Chapter 4

Chapter 4.
Sampling of
Continuous-Time Signals
4.0 Introduction
4.1 Periodic Sampling
4.2 Frequency Representation of Sampling
4 3 Reconstruction from Discrete-Time
4.3 Discrete Time Samples
4.4 Changing the Sampling Rate
4.5 Digital Processing of Analog Signals
Digital Signal Processing 4-1

4.0 Introduction
Question to answer: how to approximate a continuous
(analog) linear system by a digital system?
Notations:
Signals: continuous-time discrete-time
time-domain xc (t ) x[n ]
frequency-domain Xc( j ) X (e j )
Systems: continuous-time discrete-time
time-domain h[n ]
hc (t )
frequency-domain
Hc ( j ) H (e j )
4.1 Periodic Sampling
A typical method of obtaining a discrete-time representation of a
continuous-time signal is through periodic sampling.
x[n ] = xc (nT ), - <n< .

T is the sampling period
is the sampling frequency (samples per second)
fs = 1/ T
is the sampling frequency (radians per second)
s = 2 /T
An ideal continuous-to-discrete-time (C/D) converter

Mathematical Representation of Sampling
s(t )
Conversion
from impulse
xc (t ) train to discrete- x[n] = xc (nT )
xs (t ) time sequence
s(t ) = (t nT ) (the periodic impulse train)

n=
x s ( t ) = xc ( t ) s ( t ) = xc ( t ) (t nT ) (modulation)
n=
xs (t ) = xc ( nT ) (t nT ) (sifting property)
n=

Periodic Sampling Examples

4.2 Frequency-Domain Representation
of Sampling: Time-Domain
We modulate the periodic impulse train with the original
continuous-time signals, obtaining
xs (t ) = xc (t ) s (t )
= xc (t ) (t nT )
n=
= xc (nT ) (t nT )
n=

Frequency-Domain Representation
Given the Fourier transform of the impulse train as
2 2
s (t ) = (t nT ) S( j ) = ( k s ) (where s = )
n= T k= T
Since
1
x s ( t ) = xc ( t ) s ( t ) Xs( j ) = X c ( j ) * S( j )
2
Then
1 2
Xs( j ) = Xc( j )* ( k s )
2 T k=
1
= X c ( j( k s ))
T k=
Observations of Frequency-Domain
Representation of Sampling
This equation provides the relationship between the Fourier
transform of continuous-time signal and discrete-time signal
1
Xs( j ) = X c ( j( k s ))
T k=
X s ( j ) consists of periodically repeated and scaled copies of the

Fourier transform of xc (t ), i.e., X c ( j . )
The copies of X c ( j ) are shifted by integer multiples of the
sampling frequency s
.
All copies of replicated spectrums are superimposed to produce
the Fourier transform of the sampled signal.

Sampling Rate and Bandwidth
Given the signal of band-limited
X ( j ) = 0, > N
There is no overlap between replicated spectrums, when

we
e have
a e tthe
e sa
sampling
p g rate
ate as following
o o g
s >2 N
That means we CAN reconstruct the continuous-time signal with
an ideal low-pass filter.
There will be aliasing distortion, or aliasing when
s < 2 N
That means we CANNOT reconstruct the continuous-time signal
from its samples.

What is aliasing?
Sampling Without Aliasing Sampling With Aliasing

How to Reconstruct a Signal?
Ideal low-pass Filtering

How to Reconstruct a Signal? (Cont'd)
sampling
Original Signal Discrete-Time

Discrete Time Signal
Ideal low-pass
filtering
Ideal low-pass Filtering Reconstructed Signal

Sampling and Reconstruction Example
Given a signal xc (t ) = cos 0 t
What is the Fourier transform of the given signal?
Use the Euler equation, we know that
1 j
xc (t ) = cos 0t = e
2
( 0t
+e j 0t
)
According to continuous Fourier transform, we know
x (t ) = e j 0t
X( j ) = 2 ( 0 )
Therefore, the Fourier transform of the given signal is
Xc( j ) = ( ( 0 )+ ( + 0 ))
(No Aliasing)
Original Signal
Sampled Signal
Reconstructed Signal
2
(
1 j
xr ( t ) = e 0t
+e j 0t
) = cos 0 t
(With Aliasing)
Original Signal
Sampled Signal
Reconstructed Signal
1 j(
xr ( t ) = e
2
( s 0 )t
+e j( s 0 )t
) = cos( s 0 )t
Nyquist Sampling Theorem
Suppose that xa (t ) X a ( ) is band-limited to a frequency
interval [ N , N ], i.e.,
X ( ) = 0 for N
Then x(t)
( ) can be exactly y reconstructed from equidistant
q
2
samples xd [n ] = xa ( nTs ) = xa ( 2 n / s ), if s = 2 N ,
Ts
where T = 2 / is the sampling period, s is the
s s
sampling frequency (radians per second), N
is referred
to as the Nyquist frequency, and 2 N is called the Nyquist
rate.

