 4-33
7/30/2019 DSP Chapter4
34/66
Can you get the original signal back?e ea ow-pass er se ec s e ase pero o e resu ng
periodic Fourier transform and compensates for the)(Tj
eX
.
If the sequence x[n] has been obtained by sampling a band-
limited si nal at the N uist rate or hi her the reconstructed
signal will be equal to the original band-limited signal.
If there is aliasing during the sampling, the reconstructed signal
will be distorted, see Examples 4.2 and 4.3.
In any case, the output of the ideal D/C converter is alwaysband-limited to at most the cut-off frequency of the low-pass
filter, which is taken to one-half the sampling frequency.
DSP 4-34
7/30/2019 DSP Chapter4
35/66
4.3.2 Zero-order Hold D/A Conversion-
digital output is carefully transformed into analog form.
t tje d
aa
Comments:
ompensa on w e er or - no o .
The D/A block is not filtering - it is weighting.
d e )(a)(tga
DSP 4-35
.a
7/30/2019 DSP Chapter4
36/66
Impulse Response of Zero-order Hold,
)(tg a
1
==
n
sdda ngnxy
sT0
It holds at a constant level over each sampling period.][nyd
d
1
a
1
. .
0 1 2 3-2 -1 n-3 0 1 2 3-2 -1 n-3
DSP 4-36
Note: Z.O.H. introduces high frequencies, see sharp edges.
7/30/2019 DSP Chapter4
37/66
Frequency Response of ZOH.
2/)2/sin()( s
Tjs
a
eT
jG=
The high frequencies in the reconstructed signal (sharp steps)
are introduced from side-lobes as follows.
Ideal)(aG
Z.O.H
2
2
DSP 4-37
ss s s
Frequency response of Z.O.H
7/30/2019 DSP Chapter4
38/66
Compensation of ZOH
shit of T/2 which cannot be compensated and usually neglected.
The magnitude response can be compensated as follows.
7/30/2019 DSP Chapter4
39/66
Physical Configuration for
ZOH D/A Conversion
reconstruction filter is shown as follows.
DSP 4-39
7/30/2019 DSP Chapter4
40/66
4.4 Changing the Sampling Rate
s o en necessary o c ange e samp ng ra e o a scre e-
time signal, i.e., to obtain a new discrete-time representation of
- .
''' cc
rate that involve discrete-time operations.
]['][ nxnx
DSP 4-40
S li R t Ch E l
7/30/2019 DSP Chapter4
41/66
Sampling Rate Change Examples
(Down-sampling)
What happened during
DSP 4-41
-
4 4 1 Sampling Rate Reduction by an
7/30/2019 DSP Chapter4
42/66
4.4.1 Sampling Rate Reduction by an
Integer Factor (Down-sampling)
"sampling it" by defining a new sequence
).(][][ nMTxnMxnx cd ==
DSP 4-42
7/30/2019 DSP Chapter4
43/66
Frequency Representation of Down-Sampling [W-D]
nxnx c=
=
j k 21
= kc
TTT
==, c
=jr
XeX 21
uestions: what is the relationshi between them?
= rc
TTT
)()( jdj
eXeX
DSP 4-43
Frequency Representation of
7/30/2019 DSP Chapter4
44/66
Frequency Representation of
Down-Sampling (Cont'd)
=c
j
d
rjXeX
21)(
ris still an interger ranging
from -inf and inf
=r
)( kMir +=,
7/30/2019 DSP Chapter4
45/66
Frequency Representation of
Down-Sampling (Cont'd)
=
=
=
1
0
2211)(
M
i k
c
j
dT
k
MT
ijX
TMeX
We know that
CTFT)from(DTFT21
)(
= k
jXeX cj
21)(
= k
jXeX cj
=k =k
( )
=
=
k
c
Mij
T
k
MT
ijX
TeX
221/)2(
Therefore, we have
=1
/)2(1M
Mrjj eXeX
DSP 4-45=0r
Frequency Representation of
7/30/2019 DSP Chapter4
46/66
Frequency Representation of
Down-Sampling (Cont'd)
=
1/)2(1
MMrjj
=0r
jj
can be composed ofM copies of the periodic Fouriertransform , frequency scaled by M and shifted by integer
.d
)(
jd eX)( jeX
mu pes o .
is periodic with period 2.)( jd eX
= limited)-(band,0)( N
jeX
DSP 4-46
an e-narrowN
7/30/2019 DSP Chapter4
47/66
NT = 4/2
Ns = 4
2/ == T
DSP 4-47
4-4
Frequency Representation of
7/30/2019 DSP Chapter4
48/66
Frequency Representation of
Down-Sampling: Example
DSP 4-48
7/30/2019 DSP Chapter4
49/66
Down-sampling after Pre-filtering
- ,the band-width of signal x[n] prior to down-sampling.
- -cut-off frequency /M.
