Chapter

p 1
Sampling and Reconstruction
Ha Hoang Kha, Ph.D.Click to edit Master subtitle style
Ho Chi Minh City University of Technology
@
Email: [email protected]
 Content

™ Sampling
‰ Sampling theorem
‰ Spectrum of sampling signals

™ Antialiasing prefilter
‰ Ideal
Id l prefilter
fil
‰ Practical prefilter

™ Analog reconstruction
‰ Ideal reconstructor
‰ Practical reconstructor

Ha H. Kha 2 Sampling and Reconstruction

 Review of useful equations

x(t ) Linear system y (t ) = x(t ) ∗ h(t )

™ Linear system h(t)
X(f ) H(f) Y ( f ) = X ( f )H ( f )

Especially x(t ) = A cos(2π f 0t + θ )

™ Especially,
y (t ) = A | H ( f 0 ) | cos(2π f 0t + θ + arg( H ( f 0 )))
1
™ Fourier transform: cos(2π f 0t ) ←⎯→ [δ ( f + f 0 ) + δ ( f − f 0 )]
FT

2
1
sin(2π f 0t ) ←⎯→
← FT → j[δ ( f + f 0 ) − δ ( f − f 0 )]
2
1
™ Trigonometric formulas: cos(a) cos(b) = [cos(a + b) + cos(a − b)]
2
1
sin( a) sin(b) = − [cos( a + b) − cos( a − b)]
2
1
sin( a) cos(b) = [sin( a + b) + sin( a − b)]
2
Ha H. Kha 3 Sampling and Reconstruction
 1. Introduction

™ A typical signal processing system includes 3 stages:

™ The
Th analog
l signal
i l is
i digitalized
di i li d b by an A/D converter
™ The digitalized samples are processed by a digital signal processor.
‰ The digital processor can be programmed to perform signal processing
operations such as filtering, spectrum estimation. Digital signal processor can be
a general purpose computer, DSP chip or other digital hardware.
™ The resulting output samples are converted back into analog by a
D/A converter.

Ha H. Kha 4 Sampling and Reconstruction

 2. Analog to digital conversion

™ Analog to digital (A/D) conversion is a three-step process.

x(t) Sampler ( ) ( ) Quantizer xQ(n) Coder

x(nT)≡x(n) 11010

t=nT
A/D converter

x(t) x(n) 111 xQ(n)

110
101
100
011
t n 010 n
001
000

Ha H. Kha 5 Sampling and Reconstruction

 3. Sampling

™ Sampling is to convert a continuous time signal into a discrete time

signal The analog signal is periodically measured at every T seconds
signal.

™ x(n)≡x(nT)=x(t=nT),
( ) ( ) ( ), n=….-2,, -1,, 0,, 1,, 2,, 3……..
™ T: sampling interval or sampling period (second);
™ fs=1/T: sampling rate or sampling frequency (samples/second or
Hz)
Ha H. Kha 6 Sampling and Reconstruction
 3. Sampling-example 1

™ The analog signal x(t)=2cos(2πt) with t(s) is sampled at the rate fs=4
Hz. Find the discrete-time signal
g x(n)
( )?
Solution:

™ x(n)≡x(nT)=x(n/fs)=2cos(2πn/fs)=2cos(2πn/4)=2cos(πn/2)

n 0 1 2 3 4
x(n) 2 0 ‐2 0 2

™ Plot the signal

Ha H. Kha 7 Sampling and Reconstruction

 3. Sampling-example 2

™ Consider the two analog sinusoidal signals

7 1
x1 (t ) = 2 cos(2π t ),
) x2 (t ) = 2cos(2π t ); t (s)
8 8
These signals are sampled at the sampling frequency fs=1 Hz.
Fi d the
Find h discrete-time
di i signals
i l ?
Solution:
1 71 7
) = 2 cos(2π
x1 (n) ≡ x1 (nT ) = x1 (n n) = 2 cos( π n)
fs 81 4
1 π
= 2 cos((2 − )π n) = 2 cos( n)
4 4
1 11 1
x2 (n) ≡ x2 (nT ) = x2 (n ) = 2 cos(2π n) = 2 cos( π n)
fs 81 4
™ Observation: x1(n)=x2(n) Æ based on the discrete-time signals, we
cannot tell which of two signals are sampled ? These signals are
called “alias”
Ha H. Kha 8 Sampling and Reconstruction
 3. Sampling-example 2

