CHAPTER FOUR

FOURIER SERIES REPRESENTATION OF PERIODIC SIGNALS

4.1. Applying Complex Exponential to LTI Systems

Consider the complex exponential signal:

𝑥(𝑡) = 𝑒𝑗 𝜔0 𝑡

2𝜋
Recall that this signal is periodic with fundamental period 𝑇 = (assuming 𝜔0 > 0) . Based
𝜔0
on this complex exponential, we can define an entire harmonic family of complex exponentials,
given by

𝜙𝑘 (𝑡) = 𝑒𝑗𝑘𝜔0 𝑡 , 𝑘 ∈ ℤ

In other words, for each 𝑘 ∈ ℤ, 𝜙𝑘 (𝑡) is a complex exponential whose fundamental frequency
is 𝑘𝜔0 (i.e., 𝑘 times the fundamental frequency of 𝑥(𝑡) ). Thus, each of the signals 𝜙𝑘 (𝑡) is
periodic with period 𝑇, since

𝜙𝑘 (𝑡 + 𝑇) = 𝑒𝑗𝑘𝜔0 (𝑡+𝑇) = 𝑒𝑗𝑘𝜔0 𝑡 𝑒𝑗𝑘𝜔0 𝑇 = 𝑒𝑗𝑘 𝜔0 𝑡 𝑒𝑗𝑘2𝜋 = 𝑒𝑗𝑘 𝜔0 𝑡

Note that 𝑇 may not be the fundamental period of the signal 𝜙𝑘 (𝑡), however.

Since each of the signals in the harmonic family is periodic with period 𝑇, a linear combination
of signals from that family is also periodic. Specifically, consider the signal
∞ ∞
2𝜋
𝑥(𝑡) = ∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑡 = ∑ 𝑎𝑘 𝑒 𝑗𝑘 𝑇 𝑡
𝑘=−∞ 𝑘=−∞

The terms corresponding to 𝑘 = 1 and 𝑘 = −1 are known as the first harmonic of the signal
𝑥(𝑡). The terms corresponding to 𝑘 = 2 and 𝑘 = −2 are known as the second harmonic and
so forth.

Suppose we are given a certain periodic signal 𝑥(𝑡) with fundamental period 𝑇. Define 𝜔0 =
2𝜋
and suppose that we can write 𝑥(𝑡) as
𝑇

∞ ∞
2𝜋
𝑗𝑘 𝑡
𝑥(𝑡) = ∑ 𝑎𝑘 𝑒 𝑇 (= ∑ 𝑎𝑘 𝑒𝑗𝑘 𝜔0 𝑡 )
𝑘=−∞ 𝑘=−∞

for some sequence of coefficients 𝑎𝑘 , 𝑘 ∈ ℤ. Then the above representation is called the Fourier
Series representation of 𝑥(𝑡). The quantities 𝑎𝑘 , 𝑘 ∈ ℤ are called the Fourier Series coefficients.

1|Pa ge
Example

Consider the signal 𝑥(𝑡) = cos (𝜔0 𝑡), where 𝜔0 > 0. We have

1 𝑗 𝜔 𝑡 1 −𝑗𝜔 𝑡
𝑥(𝑡) = 𝑒 0 + 𝑒 0
2 2

This is the Fourier Series representation of 𝑥(𝑡); it has only first harmonics, with coefficients
1
𝑎1 = 𝑎−1 = 2.

Similarly, consider the signal 𝑥(𝑡) = sin (𝜔0 𝑡) . We have

1 𝑗 𝜔 𝑡 1 −𝑗𝜔 𝑡
𝑥(𝑡) = 𝑒 0 − 𝑒 0
2𝑗 2𝑗

1 1
Once again, the signal has only first harmonics, with coefficients 𝑎1 = 2𝑗
and 𝑎−1 = − 2𝑗.

Suppose that we have a periodic signal that has a Fourier Series representation
∞

𝑥(𝑡) = ∑ 𝑎𝑘𝑒𝑗𝑘 𝜔0 𝑡
𝑘=−∞

Now suppose that 𝑥(𝑡) is real, i.e., 𝑥 ∗ (𝑡) = 𝑥(𝑡). Taking the complex conjugate of both sides
of the above expression, we have
∞

𝑥 ∗ (𝑡) = ∑ 𝑎𝑘∗ 𝑒 −𝑗𝑘𝜔0 𝑡

𝑘=−∞

Equating the expressions for 𝑥(𝑡) and 𝑥 ∗ (𝑡), we have

∞ ∞

∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑡 = ∑ 𝑎𝑘∗ 𝑒 −𝑗𝑘𝜔0 𝑡

𝑘=−∞ 𝑘=−∞

Comparing the terms, we see that for any 𝑘 ∈ ℤ, the coefficient of 𝑒𝑗𝑘 𝜔0 𝑡 is 𝑎𝑘 on the left hand
side, and is 𝑎 ∗−𝑘 on the right hand side. Thus, for real signals 𝑥(𝑡) , the Fourier Series
coefficients satisfy

𝑎−𝑘 = 𝑎𝑘∗

for all 𝑘 ∈ ℤ. Substituting this into the Fourier Series representation, we have

2|Pa ge
 ∞

𝑥(𝑡) = 𝑎0 + ∑ (𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑡 + 𝑎−𝑘 𝑒 −𝑗𝑘𝜔0 𝑡 )

𝑘=1
∞

= 𝑎0 + ∑ (𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑡 + 𝑎𝑘∗ 𝑒 −𝑗𝑘𝜔0 𝑡 )

𝑘=1
∞

= 𝑎0 + ∑ 2Re {𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑡 },

𝑘=1

where Re is the real part of the given complex number. If we write 𝑎𝑘 in polar form as 𝑟𝑘𝑒𝑗 𝜃𝑘 ,
the above expression becomes
∞ ∞

𝑥(𝑡) = 𝑎0 + ∑ 2Re {𝑟𝑘𝑒𝑗 (𝑘𝜔0 𝑡+𝜃𝑘 ) } = 𝑎0 + 2 ∑ 𝑟𝑘cos (𝑘𝜔0 𝑡 + 𝜃𝑘 ).

𝑘=1 𝑘=1

This is an alternate representation of the Fourier Series for real-valued signals (known as the
trigonometric representation).

4.2. Calculating the Fourier Series Coefficients

Suppose that we are given a periodic signal 𝑥(𝑡) with period 𝑇 , and that this signal has a
Fourier Series representation
∞

𝑥(𝑡) = ∑ 𝑎𝑘𝑒𝑗𝑘 𝜔0 𝑡
𝑘=−∞

We will soon see conditions under which a signal will have such a representation, but for now,
suppose that we are just interested in finding the coefficients 𝑎𝑘 , 𝑘 ∈ ℤ. To do this, multiply
𝑥(𝑡) by 𝑒 −𝑗𝑛𝜔0 𝑡 , where 𝑛 is some integer. This gives
∞ ∞

𝑥(𝑡)𝑒 −𝑗𝑛𝜔0 𝑡 = ∑ 𝑎𝑘 𝑒𝑗𝑘 𝜔0 𝑡 𝑒 −𝑗𝑛𝜔0 𝑡 = ∑ 𝑎𝑘𝑒𝑗 (𝑘−𝑛)𝜔0 𝑡 .

𝑘=−∞ 𝑘=−∞

Now suppose that we integrate both sides of the above equation from 𝑡0 to 𝑡0 + 𝑇 for any 𝑡0 :
∞ ∞
𝑡0 +𝑇 𝑡0 +𝑇 𝑡0 +𝑇
∫ 𝑥(𝑡)𝑒 −𝑗𝑛𝜔0 𝑡 𝑑𝑡 =∫ ∑ 𝑎𝑘 𝑒𝑗 (𝑘−𝑛)𝜔0 𝑡 𝑑𝑡 = ∑ 𝑎𝑘 ∫ 𝑒𝑗 (𝑘−𝑛)𝜔0 𝑡 𝑑𝑡.
𝑡0 𝑡0 𝑘=−∞ 𝑘=−∞ 𝑡0

Now note that if 𝑛 = 𝑘, we have

𝑡0 +𝑇 𝑡0 +𝑇
∫ 𝑒𝑗 (𝑘−𝑛)𝜔0 𝑡 𝑑𝑡 = ∫ 1𝑑𝑡 = 𝑇
𝑡0 𝑡0

3|Pa ge
Otherwise, if 𝑛 ≠ 𝑘, we have

𝑡0 +𝑇 𝑡0 +𝑇
𝑗 (𝑘−𝑛)𝜔0 𝑡
1 𝑗 (𝑘−𝑛)𝜔0 𝑡
∫ 𝑒 𝑑𝑡 = 𝑒 |
𝑡0 𝑗(𝑘 − 𝑛)𝜔0 𝑡 0
1
= (𝑒𝑗 (𝑘−𝑛)𝜔0 (𝑡0 +𝑇) − 𝑒𝑗 (𝑘−𝑛)𝜔0 𝑡0 )
𝑗(𝑘 − 𝑛)𝜔0
= 0.

Thus, we have

𝑡0 +𝑇 ∞ 𝑡0 +𝑇
−𝑗𝑛𝜔0 𝑡
∫ 𝑥(𝑡)𝑒 𝑑𝑡 = ∑ 𝑎𝑘 ∫ 𝑒𝑗 (𝑘−𝑛)𝜔0 𝑡 𝑑𝑡 = 𝑎𝑛 𝑇
𝑡0 𝑘=−∞ 𝑡0

or equivalently,

1 𝑡0 +𝑇
𝑎𝑛 = ∫ 𝑥(𝑡)𝑒 −𝑗𝑛𝜔0 𝑡 𝑑𝑡
𝑇 𝑡0

where 𝑡0 is any arbitrary starting point. In other words, we obtain the Fourier coefficient 𝑎𝑛 by
multiplying the signal 𝑥(𝑡) by 𝑒 −𝑗𝑛𝜔0 𝑡 and then integrating the resulting product over any
period.

Example

Consider the signal

𝑇
0 − ≤ 𝑡 < 𝑇1
2
𝑥(𝑡) = 1 −𝑇1 ≤ 𝑡 ≤ 𝑇1
𝑇
{0 𝑇1 < 𝑡 <
2

where 𝑇1 ≤ 𝑇 and 𝑥(𝑡) is 𝑇-periodic.