How to obtain discrete-time Fourier
transform (DTFT)?
Given the sampled signal as
xs ( t ) = xc ( nT ) (t nT )
n=
Since we have the following continuous-time Fourier
transform (CTFT) pair
(t nT ) e j nT
Thus we have the continuous-time Fourier transform of the

sampled signal as
Xs( j ) = xc ( nT )e j Tn
n=

transform (DTFT)? (Cont'd)
Since we know the relationship between the sampled
signal xc (nT ) and the discrete-time sequence x[n ]
x[n ] = xc (nT )
We also have the DTFT of x[n ] is defined as
X (e j ) = x[n ]e j n
n=
By comparing with
n=

transform (DTFT)? (Cont'd)
As we compare the following two equations
X (e j ) = x[n ]e j n
n= n=
1
X s ( j ) = X (e j ) = X ( e j T ). Xs( j ) = X c ( j( k s ))
= T T k=
Continuous-time Fourier
1 transform (CTFT) of the
X (e j T
)= X c ( j( k s )) sampled signal xs (t ) .
T k=
( = T) What you can see?
1 2 k Discrete-time Fourier
X (e ) =
j
Xc j
T k= T T transform (DTFT) of the
discrete-time signal x[n ] .
Example 4.1 (Without Aliasing)
If we sample the continuous-time signal xc (t ) = cos(4000 t )
with sampling period T=1/6000.
Continuous-time Fourier transform X s ( j )
Discrete-time Fourier transform X (e j )
Problem Analysis
Fourier transform of the original signal 0 = 4000 .
Xc( j ) = ( 4000 ) + ( + 4000 )
s = 2 / T = 12000 .
Sampling frequency
Fourier transforms of the sampled signal
1 1 2
Xs( j ) = X c ( j( k s )) X (e j ) = X c ( j( k ))
T k= T k= T T

Example 4.1 (Cont'd)
( = T) ( /T) = T ( )

Example 4.2 (With Aliasing)
Problem Analysis
Fourier transform of the original signal 0 = 16000 .
Xc( j ) = ( 16000 ) + ( + 16000 )
s = 2 / T = 12000 .
Sampling frequency
Fourier transforms of the sampled signal are exactly same as the
previous one, why?
x[n ] = cos(16000 n / 6000) = cos(2 n + 4000 n / 6000) = cos(2 n / 3)

k=-2 k=1 k=-1 k=2 k=0
k=0
( = T) ( /T) = T ( )
0 = 16000

Example 4.3 (with Aliasing)
Problem Analysis
Fourier transform of the original signal
Xc( j ) = ( 4000 ) + ( + 4000 )

Sampling frequency s = 2 / T = 3000 .
The discrete-time Fourier transform is the same as previous one.
Why?
cos(4000 n / 1500) = cos(2 n + 1000 n / 1500) = cos(2 n / 3)

k=-2 k=1 k=-1 k=2 k=0
k=0
( = T) ( /T) = T ( )

4.3 Reconstruction of a Band-limited
Signal from Its Samples
If the conditions of the sampling theorem are met, and if the
modulated impulse train is filtered by an appropriate low-pass
filter, then the Fourier transform of the filter output will be
identical to the Fourier transform of the original signal.
Given a sequence of samples x[n], we form the impulse train
xs ( t ) = x[n ] (t nT )
n=
If the impulse train is the input to an ideal low-pass continuous-
time filter with impulse response hr (t )
xr (t ) = xs (t ) * hr (t ) = x[n ] (t nT ) * hr (t ) = x[n ]hr (t nT )

n= n=

4.3.1 Ideal Reconstruction System

Ideal Reconstruction System (Cont'd)
Frequency Response Impulse Response
sin( t / T ) t
hr (t ) = = sinc
t /T T

Ideal Reconstruction System (Cont'd)
The ideal reconstruction system is denoted by
xr ( t ) = x[n ]hr (t nT )
n=
sin( t / T ) t
hr (t ) = = sinc
t /T T
sin[ (t nT ) / T ]
xr ( t ) = x[n ]
n= (t nT ) / T
If x[n ] = xc (nT ) and X c ( j ) = 0 for /T = s / 2 then
we have
x r ( t ) = xc ( t )
Ideal Band-limited Interpolation
The ideal low-pass filter interpolates between the impulses
of x[n ] to construct a continuous-time signal
sin[ (t nT ) / T ]
xr ( t ) = x[n ]
n= (t nT ) / T
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 which
is not realizable in practice. Finite length low-pass filtering
will result in some reconstruction error.

Ideal Band-limited Interpolation (Cont'd)
Original continuous
continuous-time
time signal
Sampled signal
Reconstructed signal

Ideal D/C Converter

Ideal D/C Converter (Cont'd)
The properties of the ideal D/C converter are most easily
seen in the frequency-domain.
sin[ (t nT ) / T ]
xr ( t ) = x[n ]hr (t nT ) xr ( t ) = x[n ]
n= n= (t nT ) / T
Linearity of continuous-time
Xr( j ) = x[n ]H r ( j )e j Tn
. Fourier transform
n= Time shifting leads to an

exponential factor in the
Fourier transform
X r ( j ) = H r ( j ) X (e j T
). Discrete-time Fourier
Transform (DTFT) of x[n]

Can you get the original signal back?
The ideal low-pass filter selects the base period of the resulting
periodic Fourier transform X ( e j T ) and compensates for the
1/T scaling inherent in sampling.
If the sequence x[n] has been obtained by sampling a band-
limited signal
g at the Nyquist
yq rate or higher,
g , 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 always
band-limited to at most the cut-off frequency of the low-pass
filter, which is taken to one-half the sampling frequency.

89:;<= >?@A B?CD?EF<: 89::G >?@A HAIJK= 4-4

x[n] = xc (nT )
1 2 k
X (e ) = j
Xc j
T k= T T
sin[ (t nT ) / T ]
Yr (t ) = y[n]
n= (t nT ) / T
Yr ( j ) = H r ( j )Y (e j T
)
TY (e j T ),
= T
0, otherwise

>?@A ?L K<M=?EK<<N9 OPQ 89::G >?@A B?C@9:<: 1-4-4
:A?GM FGHI GM @KNOGP@KKQC R STU 4-11 =>? @A BCDDE FGHI JCDKD @EL
Y ( e ) = H ( e ) X (e )
j j j
Yr ( j ) = H r ( j ) H (e j T
) X (e j T
)
= T
1 2 k
Yr ( j ) = H r ( j ) H (e j T
) Xc j
T k= T T

