8

Sampling

&1 INTRODUCTION timne

as one which is defined for all values of
continuous-time signal time.
Earlier we had defined a defined only over a discrete set of points in
one which is (analog) signals. Analog
and a discrete-time signal as in practice are continuous-time
the signals that we encounter
analog
Most of recovery fall under the category of
processing, representation, transmission and communications, which is more
Ngnal drawbacks.In digital
Communications which have certain analog signal into a discrete-time sigat called
transform an
8'anlageous, it is required to continuous-time signal into a discrete-timne signal is
ne process of converting a time
is defined at discrete instants of time and the
mping After sampling, the signal
sampling instants is called sampling period or sampling
eval between two successive of the important factors
that we have to consider
iercal) dn the process of sampling, one
be kept sufficiently high so that the original signal can be
e Sampling rate must
eUOnstructed from its samples,

B2 SAMPLING
The
fictitious switch shown in Figure 8.1.
Sanpl ing
SWch is closedOperation
for a very
represented by a
be interval of time T(ideally, t= 0), once every Tsec during
canshort
Whçch the signal IS available at the output. Therefore, if the input is x(), then the
output x,)
AnT), n =0, EI, 2, ... and x(nT) is called the sampled sequence of x(), where T is
Galkd the sampling period or sampling interval. It is the time interval between successive
ples and the sampling frequency is given byf =(1/T) Hz. Although a mechanical switch
Shown in Figure 8.1, in
in actual practice, an electronic switch
may be used.

541
542 Signals and Systems

x(0) 4 x(nT)

x() ,()
x(n) = x(T)

Figure 8.1 Sampling operation.

8.3 SAMPLING THEOREM

The sampling theorem is one of the most useful theorems since it applies to
communication systems. The sampling theorenm states that(A band
X(0) = 0 for lol o, [i.e. Xf) =0 for f2 fm can be represented limitedintosignalandx)digtaith
determined from its samples x(nT) if the sampling frequency f,2,m where fm is the hhe
frequency component present in i.jThat is, for signal recovery, the sampling frequency me
be atleast twice the highest frequency present in the signal)
This theorem is known as uniform sampling theorem since
it pertains to te
specification of a given signal by its samples at uniform intervals of 1/2f sec.
It is also called low pass sampling theorem
because it applies to low pass signals. ie
signals for which X) = 0for all frequencies such that| Sm where f, is some finite
frequency.
Proof: The sampling operation can be represented as shown in Figure 8.2. x(t) is a
continuous-time band limited signal be sampled which has no spectral
Sm cycles per sec. That means X(0), the Fourier transform of x(t) is 0 components abov
for 0> on o) is =
impulse train which samples at a rate of f. Hz and x.(0) is the sampled signal. Tis
sampling period and f = (1/) is the sampling frequency.)
x(1)
x{() x(nT)

x,() -2TT2T
x() x(nT) -3T -TO

-2T -T 0 T 2T
Figure 8.2 Sampling
operation.
 4441

Sampling| 543
1()is
the produet of signal x() and impulse train
intervals of T sec and `,(). It is a sequence of
at regular having strength equal to the values of impulses
a'spo1ng instants. x(t) at the
A) =) ö) where S,)= ) S
-nT)
The exponential form of Fourier series of ) is:

S.u)= -nT)= C,emo where

-T/2

S,() = 8-nT) =) y ene

- T n=

x,() =x¢) S, )= T S x() em

n=-o

Taking Fourier transform on both sides, we have

F(x,()): n=

Le. X,(o) =
n=-o

X,(f) =f, Xð- nf,)