DSP 4-49
7/30/2019 DSP Chapter4
50/66
Down-sampling after Pre-filtering (Example)
Mnn /sinsinnn
n clp
==
DSP 4-50
4.4.2 Increasing the Sampling Rate
7/30/2019 DSP Chapter4
51/66
4.4.2 Increasing the Sampling Rate
by an Integer Factor
increase by a factor ofL (up-sampling).
==
otherwise,0
,][nxe
For example, ( )4321054321][
==
nnx
( )
2and9876543210
0504030201][
==
=
Ln
nxe
Or equivalently,
=
DSP 4-51
=k
e
Sampling Rate Change Examples
7/30/2019 DSP Chapter4
52/66
p g g p
(Up-sampling) Demo
-
DSP 4-52
7/30/2019 DSP Chapter4
53/66
Frequency Representation of Up-Sampling
e ourer rans orm o e up-sampe s gna s
][][)( n
n k
e ekLnkxeX
=
= =
)(][Lj
k
kLj
eXekx
== =
We can see the Fourier transform of the output of the
-transform of the input, i.e, is replaced by L.
DSP 4-53
Frequency Representation of Up-Sampling
7/30/2019 DSP Chapter4
54/66
(Example)
DSP 4-54
7/30/2019 DSP Chapter4
55/66
General System for Up-sampling
, -
therefore considered to be synonymous with interpolation.
LnnLnh clp
/][
=
=
DSP 4-55
a n
7/30/2019 DSP Chapter4
56/66
Ideal Low-pass Filtering after Up-sampling
,interpolation formula with an ideal low-pass filter as
==
==kk
lpiLkLn
kxkLnhkxnx/)(
][][][][
The impulse response of the low pass filter has properties
Lnnhi
/][
=
,....3,2,,0][ LLLnnhi
i
==
Thus for the ideal low-pass interpolation filter, we have
==
DSP 4-56
,...,,,i
Frequency Representation of Up-Sampling
7/30/2019 DSP Chapter4
57/66
and Ideal Low-pass Filtering (Example)
DSP 4-57
7/30/2019 DSP Chapter4
58/66
Linear Interpolation after Up-sampling
,/1 LnLn
= otherwise,0
lin
DSP 4-58
Linear Interpolation after Up-sampling
7/30/2019 DSP Chapter4
59/66
(Example)
2
lin
)2/sin(
)2/sin(1)(
=
L
L
eH j
DSP 4-59
Linear Interpolation after Up-sampling
7/30/2019 DSP Chapter4
60/66
(Example, Cont'd)
1]0[hi =
=
,/1 LnLn
,....3,2,,0][ LLLnnhi == otherwise,0n
= kLnhkxnx
,...3,2,,/ LLLnLnxnx ==
=k
The amount of distortion in the intervening samples can be
gauged by comparing the frequency response of the linearinterpolator with that of the ideal low-pass interpolator, as
2
)2/sin(1 =
Lj
DSP 4-60
n
)2/sin(
L
4 5 Di it l P i f A l Si l
7/30/2019 DSP Chapter4
61/66
4.5 Digital Processing of Analog Signals
Sample at a faster rate - perhaps not possible (why?).
Use an anti-aliasin filter.
DSP 4-61
Ho to red ce aliasing?
7/30/2019 DSP Chapter4
62/66
How to reduce aliasing?
- - is applied to the continuous signal prior to sampling.
The idea is sam le: remove the hi h fre uencies . The ideal frequency response of the anti-aliasing filter is an ideal
low-pass filter as ,2/s>
where the cut off >=
cF
0;1 sc
T=
7/30/2019 DSP Chapter4
63/66
Anti-aliasing: Formulation
- ,
mm
221sasad nnxnx = = = s
a
m s
a
s
dTTT
e
The repeated spectra will not fold or overlap.
If is an ideal LPF with cutoff , then
)()( aa FX
)(aF c
=
=
ca
aa
j
d
XX
mXeX
);(where
21)(
Usually, an ideal LPF cannot be realized and must be
= csms
DSP 4-63approximated.
Anti aliasing E ample 1
7/30/2019 DSP Chapter4
64/66
Anti-aliasing: Example-1
)(aX
Analog Signal Spectrum
0
)()( aa FX
Anti-aliased Spectrum
)(
j
d eX
0
c c
Sampled Signal Spectrum
2/
7/30/2019 DSP Chapter4
65/66
Anti-aliasing: Example-2
- -with aliasing
with anti-aliasing
DSP 4-65
Anti aliasing: Digital Filter Output
7/30/2019 DSP Chapter4
66/66
Anti-aliasing: Digital Filter Output
eca e overa sys em o n eres :
j
][nxd ][nyd
da a
The res onse of the di ital filte)(
aH
jeY
jeH
)(21
)( j
da
j
d eH
T
mX
T
eY
= without anti-aliasing
)(221
)( j
daa
j
d eHm
Fm
XeY
= with anti-aliasing
DSP 4-66
m sss =