@EL
Xc( j ) = 0 for
T
H (e j T
)Xc( j )
Yr ( j ) = T
0,
T
FGHI GM @KNOGP @KKQC STU JCDKD WK =AGXH SYZ BCDRKO JCDKD SPXK
.AR? SH FGKM [R\ BTML@ GM F] @^RH SDPGZ@\ _DGO BZ `DL
:A@LA SbCDM =HGc RA BM JCDKD FKL LTI @GC\@
.A?GM LTI AKGM BCDDE FGHI JCDKD
.A?GM ARAdH APGM AKGM eAR@R =GPbKD

4.4.2 Impulse Invariance
FGHI JCDKD WHZ BM L@ SfU?H =KAMC gMGC GM BCDRKO FGHI JCDKD WK JKhLRU SH AKPZ i@\
JKPZ eIGD jAGKO BCDDE

H ( j ) band lim ited H (e j ) = ? sothat H eff ( j ) = H c ( j )
kRfUM
H (e ) = H c (
j j ), ,
T
BZ eRdP BM T lGUCPL GM
H c ( j ) = 0,
T
BTML@ BCDDE JCDKD BM@m _DGO R BCDRKO JCDKD BM@m _DGO FKM @EL
A?GM @L@n@M @KI
h[n] = Thc (nT )
FGHI JCDKD IL impulse-invariant BUDP WK L@ BCDDE FGHI JCDKD
.JKKRE BCDRKO
2 k
H (e ) = j
Hc j
k= T T
H (e ) = H c (
j j )
), ,
T

4.5 Continuous –Time Processing of
Discrete-Time Signals
x[n] = xc (nT )
sin[ (t nT ) / T ]
Yr (t ) = y[n]
n= (t nT ) / T
y[n] = yc (nT )
sin[ (t nT ) / T ] sin[ (t nT ) / T ]
X c (t ) = x[n] Yc (t ) = y[n]
n= (t nT ) / T n= (t nT ) / T

X c ( j ) = TX (e j T
)
T
Yc ( j ) = H c ( j ) X c ( j )
T
1
Y (e ) = Yc ( j )
j
T T
H (e ) = H c ( j
j
)
T
oA?GMP pKdf GCqA @EL

( M + 1)
1 sin( )
2 e j M
H (e ) =
j 2
M +1 sin( )
2
oA?GM A@\ M @EL

89::G >?@A HAIJK= ?L BKIJKL 8E;@E UKE K<<N9 4-6
x[n] = xc (nT )
x'[n] = xc (nT ' )
T T'
F] IL AAsH e@LA@M BPRHP R BCDRKO FGHI =GPbKD eIGDIGM tR@ WK

tR@ FKL @A jAGuCDL A@RH LIsL FARM =] jAKL @Kv =KqA BM GHL A?GM SH
@wP @A @HL FKL JGsPL eL@M BCDDE xHGZ S?R@ BZ `DL @C lDGPH
.AR? BC\@E

V<WX JJY Z< [L:E 8L BKIJKL 8E;@E UKE H\?] 4-6-1
xd [n] = x[nM ] = xc (nMT )

e@LA@M BPRHP y@P IGD jA@?\

1 2 k
X (e ) =j
Xc j
T k= T T
1 2 r
X d (e ) = j
Xc j
T ' r= T' T'
T ' = MT
1 2 r
X d (e ) = j
Xc j
MT r= MT MT
r = i + kM
k
0 i M 1
1 M 1
$1 2 k 2 i !
X d (e ) =
j
" Xc j
M i =0 # T k = MT T MT
j( 2 i) 1 2 i 2 k
X (e M
)= Xc j
T k= MT T
M 1
1 j( 2 i )
X d (e ) =
j
X (e M M
M i =0

Frequency-domain illustration of downsampling

(a)-(c) Downsampling with aliasing. (d)-(f) Downsampling
with prefiltering to avoid aliasing.

A general system for downsampling by a factor of M
Decimator - Decimation

4.8.4 Zero-order Hold D/A Conversion
An analog system is well-approximated digitally only if the
digital output is carefully transformed into analog form.
yd [n ] %d ( e j ) g a ((t ) %a ( ) ya (t )
Comments:
Compensation with either %d ( e ) or %a ( ) - not both.
j
The D/A block g a (t ) is not filtering - it is weighting.

No compensation is needed if g a (t ) is the ideal reconstructor.

Impulse Response of Zero-order Hold
This is what is usually done in practice, here
g a (t )
1
where
ya (t ) = xd [n ]g d (t nTs )
n=
0 Ts
It holds y [n ] at a constant level over each sampling period.
d
yd [n ] Z.O.H ya (t )
1 1
-3 -2 -1 0 1 2 3 n -3 -2 -1 0 1 2 3 n
Note: Z.O.H. introduces high frequencies, see sharp edges.
Frequency Response of ZOH
Frequency response of ZOH is a sinc function.
sin( Ts / 2)
Ga ( j ) = e j Ts / 2
( / 2)
The high frequencies in the reconstructed signal (sharp steps)
are introduced from side-lobes as follows.
Ga ( ) Ideal
Z.O.H
2 2
Ts Ts Ts Ts
Frequency response of Z.O.H
Compensation of ZOH
The phase response of ZOH corresponds to an advance time
shit of T/2 which cannot be compensated and usually neglected.
The magnitude response can be compensated as follows.
( Ts / 2)
; <
%a ( ) = sin(
i ( Ts / 2) Ts %a ( )
0; else 1
2 2
Ts Ts Ts Ts
Ideal compensation reconstruction filter

Physical Configuration for
ZOH D/A Conversion
The D/A converter followed by an ideal compensated

reconstruction filter is shown as follows.