n=-oo

is the spectrum of input signal and X(o) or X,(f) is the spectrum of the
WIeTe X( a) or Xf)
Sampled signal,) shifted
transform of the sampled signal is given by an infinite sum of
Ihus, the Fourier
eplicas of the Fourier transform of the f. original signal.) n] is the shifting of X(0)
The term X[0 - (2rr/T)
Lne signal x() is band
limited to
sum of shifted replicas of (/T) X(0)
X{o) is the for
from ) = 0 to 0 = (2r/T)n( Hence..)Figure 8.3 shows the plot of X(0) and X,(0)
+1. 2, will
(n/T) 2 O, the replicasrange
ng at (2r/T) n, n = 0. [Figure 8.3(b) and (c)] if frequency
notValües of n/T. It showsthethatfrequency spectrum of TX(o) in thepassing it through a
K a p and as a result, Xí), X(o) can be recovered from X,(0) by
B, where ,
sBS u
L), n/T] is identical to with bandwidth
cutoff at w = r¯T (or will overlap and the
pass fifter which has sharp 8.3(d)], the successive frequency spectra we can say that for
,). If (z/T) < O,, [Figure Theretfore,
from the sampled signal.
recovered
original signal cannot be
Signal recovery. W, - W,,
2,,, 1.e. 0, 2
2,.
544 Signals and Systems

or f.- S, 2Sm i.e. f, 22f,,

22f
2@. i.c. nf, 22nfm. i.e. S,

ie. TS
f.
4X( 0)

-0,. 0
(a)

4 X,(0)
I/T

-o/2
2r -Qto, 0
-0-0,
(b)

+X,(@)
-W+so m
- m /T

Qwm s

-3 o 270
T

(c)

4X(0)
li/T

-0,
2 -),m 0
T T T
T (d) trequen
sn

and (d)
fespectively.
(b). (c)
Cioure 8.3 (2) Frequency spectrum of continuous-time signal x(1).(n/T) < om
of sampled signal x() for (n/T) > o,m, (r/T) = 0, and
 conclude that if the sampling interval
So we
can
but if T becomes larger than TissImall Sampling 545
hen there |is<<2f),
N,(e), 2..
fnm io) cannot be X o0) can be
wks and
recovered from an
Hrom the
previous
discussiOn, X(O)) This proves the overlap between recovered
wve can
to retneve () from A,().
Thus, we find that in general, there are
observe that
when sampling
the theorem.successive
spectra overlap. it
two basic s
from its samples.
should be
conditions to be satisfied if x 1s to

The sampling
band-limitedo, tosome frequency
frequency should be atleast
twice the
From Figure 8.3. we can observe that:
band-limiting frequency 0.

1Xí0) is a repetitive version oÌ X(0) with

O,. the sampling frequency. X(0) repeating itself at regular
intervals of
When o, > 2o,, [Figure 8.3(b)], the, spectral
hetween them, known as guard band, which replicates have alarger separation
makes process of filtering much
easier and effeetve. Even a non-ideal filter which the
does not have a sharp cutoff can
also be useds
2 When o, = 2o,, [Figure 8.3(c)), there is no separation between the replicates, so no
guard band exists, and X(0) can be obtained from X{0) by using only an ideal low
pass filter (LPF) with sharp cutoff
4 When ), < 2o, [Figure 8.3(d)], the low frequency components in X( 0) overlap on
the high frequency components of X(0), there is distortion andis X(0) cannot be
called aliasing
recovered from X(0) by using any filter. This type of distortion
Aliasing can be avoided if ,2fm or Ts (1/2fm). guard band
impossible to build filters having an infinite sharpness of cutoff.
Since it is
Detwcen f and f, - fm is preferred. processed through an ideal LPF with
sampler is he resulting
output
The impulse train at the output of the less than o, -
gan T and cutoff frequency greater than O,, and
Signal will exacty equal x).
can b
NYQUIST RATE OF SAMPLINGminimumsampling rate at distortion')
which a signal the
lt is
withoul any LPE
theoretical only anideal
Nsquist IS the
rate of samplingreconstructed from its
Sampled and still be
samples
samplingis
used,
Nyquist rate of A() fromn ,(). l
is always cqual
to

when the recover signal. and a

signal
theoretical minimum' because
can be used to extract X(0) from
i.e. to
X,(0), component present
said
in the
to be
ovei
sampled
frequencyNyquist rate is sampled.
fm whereSm is the maximum
greater than said to be
under
Signal sampled at rate is
Nyquist
Sampled at less than its
Nyquist interval
CNvquist interval is the time interval between any two aljacent samples
Nyquist ráte.) when smpl1n
Nyquist rate fv = 2/, Hz

Nyquist interval Sec.