2𝜋
Define 𝜔0 = . We have
𝑇

𝑇
1 2 1 𝑇1 2𝑇1
𝑎0 = ∫ 𝑥(𝑡)𝑑𝑡 = ∫ 𝑥(𝑡)𝑑𝑡 =
𝑇 −𝑇 𝑇 𝑇1 𝑇
2

For 𝑘 ≠ 0, we have

4|Pa ge
 𝑇
𝑇1
1 2 −𝑗𝑘𝜔0 𝑡
1 𝑇1 −𝑗𝑘𝜔 𝑡 1 1 −𝑗𝑘𝜔0 𝑡
𝑎𝑘 = ∫ 𝑇 𝑥(𝑡)𝑒 𝑑𝑡 = ∫ 𝑒 0 𝑑𝑡 = 𝑒 |
𝑇 − 𝑇 −𝑇1 𝑇 −𝑗𝑘𝜔0 −𝑇
2 1
2
= sin (𝑘𝜔0 𝑇1 )
𝑇𝑘𝜔0
sin (𝑘𝜔0 𝑇1 )
= ,
𝑘𝜋

where we used the fact that 𝑇𝜔0 = 2𝜋.

Note that in this case, 𝑎𝑘 is real for all 𝑘 ∈ ℤ, and satisfies 𝑎𝑘 = 𝑎−𝑘 . Thus we can also write
the Fourier series as
∞ ∞

𝑥(𝑡) = 𝑎0 + ∑ 𝑎𝑘 (𝑒𝑗𝑘𝜔0 𝑡 + 𝑒 −𝑗𝑘𝜔0 𝑡 ) = 𝑎0 + 2 ∑ 𝑎𝑘 cos (𝑘𝜔0 𝑡)

𝑘=1 𝑘=1

4.3. Fourier Series Representation of Continuous Time Periodic Signals

There are various different sufficient conditions that guarantee that a given signal 𝑥(𝑡) will
have a Fourier series representation. One commonly used set of conditions are known as the
Dirichlet conditions stated as follows.

A periodic signal 𝑥(𝑡) has a Fourier series representation if all three of the following conditions
are satisfied:

• The signal is absolutely integrable over one period:

𝑡0 +𝑇
∫ |𝑥(𝑡)|𝑑𝑡 < ∞
𝑡0

• In any finite interval of time 𝑥(𝑡) has bounded variation, meaning that it has only a
finite number of minima and maxima during any single period of the signal.
• In any finite interval of time, there are only a finite number of discontinuities, and each
of these discontinuities are finite.
We won't go into the proof of why these conditions are sufficient here, but it suffices to note
that signals that violate the above conditions (the last two in particular) are somewhat
pathological. The first condition guarantees that the Fourier series coefficients are finite, since

1 𝑡 +𝑇 1 𝑡0 +𝑇 1 𝑡0 +𝑇
|𝑎𝑘 | = | ∫𝑡 0 𝑥(𝑡)𝑒 −𝑗𝑘𝜔0 𝑡 𝑑𝑡| ≤ 𝑇 ∫𝑡0 |𝑥(𝑡)𝑒 −𝑗𝑘𝜔0 𝑡 |𝑑𝑡 = 𝑇 ∫𝑡0 |𝑥(𝑡)|𝑑𝑡.
𝑇 0

Thus if the signal is absolutely integrable over a period, then |𝑎𝑘 | < ∞ for all 𝑘 ∈ ℤ.

5|Pa ge
4.4. Properties of Continuous Time Fourier Series

We will now derive some useful properties of the Fourier series coefficients. Throughout we
will use the notation

FS
𝑥(𝑡) ⟷ 𝑎𝑘

to denote that the 𝑇-periodic signal 𝑥(𝑡) has Fourier series coefficients 𝑎𝑘 , 𝑘 ∈ ℤ.

4.4.1. Linearity

Suppose we have two signals 𝑥1 (𝑡) and 𝑥2 (𝑡), each of which is periodic with period 𝑇. Let

FS FS
𝑥1 (𝑡) ⟷ 𝑎𝑘 , 𝑥2 (𝑡) ⟷ 𝑏𝑘 .

For any complex scalars 𝛼, 𝛽, let 𝑔(𝑡) = 𝛼𝑥1 (𝑡) + 𝛽𝑥2 (𝑡). Then

FS
𝑔(𝑡) ⟷ 𝛼𝑎𝑘 + 𝛽𝑏𝑘 .

The above property follows immediately from the definition of the Fourier series coefficients
(since integration is linear).

Example

Consider 𝑥1 (𝑡) = cos (𝜔0 𝑡) and 𝑥2 (𝑡) = sin (𝜔0 𝑡). We have

𝛼 𝑗 𝜔 𝑡 𝛼 −𝑗 𝜔 𝑡 𝛽 𝑗 𝜔 𝑡 𝛽 −𝑗𝜔 𝑡
𝑔(𝑡) = 𝛼cos (𝜔0 𝑡) + 𝛽sin (𝜔0 𝑡) = 𝑒 0 + 𝑒 0 + 𝑒 0 − 𝑒 0
2 2 2𝑗 2𝑗
𝛼 𝛽 𝛼 𝛽
= ( + ) 𝑒𝑗 𝜔0 𝑡 + ( − ) 𝑒 −𝑗𝜔0 𝑡
2 2𝑗 2 2𝑗

Thus, we see that each Fourier series coefficient of 𝑔(𝑡) is indeed given by a linear
combination of the corresponding Fourier series coefficients of 𝑥1 (𝑡) and 𝑥2 (𝑡).

4.4.2. Time Shifting

Define 𝑔(𝑡) = 𝑥(𝑡 − 𝜏), where 𝜏 ∈ ℝ is some delay. Let the Fourier series coefficients of 𝑔(𝑡)
be given by 𝑏𝑘 , 𝑘 ∈ ℤ. Then

1 𝑡0 +𝑇 −𝑗𝑘𝜔 𝑡
1 𝑡0 +𝑇
𝑏𝑘 = ∫ 𝑔(𝑡)𝑒 0 𝑑𝑡 = ∫ 𝑥(𝑡 − 𝜏)𝑒 −𝑗𝑘𝜔0 𝑡 𝑑𝑡
𝑇 𝑡0 𝑇 𝑡0

𝑑𝑡 ‾
Define 𝑡‾ = 𝑡 − 𝜏, so that 𝑑𝑡 = 1. Then we have

1 𝑡0 −𝜏+𝑇 1 0 𝑡 −𝜏+𝑇
𝑏𝑘 = 𝑇 ∫𝑡0 −𝜏 𝑥(𝑡‾)𝑒 −𝑗𝑘𝜔0 (𝑡‾+𝜏) 𝑑𝑡‾ = 𝑒 −𝑗𝑘𝜔0 𝜏 𝑇 ∫𝑡0 −𝜏 𝑥(𝑡‾)𝑒 −𝑗𝑘𝜔0 𝑡‾ 𝑑𝑡‾ = 𝑒 −𝑗𝑘𝜔0 𝜏 𝑎𝑘 .

Thus
6|Pa ge
 FS
𝑥(𝑡 − 𝜏) ⟷ 𝑒 −𝑗𝑘𝜔0 𝜏 𝑎𝑘 .

Example

Consider 𝑥(𝑡) = cos (𝜔0 𝑡), and let 𝑔(𝑡) = 𝑥(𝑡 − 𝜏) We have

1 1 1 1
𝑔(𝑡) = cos (𝜔0 (𝑡 − 𝜏)) = 2 𝑒𝑗 𝜔0 (𝑡−𝜏) + 2 𝑒 −𝑗𝜔0 (𝑡−𝜏) = 2 𝑒 −𝑗𝜔0 𝜏 𝑒𝑗 𝜔0 𝑡 + 2 𝑒𝑗 𝜔0 𝜏 𝑒 −𝑗𝜔0 𝑡 .

1 1
Thus, we see that the coefficient of 𝑒𝑗 𝜔0 𝑡 is 2 𝑒 −𝑗𝜔0 𝜏 , and the coefficient of 𝑒 −𝑗𝜔0 𝑡 is 2 𝑒𝑗 𝜔0 𝜏 ,
as predicted by the expressions derived above.

4.4.3. Time Reversal

Define 𝑦(𝑡) = 𝑔(−𝑡). Let the Fourier series coefficients of 𝑔(𝑡) be given by 𝑏𝑘 , 𝑘 ∈ ℤ. Note
that
∞ ∞

𝑔(𝑡) = 𝑥(−𝑡) = ∑ 𝑎𝑘 𝑒 −𝑗𝑘𝜔0 𝑡 = ∑ 𝑎−𝑘𝑒𝑗𝑘 𝜔0 𝑡

𝑘=−∞ 𝑘=−∞

Thus we have
FS
𝑥(−𝑡) ⟷ 𝑎 −𝑘

i.e., the Fourier series coefficients for a time-reversed signal are just the time reversal of the
Fourier series coefficients for the original signal. Note that this is not true for the output of LTI
systems (a time reversal of the input to an LTI system does not necessarily mean that the output
is a time-reversal of the original output).

Example

Consider 𝑥(𝑡) = sin ( 𝜔0 𝑡) and define 𝑔(𝑡) = 𝑥(−𝑡). First, note that

1 𝑗 𝜔 𝑡 1 −𝑗𝜔 𝑡
𝑥(𝑡) = sin (𝜔0 𝑡) = 𝑒 0 − 𝑒 0 = 𝑎 𝑒𝑗 𝜔0 𝑡 + 𝑎 −𝑗𝜔0 𝑡
1 −1 𝑒
2𝑗 2𝑗

We have

1 −𝑗𝜔 𝑡 1 𝑗 𝜔 𝑡
𝑔(𝑡) = sin (−𝜔0 𝑡) = 𝑒 0 − 𝑒 0 = 𝑏1 𝑒𝑗 𝜔0 𝑡 + 𝑏−1 𝑒 −𝑗𝜔0 𝑡
2𝑗 2𝑗

1 1
and thus we see that 𝑏1 = − = 𝑎 −1 and 𝑏−1 = = 𝑎1 .
2𝑗 2𝑗

7|Pa ge
4.4.4. Time Scaling

Consider the signal 𝑔(𝑡) = 𝑥(𝛼𝑡), where 𝛼 ∈ ℝ>0 . Thus, 𝑔(𝑡) is a time-scaled version of 𝑥(𝑡).
𝑇
Note that the period of 𝑔(𝑡) is 2 , where 𝑇 is the period of 𝑥(𝑡). We have

𝑔(𝑡) = 𝑥(𝛼𝑡) = ∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝛼𝑡

𝑘=−∞

Thus, 𝑔(𝑡) has the same Fourier series coefficients as 𝑥(𝑡) , but the Fourier series
representation has changed: the frequency is now 𝜔0 𝛼 rather than 𝜔0 , to reflect the fact that
𝑇
the harmonic family is in terms of the new period 𝛼 rather than 𝑇.

