Chapter 5

Digital Processing of
Continuous-Time Signals
 §5.1 Digital Processing of
Continuous-Time Signals
• Digital processing of a continuous-time
signal involves the following basic steps:
(1) Conversion of the continuous-time
signal into a discrete-time signal,
(2) Processing of the discrete-time signal,
(3) Conversion of the processed discrete-
time signal back into a continuous-time
signal
 §5.1 Digital Processing of
Continuous-Time Signals
Complete block-diagram
Anti- Reconstruction
aliasing S/H A/D DSP D/A filter
filter

• Since both the anti-aliasing filter and the

reconstruction filter are analog lowpass filters,
we review first the theory behind the design of
such filters
• Also, the most widely used IIR digital filter
design method is based on the conversion of an
analog lowpass prototype
 §5.2 Sampling of
Continuous-time Signals
• The frequency-domain representation of
ga(t) is given by its continuos-time
Fourier transform (CTFT):
  jt
Ga ( j)   a
g (t )e dt
• The frequency-domain representation of
g[n] is given by its discrete-time Fourier
transform (DTFT):

G ( e j )  n g [ n ] e  j n
 §5.3 Effect of Sampling in the
Frequency Domain
• To establish the relation between Ga(j)
and G(ej) , we treat the sampling
operation mathematically as a
multiplication of ga(t) by a periodic
impulse train p(t):
 g a (t )  g p(t )
p (t )   (t  nT )
n  
p (t )
 §5.3 Effect of Sampling in the
Frequency Domain
• gp(t) is a continuous-time signal
consisting of a train of uniformly spaced
impulses with the impulse at t = nT
weighted by the sampled value ga(nT) of
ga(t) at that instant t=nT
 §5.3 Effect of Sampling in the
Frequency Domain
• The impulse train gp(t) can be expressed
as
 1  j kt 
g p (t )    e T   g a (t )
 T k   
• From the frequency-shifting property of
the CTFT, the CTFT of ej ktga(t) is given
T

by Ga(j( - kT))
 §5.3 Effect of Sampling in the
Frequency Domain
• Hence, the CTFT of gp(t) is given by


G p ( j )  1
T  Ga  j (  kT ) 
k  

• Therefore, Gp(j) is a periodic function

of  consisting of a sum of shifted and
scaled replicas of Ga(j) , shifted by integer
multiples of T and scaled by 1/T
 §5.3 Effect of Sampling in the
Frequency Domain
• The term on the RHS of the previous
equation for k = 0 is the baseband portion
of Gp(j) , and each of the remaining
terms are the frequency translated
portions of Gp(j)
• The frequency range
T T
 
2 2
is called the baseband or Nyquist band
 §5.3 Effect of Sampling in the
Frequency Domain
• Assume ga(t) is a band-limited signal with
a CTFT Ga(j) as shown below

• The spectrum P(j) of p(t) having a sampling

period T=2/T is indicated below
 §5.3 Effect of Sampling in the
Frequency Domain
• Two possible spectra of Gp(j) are
shown below
 §5.3 Effect of Sampling in the
Frequency Domain
• It is evident from the top figure on the
previous slide that if T>2 m , there is
no overlap between the shifted replicas
of Ga(j) generating Gp(j)
• On the other hand, as indicated by the
figure on the bottom, if T<2 m , there
is an overlap of the spectra of the shifted
replicas of Ga(j) generating Gp(j)
 §5.3 Effect of Sampling in the
Frequency Domain
• If T>2 m , ga(t) can be recovered
exactly from gp(t) by passing it through
an ideal lowpass filter Hr(j) with a gain
T and a cutoff frequency c greater than
m and less than T - m as shown below
 §5.3 Effect of Sampling in the
Frequency Domain
• The spectra of the filter and pertinent
signals are shown below
 §5.3 Effect of Sampling in the
Frequency Domain
Sampling theorem - Let ga(t) be a band-limited
signal with CTFT Ga(j)=0 for
| |> m
Then ga(t) is uniquely determined by its
samples ga(nT) , -n if
T  2  m
where T=2/T
 §5.3 Effect of Sampling in the
Frequency Domain
• The condition T  2 m is often
referred to as the Nyquist condition

