ZERO PADDING

It is well known that in order to increase the resolution the record length TR should be
increased. Once the sampling frequency is fixed, to increase the resolution one has to increase
N. This is achieved by Zero padding.

Let the DTFT of the finite-duration signal xN(n)={xN(0),xN(1), xN(2),,xN(N-1)} , whose

N 1
X N (e j ) = xN (n)e j n
n=0
The N-point DFT of x(n) is computed as:

2 N 1 2
j k j kn
X N (e N
) = x N ( n )e N
= X N (e j ) 2
= k
n =0 N

which can be viewed as the frequency-sampled version of the DTFT X N (e j ) of x(n) with a
2
sampling interval N
.

By padding L-N zeros to xN(n) we can artificially increase the record length to TR= L TS or
simply L. The DTFT of the zero padded L-length sequence

xL (n) = {xN (0), xN (1), xN (2),......, xN ( N -1), 0,

is given as:
L 1 N 1
X L (e j ) = xL (n)e j n = xN (n)e j n = X N (e j )
n =0 n =0

which remains the same as DTFT of the original sequence xN(n). However, the L-point DFT
of the zero padded signal is given as:

2 L 1 2 N 1 2 2
) = xL ( n ) e = xN ( n) e
j k j kn j kn j k
X L (e L L L
= X N (e L
) = X N (e j ) 2
= k
n=0 n=0 L

which can be viewed as the frequency-sampled version of the DTFT of x(n), X N (e j ) , with a
2
sampling interval L
. Since L>N, L-point DFT of the zero-padded sequence xL (n) has more
2
j k
samples than X N (e N
) in 0 2 hence yields more refined samples than N-point DFT
2
of xN(n). Thus the spacing between the samples in the DTFT has been reduced from N
to 2L
FS
and the spacing between the samples in the CTFT has been reduced from F = to
N
FS L
F = . Hence the frequency resolution has been increased times.
L N

1
Example: In Fig.1 the analog signal xa (t ) with the record length TR = 2a = 2 ms is given. The
magnitude of the CTFT of xa (t ) ,i.e.,
2
a
sin 2
2
3 sin 5.10
4
X a ( j ) = a = 10
a 4
5.10


2
is plotted in Fig.2. Fig.3 is the sampled version of xa (t ) with the sampling interval
T
Ts = R = 0.125 ms (sampling rate Fs = 8 kHz ). Fig.4 shows the 16-point DFT of x (n ) .
16
Fig.5-7 show the DFTs of the zero-padded versions of x ( n ) with L = 32, 64, 256 ,
respectively, i.e., 32-point, 64-point and 256-point DFTs.

xa(t)
1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.5 1 1.5 2
t -3
x 10
Fig.1

2
 -4 2
x 10 Xa(j*W)=a*((sin(a*W /2)./(a*W/2)).
).*exp(-j*a*W )

2 6280r/s
25120r/s
1

-2 -1 0 1 2
W 4
x 10
Fig.2
x(n)
1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12 14 16
n

Fig.3

3
 X(k) - no padded zeros (16-sample)
8

0
-8 -6 -4 -2 0 2 4 6 8
k

Fig.4
X(k) - 12 padded zeros (32-sample)
8

0
-15 -10 -5 0 5 10 15
k

Fig.5

4
 X(k) - 36 padded zeros (64-sample)
8

0
-30 -20 -10 0 10 20 30
k

Fig.6
X(k) - 240 padded zeros (256-sample)
8

0
-100 -50 0 50 100
k

Fig.7