4.4 Changing the Sampling Rate
It is often necessary to change the sampling rate of a discrete-
time signal, i.e., to obtain a new discrete-time representation of
the underlying continuous-time signals.
x[n ] = xc ( nT ) and x ' [n ] = xc ( nT ' ) where T T'
It is of interest to consider methods of changing the sampling

rate that involve discrete-time operations.
x[n ] & x ' [n ]

Sampling Rate Change Examples
(Down-sampling)
What happened during

down-sampling?

4.4.1 Sampling Rate Reduction by an
Integer Factor (Down-sampling)
The sampling rate of a sequence can be reduced by

"sampling it" by defining a new sequence
xd [n ] = x[nM ] = xc ( nMT ).

Frequency Representation of Down-Sampling
First recall that the DTFT of x[n ] = xc ( nT ) is

1 2 k
X (e ) = j
Xc j
T k= T T
Similarly the DTFT of

Similarly, xd [n ] = x[nM ] = xc ( nMT ) is
1 2 r
X d (e ) =j
Xc j
MT r= MT MT
Questions: what is the relationship between them?
X (e j ) X d (e j )
Frequency Representation of
Down-Sampling (Cont'd)
We can represent
1 2 r
X d (e ) =j
Xc j
MT r= MT MT
( r = i + kM )
1 M 1
$1 2 ( kM + i ) !
X d (e ) =
j
" Xc j
M i =0 # T k= MT MT
1 M 1
$1 2 k 2 i !
X d (e ) =
j
" Xc j
M i =0 # T k= MT T MT

We then have
1 M 1
$1 2 i 2 k !
X d (e ) =
j
" Xc j
M i =0 # T k = MT T
We know that
1 2 k
X (e j ) = Xc j (DTFT from CTFT)
T k= T T
$1 !
X (e ) = "T X c j
j( 2 i) / M 2 i 2 k
# k= MT T
Therefore, we have
X (e )
M 1
1
X d (e ) =
j j( 2 r) / M
M r =0
We can have the following conclusions by observing
X (e j ( )
M 1
1
X d (e ) =
j 2 r) / M
M r =0
There is a strong analogy between X d ( e j ) and X ( e j ).

)
X d ( e j ) can be composed of M copies of the periodic Fourier
transform X ( e j ) , frequency scaled by M and shifted by integer
multiples of 2 /M.
X d ( e j ) is periodic with period 2 .
Aliasing can be avoided if
X (e j ) = 0, N (band - limited)
2 /M 2 N (narrow - banded)
2 /T = 4 N
s =4 N
N = N T= /2
-4 4

Down-Sampling: Example

Down-sampling after Pre-filtering
If aliasing occurs during down-sampling, we need to reduce

the band-width of signal x[n] prior to down-sampling.
Signal x[n] will be pre-filtered by an ideal low-pass filter with
cut-off frequency /M.

Down-sampling after Pre-filtering (Example)
sin n sin( n / M )
hlp [n ] = c
=
n n

4.4.2 Increasing the Sampling Rate
by an Integer Factor
Consider a signal x[n] whose sampling rate we wish to

increase by a factor of L (up-sampling).
x[n / L], n = kL and k ' Z
xe [ n ] =
0, otherwise
For example, x[n ] = (1 2 3 4 5)
( n = 0, 1, 2, 3, 4)
xe [n ] = (1 0 2 0 3 0 4 0 5 0 )
( n = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 and L = 2)
Or equivalently,
xe [ n ] = x[k ] [n kL ] (not LTI convolution)
k=
Sampling Rate Change Examples
(Up-sampling)
Image 128x128 Up-sampled Image Up-sampled Image

256x256 after linear interpolation

Frequency Representation of Up-Sampling
The Fourier transform (DTFT) of the up-sampled signal is
X e (e j ) = (x[k ] [n kL]) e j n
n= k=
= x[k ]e j kL
= X (e j L )
k=
We can see the Fourier transform of the output of the

expander is a frequency-scaled version of the Fourier
transform of the input, i.e, is replaced by L.

(Example)

General System for Up-sampling
To fill missing samples, the operation of up-sampling is

therefore considered to be synonymous with interpolation.
sin n sin n / L
hlp [n ] = L ( c
=
n n/L
Gain

Ideal Low-pass Filtering after Up-sampling
As in the case of D/C converter, it is possible to obtain an

interpolation formula with an ideal low-pass filter as
sin[ ( n kL ) / L]
xi [n ] = x[k ]hlp [n kL ] = x[k ]
k= k= ( n kL ) / L
The impulse response of the low pass filter has properties

sin( n / L) hi [0] = 1
hi [n ] = &
n/L hi [n ] = 0 , n = ± L,±2 L,±3L,....
Thus for the ideal low-pass interpolation filter, we have
xi [n ] = x[n / L], n = ± L,±2 L,±3L,...

and Ideal Low-pass Filtering (Example)

Linear Interpolation after Up-sampling
Linear interpolation can be accomplished by the
1 n / L, n L
hlin =
0, otherwise

(Example)
2
1 $ sin( L / 2) !
H lin ( e j ) = "
L # sin( / 2)

(Example, Cont'd)
Please note that

1 n / L, n L hi [0] = 1
hlin [n ] = &
0, otherwise hi [n ] = 0 , n = ± L,±2 L,±3L,....
So that xi [n ] = x[k ]hlin [n kL ]

k=
xlin [n ] = x[n / L], n = ± L,±2 L,±3L,...

