1
ECE 410 DIGITAL SIGNAL PROCESSING D. Munson
University of Illinois Chapter 4
Sampling
Consider
A/D
y
a
(t)
{x
n
} {y
n
}
x
a
(t) H
d
() D/A
For now, lets ignore H
d
(); i.e. we assume that y
n
= x
n
for all n. We want to know under what
conditions we can we get back y
a
(t) = x
a
(t) exactly, and how? To do this we will study the
effects of the A/D and D/A processes on the signals in the frequency domain.
1) A/D (AnalogtoDigital, or more precisely C/D: ContinuoustoDiscrete) converter
A/D: x
a
(t)
T
Q()
x
n
In the analysis, we will neglect the quantizer. Consider
x
a
(t)
T
x
n
= x
a
(nT)
We will show:
X
d
() =
1
T
X
a
+ 2n

\

.
n=
()
extremely important
Proof:
We will use the fact that
1
n=
e
jn
2
t
=
a
(t n)
n=
(*)
This equality holds in a distributional sense, i.e.,
4.2
f
a
(t)
e
jn
2
t
n=
dt = f
a
(t)
a
(t n) dt
n=
for any continuous function f
a
(t).
Now, prove ():
X
d
() = x
n
e
jn
n=
= x
a
(nT)e
jn
n=
=
1
2
n=
X
a
() e
jnT
d e
jn
=
1
2
X
a
()
e
jn(T)
d
n=
=
1
T
X
a
(
)
a
+ 2n
T

\

.
n=
d
let t =
T
, =
2
T
in (*)
=
1
T
X
a
+ 2n
T

\

.
n=
Memorize this result! We will use it repeatedly throughout the course.
Now, what does () mean?
Example
Suppose x
a
(t) is bandlimited to rad/sec
4.3
B B
X
a
()
1
If T <
B
then terms in () dont overlap ~ "no aliasing" and get:
n = + 1 term
2
X
d
()
1/T
B 2
n = 0 term in ()
n = 1 term
i.e.,
1
T
T
X
a
Given X
a
(), you must be able to draw this picture!
How do we know the n = 0 term is confined to  T? Answer: The n = 0 term is
1
T
X
a
T

\

.
,
which first hits zero when its argument equals , i.e., when
T
= = T.
We assumed T <
B
T < , so that the n = 0 term is confined to  < as shown above. If
T > then the various terms overlap, which is called aliasing. The condition T < (no
aliasing) is equivalent to the Nyquist criterion:
T < T <
B
1
T
>
B
1
T
> 2
B
2

\

.
where 1/T is the sampling frequency in samples per second, and /2 is the bandwidth of x
a
(t) in
Hz.
Suppose in the above example we choose T =
5
3
B

\

.
, so that we have aliasing. In this case
T =
5
3
and X
d
() will be the sum of
4.4
2
2
1
/
n = 0 term n = 1 term
5
3
3
5
3
n = + 1 term
Thus, X
d
() is
2
2
X
d
()
1/T
5
3
3
5
3
4/5T
Obviously, X
d
() is no longer a periodic repetition of X
a
. This is because some of the high
frequencies in X
a
() have been moved down to low frequencies in X
d
(), i.e., high frequencies
are masquerading as low frequencies. Hence, the term aliasing.
If we choose T even larger than above, then additional terms (n = 2, etc.) in equation () will
overlap on the central interval  . In determining the shape of X
d
(), we can concern
ourselves only with  , since we know that X
d
() is periodic outside this central period.
Note: Can use equation () to prove sampling theorem. If x
a
(t) is bandlimited and you sample
above the Nyquist rate, then from the second plot on the last page:
X
d
() =
1
T
X
a
T

\

.
for 
{x
n
} x
a
(t)
So, basically
x
a
(t) = F
1
DTFT{x
n
}  
Working through the algebra gives the usual sinc reconstruction formula for x
a
(t) in terms of the
x
n
. You may see this as a homework problem!
4.5
Example
Illustrate that a higher sampling frequency shrinks the DTFT. Suppose
X
a
()
1
3000
If x
n
= x
a
(nT), sketch X
d
() for T =
1
4, 000
, T =
1
6, 000
, and T =
1
12, 000
.
X
d
()
4,000
T =
1
4,000
, T =
3
4
3
4
X
d
()
6,000
T =
1
6,000
, T =
2
2
X
d
()
12,000
4
T = , T =
4
1
12,000
So, as T decreases (sampling frequency increases), the DTFT shrinks and also grows in
amplitude.
2) D/A (DigitaltoAnalog, or more precisely D/C: DiscretetoContinuous) converter
4.6
Model as
y
a
(t) = y
n
g
a
(t nT)
n=
()
so that y
a
(t) is a weighted pulse train. Later, we will consider the case where g
a
(t) is a
rectangular pulse so that y
a
(t) is a staircase function:
1
T t
T 2T 3T 4T
t
g
a
(t)
y
a
(t)
y
0
y
1
y
2
y
3
y
4
This type of D/A is very common and is called a zeroorder hold (ZOH). An ideal D/A,
however, would use
( )
( ) sinc
a
T
g t t
y
n
g
a
(t nT) e
jt
dt
n=
= y
n
n=
g
a
(t nT)
e
jt
dt
= y
n
n=
G
a
() e
jnT
4.7
= G
a
() y
n
n=
e
jTn
Y
a
() = G
a
()Y
d
(T)
This is an important equation. You should either memorize it or remember how to quickly derive
it.
Using y
n
= x
n
and substituting for X
d
from () gives us the expression for Y
a
() in terms of X
a
:
=

.

\

+ =
2
) (
1
) (
n
a a a
T
n
X G
T
Y
( )
Suppose x
a
(t) is bandlimited to B rad/sec and we choose T <
B
. Then, supposing
B
X
a
()
gives
B
+
T n
2n
X
a
n = + 1 term
n = 0 term
n = 1 term
2
T
T
2
T
No aliasing, so that:
X
a
n=
+
2n
T

\

.
= X
a
(), 
T
Now, assume an ideal D/A so that
4.8
g
a
(t) = sinc
t
T
 

\ .
=
sin
, 0
1 , 0
t
T
t
t
T
t
Considering the Fourier transform of this pulse, we see that for an ideal D/A, G
a
() has the
shape of an ideal LPF:
G
a
() =
T
T
0 >
T
T
Now, lets picture the terms in the right hand side of
( )
:
n = + 1 term
G
a
()
2
T
T
B
T
B
2
2
T
n = 1 term
n = 0 term
i.e.,
1
T
X
a
()
X
a
+
2n
T
So:
G
a
() X
a
+
2n
T

\

.
n=
>
T
T
X T
a
0
) (
Therefore if x
a
(t) is bandlimited to
T