f2=1/8 Hz f1=7/8 Hz

fs=1 Hz

Fig: Illustration of aliasing

Ha H. Kha 9 Sampling and Reconstruction

 3. Sampling-Aliasing of Sinusoids

™ In general, the sampling of a continuous-time sinusoidal signal

x(t ) = A cos(2π f 0t + θ ) at a sampling
p g rate fs=1/T
/ results in a discrete-
time signal x(n).
™ The sinusoids xk (t ) = A cos(2π f k t + θ ) is sampled at fs , resulting in a
discrete time signal xk(n).
k=0 ±1,
™ If fk=f0+kfs, k=0, ±1 ±2,
±2 …., then x(n)=xk(n) .

Proof: ((in class))

™ Remarks: We can that the frequencies fk=f0+kfs are indistinguishable

from the frequency f0 after sampling and hence they are aliases of f0

Ha H. Kha 10 Sampling and Reconstruction

 4. Sampling Theorem-Sinusoids

™ Consider the analog signal x(t ) = A cos(Ωt ) = A cos(2π ft ) where Ω is

the frequency
q y (rad/s)
( ) of the analogg signal,
g , and f=Ω/2π is the
frequency in cycles/s or Hz. The signal is sampled at the three rate
fs=8f, fs=4f, and fs=2f.

Fi Sinusoid
Fig: Si id sampled
l d at diff
different rates
f s samples / sec samples
™ Note that = =
f cycles
l / sec cycle
l
™ To sample a single sinusoid properly, we must require f s ≥ 2 samples
f cycle
Ha H. Kha 11 Sampling and Reconstruction
 4. Sampling Theorem
™ For accurate representation of a signal x(t) by its time samples x(nT),
two conditions must be met:
1) The signal x(t) must be bandlimitted, i.e., its frequency spectrum must
be limited to fmax .

Fig:
g Typical
yp bandlimited spectrum
p

2) The sampling rate fs must be chosen at least twice the maximum

frequency fmax. f s ≥ 2 f max

™ fs=2fmax is called Nyquist rate; fs/2 is called Nyquist frequency;

[ fs/2,
[-f /2 fs/2] is
i Nyquist
N i interval.
i l
Ha H. Kha 12 Sampling and Reconstruction
 4. Sampling Theorem

™ The values of fmax and fs depend on the application

Application
pp fmax fs
Biomedical 1 KHz 2 KHz
Speech 4 KHz 8 KHz
Audio 20 KHz 40 KHz
Video 4 MHz 8 MHz

Ha H. Kha 13 Sampling and Reconstruction

 4. Sampling Theorem-Spectrum Replication
∞ ∞

™ Let x(nT ) = x (t ) = x(t ) ∑ δ (t − nT ) = x(t ) s(t ) where s(t ) = ∑ δ (t − nT )
n =−∞ n =−∞

™ s(t) is periodic, thus, its Fourier series are given by

∞
1 1 1
s (t ) = ∑
n =−∞
S n e j 2π f s nt where S n =
TT∫ δ (t )e − j 2π f s nt
dt = ∫ δ (t )dt =
TT T
1 ∞ j 2π f s nt
Thus, s(t ) = ∑ e
T n =−∞
∞
 1
which results in x (t ) = x(t ) s(t ) = ∑ x(t )e j 2π nf st
T n =−∞
  1 ∞
™ Taking the Fourier transform of x (t ) yields X ( f ) = ∑ X ( f − nf s )
T n =−∞