The amount of distortion in the intervening samples can be
gauged by comparing the frequency response of the linear
interpolator with that of the ideal low-pass interpolator, as
2
1 $ sin( L / 2) !
H lin ( e ) = "
j
L # sin( / 2)
Changing the sampling rate by a noninteger factor

Multirate Digital Signal Processing
Basic Sampling Rate Alteration Devices

FGHAPL@ tKLI\L `}s e@LA@M BPRHP y@P thGZ R tKLI\L IL jAGuCDL
=GPbKD tILA@O eG}HCDKD
Up-sampler - Used to increase the sampling rate by an

Up-
integer factor
Down--sampler - Used to decrease the sampling rate by
Down
an integer factor

Up
Up--Sampler
Time-Domain Characterization
Time-
An up-sampler with an upup--sampling factor L, where L is a
positive integer, develops an output sequence xu [n]
with a sampling rate that is L times larger than that of the
input sequence x[n]
Block-diagram representation
x[n] L xu [n ]

Up
Up--Sampler
Figure below shows the up-sampling by a factor of 3 of a

sinusoidal sequence with a frequency of 0.12 Hz
Input Sequence Output sequence up-sampled by 3

1 1
0.5 0.5
Amplitude
0 Amplitude 0
-0.5 -0.5
-1 -1
0 10 20 30 40 50 0 10 20 30 40 50
Time index n
Time index n
Up
Up--Sampler
In practice, the zero-valued samples inserted by the up-
sampler are replaced with appropriate nonzero values using
some type of filtering process
Process is called interpolation

Down
Down--Sampler
Time-Domain Characterization
Time-
An down-sampler with a down
down--sampling factor M, where
M is a positive integer, develops an output sequence y[n]
with a sampling rate that is (1/M)-th of that of the input
sequence x[n]
Block-diagram representation
x[n] M y[n]

Down
Down--Sampler
Down-sampling operation is implemented by keeping every
M-th sample of x[n] and removing M 1 in-between
samples to generate y[n]
Input-output relation
y[n] = x[nM]

Down
Down--Sampler
Figure below shows the down-sampling by a factor of 3
of a sinusoidal sequence of frequency 0.042 Hz
I t Sequence
Input S Output sequence down-sampled
down sampled by 3
1 1
0.5 0.5
Amplitude
Amplitude
0 0
-0.5 -0.5
-1 -1
0 10 20 30 40 50 0 10 20 30 40 50
Time index n Time index n
Basic Sampling Rate Alteration Devices
Sampling periods have not been explicitly shown in the
block-diagram representations of the up-sampler and the
down-sampler
This is for simplicity and the fact that the mathematical
theory of multirate systems can be understood without
b i i the
bringing h samplingli period
i d T or the
h sampling
li frequency
f FT
into the picture

Down
Down--Sampler
Figure below shows explicitly the time-dimensions for the

down-sampler
x[ n ] = xa ( nT ) M y[ n ] = xa ( nMT )
Input sampling frequency Output sampling frequency

1 ' FT 1
FT = FT = =
T M T'

Up
Up--Sampler
Figure below shows explicitly the time-dimensions for the
up-sampler
x[ n ] = xa ( nT
T) L y[n]
[ ]
xa ( nT / L ), n =0, ± L , ±2 L ,…
=
0 otherwise
Input sampling frequency Output sampling frequency

1 ' 1
FT = FT = LFT =
T T'
Basic Sampling Rate Alteration
Devices
The up
up--sampler and the down
down--sampler are linear but
time--varying discrete-
time discrete-time systems
We illustrate the time-varying
time varying property of a down-sampler
down sampler
The time-varying property of an up-sampler can be
proved in a similar manner

Basic Sampling Rate Alteration
Devices
Consider a factor-of-M down-sampler defined by
y[n] = x[nM]
Its output y1[ n] for an input x1[n] = x[n n0 ] is
then given
gi en by
b
y1[n] = x1[ Mn] = x[ Mn n0 ]
From the input-output relation of the down-sampler we obtain
y[n n0 ] = x[ M (n n0 )]
= x[ Mn Mn0 ] y1[n]
Up
Up--Sampler
Frequency
Frequency--Domain Characterization
Consider first a factor-of-2 up-sampler whose input-
output relation in the time-domain is given by
x[n / 2], n = 0, ± 2, ± 4,…

x u [n ] =
0, otherwise

Up
Up--Sampler
In terms of the z-transform, the input-output relation is

then given by
n n
X u ( z) = xu [n] z = x[n / 2] z
n= n=
n even
= x[m] z 2m
= X (z2 )
m=

Up
Up--Sampler
In a similar manner, we can show that for a factor

factor--of
of--L
up--sampler
up
L
u ( z) =
On the unit circle,Xfor X, (the
z input-output
) relation is
given
i b
by j
z=e
j j L
X u (e ) = X (e )

Up
Up--Sampler
Figure below shows the relation between and
for L =j2 in the case of a jtypical sequence x[n]
X (e ) X u (e )

Up
Up--Sampler
As can be seen, a factor-of-2 sampling rate expansion

leads to a compression of by a factor of 2 and a j2-
fold repetition in the baseband [0, 2 ] X (e )
This process is called imaging as we get an additional
“i
“image”” off the
th iinputt spectrum
t

Up
Up--Sampler
Similarly in the case of a factor-of-L sampling rate
expansion, there will be additional images of the input
spectrum in the baseband
Lowpass filtering of removes the images Land 1in
effect “fills in” the zero-valued samples in with
i
interpolated
l d sample l values
l
xu [n] L 1
xu [n]

Up
Up--Sampler
Program 10_3 can be used to illustrate the frequency-

domain properties of the up-sampler shown below for L =
4
Input spectrum Output spectrum

1 1
0.8 0.8
Magnitude
Magnitude
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
/ /

Down
Down--Sampler
Frequency
Frequency--Domain Characterization
Applying the z-transform to the input-output relation of a factor-of-M
down-sampler
we get
y[n] = x[Mn]
The expression on the right-hand side cannot be directly expressed
n
in terms of X(z)
Y ( z) = x[Mn] z
n=

