TN 207

Deterministic Signals and

Systems

Sampling
1
 Introduction
Points to be discussed in this lecture

• Instantaneous (Ideal) Sampling

• Nyquist Sampling Theorem

• Flat-Top (Zero-order Hold) Sampling

• Natural Sampling

• Practical Considerations

2
 Introduction
™ Message signals generated by analog
information sources such as speech, music,
picture, video, etc., are continuous in both,
amplitude and time/space (analog).

™ Analog message signals are converted to digital

form by an analog-to-digital (A/D) converter.

™ Analog to digital conversion involves sampling

to produce a discrete time signal, quantizing to
generate a discrete-amplitude signal, and
encoding.

3
 Introduction
™ Sampling plays a role as a bridge between
continuous-time signals and discrete-time signals.

™ Processing of discrete-time signals is more

flexible and preferable to processing continuous-
time signals.

™ Discrete-time systems are inexpensive,

lightweight, programmable, and easily producible
than continuous-time systems.

4
 Introduction
™ Sampling plays a role as a bridge between
continuous-time signals and discrete-time signals.

™ Processing of discrete-time signals is more

flexible and preferable to processing continuous-
time signals.

™ Discrete-time systems are inexpensive,

lightweight, programmable, and easily producible
than continuous-time systems.

5
 Instantaneous (Ideal) Sampling
™ Consider a signal xs ( t ) obtained by
instantaneous sampling a signal x ( t ) at a periodic
interval Ts . Ts is sampling period and f s = 1T is the
s
sampling rate.
x (t )

xs ( t )

t
−Ts Ts
6
 Instantaneous (Ideal) Sampling
™ The instantaneous sampled signal xs ( t ) can be
expressed in the time domain by
∞
xs ( t ) = ∑ x ( nT )δ ( t − nT )
n =−∞
s s

™ The sampled signal xs ( t ) can be obtained by a

product of x ( t ) and a periodic impulse train δ Ts ( t )
xs ( t ) = x ( t ) δ Ts ( t )
∞
= x ( t ) ∑ δ ( t − nTs )
n =−∞

7
Instantaneous (Ideal) Sampling

x (t ) xs ( t ) = δ Ts ( t ) × x ( t )

∞
δT (t ) =
s ∑ δ ( t − nT )
n =−∞
s

8
 Instantaneous (Ideal) Sampling
™ Using frequency convolution property, it follows
that
1 ∞ ⎛ 1⎞
Xs ( f ) = X ( f )∗ ∑ δ ⎜ f − n ⎟
Ts n=−∞ ⎝ Ts ⎠
Prove!
1 ∞ ⎛ 1⎞
= ∑ X ⎜ f −n ⎟
Ts n=−∞ ⎝ Ts ⎠

™ Note that:
1. X ( f ) is a continuous spectrum.
2. X s ( f ) is periodic with a period equal to 1T
s

9
 Instantaneous (Ideal) Sampling
™ Alternatively, X s ( f ) can be derived using the time-
shift property of Fourier transform i.e.,
∞
1
Xs ( f ) =
Ts
∑ x ( nT ) exp ( − j 2π fnT )
n =−∞
s s

™ This is the complex exponential Fourier series

expansion of the spectrum of xs ( t ) since X s ( f ) is
periodic in f with coefficients x ( nTs ) .

10
 Sampling Theorem
™ Let the spectrum of x ( t ) be strictly band-limited to
B Hz as shown below
| X ( f )|

f
−B B

™ Let the sampling period Ts = 1 2 B , which implies that

the sampling rate f s = 2 B .

11
 Sampling Theorem
™ In this case,
∞
X s ( f ) = 2 B ∑ X ( f − 2 Bn )
n =−∞

| Xs ( f )|
2B

f
−3B −B B 2B 3B

12
 Sampling Theorem
™ Therefore, the spectrum of x ( t ) , X ( f ) can be
derived from X s ( f ) as

1
X(f )= Xs ( f ) −B≤ f ≤B
2B
1 ∞ ⎛ n ⎞ ⎛ π nf ⎞
= ∑ x⎜ ⎟ exp ⎜ − j
2 B n=−∞ ⎝ 2 B ⎠ ⎝
⎟
B ⎠
−B≤ f ≤B

™ It can be concluded that X ( f ) is uniquely

determined by sample values x ( n / 2 B ) and so is x ( t ) .

13
 Sampling Theorem
™ Reconstruction of x ( t ) from xs ( t ) can be derived
from the inverse Fourier transform of X ( f ).

∞
x ( t ) = ∫ X ( f ) exp ( j 2π ft ) df
−∞

1 ∞ ⎛ n ⎞ ⎛ π nf ⎞
∑ ⎟ exp ( j 2π ft ) df
B
=∫ x⎜ ⎟ exp ⎜ − j
n =−∞ ⎝ 2 B ⎠ ⎝ ⎠
− B 2B B
∞
⎛ n ⎞
= ∑ x⎜ ⎟sinc ( 2 Bt − n )
n =−∞ ⎝ 2 B ⎠

™ This is an interpolation formula.

14
 Sampling Theorem
™ Sampling theorem states that:

1. A signal of finite energy band-limited to B Hz is

completely described by specifying the values
of the signal at instants of time separated by
1/ 2B sec.

