ELG 3125 Signals and Systems Chapter 7

ELG3125 Signal and System Analysis

Chapter 7 Sampling

Lecture 17

1
ELG 3125 Signals and Systems Chapter 7

An analog to digital converter (ADC)

2
ELG 3125 Signals and Systems Chapter 7

3
 ELG 3125 Signals and Systems Chapter 7

The sampled output can be mathematically expressed as

∞ ∞
xp = (t ) x(t ) ∑ δ (t − kT
(t ) x(t ) p= = ) ∑ x(kT )δ (t − kT )
k =−∞ k =−∞

From the multiplication property (Section 4.5), we know that

1
X p ( jω )
= X ( jω ) ∗ P( jω )
2π

The sampling impulse train is given by

∞
=
p (t ) ∑ δ (t − kT )
k =−∞

4
 ELG 3125 Signals and Systems Chapter 7

2π
Since it is periodic and the period is T (or frequency of ωs = ), it can be represented by
T
Fourier Series. The Fourier Series coefficients are calculated by

1 T /2 − jkωs t 1 T /2 − jkωs 0 1 T /2 1
ak
T ∫− T / 2=
δ ( t )e dt
T ∫− T / 2=
δ ( t )e dt = ∫
T −T / 2
δ ( t )dt
T
The Fourier Series representation is
∞ ∞
2π ∞
=
p (t ) ∑ k
a
k =−∞
e ikωs t
jω ) ∑ 2π akδ (t − k=
⇔ P (=
k =−∞
ωs )
T
∑ δ (t − kω )
k =−∞
s

1 1 2π ∞
1 ∞
jω )
X p (=
2π
X ( jω ) * P(=
jω )
2π
X ( jω ) *
T
∑
k =−∞
δ (ω − k=
ωs ) ∑
T k =−∞
X (ω − kωs )

5
ELG 3125 Signals and Systems Chapter 7

A bandlimited signal to be sampled. Its maximum

frequency is ωM

Spectrum of a sampling impulse

train with a sampling period of T
or frequency of ωs

1 ∞
X p ( jω )
= ∑
T k =−∞
X (ωs − kωs )

Spectrum of the sampled

signal- it is periodic and the
spacing is ωs
6
 ELG 3125 Signals and Systems Chapter 7

As can be seen, we can recover the original signal by using a lowpass filter, to select the
spectrum centered at ω = 0.

H LP ( jω)

−ωc ωc ω
X r ( jω)

−ωM ωM ω
7
 ELG 3125 Signals and Systems Chapter 7

If the sampling frequency is smaller than 2ωM , the adjacent spectrums will overlap,

H LP ( jω)

−ωc ωc ω
X r ( jω)

−ωc ωc ω
If ωs < 2ωM , the original signal cannot be recovered with no distortions.
8
 ELG 3125 Signals and Systems Chapter 7

The frequency = 2ωM is the Nyquist rate.

As discussed, an ideal lowpass filter cannot be practically implemented due to the anti-
causality of the filter. In practice, a nonideal lowpass filter that approximated the desired
frequency characteristic would be used. To ensure the recovery of the original signal, the
sampling frequency is usually higher than two times of the maximum frequency of the
signal to be sampled.
9
 ELG 3125 Signals and Systems Chapter 7

For example, a high-quality music signal has a highest frequency of 20 kHz, to recover
the signal, a sampling frequency of 40 kHz would be fine. However, in practical Hi-Fi
systems, the sampling frequency is higher than 40 kHz. The following is copied from
Wikipedia.

10
 ELG 3125 Signals and Systems Chapter 7

The sampling theorem, which is most easily explained in terms of impulse-

train sampling, establishes the fact that a band-limited signal is uniquely
represented by its samples.

In practice, however, narrow, large-amplitude pulses, which approximate

impulses, are also relatively difficult to generate and transmit, and it is often
more convenient to generate the sampled signal in a form referred to as a zero-
order hold.

Such a system samples x(t) at a given instant and holds that value until the
next instant at which a sample is taken, as illustrated in Figure 7.5.

11
 ELG 3125 Signals and Systems Chapter 7

The implementation of zero-order hold: As can be seen a filter with an

impulse response of h0 (t ) is connected at the output:

12
ELG 3125 Signals and Systems Chapter 7

Note: a signal convolves with a delta

function equals to the signal itself except a
time shift due to shifted delta function.

13
 ELG 3125 Signals and Systems Chapter 7

To recover the original signal, a filter hr (t ) should be used.