™ Observation: The spectrum of discrete-time signal is a sum of the

original spectrum of analog signal and its periodic replication at the
i
interval
l fs.
Ha H. Kha 14 Sampling and Reconstruction
 4. Sampling Theorem-Spectrum Replication

™ fs/2 ≥ fmax

Fig: Spectrum replication caused by sampling

Fi Typical
Fig: T i l badlimited
b dli i d spectrum
™ fs/2 < fmax

Fig: Aliasing caused by overlapping spectral replicas

Ha H. Kha 15 Sampling and Reconstruction
 5. Ideal Analog reconstruction

Fig: Ideal reconstructor as a lowpass filter

™ An ideal reconstructor acts as a lowpass filter with cutoff frequency

equal to the Nyquist frequency fs/2.
⎧T f ∈ [− f s / 2, f s / 2]
™ An ideal reconstructor (lowpass filter) H ( f ) = ⎨
⎩0 otherwise
othe wise
Then  
X a ( f ) = X ( f )H ( f ) = X ( f )

Ha H. Kha 16 Sampling and Reconstruction

 5. Analog reconstruction-Example 1

™ The analog signal x(t)=cos(20πt) is sampled at the sampling

frequency fs=40 Hz.
a) Plot the spectrum of signal x(t) ?
b) Find
Fi d the
h discrete
di time
i signal
i l x(n)
( )?
c) Plot the spectrum of signal x(n) ?
d) The signal x(n) is an input of the ideal reconstructor,
reconstructor find the
reconstructed signal xa(t) ?

Ha H. Kha 17 Sampling and Reconstruction

 5. Analog reconstruction-Example 2

™ The analog signal x(t)=cos(100πt) is sampled at the sampling

frequency fs=40 Hz.
a) Plot the spectrum of signal x(t) ?
b) Find
Fi d the
h discrete
di time
i signal
i l x(n)
( )?
c) Plot the spectrum of signal x(n) ?
d) The signal x(n) is an input of the ideal reconstructor,
reconstructor find the
reconstructed signal xa(t) ?

Ha H. Kha 18 Sampling and Reconstruction

 5. Analog reconstruction

™ Remarks: xa(t) contains only the frequency components that lie in the
Nyquist interval (NI) [-f
[ fs//2,
//2 fs/2].
/2]

sampling
p g at fs ideal reconstructor
™ x(t), f0 ∈ NI ------------------> x(n) ----------------------> xa(t), fa=ff0

sampling at fs ideal reconstructor

™ xk(t), fk=f0+kfs------------------> x(n) ----------------------> xa(t), fa=f0

™ The
Th frequency
f fa off reconstructedd signal
i l xa(t)
( ) is
i obtained
b i d byb adding
ddi
to or substracting from f0 (fk) enough multiples of fs until it lies
within the Nyquist interval [[-ffs//2,
//2 fs/2]..
/2] That is
f a = f mod( f s )

Ha H. Kha 19 Sampling and Reconstruction

 5. Analog reconstruction-Example 3

™ The analog signal x(t)=10sin(4πt)+6sin(16πt) is sampled at the rate 20

Hz. Findd the
h reconstructed
d signall xa(t) ?

Ha H. Kha 20 Sampling and Reconstruction

 5. Analog reconstruction-Example 4

™ Let x(t) be the sum of sinusoidal signals

x(t)=4+3cos(πt)+2cos(2πt)+cos(3πt)
x(t) 4+3cos(πt)+2cos(2πt)+cos(3πt) where t is in milliseconds.
milliseconds
a) Determine the minimum sampling rate that will not cause any
aliasing effects ?
b) To observe aliasing effects, suppose this signal is sampled at half its
Nyquist rate. Determine the signal xa(t) that would be aliased with
x(t) ? Plot the spectrum of signal x(n) for this sampling rate?

