Professor Fearing DTFT from FT v1.

0 Fall 2016

Introduction
The discrete time Fourier Transform can be found by taking the continuous time (CT) Fourier Transform of
a sampled signal. The basic DTFT is mostly straight forward, but there are a few subtle points considered
in this handout.
• What are the steps in finding the DTFT using CTFT operations?

• What is the difference between ω in the CTFT and Ω in the DTFT?

• How does the sample period T effect the height of the DTFT?
• In general the DTFT is just the CTFT of the sampled signal multiplied by T , and frequency normalized.
Why does this rule not work for FT which contain δ(ω)?

Deriving discrete time Fourier Transform

Notation: the continuous time Fourier Transform X(jω) has frequency −∞ < ω < ∞. The normalized
frequency Ω for the DTFT X(ejΩ ) has 0 ≤ Ω < 2π (or any period of 2π).

Consider a continuous time signal x(t) with Fourier Transform X(jω). The signal is sampled at rate Ts ,
giving
X∞ X∞
xδ (t) = x(t) · δ(t − nTs ) = x(nTs )δ(t − nTs ) (1)
n=−∞ n=−∞

The CT spectrum for the sampled signal (using FT of the product) is:

F{xδ (t)} = Xδ (jω) (2)

∞
1 X k2π
= X(jω) ∗ δ(ω − ) (3)
Ts Ts
k=−∞
∞
1 X k2π
= X(jω − ) (4)
Ts Ts
k=−∞

or, taking FT of each term in the sum for xδ (t) (eq: 1):
∞
X
Xδ (jω) = x(nTs )e−jωnTs
n=−∞

2π
To calculate the frequency normalized DTFT, let Ω = ωTs . (Check that ω = Ts corresponds to Ω = 2π.)
Thus

∞
jΩ X
Xδ ( ) = x(nTs )e−jΩn
Ts n=−∞

This is just the DTFT analysis equation. The frequency normalization can cause some issues when the FT
contains δ(ω) as shown in Example 1 and 2 below.
∞
X jΩ
X(ejΩ ) = x[n]e−jΩn = Xδ ( )
n=−∞
Ts
.

1
 Time Domain Frequency Domain

x0(t) 1/2 X0(j ω)

1

0 2 t −π/2 0 π/2 2π 3π 4π
ω
1/2
x 0δ(t) X 0δ(j ω) 2
... ...

0 2 t −π/2 0 π/2 2π 3π 4π
ω
1/2 X 0 (e j Ω )
x0[n]
... 2 ...

0 4 n −π/4 0 π/4 π 3π/2 2π

ω
Example 0
Let x0 (t) = sinπt
πt/2
, then X0 (jω) = Π( ωπ ). With sampling rate T0 the sampled signal x0 (nT0 ) = sin πnT0 /2
πnT0
has FT:
∞
1 X k2π
X0δ (jω) = X0 (jω) ∗ δ(ω − ) (5)
T0 T0
k=−∞
∞
1 X k2π
= X0 (jω − ) (6)
T0 T0
k=−∞
∞ k2π
1 X ω− T0
= Π( ) (7)
T0 π
k=−∞

Substituting ω = Ω/T0 , to normalize frequency range, eq. 7 becomes

∞ Ω
Ω 1 X − k2π
X0δ (j ) = Π( T0 T0
) (8)
T0 T0 π
k=−∞
∞
1 X Ω − k2π
= Π( ) (9)
T0 πT0
k=−∞

We can show that the FT of the sampled signal x0δ (t) is the same as the DTFT of x0 [n], that is that
X0δ (j TΩ0 ) = X0 (ejΩ ). The DTFT is

∞
X
X0 (ejΩ ) = x0 [n]e−jΩn (10)
n=−∞
∞
1 X sin πT2 0 n −jΩn
= e (11)
T0 n=−∞
πn
∞
1 X Ω − 2πk
= Π( ) (12)
T0 πT0
k=−∞
1 Ω
Where x0 [n] was found from the IDTFT of X0 (ejΩ ) over one period which is T0 Π( πT0 ). Thus
πT0
1 2 sin πT2 0 n sin πT2 0 n
Z
1 2 1 jΩn
x0 [n] = e dΩ = = (13)
2π −πT0
2
T0 2πT0 n πnT0

We have shown that X0δ (j TΩ0 ) = X0 (ejΩ ). Also the result agrees with Table 5.2. This example is shown in
the figure above, with T0 = 1/2.

2
 Example 1
Let x1 (t) = cos(2πt) with sampling rate T1 = 18 . Then X1 (jω) = π(δ(ω − 2π) + δ(ω + 2π)). The sampled
signal has FT:
∞
1 X k2π
X1δ (jω) = X(jω − ) (14)
T1 T1
k=−∞
∞
X k2π k2π
= 8π δ(ω − 2π − ) + δ(ω + 2π − ) (15)
T1 T1
k=−∞

Substituting ω = Ω/T1 , eq. 15 becomes

∞
Ω X Ω k2π Ω k2π
X1δ (j ) = 8π δ( − 2π − ) + δ( + 2π − ) (16)
T1 T1 T1 T1 T1
k=−∞
∞
X Ω − 2πT1 − k2π Ω + 2πT1 − k2π
= 8π δ( ) + δ( ) (17)
T1 T1
k=−∞

1 −k2π
Now since δ(at) = 1
|a| δ(t), δ( Ω+2πT
T1 ) = T1 δ(Ω + 2πT1 − k2π).
Then
∞
Ω X
X1δ (j ) = 8πT1 δ(Ω + 2πT1 − k2π) + δ(Ω − 2πT1 − k2π) (18)
T1
k=−∞
∞
X π π
= π δ(Ω + − k2π) + δ(Ω − − k2π) (19)
4 4
k=−∞

= X1δ (ejΩ ). (20)

This agrees with Table 5.2 in OW.

3
 Time Domain Frequency Domain
1
x(t) X(j ω) π
... ...

1 −2π 0 2π ω
0 t

1
Xδ (j ω) 4π
x (t)
... δ ... ... ...

1 −10π −6π −2π 0 2π 6π 10π 14π 18π ω

0 t

X(ej Ω ) π
... x[n]
1 ... ... ...
2
0 1 3 4 π/2 3π Ω
n 0 π 2π
2

Example 2
Let x2 (t) = cos(2πt) with sampling rate T2 = 14 . Then X2 (jω) = π(δ(ω − 2π) + δ(ω + 2π)). The sampled
signal has FT:
∞
X k2π k2π
X2δ (jω) = 4π δ(ω − 2π − ) + δ(ω + 2π − ) (21)
T2 T2
k=−∞

Substituting ω = Ω/T , eq. 21 becomes

∞
Ω X
X2δ (j ) = 4πTs δ(Ω + 2πT2 − k2π) + δ(Ω − 2πT2 − k2π) (22)
T
k=−∞
∞
X π π
= π δ(Ω + − k2π) + δ(Ω − − k2π) (23)
2 2
k=−∞

= X2δ (ejΩ ). (24)

This agrees with Table 5.2 in OW.

Comparing example 1 and example 2, it is seen that the DTFT amplitude does not depend on sample
period Ti for this example, due the scaling property of the δ() function.