Example

Consider 𝑥(𝑡) = cos (𝜔0 𝑡), which has series representation

1 𝑗 𝜔 𝑡 1 −𝑗𝜔 𝑡
𝑥(𝑡) = 𝑒 0 + 𝑒 0
2 2

Then we have

1 1
𝑔(𝑡) = 𝑥(𝛼𝑡) = cos (𝜔0 𝛼𝑡) = 𝑒𝑗 𝜔0 𝛼𝑡 + 𝑒 −𝑗𝜔0 𝛼𝑡 .
2 2

4.4.5. Multiplication

Let 𝑥1 (𝑡) and 𝑥2 (𝑡) be 𝑇-periodic signals with Fourier series 𝑎𝑘 and 𝑏𝑘 respectively. Consider
the signal 𝑔(𝑡) = 𝑥1 (𝑡)𝑥2 (𝑡), and note that 𝑔(𝑡) is also 𝑇-periodic. We have
∞ ∞ ∞ ∞

𝑔(𝑡) = 𝑥1 (𝑡)𝑥2 (𝑡) = ∑ 𝑎𝑙 𝑒𝑗𝑙𝜔0 𝑡 ∑ 𝑏𝑛 𝑒𝑗𝑛𝜔0 𝑡 = ∑ ∑ 𝑎𝑙 𝑏𝑛 𝑒𝑗 (𝑙+𝑛)𝜔0 𝑡

𝑙=−∞ 𝑛=−∞ 𝑙=−∞ 𝑛=−∞

Define 𝑘 = 𝑙 + 𝑛, so that
∞ ∞ ∞ ∞

𝑔(𝑡) = ∑ ∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑙𝜔0 𝑡 = ∑ ( ∑ 𝑎𝑙 𝑏𝑘−𝑙 )𝑒𝑗𝑘 𝜔0 𝑡

𝑙=−∞ 𝑙=−∞ 𝑘=−∞ 𝑙=−∞

Thus the Fourier series coefficients of 𝑔(𝑡) are given by

∞
FS
𝑔(𝑡) = 𝑥1 (𝑡)𝑥2 (𝑡) ⟷ ∑ 𝑎𝑙 𝑏𝑘−𝑙 .
𝑙=−∞

8|Pa ge
In other words, the Fourier series coefficients of a product of two signals are given by the
convolution of the corresponding Fourier series coefficients.

Example

Consider the signals 𝑥1 (𝑡) = cos (𝜔0 𝑡) and 𝑥2 (𝑡) = sin (𝜔0 𝑡) . Define 𝑔(𝑡) = 𝑥1 (𝑡)𝑥2 (𝑡) .
The Fourier series representations of 𝑥1 (𝑡) and 𝑥2 (𝑡) are given by

1 𝑗 𝜔 𝑡 1 −𝑗𝜔 𝑡 1 1
𝑥1 (𝑡) = 𝑒 0 + 𝑒 0 , 𝑥2 (𝑡) = 𝑒𝑗 𝜔0 𝑡 − 𝑒 −𝑗𝜔0 𝑡
2 2 2𝑗 2𝑗

1
Denote the Fourier series coefficients of 𝑥1 (𝑡) by the sequence 𝑎𝑘 , with 𝑎 −1 = 𝑎1 = 2, and
𝑎𝑘 = 0 otherwise. Similarly, denote the Fourier series coefficients of 𝑥2 (𝑡) by the sequence 𝑏𝑘 ,
1 1
with 𝑏−1 = − 2𝑗 , 𝑏(1) = 2𝑗 , and 𝑏𝑘 = 0 otherwise. Denote the Fourier series coefficients of
𝑔(𝑡) by 𝑐𝑘, 𝑘 ∈ ℤ. Then we have
∞

𝑐𝑘 = 𝑎𝑘 ∗ 𝑏𝑘 = ∑ 𝑎𝑙 𝑏𝑘−𝑙
𝑙=−∞

1 1
Convolving the two sequences given above, we see that 𝑐−2 = − 4𝑗 , 𝑐2 = 4𝑗 , and 𝑐𝑘 = 0
otherwise. Thus

1 𝑗2𝜔 𝑡 1 −𝑗2𝜔 𝑡
𝑔(𝑡) = 𝑥1 (𝑡)𝑥2 (𝑡) = cos (𝜔0 𝑡)sin (𝜔0 𝑡) = 𝑒 0 − 𝑒 0
4𝑗 4𝑗

1
Noting that cos (𝜔0 𝑡)sin (𝜔0 𝑡) = 2 sin (2𝜔0 𝑡) , we see that the above expression is, in fact,
correct.

4.4.6. Parseval’s Theorem of Continuous time signals

Consider a 𝑇-periodic signal 𝑥(𝑡), and let 𝑎𝑘 , 𝑘 ∈ ℤ be its Fourier series coefficients. Now let
us consider the average power of 𝑥(𝑡) over one period, defined as

1 𝑡0 +𝑇
∫ |𝑥(𝑡)| 2 𝑑𝑡
𝑇 𝑡0

Substituting |𝑥(𝑡)| 2 = 𝑥 ∗ (𝑡)𝑥(𝑡) and the Fourier series for 𝑥(𝑡) we have
∞ ∞
1 𝑡0 +𝑇 2
1 𝑡0 +𝑇
∫ |𝑥(𝑡)| 𝑑𝑡 = ∫ ( ∑ 𝑎∗𝑘 𝑒 −𝑗𝑘𝜔0 𝑡 ) ( ∑ 𝑎∗𝑛 𝑒𝑗𝑛𝜔0 𝑡 ) 𝑑𝑡
𝑇 𝑡0 𝑇 𝑡0
𝑘=−∞ 𝑛=−∞
∞ ∞
𝑡0 +𝑇
1
= ∑ ∑ 𝑎𝑘∗ 𝑎𝑛 ∫ 𝑒 −𝑗𝑘𝜔0 𝑡 𝑒𝑗𝑛𝜔0 𝑡 𝑑𝑡
𝑇 𝑡0
𝑘=−∞ 𝑛=−∞

9|Pa ge
Using the fact that 𝑒𝑗𝑘 𝜔0 𝑡 and 𝑒𝑗𝑛𝜔0 𝑡 are orthogonal for 𝑘 ≠ 𝑛, we have
∞ ∞
1 𝑡0 +𝑇
∫ |𝑥(𝑡)|2 𝑑𝑡 = ∑ 𝑎𝑘∗ 𝑎𝑘 = ∑ |𝑎𝑘 | 2
𝑇 𝑡0
𝑘=−∞ 𝑘=−∞

This leads to Parseval's Theorem: for a 𝑇-periodic signal 𝑥(𝑡) with Fourier series coefficients
𝑎𝑘 , 𝑘 ∈ ℤ, we have
∞
1 𝑡0 +𝑇
∫ |𝑥(𝑡)|2 𝑑𝑡 = ∑ |𝑎𝑘 |2
𝑇 𝑡0
𝑘=−∞

Example

1
Consider the signal 𝑥(𝑡) = cos (𝜔0 𝑡) , with Fourier coefficients 𝑎1 = 𝑎−1 = and 𝑎𝑘 = 0
2
otherwise. We have

1 𝑡0 +𝑇 1 𝑡0 +𝑇 1 𝑡0 +𝑇 1 1
∫𝑡0 |𝑥(𝑡)|2 𝑑𝑡 = 𝑇 ∫𝑡0 cos2 (𝜔0 𝑡)𝑑𝑡 = 𝑇 ∫𝑡0 (1 + cos (2𝜔0 𝑡))𝑑𝑡 = 2.
𝑇 2

We also have
∞
1 1 1
∑ |𝑎𝑘 | 2 = 𝑎21 + 𝑎2−1 = + =
4 4 2
𝑘=−∞

which agrees with the direct calculation of average power, as indicated by Parseval's Theorem.

4.5. Fourier Series Representation of Discrete-Time Periodic Signals

We now turn our attention to the discrete-time Fourier series. Specifically, consider the
discrete-time signal 𝑥[𝑛], 𝑛 ∈ ℤ , and suppose it is 𝑁 -periodic. Let
2𝜋
𝜔0 = , and note that the signal 𝜙𝑘[𝑛] = 𝑒𝑗𝑘 𝜔0 𝑛 is also 𝑁-periodic for any 𝑘 ∈ ℤ. Thus, as in
𝑁
the case for continuous-time signals, we would like to write 𝑥[𝑛] as a linear combination of
signals from the harmonic family 𝜙𝑘[𝑛], 𝑘 ∈ ℤ, i.e.,
∞ ∞
2𝜋
𝑥[𝑛] = ∑ 𝑎𝑘 𝑒𝑗𝑘 𝜔0 𝑛 = ∑ 𝑎𝑘 𝑒 𝑗𝑘 𝑁 𝑛
𝑘=−∞ 𝑘=−∞

At this point, we encounter the first main difference between the discrete-time and continuous-
time Fourier series. Recall that a discrete-time complex exponential with frequency 𝜔0 is the
same as a discrete-time complex exponential with frequency 𝜔0 + 2𝜋. Specifically, for any
𝑘 ∈ ℤ, consider

10 | P a g e
 𝜙𝑘+𝑁 [𝑛] = 𝑒𝑗 (𝑘+𝑁)𝜔0 𝑛 = 𝑒𝑗𝑘 𝜔0 𝑛 𝑒𝑗𝑁 𝜔0 𝑛 = 𝑒𝑗𝑘𝜔0 𝑛

since 𝑁𝜔0 = 2𝜋. Thus, there are only 𝑁 different complex exponentials in the discrete-time
harmonic family for the fundamental frequency 𝜔0 , and so we have the following.

The discrete-time Fourier series for an 𝑁-periodic signal 𝑥[𝑛] is given by

𝑛0 +𝑁−1 𝑛0 +𝑁−1
2𝜋
𝑥[𝑛] = ∑ 𝑎𝑘 𝑒𝑗𝑘 𝜔0 𝑛 = ∑ 𝑎𝑘 𝑒 𝑗𝑘 𝑁 𝑛
𝑘=𝑛0 𝑘=𝑛0

where 𝑛0 is any integer.

In other words, the discrete-time signal can be written in terms of any 𝑁 contiguous multiples
of the fundamental frequency 𝜔0 . Since there are only 𝑁 coefficients 𝑎𝑘 in this representation,
and since it does not matter which contiguous 𝑁 members of the harmonic family we choose,
the Fourier series coefficients are 𝑁-periodic as well, i.e., 𝑎𝑘 = 𝑎𝑘+𝑁 for all 𝑘 ∈ ℤ.