2. A signal of finite energy band-limited to B Hz

may be completely recovered from knowledge
of its samples taken at the rate 2B per second,
which is called the Nyquist rate.

15
 Sampling Theorem
™ Reconstruction of x ( t ) from xs ( t ) is implemented by
means of a low-pass reconstruction filter.

™ For realizable reconstruction LPF, the sampling

rate f s must be larger than 2B .

™ If f s < 2 B , frequency-shifted versions of X ( f )

overlap. The overlap results in aliasing effect and
the original signal x ( t ) cannot be recovered from
xs ( t )

16
 Sampling Theorem
| Xs ( f )|
2B

f
− f s < −2 B −B B fs > 2B

| Xs ( f )|

2B

f
− fs > 2B −B B fs < 2B
17
 Flat-Top (Zero-Order Hold) Sampling
™ Flat-top sampling of x ( t ) is obtained by
instantaneous sampling at sampling period Ts and
holding the sample value for a duration of T sec.
Flat-top sampled signal is defined by
∞
xs ( t ) = ∑ x ( nT ) g ( t − nT )
n =−∞
s s

∞
= g (t ) ∗ ∑ x ( nT )δ ( t − nT )
n =−∞
s s

∞
= g ( t ) ∗ x ( t ) ∑ δ ( t − nTs )
n =−∞

18
 Flat-Top (Zero-Order Hold) Sampling
x ( nTs ) n-th sample of message signal x ( t ) .
Ts Sampling period
1/ Ts ≥ 2 B Sampling rate

⎧1, 0≤t ≤T
g (t ) = ⎨
⎩0, elsewhere

™ Fourier transform of xs ( t ) is given by

∞
X s ( f ) = f sG ( f ) ∑ X ( f − nf s )
n =−∞
G ( f ) = Tsinc ( fT ) exp ( − jπ fT )

19
 Flat-Top (Zero-Order Hold) Sampling

x (t )
g (t )
xs ( t ) =
⎡⎣δTs ( t ) × x( t )⎦⎤ ∗ g( t )
∞
δT (t ) =
s ∑ δ ( t − nT )
n =−∞
s

20
 Flat-Top (Zero-Order Hold) Sampling
xs ( t )

t
X(f)

Xs ( f )
f

f
−1/T 1/T 21
 Flat-Top (Zero-Order Hold) Sampling
™ Recovery of x ( t ) is accomplished by the
reconstruction LPF.

™ The reconstructed signal is processed by an

equalizer to minimize the aperture effect caused by
G ( f ).

™ Frequency response of the equalizer H eq ( f ) is

Ts
H eq ( f ) =
Tsinc (Tf )

22
 Natural Sampling
™ A natural sampled signal xs ( t ) can be expressed
in time domain by
∞
⎧1 0 ≤ t ≤τ
xs ( t ) = x ( t ) × ∑ h ( t − nTs ), h ( t ) = ⎨
n =−∞ ⎩0 otherwise

∞
™ The switching signal p (t ) = ∑ h ( t − nT )
n =−∞
s is a
periodic rectangular pulse train.

23
 Natural Sampling

x (t ) xs ( t ) = p ( t ) × x ( t )

∞
p (t ) = ∑ h ( t − nT )
n =−∞
s

24
 Natural Sampling
x (t )

p (t )

t
s (t )

25
 Natural Sampling
™ Complex exponential Fourier series of the
switching signal p ( t ) is expresses as:
∞
p (t ) = ∑c
n =−∞
n exp ( j 2π nf s t ), cn =τ f s sinc ( nτ f s ) exp ( − jπ nf s t )

™ Using the Fourier series expression for p ( t ) , it

follows that
∞
xs ( t ) = x ( t ) ∑ cn exp ( j 2π nf s t )
n =−∞

26
 Natural Sampling
™ From the frequency-shifting property, Fourier
transform of xs ( t ) becomes
∞
Xs ( f ) = ∑ c X ( f − nf )
n =−∞
n s

™ Note

1. X s ( f ) consists of an infinite number of copies of

X ( f ) shifted every f s Hz .
2. The nth copy is scaled by cn .

27
 Natural Sampling
| Xs ( f )|

−3B −B B 3B f

™ Reconstruction of x ( t ) from xs ( t ) is implemented

by means of a low-pass filter (LPF) called
reconstruction filter if f s ≥ 2 B . (Note that c0 = 1.0 ).

28
 Practical Considerations
™ Practical signals are time-limited which implies
that they are not band-limited.

™ To avoid aliasing, an anti-aliasing (prefilter) low-

pass filter processes a signal with cut-off
frequency equal to half the Nyquist rate.

™ Realizable filters require a nonzero transition

bandwidth which implies f s > 2 B (i.e sampling rate
f s must be much larger than baseband signal
bandwidth B .)

29
 Practical Considerations
™ To minimize the sampling rate, which implies
lower transmission rates and less storage memory,
small transition bandwidth of filters is desired.

™ A good engineering balance is to allow a

transition bandwidth 20% of the baseband signal
bandwidth such that f s ≥ 2.2 B .

™ 44,100 samples/s is used for a high quality

compact disc (CD) digital audio system for a music
signal with bandwidth of 20 kHz.

30
 Practical Considerations
™ A sampling rate of 8000 samples/s is used for
digital telephone systems for telephone quality
speech with bandwidth of 4 kHz.

31