8.5 EFFECTS OF UNDER SAMPLING-ALIASING

When o, <20,, i.e. when the signal is under sampled, X(0), the
longer replicated in X,(0), and thus is no longer recoverable by low passspectrum of Txt)
in which the individual terms in equation X,(0) =
filtering This et
) X(0-n
T
n=-o0
n),) overlap is refete.
as aliasing. This process of spectral overlap' is also called frequency folding effer let
aliasing is defined as the phenomenon n whicFgA requency
spectrum of signal takes identity of a lower frequency component component in the fren
in the spectrum of
sampled signal
Aliasing can occur if either of the following conditions exists:
KThe signal is not band-limited to
finite range.
2 The sampling rate is too low.
Theoretically if the signal is not band-limited, there is no way of avoiding the alas;
problem with the basic sampling scheme employed. However, the spectra of most
signals are such that they may be assumed to be band-limited rdl it
Further,, a common praxt:
employed in many sampled data systems is to filter the continuous-time
sampling to ensure that it does meet the band-limited criterion ctosety enough signals betr:
for all prat
purposes
Tø avoid aliasing, it should be
ensured that:
1. x(t) is strioly band-limited (this can be ensured by using anti-aliasing filter eir
the sampler).
2. S, is greater than 2fm:

8.6 ANTI-ALIASING FILTER

Sampling theorem states that asignal can be perfectly reconstructed from itssamples ony
it is band-limited.Un practice no signal is strictly bandI-limited, i.e. in general signalshs:

frequency spectra
components. Whenconsisting of low frequency as high
frequency Ireguen
asignal is
sampled, with components f allsignals with alastt
as well
range higher than o,/2 appear as signal sampling frequency. Q72creating
frequencies
Therefore, to avoid afiasing errors caused berween high
by the undesired 0 andfrequency signals,
necessary to first band-limit x() LPF suchth
to an
some appropriate frequency f bY using
 Sampling 547
This LPF used lor band-limiting a signal before
retained.
of the
energy is generally referred to as an anti-aliasing filter since itsampling.
part 8.4, is is used
smst i n Figure
shOWn
for preventing aliasing.

() Anti-aliasing filter xn)

Sampler
Figure 8.4 Anti-aliasing filter.

7SAMPLING TECHNIQUES
is done in several ways. Basically there are three types of sarnpling
Sanmpling of a signal
mhniques: sampling
KInstantaneous sampling or impulse
2Natural sampling
A Flat top sampling impulse sampling is also called ideal
instantaneous or
Out of these three methods, Flat top sampling are called practical sampling
sampling and the Ssubsequent sections.
sAmpling. whereas the natural
techniques are discussed in detail in
methods. These sampling
Sampling
87.1 ldeal or Impulse kth element of the sequence
done
instantaneously so that the operation is shown in
be kT. The
ldeally, sampling should represents the value of x(1)
at =
for
every
zero time once in time
closes almost
0alned by sampling that the fictitious sampler output for a very very short an
Tgure 8.5(a).
Assume
the input signal to the switch can be replaced by
Tsec. It is equivalent to
transmitting mechanical Figure 8.5(b).
sec. The as shown in
almost zero time) once in
every T Amplitude Modulator
clectronic switch which is
basically a Pulse
x,()
x(nT)
n (b)

(a)

4 x(0)

2T
-27-T 0 T

(c)
8.5 Idealsampling.
 548 Signals and Systems

Now. the operation is cquivalent to multiplying the inpul ignal x(t) by an impule
is a train of
as shown in Figure 8.5(c). So the output of the sampler
to the instantancous value of the input signal at the sampling
instant
The impulse train, also called the sampling uncion IS represented ..
impulses
of