To find the Fourier series coefficients in (4.2), we use a similar trick as in the continuous-time
case. Specifically, first multiply both sides of (4.2) by 𝑒 −𝑗𝑟𝜔0 𝑛 , where 𝑟 is any integer, and sum
both sides over 𝑁 terms. This gives

𝑛1 +𝑁−1 𝑛1 +𝑁 −1 𝑛0 +𝑁−1

∑ 𝑥[𝑛]𝑒 −𝑗𝑟𝜔0 𝑛 = ∑ ∑ 𝑎𝑘 𝑒𝑗 (𝑘−𝑟)𝜔0 𝑛

𝑛=𝑛1 𝑛=𝑛1 𝑘=𝑛0

where 𝑛1 is any integer. Interchanging the summations, we have

𝑛1 +𝑁−1 𝑛0 +𝑁−1 𝑛1 +𝑁 −1

∑ 𝑥[𝑛]𝑒 −𝑗𝑟𝜔0 𝑛 = ∑ 𝑎𝑘 ∑ 𝑒𝑗 (𝑘−𝑟)𝜔0 𝑛

𝑛=𝑛1 𝑘=𝑛0 𝑛=𝑛1

Now suppose that 𝑘 − 𝑟 is a multiple of 𝑁. In this case we obtain

𝑛1 +𝑁 −1 𝑛1 +𝑁−1

∑ 𝑒𝑗 (𝑘−𝑟)𝜔0 𝑛 = ∑ 1=𝑁
𝑛=𝑛1 𝑛=𝑛1

On the other hand, if 𝑟 − 𝑘 is not a multiple of 𝑁, we use the finite sum formula to obtain
𝑛1 +𝑁−1
𝑒𝑗 (𝑘−𝑟)𝜔0 𝑛1 − 𝑒𝑗 (𝑘−𝑟)𝜔0 (𝑛1 +𝑁)
∑ 𝑒𝑗 (𝑘−𝑟)𝜔0 𝑛 = =0
𝑒𝑗 (𝑘−𝑟)𝜔0 − 1
𝑛=𝑛1

Thus, we have the following.

The discrete-time Fourier series coefficients are given by

11 | P a g e
 𝑛1 +𝑁−1
1
𝑎𝑘 = ∑ 𝑥[𝑛]𝑒 −𝑗𝑘𝜔0 𝑛
𝑁
𝑛=𝑛1

where 𝑛1 is any integer. The Fourier series coefficients are

𝑁-periodic, i.e., 𝑎𝑘 = 𝑎𝑘+𝑁 .

Example

Consider the 𝑁-periodic signal 𝑥[𝑛], which is equal to 1 for −𝑁1 ≤ 𝑛 ≤ 𝑁1 and zero otherwise
(modulo the periodic constraints).

We have
𝑁1
1
𝑎𝑘 = ∑ 𝑒 −𝑗𝑘𝜔0 𝑛
𝑁
𝑛=−𝑁1

2𝑁1 +1
If 𝑘 = 0, we have 𝑎0 = . For 𝑘 ∈ {1,2, … , 𝑁 − 1}, we have (via the finite sum formula)
𝑁

1 1 1
1 𝑒𝑗𝑘 𝜔0 𝑁1 − 𝑒𝑗𝑘 𝜔0 (𝑁1 +1) 1 𝑒 −𝑗𝑘𝜔0 2 𝑒 𝑗𝑘𝜔0 (𝑁1 +2) − 𝑒 𝑗𝑘𝜔0 (𝑁1 +2)
𝑎𝑘 = =
𝑁 1 − 𝑒 −𝑗𝑘𝜔0 𝑁 −𝑗𝑘𝜔0 12 1 1
𝑒 𝑒 𝑗𝑘𝜔0 2 − 𝑒 −𝑗𝑘𝜔0 2
1
sin (𝑘𝜔0 (𝑁1 + 2))
1
= 𝜔
𝑁 sin (𝑘 20 )

It is of interest to note that we do not have to worry about convergence conditions for discrete-
time Fourier series, as we did in the continuous-time case. Specifically, for an 𝑁 -periodic
discrete-time signal 𝑥[𝑛], we only require 𝑁 numbers to completely specify the entire signal.
The Fourier series coefficients 𝑎𝑘 , 𝑘 ∈ {0,1, … , 𝑁 − 1} thus contain as much information as
the signal itself, and form a perfect representation of the signal. In other words, for discrete-
time signals, the Fourier series representation is just a transformation of the signal into another
form; we do not encounter discrepancies like the Gibbs phenomenon in discrete-time.

4.6. Properties of Discrete Time Fourier Series

The discrete-time Fourier series has properties that can be derived in almost the same way as
the properties for the continuous-time Fourier series (linearity, time-shifting, etc.). Here we
will just discuss the multiplication property and Parseval's theorem.

4.6.1. Multiplication of Discrete-Time periodic signals

12 | P a g e
First, let's start with an example. Consider two 2-periodic signals 𝑥1 [𝑛] and 𝑥2 [𝑛]. We know
that the Fourier series respresentations can be uniquely specified by two coefficients.
Specifically,

𝑥1 [𝑛] = 𝑎0 + 𝑎1 𝑒𝑗 𝜔0 𝑛 , 𝑥2 [𝑛] = 𝑏0 + 𝑏1 𝑒𝑗 𝜔0 𝑛

Now consider the product

𝑔[𝑛] = 𝑥1 [𝑛]𝑥2 [𝑛] = (𝑎0 + 𝑎1 𝑒𝑗 𝜔0 𝑛 )(𝑏0 + 𝑏1 𝑒𝑗𝜔0 𝑛 )

= 𝑎0 𝑏0 + (𝑎0 𝑏1 + 𝑎1 𝑏0 )𝑒𝑗 𝜔0 𝑛 + 𝑎1 𝑏1 𝑒𝑗2𝜔0 𝑛

2𝜋
Now, note that 𝜔0 = , and thus 𝑒𝑗2𝜔0 𝑛 = 1. This gives
2

𝑔[𝑛] = (𝑎0 𝑏0 + 𝑎1 𝑏1 ) + (𝑎0 𝑏1 + 𝑎1 𝑏0 )𝑒𝑗 𝜔0 𝑛

Thus, the Fourier series coefficients of 𝑔[𝑛] are given by 𝑐0 = 𝑎0 𝑏0 + 𝑎1 𝑏1 and 𝑐1 = 𝑎0 𝑏1 +

𝑎1 𝑏0 . We can write these in a unform way as follows:
1

𝑐0 = 𝑎0 𝑏0 + 𝑎1 𝑏1 = 𝑎0 𝑏0 + 𝑎1 𝑏−1 = ∑ 𝑎𝑙 𝑏−𝑙
𝑙=0
1

𝑐1 = 𝑎0 𝑏1 + 𝑎1 𝑏0 = ∑ 𝑎𝑙 𝑏1−𝑙 ,
𝑙=0

where we used the fact that 𝑏1 = 𝑏−1 by the periodic nature of the discrete-time Fourier series
coefficients. The above expressions show that the Fourier series coefficients of the product of
the two signals are given by a form of convolution of the coefficients of those signals; however,
the convolution is over a finite number of terms, as opposed to over all time-indices.

Let us generalize this to functions with a larger period. Let 𝑥1 [𝑛] and 𝑥2 [𝑛] be two N-periodic
discrete-time signals, with discrete-time Fourier series coefficients 𝑎𝑘 and 𝑏𝑘 , respectively. We
have

𝑁−1 𝑁−1

𝑥1 [𝑛] = ∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑛 , 𝑥2 [𝑛] = ∑ 𝑏𝑘 𝑒𝑗𝑘𝜔0 𝑛

𝑘=0 𝑘=0

Define the signal 𝑔[𝑛] = 𝑥1 [𝑛]𝑥2 [𝑛]. We have

𝑁−1 𝑁−1
𝑗𝑙𝜔0 𝑛
𝑔[𝑛] = 𝑥1 [𝑛]𝑥2 [𝑛] = (∑ 𝑎𝑙 𝑒 ) (∑ 𝑏𝑟 𝑒𝑗𝑟𝜔0 𝑛 )
𝑙=0 𝑟=0
𝑁−1 𝑁−1

= ∑ ∑ 𝑎𝑙 𝑏𝑟𝑒𝑗 (𝑙+𝑟)𝜔0 𝑛 ,
𝑙=0 𝑟 =0

13 | P a g e
where we have used 𝑙 and 𝑟 as the indices in the Fourier series in order to keep the terms in the
two sums distinct.

Define the new variable 𝑘 = 𝑙 + 𝑟. Substituting into the above expression, this gives

𝑁−1 𝑙+𝑁−1

𝑔[𝑛] = ∑ ∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛

𝑙=0 𝑘=𝑙
𝑁−1 𝑁−1 𝑙+𝑁−1

= ∑ (∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛 + ∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛 )

𝑙=0 𝑘 =𝑙 𝑘=𝑁
𝑁−1 𝑁−1 𝑙 −1

= ∑ (∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛 + ∑ 𝑎𝑙 𝑏𝑘+𝑁−𝑙 𝑒𝑗(𝑘+𝑁 )𝜔0 𝑛 )

𝑙=0 𝑘 =𝑙 𝑘=0
𝑁−1 𝑁−1 𝑙 −1

= ∑ (∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛 + ∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛 )

𝑙=0 𝑘 =𝑙 𝑘=0
𝑁−1 𝑁−1

= ∑ ∑ 𝑎𝑙 𝑏𝑘−𝑙 𝑒𝑗𝑘𝜔0 𝑛
𝑙=0 𝑘=0
𝑁−1 𝑁−1

= ∑ (∑ 𝑎𝑙 𝑏𝑘−𝑙 )𝑒𝑗𝑘 𝜔0 𝑛
𝑘=0 𝑙=0

Thus, the Fourier series coefficients of 𝑔[𝑛] are given by

𝑁−1

𝑐𝑘 = ∑ 𝑎𝑙 𝑏𝑘−𝑙
𝑙=0

for 0 ≤ 𝑘 ≤ 𝑁 − 1 . The above convolution is known as the periodic convolution of two

periodic signals; for any 𝑘, there are only 𝑁 terms in the summation. We saw an example of
this with the 2-periodic signals that we had above. In essence, the above calucations are simply
multiplying together the discrete-time Fourier series representations of the two signals (each of
which has 𝑁 terms), and then using the periodicity of the discrete-time complex exponentials
in the frequencies to combine terms together. Note that we can actually perform the sum over
any continuous 𝑁 values of 𝑙, since all of the signals involved are periodic.

4.6.2. Parseval’s theorem for discrete-time Signals

Let 𝑥[𝑛] be an 𝑁-periodic discrete-time signal, with Fourier series coefficients 𝑎𝑘 , 0 ≤ 𝑘 ≤

𝑁 − 1. The average power of 𝑥[𝑛] over one period is

𝑛0 +𝑁−1
1
∑ |𝑥[𝑛]|2
𝑁
𝑛=𝑛0

where 𝑛0 is any integer. Parseval's theorem for discrete-time signals states the following.