As can be seen, the overall frequency response of the two cascaded filters
should be identical to that of the lowpass filter H LP (ω ) . That is,
H LP (ω ) = H O (ω ) H r (ω )

14
 ELG 3125 Signals and Systems Chapter 7

The frequency response of h0 (t ) is 1, t < T1 2sin ωT1

=x(t )  ⇔
t > T1 ω
 2sin(ωT / 2) 
− jωT /2
0,
H o (ω ) = e  
ω
(based on the shifting property of FT)
H o (ω ) H r (ω ) = H LP (ω )
H LP (ω ) e jωT /2 H LP (ω )
H r (ω ) =
=
H o (ω ) 2sin(ωT / 2)
ω
Once again, in practice the frequency response H r (ω ) cannot be exactly realized, and
thus an adequate approximation to it must be designed. In fact, in many situations, the
output of the zero-order hold is considered an adequate approximation to the original
signal by itself, without any additional lowpass filtering, and in essence represents a
possible, although admittedly very coarse, interpolation between the sample values.
15
ELG 3125 Signals and Systems Chapter 7

16
 ELG 3125 Signals and Systems Chapter 7

Example:

π π
sin ω0t ⇔ δ (ω − ω0 ) − δ (ω + ω0 )
j j

X ( jω)
2πδ(ω)
π
δ(ω − 4000π)
πδ(ω + 2000π) πδ(ω − 2000π) j

Solution:
(a) Apply FT to x(t), we have
As can be seen ωM = 4000π π
ω
− δ(ω + 4000π)
j

The Nyquist rate is ω

= NR ωM 8000π
2=
17
 ELG 3125 Signals and Systems Chapter 7

sin(4000π t ) 1, ω < 4000π

(b) x(t ) = , its FT is X ( jω ) = 
πt 0, ω > 4000π
sin Wt 1, ω <W
(Based on ⇔ X ( jω ) =  )
πt 0, ω >W
Since ωM = 4000π , the Nyquist rate is ω = NR ωM 8000π
2=

 sin(4000π t )   sin(4000π t )   sin(4000π t ) 

2

=
(c) x(t )  =    × 
 πt   πt   πt
1
Based on the convolution property, x1 (t ) × x1 (t ) ↔ X ( jω ) ∗ X ( jω )
2π
 sin(4000π t )  1, ω < 4000π
  ↔ X 1 ( jω ) =

πt  0, ω > 4000π
18
ELG 3125 Signals and Systems Chapter 7

 sin(4000π t )   sin(4000π t )  1
x=
(t )   ×  ↔ X ( jω ) ∗ X ( jω )
 πt   πt  2π
X 1 ( jω)

ω
∗
−4000π 4000π

X 1 ( jω)

−4000π 4000π
ω The convolution of two identical
spectrums will make the total spectrum

⇓ have a doubled spectral width.

X ( jω)

ωM = 2 × 4000π =
8000π The Nyquist
rate is ω
= ωM 16000π
2=
−8000π 8000π
ω NR

19
 ELG 3125 Signals and Systems Chapter 7

(final exam 2007)

Solution:
(a) The plot of X ( jω ) :
X ( jω)

ω0

0 ω0 ω
20
 ELG 3125 Signals and Systems Chapter 7

2π 3ω0
=
(b) The sampling frequency is ωs = . Assume the spectrum of the
( 4π / 3ω0 ) 2
2π ∞ 3ω0 3ω0
sampling impulse train is given by P= ( jω ) ∑
T k =−∞
δ (ω −
4
−k
2
) , which is
shifted by 3ω0 / 4 due to the spectrum of the input signal is not centered at the origin.
The sampled output:

1 X P ( jω)
=X p ( jω ) X ( jω ) ∗ P( jω )
2π
ω0
1 2π ∞  3ω0 3ω0 
= X ( jω ) ∗ ∑ δ  jω − −k 
2π T k =−∞  4 2 
−
9ω0 3ω 3ω
− 0− 0 0
3ω0 3ω0 9ω0 ω
1 ∞  3ω0 3ω0 
∑
2 4 4 4
X  jω −
4 2
−k 
T k =−∞  4 2 
21
 ELG 3125 Signals and Systems Chapter 7

(final 2002)
1
x(t )cos(5ω0t ) ↔ X ( jω ) * [πδ (ω − 5ω0 ) + πδ (ω + 5ω0 ) ]
2π
Solution:
1 1
= X ( jω − 5ω0 ) + X ( jω + 5ω0 )
2 2
X ( jω) Y ( jω)

−ωM = −ω0 / 2 ωM =ω0 / 2 ω −ω0 / 2 − 5ω0 ω0 / 2 + 5ω0

ω
The Nyquist rate = 2ωM =
2(ω0 / 2 + 5ω0 ) =
= 11ω0
22
 ELG 3125 Signals and Systems Chapter 7

Solution: from X ( jω ) ∗ X ( jω ) = 0, for ω > 15000π , we know

X ( jω ) 0, for ω > 7500π (convolution of two identical spectrums will make the
=
spectrum width doubled)

2π 2π
The sampling rate is T=10 s, the sampling frequency is ωs =
-4
= −4
= 2 × 10 4
π
T 10
To guarantee a correct recovery for the sampled signal, the sampling rate must be at least
two times the maximum frequency, that this, ωs ≥ 2ωM = 2 × 7500π = 15000π .
Considering that ωs = 2 × 104 π > 2ωM , the original signal can be correctly recovered.
23