S,() = X St -nT)
The sampled signal is given by
x,(0) = x() S,() =x() 8-nT) = x(nT) S(1 -nT)
n=-oo n=-00

x,(@) = 2 x(0- na, )

n=-o0

X,()=f. n=-oo
xf-nf,)
This equation gives the spectrum of ideally
XÁo) is an infinite sum of shifted repticas of X( sampled signal. It shows that the ste
0) spaced no, apart, where n=t1. 1
and scaled by a factor 1/7) However, it may be
noted that ideal or instantaneous samplin:
possible only in theory because it is impossible to have a pulse
zero, Practically, the flat top sampling or natural sampling is used.with pulse width approachng
8.7.2 Natural Sampling
Natural sampling, also called sampling, using a
of accomptishing sampling of a sequence of pulses the most pracical w
signal x) with a pulse train P) as band-limited signal. This is achieved by muliply1ng
shown in
duration t and occurs at a sampling period of T Figure 8.6. Each pulse of Pr) 1S0
input duning that short duration, t. sec. The output of the
Hence is termed as natural
it sampler 1s sal
sampling.

,() =)pl)

Figure 8.6
Figure 8.7 explains the process of naturalNatural
sampling.
sampler.Figure 8.7(a) is the signalxt) N
sampled, and Figure 3.7(b) is its
Figure 8.7(d) is its spectrum Pf).spectrum
Figure X(f),
8.7(e) Figure
is the 8.7(c)
outputis of hepulse
Sampler
train, rl0),sol
the
 Sampling 549

output spectrum X(). From Figure 8.7), it is clear that Xf) can be
is the x() can be recovered from x(0), if ,> 2fm by using an LPF
X.¯). ic.atleast
from constant upto f = f. and whose cutoff frequency B is such that
aerd

A) 4X)

Jm
0
(b)
(a)
4PA) 4Pð)

(d)
(C)
4X,(6)
4x,(1)

0 ()

(e) sampling.
Figure 8.7 Natural

Ihe output of the sampler is: ()

x,() = x() p,
pu-nT)
where P,() = expansOn
series
Fourier
its
us write
As pAt) is a periodic pulse train, let Gul!
)
pt-n)=
Pr()=
550 Signals and Systems

where
-2
much less than T
Since the width of p). single pulse in P, () is very
D2, we may write and pt)

-D2

C, =f,Pnf,)
where Puf,) = F[p),.=nf,

Pr() = f, Pf)e2rnl4
n=-oo

and
n=-0o

-f, ) P(nf,) X(/) SG -nf,)

n=-o

Since F(eZ# ]= 8(f -nf,)

Hence
x,f)= f, Pnf,) X(f -n)
If x() has a spectrum Xf), as shown in Figure 8.7(b), then X(f), the spectrum ot te
sampled version of xt) will appear as shown in Figure 8.7().

8.7.3 Flat Top Sampling

This is the simplest and most popular sampling
method
circuit with flat top samples This is also called that uses
practical the sample
sampling. Here and
the topoft
samples remajn constant which is equal to the instantaneous value of the base bandsignal
the beginning of sampling. The duration or width of each sample is r and the samplin
rate. f, = /T.)
The schematic of a 'sample and hold' (S/H) circuit is shown in Figure 8.8a),uad
typical output waveform from an S/H circuit is
The S/H circuit shown in Figure 8.8(b).
connected as shown in essentially consists of two switches S; and S, and brief a penid
Figure 8.8(a), With Verylothevalx
each sampling instant. The S; open, S, is closed for acqual
the input signal xti) at the capacitor then gets charged to a voltage
C whuhS
end of
is closed to allow the sampling instant and holds it for a period t at the Tepeatedt th
capacitor to discharge. This sequence of operations is
 Sampling 551
l subscquentsampling instants. The switches S, and S, are generally FET switches
by giving appropriate pulscs (o their gates. An actual S/H circuit uses one or
The voltage acrOSs Cappears as x() and is
sketched in Figure 8.8(b).

x,()