Down
Down--Sampler
To get around this problem, define a new sequence

:
xint [n]
x[n], n = 0, ± M , ± 2 M ,…
Thenxint [n] =
0, otherwise
th i
n n
Y ( z) = x[Mn] z = xint [Mn] z
n= n=
k/M 1/ M
= xint [k ] z = X int ( z )
k=
Down
Down--Sampler
Now, xint [can

n] be formally related to x[n] through
where
xint [n] = c[n] ( x[n]
A convenient representation
p of c[n]
[ ] is g
given byy
1, n = 0, ± M , ± 2 M ,…
c[n] =
where 0, otherwise
1 M 1 kn
c[n] = WM
M k =0
j2 / M
WM = e
Down
Down--Sampler
Taking the z-transform of = cmaking

xint [n]and [n] ( x[n]
use of
1 M 1
kn
c[n] = WM
we
eaarrive
e at M k =0
1 M 1
n
X int ( z ) = c[n]x[n] z = WMkn x[n] z n
n= M n= k =0
( )
1 M 1 1 M 1
kn n k
= x[n]WM z = X z WM
M k =0 n = M k =0
Down
Down--Sampler
Consider a factor-of-2 down-sampler with an input x[n]
whose spectrum is as shown below
The DTFTs of the output and the input sequences of this

down-sampler are then related as
1
Y (e j ) = { X (e j / 2) + X( ej / 2 )}
Digital Signal Processing 2 4-107
Down
Down--Sampler
Now X ( e j / 2 ) = X (e j ( implying
2 ) / 2 )that the
second term in the previous
j / 2 equation is
X (firste term )
simply obtained by shifting the to the
right by an amount 2 as shown below
X (e j /2)

Down
Down--Sampler
The plots of the two terms have an overlap, and hence, in
general, the original “shape” of is lost when x[n] is
down-sampled as indicated below
X (e j )

Down
Down--Sampler
This overlap causes the aliasing that takes place due to
under-sampling
There is no overlap, i.e., no aliasing, only if
Note:
ote iss indeed
deed pe
periodic
od c with
t a pe
period
od 2 , e
even
e
though the stretched version of is periodic with a
j
period 4 X (e ) = 0 for /2
Y (e j )
X (e j )

Down
Down--Sampler
For the general case, the relation between the DTFTs of the
output and the input of a factor-of-M down-sampler is given
by
is a sum of
M M1uniformly shifted and stretched
versionsYof(e j ) =and1 scaled by aj ( factor
2 kof) /1/M
M)
X (e
M k =0
Y (e j )
X (e j )

Down
Down--Sampler
Aliasing is absent if and only if
X ( e j ) = 0 for /M
as shown below for M = 2
X (e j ) = 0 for /2

Down
Down--Sampler
Program 10_4 can be used to illustrate the frequency-

domain properties of the up-sampler shown below for M
=2

1 0.5
0.8 0.4
Magnitude
Magnitude
0.6 0.3
0.4 0.2
0.2 0.1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
/ /

Down
Down--Sampler
The input and output spectra of a down-sampler with M =
3 obtained using Program 10-4 are shown below

1 0.5
0.8 0.4
Magnitude
Magnitude
0.6 0.3
Effect
0.4 of aliasing can be clearly0.2seen
0.2 0.1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
/ /

Cascade Equivalences
A complex multirate system is formed by an

interconnection of the up-sampler, the down-sampler,
and the components of an LTI digital filter
In many applications these devices appear in a cascade
form
An interchange of the positions of the branches in a
cascade often can lead to a computationally efficient
realization

To implement a fractional change in the sampling rate we

need to employ a cascade of an up-sampler and a down-
sampler
Consider the two cascade connections shown below
x[n ] M L y1 [ n ]
x[n ] L M y2 [ n]

A cascade of a factor-of-M down-sampler and a factor-of-L
up-sampler is interchangeable with no change in the input-
output relation:
if and only if M and L are relatively prime,

prime i.e., M and L do
not have any common factor that is an integer k > 1
y1[n] = y2[n]

Two other cascade equivalences are shown below
Cascade equivalence #1
#1
x[n ] M H (z ) y1 [ n ]
* x[n ] H (z M ) M y1 [ n ]
Cascade equivalence #2
x[n ] L H (z L ) y2 [ n]
* x[n ] H (z ) L y2 [ n]
Filters in Sampling Rate Alteration
Systems
From the sampling theorem it is known that a sampling
rate of a critically sampled discrete-time signal with a
spectrum occupying the full Nyquist range cannot be
reduced any further since such a reduction will introduce
aliasing
Hence the bandwidth
Hence, band idth of a critically
criticall sampled signal must
m st
be reduced by lowpass filtering before its sampling rate is
reduced by a down-sampler

Filters in Sampling Rate Alteration
Systems
Likewise, the zero-valued samples introduced by an up-
sampler must be interpolated to more appropriate values
for an effective sampling rate increase
We shall show next that this interpolation can be
achieved simply by digital lowpass filtering
We now develop the frequency response specifications
of these lowpass filters

Filter Specifications
Since up-sampling causes periodic repetition of the basic

spectrum, the unwanted images in the spectra of the up-
sampled signal must be removed by using a
lowpass filter H(z), called the interpolation filter
filter, as
indicated below
xu [n]
The above system is called an interpolator

xu [n]
x[n ] L H (z) y[n ]

Filter Specifications
On the other hand, prior to down-sampling, the signal

v[n] should be bandlimited to
by means of a lowpass filter, called the decimation filter,
filter
< / Mbelow to avoid aliasing caused by down-
as indicated
sampling
The above system is called a decimator
x[n ] H (z) M y[n ]