14 | P a g e
 𝑛0 +𝑁−1 𝑁−1
1
∑ |𝑥[𝑛]|2 = ∑ | 𝑎𝑘 |2
𝑁
𝑛=𝑛0 𝑘=0

The following example illustrates the application of the various facts that we have seen about
the discrete-time Fourier series.

Example

Suppose we are told the following facts about a discrete-time signal 𝑥[𝑛].

• 𝑥[𝑛] is periodic with period 𝑁 = 6.

• ∑5𝑛=0 𝑥[𝑛] = 2.
• ∑7𝑛=2 (−1) 𝑛 𝑥[𝑛] = 1.
• 𝑥[𝑛] has the minimum power per period of all signals satisfying the preceding three
conditions.
The above facts are sufficient for us to uniquely determine 𝑥[𝑛]. First, note that the Fourier
series representation of 𝑥[𝑛] is

5 5
𝜋
𝑥[𝑛] = ∑ 𝑎𝑘 𝑒𝑗𝑘 𝜔0 𝑛 = ∑ 𝑎𝑘 𝑒 𝑗𝑘 3 𝑛
𝑘=0 𝑘=0

where we used the fact that the signal is 6-periodic.

From the second fact, we have

𝑁−1
1 2 1
𝑎0 = ∑ 𝑥[𝑛] = =
𝑁 6 3
𝑘=0

𝜋
To use the third fact, note that (−1) 𝑛 = 𝑒 −𝑗𝜋𝑛 = 𝑒 −𝑗33 𝑛 . Thus, the third fact seems to be
related to the Fourier series coefficient for 𝑘 = 3. Specifically, we have
7
1 1
𝑎3 = ∑ 𝑥[𝑛]𝑒 −𝑗𝜋𝑛 =
𝑁 6
𝑘=2

To use the last fact, note from Parseval's theorem that the power of the signal over one period
is given by

5 5
1
∑ |𝑥[𝑛]| 2 = ∑ |𝑎𝑘 | 2
𝑁
𝑛=0 𝑘=0
= |𝑎0 |2 + | 𝑎1 |2 + |𝑎2 |2 + | 𝑎3 |2 + |𝑎4 | 2 + |𝑎5 |2

15 | P a g e
We are told that 𝑥[𝑛] has the minimum average power over all signals that satisfy the other
three conditions. Since the other three conditions have already set 𝑎0 and 𝑎1 , the average power
is given by setting all of the other Fourier series coefficients to 0 . Thus, we have
5
𝜋 1 1
𝑥[𝑛] = ∑ 𝑎𝑘𝑒 𝑗𝑘 3 𝑛 = 𝑎0 + 𝑎3 𝑒𝑗𝜋𝑛 = + (−1) 𝑛
3 6
𝑘=0

16 | P a g e
 Chapter 5

Fourier Transform

5.1. Introduction

Suppose that we are given a signal 𝑥(𝑡) that is aperiodic. As a concrete example, suppose that
𝑥(𝑡) is a solitary square pulse, with 𝑥(𝑡) = 1 if −𝑇1 ≤ 𝑡 ≤ 𝑇1 , and zero elsewhere. Clearly 𝑥(𝑡)
is not periodic. Now define a new signal 𝑥̃(𝑡) which is a periodic extension of 𝑥(𝑡) with period
𝑇. In other words, 𝑥̃(𝑡) is obtained by repeating 𝑥(𝑡), where each copy is shifted 𝑇 units in time.
This 𝑥̃(𝑡) has a Fourier series representation, which we found in the last chapter to be:

2𝑇1 2sin (𝑘𝜔0 𝑇1 )

𝑎0 = , 𝑎𝑘 =
𝑇 𝑘𝜔0 𝑇

for 𝑘 ∈ ℤ.

Now recall that the Fourier series coefficients are calculated as follows:
𝑇
1 2
𝑎𝑘 = ∫ 𝑥̃(𝑡)𝑒 −𝑗𝑘𝜔0 𝑡 𝑑𝑡
𝑇 𝑇
−
2

However, we note that 𝑥(𝑡) = 𝑥̃(𝑡) in the interval of integration, and thus:
𝑇
1 2
𝑎𝑘 = ∫ 𝑥(𝑡)𝑒 −𝑗𝑘𝜔0 𝑡 𝑑𝑡
𝑇 𝑇
−
2

Furthermore, since 𝑥(𝑡) is zero for all 𝑡 outside the interval of integration, we can expand the
limits of the integral to obtain:
∞
1
𝑎𝑘 = ∫ 𝑥(𝑡)𝑒 −𝑗𝑘𝜔0 𝑡 𝑑𝑡
𝑇 −∞

Let us define
∞
𝑋(𝑗𝜔 ) = ∫ 𝑥(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞

This is called the Fourier transform of the signal 𝑥(𝑡), and the Fourier series coefficients can
1
be viewed as samples of the Fourier transform, scaled by 𝑇 , i.e.,

1
𝑎𝑘 = 𝑋(𝑗𝑘𝜔0 ), 𝑘 ∈ ℤ
𝑇

17 | P a g e
Now consider the fact that
∞ ∞
𝑗𝑘𝜔 0 𝑡
1
𝑥̃(𝑡) = ∑ 𝑎𝑘 𝑒 = ∑ 𝑋( 𝑗𝑘𝜔0 )𝑒𝑗𝑘 𝜔0𝑡
𝑇
𝑘=−∞ 𝑘=−∞

2𝜋
Since 𝜔0 = , this becomes
𝑇

∞
1
𝑥̃(𝑡) = ∑ 𝑋(𝑗𝑘𝜔0 )𝑒𝑗𝑘𝜔0 𝑡𝜔0
2𝜋
𝑘=−∞

Now consider what happens as the period 𝑇 gets bigger. In this case, 𝑥̃(𝑡) approaches 𝑥(𝑡), and
so the above expression becomes a representation of 𝑥(𝑡). As 𝑇 → ∞ , we have 𝜔0 → 0. Since
each term in the summand can be viewed as the area of the rectangle whose height is
𝑋( 𝑗𝑘𝜔0 ) 𝑒𝑗𝑘𝜔0 𝑡 and whose base goes from 𝑘𝜔0 to (𝑘 + 1)𝜔0 , we see that as 𝜔0 → 0, the sum on
the right hand side approaches the area underneath the curve 𝑋(𝑗𝜔)𝑒𝑗𝜔𝑡 (where 𝑡 is held fixed).
Thus, as 𝑇 → ∞ we have
∞
1
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔)𝑒𝑗𝜔𝑡 𝑑𝜔
2𝜋 −∞

Thus we have the following.

Given a continuous-time signal 𝑥(𝑡), the Fourier

Transform of the signal is given by

∞
𝑋(𝑗𝜔 ) = ∫ 𝑥(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞

The Inverse Fourier Transform of the signal is given by

1 ∞
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔)𝑒𝑗𝜔𝑡 𝑑𝜔
2𝜋 −∞

The Fourier transform 𝑋(𝑗𝜔) is also called the spectrum of the signal, as it represents the
contribution of the complex exponential of frequency 𝜔 to the signal 𝑥(𝑡).

Example 5.1. Consider the signal 𝑥(𝑡) = 𝑒 −𝑎𝑡 𝑢(𝑡), 𝑎 ∈ ℝ>0 . The Fourier transform of this
signal is
∞ ∞
𝑋(𝑗𝜔) = ∫ 𝑥(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡 = ∫ 𝑒 −𝑎𝑡 𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞ 0
∞
1 −(𝑎+𝑗𝜔 )𝑡 |
=− 𝑒
𝑎 + 𝑗𝜔 0
1
= .
𝑎 + 𝑗𝜔

18 | P a g e
To visualize 𝑋(𝑗𝜔), we plot its magnitude and phase on separate plots (since 𝑋(𝑗𝜔) is complex
valued in general). We have

1 𝜔
|𝑋(𝑗𝜔)| = , ∠𝑋(𝑗𝜔) = −tan−1 ( ).
√𝑎2 + 𝜔2 𝑎

Example 5.2. Consider the signal 𝑥(𝑡) = 𝛿(𝑡). We have

∞
𝑋(𝑗𝜔 ) = ∫ 𝛿(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡 = 1
−∞

In other words, the spectrum of the impulse function has an equal contribution at all
frequencies.

Example 5.3. Consider the signal 𝑥(𝑡) which is equal to 1 for −𝑇1 ≤ 𝑡 ≤ 𝑇1 and zero
elsewhere. We have
𝑇1
1 2sin (𝜔𝑇1 )
𝑋(𝑗𝜔) = ∫ 𝑒 −𝑗𝜔𝑡 𝑑𝑡 = (𝑒𝑗𝜔 𝑇1 − 𝑒 −𝑗𝜔 𝑇1 ) =
−𝑇1 𝑗𝜔 𝜔

Example 5.4. Consider the signal whose Fourier transform is

1, |𝜔| ≤ 𝑊
𝑋(𝑗𝜔) = {
0 |𝜔| > 𝑊

We have
∞ 𝑊
1 1
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔 )𝑒𝑗𝜔𝑡 𝑑𝜔 = 𝑒𝑗𝜔𝑡 𝑑𝜔
∫
2𝜋 −∞ 2𝜋 −𝑊
1 1 𝑗𝜔𝑡 𝑊
= 𝑒 |
2𝜋 𝑗𝑡 −𝑊
sin (𝑊𝑡)
=
𝜋𝑡

The previous two examples showed the following. When 𝑥(𝑡) is a square pulse,

2sin (𝜔 𝑇1 ) sin (𝑊𝑡)

then 𝑋(𝑗𝜔 ) = and when 𝑋(𝑗𝜔) is a square pulse, 𝑥(𝑡) = . This is an example of
𝜔 𝜋𝑡
the duality property of Fourier transforms, which we will see later.

sin (𝑊𝑡)
Functions of the form will show up frequently and are called sinc functions.
𝜋𝑡
Specifically

sin (𝜋𝜃)
sinc (𝜃) =
𝜋𝜃

19 | P a g e
 2sin (𝜔 𝑇1 ) 𝜔𝑇1 sin (𝑊𝑡) 𝑊 𝑊𝑡
Thus 𝜔
= 2𝑇1 sinc (
𝜋
) and
𝜋𝑡
=
𝜋
sinc (
𝜋
).

5.2. Existence of Fourier Transform

Just as we saw with the Fourier series for periodic signals, there are some rather mild conditions
under which a signal 𝑥(𝑡) is guaranteed to have a Fourier transform (such that the inverse
Fourier transform converges to the true signal). Specifically, there are a set of sufficient
conditions (also called Dirichlet conditions) under which a continuous-time signal 𝑥(𝑡) is
guaranteed to have a Fourier transform:
∞
1. 𝑥(𝑡) is absolutely integrable: ∫−∞ |𝑥(𝑡)|𝑑𝑡 < ∞ .