• The frequency T/2 is usually

referred to as the folding frequency
 §5.3 Effect of Sampling in the
Frequency Domain
• Given {ga(nT)}, we can recover exactly
ga(t) by generating an impulse train

g p (t )  g
n  
a (nT ) (t  nT )

and then passing it through an ideal

lowpass filter Hr(j) with a gain T and a
cutoff frequency c satisfying
m < c < (T - m )
 §5.3 Effect of Sampling in the
Frequency Domain
• The highest frequency m contained in
ga(t) is usually called the Nyquist
frequency since it determines the
minimum sampling frequency T =2m
that must be used to fully recover ga(t)
from its sampled version
• The frequency 2m is called the Nyquist
rate
 §5.3 Effect of Sampling in the
Frequency Domain
• Oversampling - The sampling frequency is
higher than the Nyquist rate
• Undersampling - The sampling frequency
is lower than the Nyquist rate
• Critical sampling - The sampling
frequency is equal to the Nyquist rate
• Note: A pure sinusoid may not be
recoverable from its critically sampled
version
 §5.4 Recovery of the Analog Signal

• We now derive the expression for the

^ (t )
g
output a of the ideal lowpass
reconstruction filter Hr(j) as a function
of the samples g[n]
• The impulse response hr(t) of the lowpass
reconstruction filter is obtained by
taking the inverse DTFT of Hr(j)
 T ,   c
H r ( j)  
 0,   c
 §5.4 Recovery of the Analog Signal

• Thus, the impulse response is given by

hr (t )  1  H ( j) e jt d  T c e jt d

2   r 
2   c
sin(ct )
 ,  t 
T t / 2
• The input to the lowpass filter is the
impulse train gp(t):

g p (t )  n   g[n] (t  nT )
 §5.4 Recovery of the Analog Signal
• Therefore, the output g^ a (t ) of the ideal
lowpass filter is given by: 
g^ a (t )  hr (t ) * g p (t )   g[n]hr (t  nT )
n  
Substituting hr(t)=sin(ct)/(Tt/2) in the
above and assuming for simplicity
c= T/2= /T , we get
 sin[(t  nT ) / T ]
g a (t )   g[n]
^
n   (t  nT ) / T
which is called Poisson sum formula
 §5.4 Recovery of the Analog Signal

• The ideal bandlimited interpolation

process is illustrated below

Illustration of Poisson sum formula

 §5.6 Sampling of Bandpass Signals

• The conditions developed earlier for the

unique representation of a continuous-time
signal by the discrete-time signal obtained by
uniform sampling assumed that the
continuous-time signal is bandlimited in the
frequency range from DC to some frequency
m
• Such a continuous-time signal is commonly
referred to as a lowpass signal
 §5.6 Sampling of Bandpass Signals

• There are applications where the continuous-

time signal is bandlimited to a higher
frequency range L ||  H with L >0
• Such a signal is usually referred to as the
bandpass signal
• To prevent aliasing a bandpass signal can of
course be sampled at a rate greater than twice
the highest frequency, i.e. by ensuring
T 2 H
 §5.6 Sampling of Bandpass Signals

• However, due to the bandpass spectrum

of the continuous-time signal, the
spectrum of the discrete-time signal
obtained by sampling will have spectral
gaps with no signal components present
in these gaps
• Moreover, if H is very large, the
sampling rate also has to be very large
which may not be practical in some
situations
 §5.6 Sampling of Bandpass Signals

• A more practical approach is to use

under-sampling
• Let = H - L define the bandwidth
of the bandpass signal
• Assume first that the highest frequency
H contained in the signal is an integer
multiple of the bandwidth, i.e.,
H = M()
 §5.6 Sampling of Bandpass Signals

• We choose the sampling frequency T to

satisfy the condition
T = 2() = 2H/M
which is smaller than 2H , the Nyquist
rate
• Substitute the above expression in