(a)
x(0)
x(0)
a(-7) x(7)

(-27)
x(2T)

-1 2T 37
-2T
(b)
output of S/H circuit.
Figure 8.8 (a) Schematic of an S/H circuit, (b) Signal x() and
x,(t) consists of a sequence of
From the Figure, it is obvious that the sampled version, = kT and the amplitude of the
at
fectangular pulses, the leading edge of the kth pulse being
pulse being the value of x(t) at t = kT, i.e. x(kT). ideally
signal x.(t) is the convolution of rectangular pulses p() and the
Ine sampled
sampled version of x(t), i.e. of x()
X,(f) = PS) Xsð)
8.9.
Assume thaat x(t) has a spectrum as shown in Figure
4Xð)

JM X).
Figure 8.9 Assumed shape of which is a sinc
transtorm, P) only at
width t, its Fourier have its first zero values away
Since p(t) is a of will far
rectangular pulse in Figure 8.10and occur at t(l/r), will be
8.10(b).
funcúon will have a Shape as shown
values of |POl
Since t <T, these zero Xsf), its plot will be
which
as shown in
Figure

Irom f, and Since X,f) =

Pè)
 Sampling553

M)= x{nT) 8(2 - nT) ht - A) dl

chaangingthe
order of summation and integration,
(a UNon
.

Pð) XA) =X()

v)= x(nT) h(t - nT)

-f. Filter
(b)
J 8A1 ldeal Reconstruction
Figure 8.10 (a) Plot of w0ksampled at a frequency exceeding the Nyquist rate and if the sampled signal x4) is
PO) and Xf). (b)
Observe asthat the Plot of X(f) = dthrough an ideal LPF, with bandwidth greater than f, but less than f, - hand apass
reduced compared magnitudes of the high P{)XAfy nd amplitude response of 7, the tilter output is x(). We choose the bandwidth of the ideal
of Xsf) bythe magnitudes of frequeney
to onstriction filter to be 0.5f,. The ransfer function of this ideal reconstruction filter is.
multiplication
x(), by passing x,(t) low components
Pð). So we can only get a frequency in Xf) are releiod
the
componentof xt),
s becaue
therelore.
This through an LPF. distorted version H)= lfl<0.5f.
relative to distortion, wherein the
the amplitudes of the amplitudes of the high
but o eu
0, otherwise
obtained from the flat top low
frequency frequency components are reðut! 2s shown in Figure 8.11.
This aperture effect
can besampled version components,
of the signal is in the reconstructed signal zs
referred to as the aperture fet
cascade with the reduced by
reconstruction filter and using an
adjusting equalizer with transfer function HÊna
H¯) so that
Ideal reconstruction filter

+Hð)
Zx(nT) 8 -nT) T

8.8 DATA RECONSTRUCTION 0

The process of obtaining the Figure 8.11 Reconstruction filtering

reconstruction or interpolation.)analog
We
signal x(i) from the sampled signal
know that x,) is calileu e impulse response of the ideal
reconstruction filter is given by
x,() =x() &, () =x) ) 8-nT) / h(t) =
n=-oo

or
x,()= X x(nT) t - nT) Which is
h) = T
since S(t - nT) is zero except at the sampling instants =nT. The reconstruction filter.
is assumed to be linear and time invariant, has unit impulse response h), The reconstruct sin r/,
filter output, y() is given by the convolution, 2j
554 Sienals und Systems
h() = sinc f
get
Substituting this value of h() in the expression for output y(), we

v) =x() = x(nT) sinc f, (t -nT)

expression, which is often referred to as an interS

A more convenient form for this
formula is:

x() = x(nT) sinc

n=-o

This shows that the original data signal can be reconstructed by weighing each samçle
sinc function centred at the sample time and summing.
The reconstruction filter discussed above is non-causal and the
limited. So it cannot be used for real time applications. impulse response
In practice, several other methods are used to reconstruct the signal. Some
important ones among them are:
Zero order hold
First order hold
Linear interpolator
Most commonly the signal is reconstructed using zero order hold.
8.8.2 Zero Order Hold
One of the most widely used interpolator is the zero order hold (ZOH).The ZOH reconst
the continuous-time signal from its samples by holding the given sample for an interu
the next sample is received) as shown in Figure 8.12. So the ZOH generaie
approximations.
Mathematically.
R) = x(n) for n+T Sns (n + 1)T
In particular.