2. 𝑥(𝑡) has a finite number of maxima and minima in any finite interval.

3. 𝑥(𝑡) has a finite number of discontinuities in any finite interval, and each of these
discontinuities is finite.
If all the above conditions are satisfied, 𝑥(𝑡) is guaranteed to have a Fourier transform. Note
that this only a sufficient set of conditions, and not necessary.

An alternate sufficient condition is that the signal have finite energy (i.e., that it be square
integrable):
∞
∫ |𝑥(𝑡)|2 𝑑𝑡 < ∞
−∞

1
For example, the signal 𝑥(𝑡) = 𝑡 𝑢(𝑡 − 1) is square integrable, but not integrable. Thus, the
finite energy condition guarantees that 𝑥(𝑡) will have a Fourier transform, whereas the
Dirichlet conditions do not apply.

5.3. Fourier Transform of Periodic Signals

The Fourier transform can also be applied to certain periodic signals (although such signals
will not be integrable or square integrable over the entire time-axis). A direct application of the
Fourier transform equation to such signals will not necessarily yield a meaningful answer, due
to the fact that periodic signals do not die out. Instead, we will work backwards by starting with
a frequency domain signal and doing an inverse Fourier transform to see what pairs arise.

Thus, consider the signal 𝑥(𝑡) whose Fourier transform is:

𝑋(𝑗𝜔) = 2𝜋𝛿(𝜔 − 𝜔0 )

i.e., the frequency domain signal is a single impulse at 𝜔 = 𝜔0 , with area 2𝜋. Using the
inverse Fourier transform, we obtain

20 | P a g e
 ∞ ∞
1
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔 )𝑒𝑗𝜔𝑡 𝑑𝜔 =∫ 𝛿(𝜔 − 𝜔0 )𝑒𝑗𝜔𝑡 𝑑𝜔
2𝜋 −∞ −∞
= 𝑒𝑗 𝜔 0 𝑡 .

Thus, the Fourier transform of 𝑥(𝑡) = 𝑒𝑗𝜔0 𝑡 is 𝑋(𝑗𝜔 ) = 2𝜋𝛿 (𝜔 − 𝜔0 ). Similarly, if

∞

𝑋(𝑗𝜔 ) = ∑ 𝑎𝑘 2𝜋𝛿( 𝜔 − 𝑘𝜔0 )

𝑘=−∞

then an application of the inverse Fourier transforms gives

∞

𝑥(𝑡) = ∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑡
𝑘=−∞

In other words, if 𝑥(𝑡) is a periodic signal with Fourier series coefficients 𝑎𝑘 , then the Fourier
transform of 𝑥(𝑡) consists of a sequence of impulse functions, each spaced at multiples of 𝜔0 ;
the area of the impulse at 𝑘𝜔0 will be 2𝜋𝑎𝑘 .

Example 5.5. Consider the signal 𝑥(𝑡) = 𝑐𝑜𝑠 (𝜔0 𝑡) . The Fourier series coefficients are 𝑎1 =
1
𝑎−1 = . Thus, the Fourier transform of this signal is given by
2

𝑋(𝑗𝜔) = 𝑎1 2𝜋𝛿(𝜔 − 𝜔0 ) + 𝑎−1 2𝜋𝛿(𝜔 + 𝜔0 ) = 𝜋𝛿 ( 𝜔 − 𝜔0 ) + 𝜋𝛿 (𝜔 + 𝜔0 )

Example 5.6. Consider the periodic signal:

∞

𝑥(𝑡) = ∑ 𝛿(𝑡 − 𝑛𝑇)

𝑛=−∞

The Fourier series coefficients for this signal are given by

𝑇 ∞
1 2 1
𝑎0 = ∫ ∑ 𝛿(𝑡 − 𝑛𝑇)𝑑𝑡 =
𝑇 𝑇 𝑇
− 𝑛=−∞
2
𝑇 ∞
1 2 1
𝑎𝑘 = ∫ ∑ 𝛿(𝑡 − 𝑛𝑇)𝑒 −𝑗𝑘𝜔0 𝑡 𝑑𝑡 = .
𝑇 − 𝑇 𝑇
2 𝑛=−∞

Thus, the Fourier transform of this signal is given by

∞ ∞
2𝜋 2𝑘𝜋
𝑋(𝑗𝜔) = ∑ 𝑎𝑘 2𝜋𝛿(𝜔 − 𝑘𝜔0 ) = ∑ 𝛿 (𝜔 − )
𝑇 𝑇
𝑘=−∞ 𝑘=−∞

Thus, if 𝑥(𝑡) is an impulse train with period 𝑇, its Fourier transform is also an impulse train in
2𝜋
the frequency domain, except with period . Once again, we see that if 𝑇 increases (i.e., the
𝑇
period increases in the time-domain) we obtain a time-shrinking in the frequency domain.

21 | P a g e
5.4. Properties of Continuous-Time Fourier Transform

We will now discuss various properties of the Fourier transform. As with the Fourier series, we
will find it useful to introduce the following notation. Suppose 𝑥(𝑡) is a time-domain signal,
and 𝑋(𝑗𝜔) is its Fourier transform. We then say

𝑋(𝑗𝜔) = ℱ{𝑥(𝑡)}
𝑥(𝑡) = ℱ −1 {𝑋(𝑗𝜔)}.

We will also use the notation.

F
𝑥(𝑡) ⟷ 𝑋(𝑗𝜔)

to indicate that 𝑥(𝑡) and 𝑋(𝑗𝜔) are Fourier transform pairs.

5.4.1. Linearity

The first property of Fourier transforms is easy to show:

ℱ{𝛼𝑥1 (𝑡) + 𝛽𝑥2 (𝑡) } = 𝛼ℱ{𝑥1 (𝑡)} + 𝛽ℱ {𝑥2 (𝑡) },

which follows immediately from the definition of the Fourier transform.

5.4.2. Time-Shifting

Suppose 𝑥(𝑡) is a signal with Fourier transform 𝑋(𝑗𝜔). Define 𝑔(𝑡) = 𝑥(𝑡 − 𝜏) where 𝜏 ∈ ℝ.
Then we have
∞ ∞
−𝑗𝜔𝑡
𝐺(𝑗𝜔) = ∫ 𝑔(𝑡)𝑒 𝑑𝑡 = ∫ 𝑥(𝑡 − 𝜏)𝑒 −𝑗𝜔𝑡 𝑑𝑡 = 𝑒 −𝑗𝜔𝜏 𝑋(𝑗𝜔).
−∞ −∞

Thus

ℱ{𝑥(𝑡 − 𝜏)} = 𝑒 −𝑗𝜔𝜏 𝑋(𝑗𝜔 ).

Note the implication: if we time-shift a signal, the magnitude of its Fourier transform is not
affected. Only the phase of the Fourier transform gets shifted by −𝜔𝜏 at each frequency 𝜔 .

5.4.3. Conjugation

Consider a signal 𝑥(𝑡). We have

22 | P a g e
 ∞ ∗ ∞
∗ −𝑗𝜔𝑡
𝑋 (𝑗𝜔) = (∫ 𝑥(𝑡)𝑒 𝑑𝑡) = ∫ 𝑥 ∗ (𝑡)𝑒𝑗𝜔𝑡 𝑑𝑡
−∞ −∞

Thus,
∞
𝑋 ∗ (−𝑗𝜔) = ∫ 𝑥 ∗ (𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡 = ℱ{𝑥 ∗ (𝑡)}
−∞

The above is true for any signal 𝑥(𝑡) that has a Fourier transform. Now suppose additionally
that 𝑥(𝑡) is a real-valued signal. Then we have 𝑥 ∗ (𝑡) = 𝑥(𝑡) for all 𝑡 ∈ ℝ. Thus

𝑋 ∗ (−𝑗𝜔 ) = ℱ {𝑥 ∗ (𝑡)} = ℱ{𝑥(𝑡)} = 𝑋(𝑗𝜔 ).

Based on the above relationship between 𝑋 ∗ (−𝑗𝜔 ) and 𝑋(𝑗𝜔) for real-valued signals, we see
the following. Write 𝑋(𝑗𝜔 ) in polar form as

𝑋(𝑗𝜔) = |𝑋(𝑗𝜔)|𝑒𝑗∠𝑋(𝑗𝜔)

Then we have

𝑋(−𝑗𝜔) = 𝑋 ∗ (𝑗𝜔 ) = |𝑋(𝑗𝜔)|𝑒 −𝑗∠𝑋(𝑗𝜔 ) .

Thus, for any 𝜔, 𝑋(−𝑗𝜔 ) has the same magnitude as 𝑋(𝑗𝜔 ), and the phase of 𝑋(−𝑗𝜔 ) is the
negative of the phase of 𝑋(𝑗𝜔) . Thus, when plotting 𝑋(𝑗𝜔) , we
only have to plot the magnitude and phase for positive values of 𝜔 , as the plots for negative
values of 𝜔 can be easily recovered according to the relationships described above.

Example 5.7. Consider again the signal 𝑥(𝑡) = 𝑒 −𝑎𝑡 𝑢(𝑡); we saw earlier that the Fourier
transform of this signal is

1
𝑋(𝑗𝜔) = .
𝑎 + 𝑗𝜔

It is easy to verify that

1
𝑋(−𝑗𝜔) = = 𝑋 ∗ (𝑗𝜔)
𝑎 − 𝑗𝜔

as predicted. Furthermore, we can see from the plots of the magnitude and phase of 𝑋(𝑗𝜔 ) that
the magnitude is indeed an even function, and the phase is an odd function.

Suppose further that 𝑥(𝑡) is even (in addition to being real-valued). Then we have 𝑥(𝑡) = 𝑥(−𝑡).
Then we have
∞ ∞ ∞
𝑗𝜔𝑡 𝑗𝜔𝑡
𝑋(−𝑗𝜔) = ∫ 𝑥(𝑡)𝑒 𝑑𝑡 = ∫ 𝑥(−𝑡)𝑒 𝑑𝑡 =∫ 𝑥(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞ −∞ −∞
= 𝑋(𝑗𝜔)

23 | P a g e
This, together with the fact that 𝑋(−𝑗𝜔) = 𝑋 ∗ (𝑗𝜔) for real-valued signals indicates that 𝑋(𝑗𝜔)
is real-valued and even.

Similarly, if 𝑥(𝑡) is real-valued and odd, we have 𝑋(𝑗𝜔) is purely imaginary and odd.