Interpolation Filter Specifications
Assume x[n] has been obtained by sampling a continuous-

time signal at the Nyquist rate
If and
xa (t )
denote the Fourier transforms of
and x[n], respectively, then it can be shown
j
Xa( j ) X (e )
where
xa (t )
is the sampling period
j 1 j j2 k
X (e )= Xa
To k = To
To
Since the sampling is being performed at the Nyquist rate,

rate
there is no overlap between the shifted spectras of
If we instead sample at a much higher rate
X ( j / To ) is related to
yielding y[n], its Fourier transform
th
throughh
xa (t )
T = L ( To
j
Y (e ) Xa( j )
j 1 j j2 k L j j2 k
Y (e )= Xa = Xa
T k= T To k = To / L
On the other hand, if we pass x[n] through a factor-of-L
up-sampler generating , the relation between the
Fourier transforms of x[n] and xuby
are given [n]
It therefore
th f xllu [n
[ ] that
ffollows th t if i passed
is d through
th h an
ideal lowpass filter H(z) with a cutoff at /L and a gain of
j j L
X
L, the output of the ( e ) = X ( e ) y[n]
u filter will be precisely
xu [n]

In practice, a transition band is provided to ensure the
realizability and stability of the lowpass interpolation filter
H(z)
Hence, the desired lowpass filter should have a stopband
edge at and a passband edge close to to
reduce
d the
h distortion
di i off theh spectrum off x[n]
[ ]
s = /L
p s

If is the highest frequency that needs to be preserved in
x[n], then
c
Summarizing the specifications of the lowpass interpolation
filter are thus given by = /L
p c
j L, c/L
H (e )=
0, /L
Decimation Filter Specifications
In a similar manner, we can develop the specifications for
the lowpass decimation filter that are given by
j 1, c /M
H (e )=
0, /M

Filter Design Methods
The design of the filter H(z) is a standard IIR or FIR lowpass
filter design problem
Any one of the techniques outlined in Chapter 7 can be
applied for the design of these lowpass filters

Filters for Fractional Sampling Rate
Alteration
A fractional change in the sampling rate can be achieved by
cascading a factor-of-M decimator with a factor-of-L
interpolator, where M and L are positive integers
Such a cascade is equivalent to a decimator with a
decimation factor of M/L or an interpolator with an
i
interpolation
l i factor
f off L/M

Alteration
There are two possible such cascade connections as
indicated below
H d (z) M L H u (z)
L H u (z) H d (z) M
The second scheme is more computationally efficient

since only one of the filters, H u (z ) or H d (z ) , is
adequate to serve as both the interpolation and the
decimation filter

Alteration
Hence, the desired configuration for the fractional

sampling rate alteration is as indicated below where the
lowpass filter H(z) has a stopband edge frequency given
by
s = min ,
L M
L H (z) M

Computational Requirements
The lowpass decimation or interpolation filter can be

designed either as an FIR or an IIR digital filter
In the case of single-rate digital signal processing, IIR
digital filters are, in general, computationally more
efficient than equivalent FIR digital filters, and are
ttherefore
eeoep preferred
e e ed where
e e co
computational
putat o a cost needs
eeds to
be minimized

This issue is not quite the same in the case of multirate

digital signal processing
To illustrate this point further, consider the factor-of-M
decimator shown below
If the decimation filter H(z) is an FIR filter of length N

v[n ] y[n ]
implemented
x[n ] in a H (z) form, then
direct M
N 1
v[n] = h[m] x[n m]
Digital Signal Processing m =0 4-134
consider the factor-of-M decimator shown below
If the decimation filter H(z) is an FIR filter of length N

implemented in a direct form, then
v[n ] y[n ]
x[n ] H (z) M
N 1
v[n] = h[m] x[n m]
m =0

Now, the down-sampler keeps only every M-th sample of
v[n] at its output
Hence, it is sufficient to compute v[n] only for values of n
that are multiples of M and skip the computations of in-
between samples
This leads to a factor of M savings in the computational
complexity

Now assume H(z) to be an IIR filter of order K with a
transfer function
where
ee V ( z) P( z )
= H ( z) =
X ( z) D( z )
K
n
P( z ) = pn z
n =0
K
n
D( z ) = 1 + dn z
Digital Signal Processing n =1 4-137
Its direct form implementation is given by
w[n] = d1w[n 1] d 2 w[n 2]
d K w[n K ] + x[n]
v[n] = p0 w[n] + p1w[n 1] + + pK w[n K ]
Since v[n] is being down-sampled, it is sufficient to compute

v[n] only for values of n that are integer multiples of M

However, the intermediate signal w[n] must be computed
for all values of n
For example, in the computation of
v[M ] = p0 w[M ] + p1w[M 1] + + pK w[M K]
K+1 successive values of w[n] are still required
As a result, the savings in the computation in this case is
going to be less than a factor of M

For the case of interpolator design, very similar arguments
hold
If H(z) is an FIR interpolation filter, then the computational
savings is by a factor of L (since v[n] has L 1 zeros
between its consecutive nonzero samples)
On the other hand, computational savings is significantly
less with IIR filters

Sampling Rate Alteration Using
MATLAB
The function decimate can be employed to reduce the

sampling rate of an input signal vector x by an integer
factor M to generate the output signal vector y
The decimation of a sequence by a factor of M can be
obtained using Program 10_5 which employs the function
decimate

MATLAB
Example - The input and output plots of a factor-of-2
decimator designed using the Program 10_5 are shown
below
Input sequence Output sequence

2 2
1 1
Amplitude
Amplitude
0 0
-1 -1
-2 -2
0 20
40 60 80 100 0 10 20 30 40 50
Sampling Rate Alteration Using MATLAB
The function interp can be employed to increase the
sampling rate of an input signal x by an integer factor L
generating the output vector y
The lowpass filter designed by the M-file is a symmetric FIR
filter