Ha H. Kha 21 Sampling and Reconstruction

 6. Ideal antialiasing prefilter

™ The signals in practice may not bandlimitted, thus they must be

f l d by
filtered b a lowpass
l fl
filter

Fi Ideal
Fig: Id l antialiasing
ti li i prefilter
p filt

Ha H. Kha 22 Sampling and Reconstruction

 6. Practical antialiasing prefilter

™ A lowpass filter: [-fpass, fpass] is the frequency range of interest for the
application
l (ffmax=ffpass)
™ The Nyquist frequency fs/2 is in the middle of transition region.
™ The stopband frequency fstop and the minimum stopband attenuation
Astop dB must be chosen appropriately to minimize the aliasing
effects.
effects

f s = f pass + f stop

Fig: Practical antialiasing lowpass prefilter

Ha H. Kha 23 Sampling and Reconstruction
 6. Practical antialiasing prefilter

™ The attenuation of the filter in decibels is defined as

H( f )
A( f ) = −20 log10 (dB)
H ( f0 )

where f0 is a convenient reference frequency, typically taken to be at

p filter.
DC for a lowpass
™ α10 =A(10f)-A(f) (dB/decade): the increase in attenuation when f is
g byy a factor of ten.
changed
™ α2 =A(2f)-A(f) (dB/octave): the increase in attenuation when f is
changed by a factor of two.
™ Analog filter with order N, |H(f)|~1/fN for large f, thus α10 =20N
(dB/decade) and α10 =6N (dB/octave)

Ha H. Kha 24 Sampling and Reconstruction

 6. Antialiasing prefilter-Example

™ A sound wave has the form

x(t ) = 2 A cos(10π t ) + 2 B cos(30π t ) + 2C cos(50π t )
+ 2 D cos(60π t ) + 2 E cos(90π t ) + 2 F cos(125π t )

where t is in milliseconds. What is the frequency content of this

g ? Which parts
signal p of it are audible and whyy ?
This signal is prefilter by an anlog prefilter H(f). Then, the output y(t)
of the p
prefilter is sampled
p at a rate of 40KHz and immediatelyy
reconstructed by an ideal analog reconstructor, resulting into the final
analog output ya(t), as shown below:

Ha H. Kha 25 Sampling and Reconstruction

 6. Antialiasing prefilter-Example

Determine the output signal y(t) and ya(t) in the following cases:
a)When there is no prefilter, that is, H(f)=1 for all f.
b)When H(f) is the ideal prefilter with cutoff fs/2=20 KHz.
c)When H(f) is a practical prefilter with specifications as shown
below:

The filter’s phase response is assumed to be ignored in this example.

Ha H. Kha 26 Sampling and Reconstruction

 7. Ideal and practical analog reconstructors

™ An ideal reconstructor is an ideal lowpass filter with cutoff Nyquist

f
frequency f/
fs/2.

Ha H. Kha 27 Sampling and Reconstruction

 7. Ideal and practical analog reconstructors

sin(π f s t)
™ The ideal reconstructor has the impulse response: h(t ) =
which
h h is not realizable l response is not casualπl f s t
l bl since its impulse

™ It is practical to use a
staircase reconstructor

Ha H. Kha 28 Sampling and Reconstruction

 7. Ideal and practical analog reconstructors

Fig: Frequency response of staircase recontructor

Ha H. Kha 29 Sampling and Reconstruction

 7. Practical reconstructors-antiimage postfilter

™ An analog lowpass postfilter whose cutoff is Nyquist frequency fs/2

is usedd to remove the
h surviving spectrall replicas.
l

Fig: Analog anti-image postfilter

Fi Spectrum
Fig: S after
f postfilter
fil
Ha H. Kha 30 Sampling and Reconstruction
 8. Homework

™ Problems: 1.2, 1.3, 1.4, 1.5, 1.9

Ha H. Kha 31 Sampling and Reconstruction