G p ( j )  1
T  Ga  j (  k T )
k  
 §5.6 Sampling of Bandpass Signals

• This leads to
G p ( j )  1  
k   Ga  j   j 2 k (  ) 
T
As before, Gp(j) consists of a sum of Ga(j) and
replicas of Gp(j) shifted by integer multiples of
twice the bandwidth  and scaled by 1/T
• The amount of shift for each value of k ensures
that there will be no overlap between all
shifted replicas
no aliasing
 §5.6 Sampling of Bandpass Signals

• Figure below illustrate the idea behind

Ga ( j)

 H  L 0 L H


G p ( j )


 H  L 0 L H
 §5.6 Sampling of Bandpass Signals

• As can be seen, ga(t) can be recovered from gp(t)

by passing it through an ideal bandpass filter
with a passband given by L ||  H and a
gain of T
• Note: Any of the replicas in the lower frequency
bands can be retained by passing through
bandpass filters with passbands
L- k()  ||  H - k() , 1  k  M-1
providing a translation to lower
frequency ranges
§5.7 Analog Lowpass Filter Specifications

• Typical magnitude response |Ha(j)| of

an analog lowpass filter may be given as
indicated below
§5.7 Analog Lowpass Filter Specifications

• In the passband, defined by 0   p , we require

1-p  |Ha(j)|  1+ p , ||  p
i.e., |Ha(j)| approximates unity within an error of
p
• In the stopband, defined by s    , we require
|Ha(j)|  s s    
i.e., |Ha(j)| approximates zero within an error of
s
 §5.7 Analog Lowpass Filter Specifications

 p - passband edge frequency

 s - stopband edge frequency
 p - peak ripple value in the passband
 s - peak ripple value in the stopband
• Peak passband ripple
 p   20 log10 (1   p ) dB
• Minimum stopband attenuation
 s   20 log10 ( s ) dB
§5.7 Analog Lowpass Filter Specifications

• Magnitude specifications may

alternately be given in a normalized
form as indicated below
§5.7 Analog Lowpass Filter Specifications

• Here, the maximum value of the magnitude in

the passband assumed to be unity

• 1 / (1 + 2) - Maximum passband deviation,

given by the minimum value of the magnitude
in the passband

• 1/A - Maximum stopband magnitude

§5.8 Analog Lowpass Filter Design

• Two additional parameters are defined -

(1) Transition ratio k = p/ s

For a lowpass filter k<1

(2) Discrimination parameter k1 = / (A2 -1)

Usually k1<<1
 §5.8.1 Butterworth Approximation
• The magnitude-square response of an N-th
order analog lowpass Butterworth filter is given
by
2 1
H a ( j )  2N
1  ( /  c )
First 2N - 1 derivatives of |Ha(j)|2 at  = 0 are
equal to zero
• The Butterworth lowpass filter thus is said to
have a maximally-flat magnitude at  = 0
 §5.8.1 Butterworth Approximation

• Gain in dB is
G()=10log10|Ha(j)|2
As G(0)=0 and
G(c)=10log10(0.5)=-3.0103-3 dB
c is called 3-dB cutoff frequency
 §5.8.1 Butterworth Approximation

• Two parameters completely

characterizing a Butterworth
lowpass filter are c and N
• These are determined from the
specified bandedges p and s , and
minimum passband magnitude 1/(1
+ 2) , and maximum stopband ripple
1/A
 §5.8.1 Butterworth Approximation

The order of the filter

1 log10[( A2  1) /  2 ] log10 (1/ k1 )
N  (4.35)
2 log10 ( s /  p ) log10 (1/ k )

So ,Ωc is called 3-dB cutoff frequency.