MATLAB
The filter allows the original input samples to appear as is in
the output and finds the missing samples by minimizing the
mean-square errors between these samples and their ideal
values
The interpolation of a sequence x by a factor of L can be
obtained using the Program 10_6 which employs the
function interp

Example - The input and output plots of a factor-of-2
interpolator designed using Program 10_6 are shown below

2 2
1 1
Amplitude
Amplitude
0 0
-1 -1
-2 -2
0 10 20 30 40 50 0 20 40 60 80 100
MATLAB
The function resample can be employed to increase the
sampling rate of an input vector x by a ratio of two
positive integers, L/M, generating an output vector y
The M-file employs a lowpass FIR filter designed using
fir1 with a Kaiser window
The fractional interpolation of a sequence can be
obtained using Program 10_7 which employs the function
resample

Example - The input and output plots of a factor-of-5/3
interpolator designed using Program 10_7 are given below

2 2
1 1
Amplitude
Amplitude
0 0
-1 -1
-2 -2
010 20 30 0 10 20 30 40 50
Digital Processing of Analog Signals
There are only two approaches to avoiding aliasing

Sample at a faster rate - perhaps not possible (why?).
Use an anti-aliasing filter.

How to reduce aliasing?
An anti-aliasing filter is a low-pass analog filter (LPF) that
is applied to the continuous signal prior to sampling.
The idea is simple: remove the high frequencies .
The ideal frequency response of the anti-aliasing filter is an ideal
low-pass filter as > /2 ,
s
where the cut off 1; c 1

F = c < s =
0; > c
2 Ts
Even the LPF destroys information, it is better than the aliasing

effect. Ideal sampler
xa (t ) Fa ( ) xd (n)
Anti-aliasing filter
(low-pass)
Ts
Anti-aliasing: Formulation
With anti-aliasing, the sampled signal becomes
1 2 m 2 m
xd [n ] = xa (nTs ) * f a (nTs ) X d (e j ) = Xa Fa
Ts m= Ts Ts
The repeated spectra X a (() Fa (() will not fold or overlap.

If Fa ( ) is an ideal LPF with cutoff c , then
1 2 m X a ( ); c
X d (e ) =
j
Xa where X a =
Ts m= Ts 0; > c
Usually, an ideal LPF cannot be realized and must be

approximated.
Anti-aliasing: Example-1
Xa( )
Analog Signal Spectrum
0
X a ( ) Fa ( )
Anti-aliased Spectrum
0
c
c
Sampled Signal Spectrum X d (e j )
(without aliasing) c < s /2
0
cTs T
2 c s
2
Anti-aliasing: Example-2
Original image Down-sampling Down-sampling

with aliasing with anti-aliasing

Anti-aliasing: Digital Filter Output
Recall the overall system of interest:

xd [n ] yd [n ]
xa (t ) H d (e j ) ya (t )
Ha ( )
The response Yd (e j ) of the digital filter H d ( e j )

1 2 m
Yd ( e j ) = Xa H d (e j ) without anti-aliasing
Ts m= Ts
1 2 m 2 m
Yd ( e j ) = Xa Fa H d (e j ) with anti-aliasing
Ts m= Ts Ts

Chapter 4

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Chapter 4

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 4

Uploaded by

Copyright:

Available Formats

Chapter 4.

Digital Signal Processing 4-1

x[n ] = xc (nT ), - <n< .

Digital Signal Processing 4-3

s(t ) = (t nT ) (the periodic impulse train)

Digital Signal Processing 4-4

Digital Signal Processing 4-5

Digital Signal Processing 4-6

X s ( j ) consists of periodically repeated and scaled copies of the

Digital Signal Processing 4-8

There is no overlap between replicated spectrums, when

Digital Signal Processing 4-10

Sampling Without Aliasing Sampling With Aliasing

Digital Signal Processing 4-11

Ideal low-pass Filtering

Digital Signal Processing 4-12

Original Signal Discrete-Time

Ideal low-pass Filtering Reconstructed Signal

Digital Signal Processing 4-13

Digital Signal Processing 4-17

Thus we have the continuous-time Fourier transform of the

Digital Signal Processing 4-18

Digital Signal Processing 4-19

Digital Signal Processing 4-21

Digital Signal Processing 4-22

x[n ] = cos(16000 n / 6000) = cos(2 n + 4000 n / 6000) = cos(2 n / 3)

Digital Signal Processing 4-23

Digital Signal Processing 4-24

Xc( j ) = ( 4000 ) + ( + 4000 )

cos(4000 n / 1500) = cos(2 n + 1000 n / 1500) = cos(2 n / 3)

Digital Signal Processing 4-25

Digital Signal Processing 4-26

xr (t ) = xs (t ) * hr (t ) = x[n ] (t nT ) * hr (t ) = x[n ]hr (t nT )

Digital Signal Processing 4-27

Digital Signal Processing 4-28

Frequency Response Impulse Response

Digital Signal Processing 4-29

If there is no aliasing, the ideal low-pass filter interpolates

Digital Signal Processing 4-31

Digital Signal Processing 4-32

Digital Signal Processing 4-33

n= Time shifting leads to an

Digital Signal Processing 4-34

Digital Signal Processing 4-35

Digital Signal Processing 4-36

Digital Signal Processing 4-37

Digital Signal Processing 4-38

Digital Signal Processing 4-40

Digital Signal Processing 4-41

Digital Signal Processing 4-43

Digital Signal Processing 4-45

Digital Signal Processing 4-47

oA?GM A@\ M @EL

F] IL AAsH e@LA@M BPRHP R BCDRKO FGHI =GPbKD eIGDIGM tR@ WK

Digital Signal Processing 4-50

xd [n] = x[nM ] = xc (nMT )

Digital Signal Processing 4-51

Digital Signal Processing 4-53

Digital Signal Processing 4-54