Example 5.8. Consider the signal 𝑥(𝑡) = 𝑒 −𝑎|𝑡| , where 𝑎 is a positive real number. This signal
is real-valued and even. We have
∞ 0 ∞
𝑋(𝑗𝜔 ) = ∫ 𝑥(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡 =∫ 𝑒 𝑎𝑡 𝑒 −𝑗𝜔𝑡 𝑑𝑡 + ∫ 𝑒 −𝑎𝑡 𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞ −∞ 0
1 1
= +
𝑎 − 𝑗𝜔 𝑎 + 𝑗𝜔
2𝑎
= 2 .
𝑎 + 𝜔2

As predicted, 𝑋(𝑗𝜔) is real-valued and even.

5.4.4. Differentiation

Consider the inverse Fourier transform

∞
1
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔)𝑒𝑗𝜔𝑡 𝑑𝜔
2𝜋 −∞

Differentiating both sides with respect to 𝑡 , we obtain

∞
𝑑𝑥(𝑡) 1
= ∫ 𝑗𝜔𝑋(𝑗𝜔)𝑒𝑗𝜔𝑡 𝑑𝜔
𝑑𝑡 2𝜋 −∞

Thus, we see that

𝑑𝑥(𝑡)
ℱ{ } = 𝑗𝜔𝑋 (𝑗𝜔 )
𝑑𝑡

5.4.5. Time and Frequency Scaling

Let 𝑎 be a nonzero real number and consider the signal 𝑔(𝑡) = 𝑥(𝑎𝑡) (i.e., a time-scaling of
𝑥(𝑡) ). We have

∞
ℱ{𝑔(𝑡)} = ∫ 𝑥(𝑎𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞

If we perform the substitution 𝜏 = 𝑎𝑡 , we have

24 | P a g e
 ∞
1 𝜔
−𝑗 𝜏
∫ 𝑥(𝜏)𝑒 𝑎 𝑑𝜏, 𝑎>0
𝑎 −∞
ℱ{𝑔(𝑡)} = .
1 ∞ 𝜔
− ∫ 𝑥(𝜏)𝑒 −𝑗 𝑎 𝜏 𝑑𝜏, 𝑎<0
{ 𝑎 −∞

This can be written in a uniform way as

1 𝜔
ℱ{𝑥(𝑎𝑡)} = 𝑋 (𝑗 )
|𝑎| 𝑎

Thus we see again that shrinking a signal in the time-domain corresponds to expanding it in
the frequency domain, and vice versa.

5.4.6. Duality

We have already seen a few examples of the duality property: suppose 𝑥(𝑡) has Fourier
transform 𝑋(𝑗𝜔). Then if we have a time-domain signal that has the same form as 𝑋(𝑗𝜔), the
Fourier transform of that signal will have the same form as 𝑥(𝑡). For example, the square pulse
in the time-domain had a Fourier transform in the form of a sinc function, and a sinc function
in the time-domain had a Fourier transform in the form of a square pulse.

We can consider another example. Suppose 𝑥(𝑡) = 𝑒 −|𝑡| . Then one can verify that

2
ℱ {𝑒 −|𝑡| } =
1 + 𝜔2

Specifically we have
∞
1 2
𝑒 −|𝑡| = ∫ 2
𝑒 −𝑗𝜔𝑡 𝑑𝜔
2𝜋 −∞ 1 + 𝜔

If we multiply both sides by 2𝜋 and interchange 𝜔 and 𝑡 , we obtain

∞
2
2𝜋𝑒 −|𝜔| = ∫ 𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞ 1 + 𝑡2

Thus, we have

2
ℱ{ } = 2𝜋𝑒 −|𝜔 |
1 + 𝑡2

Duality also applies to properties of the Fourier transform. For example, recall that
differentiation in the time-domain corresponds to multiplication by 𝑗𝜔 in the frequency domain.
We will now see that differentiation the frequency domain corresponds to multiplication by a
certain quantity of 𝑡 in the time-domain. We have

25 | P a g e
 ∞
𝑑𝑋(𝑗𝜔)
=∫ 𝑥(𝑡)(−𝑗𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
𝑑𝜔 −∞

Thus, differentiation in the frequency domain corresponds to multiplication by −𝑗𝑡 in the

time-domain.

5.4.7. Parseval’s Theorem

Just as with periodic signals, we have the following.

∞
1 ∞
∫ |𝑥(𝑡)|2 𝑑𝑡 = ∫ |𝑋(𝑗𝜔)|2 𝑑𝜔
−∞ 2𝜋 −∞

To derive this, note that

∞ ∞ ∞ ∞
1
∫ |𝑥(𝑡)|2 𝑑𝑡 = ∫ 𝑥(𝑡)𝑥 ∗ (𝑡)𝑑𝑡 = ∫ 𝑋 ∗ (𝑗𝜔)𝑒 −𝑗𝜔𝜏 𝑑𝜔𝑑𝑡
𝑥(𝑡) ∫
−∞ −∞ 2𝜋 −∞ −∞
1 ∞ ∗ ∞
= ∫ 𝑋 (𝑗𝜔) ∫ 𝑥(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡𝑑𝜔
2𝜋 −∞ −∞
1 ∞ ∗
= ∫ 𝑋 (𝑗𝜔 )𝑋(𝑗𝜔)𝑑𝜔
2𝜋 −∞
1 ∞
= ∫ |𝑋(𝑗𝜔)|2 𝑑𝜔
2𝜋 −∞

5.4.8. Convolution

Consider a signal 𝑥(𝑡) with Fourier transform 𝑋(𝑗𝜔). From the inverse Fourier transform, we
have

1 ∞
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔)𝑒𝑗𝜔𝑡 𝑑𝜔
2𝜋 −∞

This has the interpretation that 𝑥(𝑡) can be written as a superposition of complex exponentials
(with frequencies spanning the entire real axis). From earlier in the course, we know that if
the input to an LTI system is 𝑒𝑗𝜔𝑡 , then the output will be 𝐻(𝑗𝜔)𝑒𝑗𝜔𝑡 , where
∞
𝐻(𝑗𝜔) = ∫ ℎ(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞

In other words, 𝐻(𝑗𝜔) is the Fourier transform of the impulse response ℎ(𝑡). This, together
with the LTI property of the system, implies that
∞ ∞
1 𝑗𝜔𝑡
1
𝑥(𝑡) = ∫ 𝑋(𝑗𝜔)𝑒 𝑑𝜔 ⇒ ∫ 𝑋(𝑗𝜔)𝐻(𝑗𝜔 )𝑒𝑗𝜔𝑡 𝑑𝜔 = 𝑦(𝑡).
2𝜋 −∞ 2𝜋 −∞

Thus, we see that the Fourier transform of the output 𝑦(𝑡) is given by

26 | P a g e
 𝑌(𝑗𝜔) = 𝐻(𝑗𝜔 )𝑋(𝑗𝜔 )

In other words:

The Fourier transform of the output of an LTI system is given by the product of the Fourier
transforms of the input and the impulse response.

This is potentially the most important fact pertaining to LTI systems and frequency domain
analysis. Let's derive this another way just to reinforce the fact.

Suppose that we have two signals 𝑥(𝑡) and ℎ(𝑡), and define
∞
𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡) = ∫ 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏
−∞

We have
∞ ∞ ∞
𝑌(𝑗𝜔) = ∫ 𝑦(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡 =∫ [∫ 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏] 𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞ −∞ −∞
∞ ∞
=∫ 𝑥(𝜏) [∫ ℎ(𝑡 − 𝜏)𝑒 −𝑗𝜔𝑡 𝑑𝑡] 𝑑𝜏
−∞ −∞
∞
=∫ 𝑥(𝜏)𝑒 −𝑗𝜔𝜏 𝐻(𝑗𝜔)𝑑𝜏
−∞
∞
= 𝐻(𝑗𝜔) ∫ 𝑥(𝜏)𝑒 −𝑗𝜔𝜏 𝑑𝜏
−∞
= 𝐻(𝑗𝜔)𝑋(𝑗𝜔 ).

In the third line, we used the time-shifting property of the Fourier transform. Thus we see that
convolution of two signals in the time-domain corresponds to multiplication of the signals in
the frequency domain, i.e.,

ℱ{𝑥(𝑡) ∗ ℎ(𝑡)} = ℱ{𝑥(𝑡)}ℱ{ℎ(𝑡)}.

One thing to note here pertains to the existence of the Fourier transform of ℎ(𝑡). Specifically,
recall that the LTI system is stable if and only if
∞
∫ |ℎ(𝑡)|𝑑𝑡 < ∞
−∞

This is precisely the first condition in the Dirichlet conditions; thus, as long as the system is
stable and the impulse response satisfies the other two conditions (which almost all real systems
would), the Fourier transform is guaranteed to exist. If the system is unstable, we will need the
machinery of Laplace transforms to analyze the input-output behavior, which we will defer to
a later discussion.

The convolution - multiplication property is also very useful for analysis of interconnected
linear systems. For example, consider the series interconnection shown in Fig. 5.1.

27 | P a g e
 Figure 5.1: A series interconnection of systems.

We have

𝑦(𝑡) = 𝑦1 (𝑡) ∗ ℎ2 (𝑡) = (𝑥(𝑡) ∗ ℎ1 (𝑡)) ∗ ℎ2 (𝑡) = 𝑥(𝑡) ∗ (ℎ1(𝑡) ∗ ℎ2 (𝑡) ).

Taking Fourier transforms, we obtain

𝑌(𝑗𝜔) = 𝑋(𝑗𝜔)𝐻1 (𝑗𝜔 )𝐻2 (𝑗𝜔).

This reinforces what we saw earlier that the series interconnection of LTI systems can be
lumped together in a single LTI system whose impulse response is the convolution of the
impulse responses of the individual systems. In the frequency domain, their Fourier transforms
get multiplied together. One of the important implications of the convolution property is that it
allows us to investigate the effect of systems on signals in the frequency domain. For example,
this facilitates the design of appropriate filters for signals, as illustrated in the following
example.

Example 5.9. Consider a signal 𝑣(𝑡) which represents a measurement of some relatively low
frequency content (such as a voice signal). Suppose that we measure this signal via a
microphone, whose output is given by

𝑥(𝑡) = 𝑣(𝑡) + 𝑛(𝑡)

where 𝑛(𝑡) is high-frequency noise. Note that 𝑋(𝑗𝜔) = 𝑉(𝑗𝜔) + 𝑁(𝑗𝜔). We would like to take
the measured signal 𝑥(𝑡) and remove the noise; unfortunately, we do not have access to 𝑛(𝑡) to
subtract it out. Instead, we can work in the frequency domain. Suppose we design a filter (an
LTI system) whose impulse response as the following Fourier transform:

1 |𝜔| ≤ 𝑊
𝐻(𝑗𝜔 ) = {
0 |𝜔| > 𝑊

where 𝑊 is the highest frequency of the underlying signal 𝑣(𝑡). If we feed 𝑥(𝑡) into this filter,
the output will have Fourier transform given by

𝑌(𝑗𝜔 ) = 𝑋(𝑗𝜔)𝐻(𝑗𝜔 ) = 𝑉(𝑗𝜔)𝐻(𝑗𝜔 ) + 𝑁(𝑗𝜔)𝐻(𝑗𝜔)

If all of the frequency content of the noise 𝑛(𝑡) occurs at frequencies larger than 𝑊 , then we
see that 𝑁(𝑗𝜔)𝐻(𝑗𝜔 ) = 0 , and thus

𝑌(𝑗𝜔) = 𝑉(𝑗𝜔)𝐻(𝑗𝜔) = 𝑉(𝑗𝜔).