i,()= x(0) for 0 <tsT

= x(T) for TSS2T
= x(2T) for 2T STS3T
:

The impulse response of a zero order hold is given by

h(t) =1 0stsT
=0
otherwise
 Sampling 555

Zeo order hold

h)

Figure 8.12 Zero order hold.

.*t Transfer Function of a Zero Order Hold

of a zero order hold is the convolution of its input x(nT) and its impulse
se h), ie.

i,)=x(nT) * h() =) x(nT) h (t - nT)

Ax zero order hold,

h(t) = u) u(t -T)

h(t nT) =u(t -nT) - u[t-(n+1)T]

i,()= x(nT)[u(t -nT)- u(t - (n+)T]

laing Laplace transform on both sides, we have

+DT}
x(nT) {u(t -nT) - u( - (n
LIi,0)= *,0) =.|
n+)S

|-e

zero order hold haronKs To

ITanster function of hugher order sNxh
consists of tends o
of steps,
il
LPE) This tiller ofiencalled
consists to an Hewethis tilter is
Since the Output of the Z0H of ZOH
is applied
Z0H.
generated by the
Temovee corners on
these harmonics, the output
approximations
the slep
Smoothing filter,
556 Signals and Svstemns

EXAMPLE 8.1 Deterninc the Nyquist rate coresponding to cach of the

()=ltcos 2000aI t sin 400047
(b) x()= sin(400)x)
.(a)
sin 4000z)

Solution:
(a) Given
x)=1 +cos 20001 +sin 4000t

Highest frequency component in 1 is zero

Highest frequency component in cos 2000z= cos Wm 1S W,m = 2000
Highest tfrequency component in sin 4000TT = Sin Om2 is O2 = 40007
So the maximum frequency component in x() 1s Qm = 40007 [highest of 0
4000).
2rf = 4000x
4000x
O Jm = 2000 Hz,
2
Nyquist rate fy = fm
= 2 x 2000 = 4000 HZ
1
and Nyquist interval =

4000
sec =0.25 ms
Aliternative method
Given x() =1+ cos 2000 + sin 4000z
Taking Fourier transform on both sides, we get
X(O) = 2rÖ w) + n| Sw+2000z) + Sw- 20007))

+ jn|O w+4000z) - So-4000z))

The frequency spectrum is shown in Figure 8.13.
X(0) 4 2r

4000 20000r 0 4000r

2000x
Figure 8.13 Spectrum of Example 8.1(a).
557 function
a Figure
oftransform

Sampling 4000.
sinc
Fourier
si4000
nc(4000z1)
o,,
a
is 8.14(b).
4000z. is
pulse
the
Hz
O,t
4000r.
Hz
4000
ms
=0.25 rectangular
property,
Figure
= sin
4000 =0.25
ms 2/m
fN
G)
is Hz
2000
= fN
2fm
=
4000It
=
2000
Hz =
2000 4000z1
inshownsin 4000xt
#2)).
= sin(4000xt) 4000 duality 40001
8.13 2000 x(0) 4000
=
2fm =2fm=2x 1 a sinc[r(
4000 in = =interval
Nyquist of
Figure 2f, =2x 1 sin component4000z theUsingas
transform
Fourier
= interval
Nyquist 27
f pulse 4000
= t function
Sinc 2
x(t) incomponent
component2T 40002
in Wm f
rate
= rate 14(a).
rectangular
a
form

frequency Nyquist 4000i

Sin the
Nyquist 8.
of
frequency frequency Allernative
method
maximum thatin is x1)=
highest highest know
function
represented
the Given a
The and Given
b) The So We Sinc is
It