  ( p /  c ) N
A 2  1  ( s /  c ) 2 N
 §5.8.1 Butterworth Approximation

• Typical magnitude responses with c =1

Butterworth Filter

N=2
1 N=4
0.8 N = 10
Magnitude

0.6
0.4
0.2
0
0 1 2 3
W
 §5.8.2 Chebyshev Approximation
• The magnitude-square response of an N-
th order analog lowpass Type 1
Chebyshev filter is given by
2 1
H a ( s) 
1   2TN2 ( /  p )
where TN() is the Chebyshev polynomial
of order N:
 cos( N cos1 ),  1
TN ()   1
cosh( N cosh ),   1
 §5.8.2 Chebyshev Approximation

• Typical magnitude response plots of the

analog lowpass Type 1 Chebyshev filter
are shown below
Type 1 Chebyshev Filter

N=2
1 N=3
0.8 N=8
Magnitude

0.6
0.4
0.2
0
0 1 2 3
W
 §5.8.2 Chebyshev Approximation
• If at  = c the magnitude is equal to 1/A,
then
2
H a ( j s )  1  1
1   2TN2 ( s /  p ) A2
Solving the above we get
cosh 1 ( A2  1 /  ) cosh 1 (1 / k1 )
N 1

cosh ( s /  p ) cosh 1 (1 / k )
• Order N is chosen as the nearest integer greater
than or equal to the above value
 §5.8.2 Chebyshev Approximation
• The magnitude-square response of an N-th
order analog lowpass Type 2 Chebyshev (also
called inverse Chebyshev) filter is given by

2 1
H a ( j )  2
2 TN ( s
/  p )
1   
T (
 N s  /  ) 
where TN() is the Chebyshev polynomial of
order N
 §5.8.2 Chebyshev Approximation

• Typical magnitude response plots of the

analog lowpass Type 2 Chebyshev filter
are shown below
Type 2 Chebyshev Filter

N=3
1 N=5
0.8 N=7
Magnitude

0.6
0.4
0.2
0
0 1 2 3
W
 §5.8.2 Chebyshev Approximation
• The order N of the Type 2 Chebyshev filter is
determined from given , s, and A using
cosh 1 ( A2  1 /  ) cosh 1 (1 / k1 )
N 1
 1
cosh ( s /  p ) cosh (1 / k )
Example - Determine the lowest order of a
Chebyshev lowpass filter with a 1-dB cutoff
frequency at 1 kHz and a minimum attenuation
of 40 dB at 5 kHz -
1
cosh (1 / k1 )
N 1
 2.6059
cosh (1 / k )
 §5.8.3 Elliptic Approximation

• The square-magnitude response of an

elliptic lowpass filter is given by
2 1
H a ( j )  2 2
1   RN (  /  p )
where RN() is a rational function of order N
satisfying RN(1/)=1/ RN() , with the roots of
its numerator lying in the interval 1< <1 and
the roots of its denominator lying in the
interval 1<  < 
 §5.8.3 Elliptic Approximation
• For given p, s, , and A, the filter
order can be estimated using
2 log10 (4 / k1 )
N
log10 (1 /  )
2
where k' 1 k
0  1  k '
2(1  k ')
  0  2( 0 )5  15( 0 )9  150( 0 )13
 §5.8.3 Elliptic Approximation
• Example - Determine the lowest order of a elliptic
lowpass filter with a 1-dB cutoff frequency at 1 kHz
and a minimum attenuation of 40 dB at 5 kHz
Note: k = 0.2 and 1/k1=196.5134
• Substituting these values we get
k’=0.979796, 0=0.00255135,
=0.0025513525
• and hence N = 2.23308
• Choose N = 3
 §5.8.3 Elliptic Approximation

• Typical magnitude response plots with

are shown below
Elliptic Filter

N=3
1 N=4
0.8
Magnitude

0.6
0.4
0.2
0
0 1 2 3
W
 Design of Other Analog Filter
(Spectral transformation in analog frequency)
Type Spectral transformation
^
LP LP S
S  ^
c
LP HP ^ ^
S= Ωc/ S

^2 ^ 2
S 
LP BP S  p ^ ^ 0
^
S ( p 2   p1 )
third-order normalized lowpass Butterworth
transfer function
1 1
H an ( s )  
( s  1)( s  s  1) 1  2s  2 s 2  s 3
2

which has 3-dB frequency at c  1 .

 Homework
• Problems
5.2, 5.4, 5.5, 5.8, 5.11, 5.17, 5.25