The Cooper Union

Department of Electrical Engineering

ECE114 Digital Signal Processing
Lecture Notes: Sinc Functions and Sampling Theory
October 7, 2011

Sinc Functions
A rectangular pulse in time/frequency corresponds to a sinc function in frequency/time.
Two sinc functions arise: the “ordinary”sinc, essentially sin = , which extends from 1 to
1 and has equally spaced zero crossings, and the Dirichlet sinc, which is periodic and also
has equally spaced zero crossings.
Here, we take:
sin
sinc =
sin x
Note that sometimes the sinc function is de…ned as sinc (x) = x
, as in MATLAB. The
Dirichlet sinc of order N is de…ned as:
sin (N !=2)
DN (!) =
N sin (!=2)

Continuous-Time Rectangular Pulse

Let x (t) = 1, 0 t T , and 0 otherwise. Then:
j fT sin f T
X (f ) = e T
fT
Notice the linear-phase factor arises because x (t) is symmetric about time T =2. That is,
x (t) = x0 (t T =2) where X0 (f ) is zero-phase. Also observe that:
X (mf0 ) = T (m)
where here ( ) is the discrete-time impulse.
f0 = 1=T
We verify the result:
Z T
j2 f t
X (f ) = e dt
0
1 T
= e j2 f t 0
j2 f
1 e j2 f T
=
j2 f
ej f T e j f T
= e j fT T
2j ( f T )
sin f T
= e j fT T
fT

1
Ideal Analog Lowpass Filter
Let H (f ) = 1, jf j fc , H (f ) = 0 otherwise. Then:
sin t=T
h (t) = 2fc
t=T
where T = 1= (2fc ). Observe that:
h (nT ) = (n)
where ( ) is the discrete-time impulse.
We verify the result:
Z fc
h (t) = ej2 ft
df
fc
1
= ej2 fc t e j2 fc t
j2 t
sin (2 fc t)
= 2fc
2 fc t

Ideal Digital Lowpass Filter

Let H (!) = 1, j!j ! c , H (!) = 0 otherwise. Then:
! c sin (n! c )
h (n) =
n! c
Observe that h (n) is a sampled sinc function, but does not necessarily have exact zero-
crossings (if those zero-crossings would not occur at integer time points).
We verify the result:
Z !c
1
h (n) = ej!n d!
2 !c
1
= ej!c n e j!c n
2 jn
! c sin (n! c )
=
n! c

Discrete-Time Rectangular Pulse

Let h (n) = 1, 0 n N 1, and h (n) = 0 otherwise. This is a discrete-time rectangular
pulse. We would expect its DTFT H (!) to be a sinc function. However, H (!) must be
periodic. Therefore, it is a Dirichlet sinc. Speci…cally:
N 1 sin (N !=2) N 1
H (!) = e j ( 2 )! N = e j ( 2 )! N DN (!)
N sin (!=2)
Observe that:
N k = 0; N; 2N;
H (k! 0 ) = N (k mod N ) =
0 otherwise

2
where ( ) is the discrete impulse and ! 0 = 2 =N . In other words, H (!) is zero at all the
DFT bin frequencies, other than DC.
To see this:
X1
N
j!n
H (!) = e
0
1 e j!N
=
1 e j!
e j!N=2 ej!N=2 e j!N=2
=
e j!=2 ej!=2 e j!=2
N 1 sin (N !=2)
= e j ( 2 )!
sin (!=2)
N 1
= e j ( 2 )! N DN (!)

Other cases
Suppose we have a periodic continuous-time pulse train with period T , with x (t) = 1,
0 t 0 , 0 for 0 < t T . Then the line spectrum are samples of a sinc. Which sinc?
Suppose we have a set of DFT coe¢ cients given by X (k) = 1, 0 k M , and X (k) = 0,
M + 1 k N 1. Then x (n) are samples of a sinc. Which sinc?

Impulse Train
Consider a continuous-time periodic impulse train:
X
1
(t nT )
n= 1

Theorem 1 The Fourier transform of a periodic impulse train is a periodic impulse train.
Speci…cally:
X 1 X1
(t nT ) ! fs (f mfs )
n= 1 m= 1

where fs = 1=T .

Proof. The Fourier transform of the impulse train is:

X
1 X
1
j2 f nT
F (t nT ) = e
n= 1 n= 1

Now,
P1 consider a discrete-time signal x (n) = 1 for all n. Then its DTFT is X (!) =
j!n
n= 1 e . We know this to be 2 (!) via the IDTFT formula:
Z
1
1= [2 (!)] ej!n d!
2

3
However, let us be more precise about this. On the one hand:
X
1
j!n
e =2 (!) , !
1

On the other hand, we know this function is periodic with period 2 . Thus, the more precise
result is:
X
1 X1
e j!n = 2 (! 2 m) , 1 ! 1
1 m= 1

Now, let us consider (a ) in terms of ( ), for a 6= 0. Observe:

Z 1 Z 1
1 1
(a ) f ( ) d = ( ) f ( =a) d = f (0)
= 1 jaj = 1 jaj
where we make the substitution = a . Thus:
1
(a ) = ( )
jaj
Therefore:
X
1 X
1
j2 f nT
e = 2 (2 f T 2 m)
n= 1 m= 1

1 X
1
= 2 (f 2 m=2 T )
2 T m= 1
X
1
= fs (f mfs )
m= 1

with fs = 1=T .

Sinc Interpolation Formula

The basic sampling theorem is that if:
xa (t) ! Xa (f )
and x (n) = xa (nT ), and:
x (n) ! X (!)
then:
X
1
X (!) = fs Xa (f mfs )j!=2 f =fs
m= 1

where fs = 1=T . In terms of analog radian frequency = 2 f , with s = 2 fs = 2 =T :

X
1
X (!) = fs Xa ( m s )j!= T
m= 1

One question we may have is what is the analog signal whose spectrum is periodic. The
answer is an impulse train, scaled by the samples of xa (t). That is:

4
 P
Theorem 2 The inverse CTFT of fs 11 Xa (f mfs ), i.e., the analog signal whose spec-
trum is the periodized version of X (f ), is:

X
1
x (n) (t nT )
n= 1

P P
Proof. Consider fs 1 m= 1 (f mfs ). Its ICTFT is 1
n= 1 (t nT ). Multiplying the
impulse train in the time domain by xa (t) yields:

X
1 X
1 X
1
xa (t) (t nT ) = xa (nT ) (t nT ) = x (n) (t nT )
n= 1 n= 1 n= 1

But this corresponds to convolution in the frequency domain:

X
1 X
1
Xa (f ) fs (f mfs ) = fs X (f mfs )
m= 1 m= 1

Assuming Xa (f ) is bandlimited to the range jf j fs =2, we have:

X
1
Xa (f ) = Xa (f mfs ) , fs =2 f fs =2
1

Now suppose we apply the ideal brickwall …lter H (f ) :

1 jf j fs =2
H (f ) =
0 otherwise

Then:
X
1
Xa (f ) = H (f ) Xa (f mfs ) , 1<f <1
1

Taking the inverse Fourier transform yields:

Theorem 3 (Sinc Interpolation Formula) If xa (t) is bandlimited to jf j fs =2, then it

can be perfectly reconstructed from its samples x (n) = xa (nT ) via:

X
1
xa (t) = x (n) (t nT )
n= 1

where:
sin ( t=T )
(t) =
t=T