28 | P a g e
In other words, we have recovered the voice signal 𝑣(𝑡) by passing 𝑥(𝑡) through the low-pass
filter.

Recall that the inverse Fourier transform of the given 𝐻(𝑗𝜔) is

sin (𝑊𝑡)
ℎ(𝑡) =
𝜋𝑡

However, there are various challenges with implementing an LTI system with this impulse
response. One is that this is noncausal, and thus one must potentially include a sufficiently large
delay (followed by a truncation of the signal) in order to apply it. Another problem is that it
contains many oscillations, which may not be desirable for an impulse response.

Instead of the above filter, suppose consider another filter whose impulse response is

ℎ2 (𝑡) = 𝑒 −𝑎𝑡 𝑢(𝑡)

This filter can be readily implemented with an RC circuit (with the input signal being applied
as an input voltage, and the output signal being the voltage across the capacitor). The Fourier
transform of this impulse response is

1
𝐻2 (𝑗𝜔) =
𝑗𝜔 + 𝑎

The magnitude plot of this Fourier transform has content at all frequencies, and thus this filter
will not completely eliminate all of the high frequency noise. However, by tuning the value of
𝑎 , one can adjust how much of the noise affects the filtered signal. Note that this filter will also
introduce phase shifts at different frequencies, which will also cause some distortion of the
recovered signal.

5.4.9. Multiplication

We just saw that multiplication in the time domain corresponds to convolution in the frequency
domain. By duality, we obtain that multiplication in the frequency domain corresponds to
convolution in the time-domain. Specifically, consider two signals 𝑥1 (𝑡) and 𝑥2 (𝑡), and define
𝑔(𝑡) = 𝑥1 (𝑡)𝑥2 (𝑡). Then we have

∞ ∞
𝐺(𝑗𝜔) = ∫ 𝑔(𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡 =∫ 𝑥1 (𝑡)𝑥2 (𝑡)𝑒 −𝑗𝜔𝑡 𝑑𝑡
−∞ −∞
∞ ∞
1
= ∫ 𝑥2 (𝑡) ∫ 𝑋1 (𝑗𝜃)𝑒𝑗𝜃𝑡 𝑑𝜃𝑒 −𝑗𝜔𝑡 𝑑𝑡
2𝜋 −∞ −∞
∞ ∞
1
= ∫ 𝑋1 (𝑗𝜃) ∫ 𝑥2 (𝑡)𝑒 −𝑗(𝜔−𝜃)𝑡 𝑑𝑡𝑑𝜃
2𝜋 −∞ −∞
1 ∞
= ∫ 𝑋1 (𝑗𝜃)𝑋2 (𝑗(𝜔 − 𝜃))𝑑𝜃 .
2𝜋 −∞

29 | P a g e
Thus,
∞
1 1
ℱ{ 𝑥1(𝑡)𝑥2 (𝑡)} = (𝑋1 (𝑗𝜔) ∗ 𝑋2 (𝑗𝜔) ) = ∫ 𝑋1 (𝑗𝜃)𝑋2 (𝑗(𝜔 − 𝜃))𝑑𝜃
2𝜋 2𝜋 −∞

Multiplication of one signal 𝑥1(𝑡) by another signal 𝑥2 (𝑡) can be viewed as modulating the
amplitude of one signal by the other. This plays a key role in communication systems.

Example 5.10. Consider a signal 𝑠(𝑡) whose frequency spectrum lies in some interval
[−𝑊, 𝑊] . Consider the signal 𝑝(𝑡) = 𝑐𝑜𝑠 (𝜔0 𝑡). The Fourier transform of 𝑝(𝑡) is given by

𝑃(𝑗𝜔) = 𝜋𝛿(𝜔 − 𝜔0 ) + 𝜋𝛿(𝜔 + 𝜔0 ).

Now consider the signal 𝑥(𝑡) = 𝑠(𝑡)𝑝(𝑡), with Fourier transform given by
∞
1
𝑋(𝑗𝜔) = ∫ 𝑆(𝑗𝜃 )𝑃(𝑗(𝜔 − 𝜃))𝑑𝜃
2𝜋 −∞
1 ∞
= ∫ 𝑆(𝑗𝜃 )𝛿 (𝜔 − 𝜃 − 𝜔0 ) 𝑑𝜃
2 −∞
1 ∞
+ ∫ 𝑆(𝑗𝜃)𝛿( 𝜔 − 𝜃 + 𝜔0 )𝑑𝜃
2 −∞
1 1
= 𝑆(𝑗( 𝜔 − 𝜔0 )) + 𝑆(𝑗 (𝜔 + 𝜔0 )).
2 2

Thus, multiplying the signal 𝑠(𝑡) by 𝑝(𝑡) results in a signal 𝑥(𝑡) whose frequency spectrum
consists of two copies of the spectrum of 𝑠(𝑡), cenetered at the frequencies 𝜔0 and −𝜔0 and
1
scaled by .
2

The above example illustrates the principle behind amplitude modulation (AM) in
communication and radio systems. A low frequency signal (such as voice) is amplitude
modulated to a higher frequency that is reserved for that signal. It is then transmitted at that
frequency to the receiver. The following example illustrates how the receiver can recover the
transmitted signal.

Example 5.11. Consider the signal 𝑥(𝑡) = 𝑠(𝑡)𝑝(𝑡) from the previous example.

Its frequency spectrum has two copies of the spectrum of 𝑠(𝑡), located at ±𝜔0 . We want to
recover the original signal 𝑠(𝑡) from 𝑥(𝑡). To do this, suppose we multiply 𝑥(𝑡) by cos (𝜔0 𝑡)
again, to obtain

𝑦(𝑡) = 𝑥(𝑡)cos (𝜔0 𝑡)

As above, we have

1 1
𝑌(𝑗𝜔) = 𝑋(𝑗(𝜔 − 𝜔0 )) + 𝑋(𝑗( 𝜔 + 𝜔0 )).
2 2

30 | P a g e
By drawing this, we see that the frequency spectrum of 𝑦(𝑡) contains three copies of the
1 1
spectrum of 𝑠(𝑡) : one copy centered at 𝜔 = 0 (and scaled by 2 ), one copy at 2𝜔0 scaled by 4,
1
and one copy at −2𝜔0 , scaled by 4. Thus, if we want to recover 𝑠(𝑡) from 𝑦(𝑡), we simply apply
a low pass filter to it (and scale it by 2 ).

5.5 The Discrete-Time Fourier Transform

Consider a general signal 𝑥[𝑛] which is nonzero on some interval −𝑁1 ≤ 𝑛 ≤ 𝑁2 and zero
elsewhere. We create a periodic extension 𝑥̃[𝑛] of this signal with period 𝑁 (where 𝑁 is large
enough so that there is no overlap). As 𝑁 → ∞ , 𝑥̃[𝑛] becomes equal to 𝑥[𝑛] for each finite value
of 𝑛 .

Since 𝑥̃[𝑛] is periodic, it has a discrete-time Fourier series representation given by

𝑁−1

𝑥̃[𝑛] = ∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑛
𝑘=0

2𝜋
where 𝜔0 = . The Fourier series coefficients are given by
𝑁

𝑛0 +𝑁−1
1
𝑎𝑘 = ∑ 𝑥̃[𝑛]𝑒 −𝑗𝜔0 𝑘𝑛
𝑁
𝑛=𝑛0

where 𝑛0 is any integer. Suppose we choose 𝑛0 so that the interval [ −𝑁1 , 𝑁2 ] is contained in
[𝑛0 , 𝑛0 + 𝑁 − 1]. Then since 𝑥̃[𝑛] = 𝑥[𝑛] in this interval, we have

𝑛0 +𝑁−1 ∞
1 −𝑗𝜔 0 𝑘𝑛
1
𝑎𝑘 = ∑ 𝑥[𝑛]𝑒 = ∑ 𝑥[𝑛]𝑒 −𝑗𝜔0 𝑘𝑛
𝑁 𝑁
𝑛=𝑛0 𝑛=−∞

Let us now define the discrete-time Fourier transform as

∞
𝑗𝜔 )
𝑋(𝑒 ≜ ∑ 𝑥[𝑛]𝑒 −𝑗𝜔𝑛
𝑛=−∞

1
From this, we see that 𝑎𝑘 = 𝑋(𝑒𝑗𝑘𝜔0 ), i.e., the discrete-time Fourier series coefficients are
𝑁
obtained by sampling the discrete-time Fourier transform at periodic intervals of 𝜔0 . Also note
that 𝑋(𝑒𝑗𝜔 ) is periodic in 𝜔 with period 2𝜋 (since 𝑒 −𝑗𝜔𝑛 is 2𝜋-periodic).

Using the Fourier series representation of 𝑥̃[𝑛] , we now have

𝑁−1 𝑁−1 𝑁−1
𝑗𝑘𝜔 0 𝑛
1 1
𝑥̃[𝑛] = ∑ 𝑎𝑘 𝑒 = ∑ 𝑋(𝑒𝑗𝑘 𝜔0 )𝑒𝑗𝑘𝜔0 𝑛 = ∑ 𝑋( 𝑒𝑗𝑘𝜔0 )𝑒𝑗𝑘𝜔0𝑛 𝜔0 .
𝑁 2𝜋
𝑘=0 𝑘=0 𝑘=0

31 | P a g e
Once again, we see that each term in the summand represents the area of a rectangle of width
𝜔0 obtained from the curve 𝑋 (𝑒𝑗𝜔 )𝑒𝑗𝜔 . As 𝑁 → ∞, we have 𝜔0 → 0. In this case, the sum of the
areas of the rectangles approaches the integral of the curve 𝑋( 𝑒𝑗𝜔 )𝑒𝑗𝜔𝑛 , and since the sum was
over only 𝑁 samples of the function, the integral is only over one interval of length 2𝜋. Since
𝑥̃[𝑛] approaches 𝑥[𝑛] as 𝑁 → ∞ , we have

1
𝑥[𝑛] = ∫ 𝑋(𝑒𝑗𝜔 ) 𝑒𝑗𝜔𝑛 𝑑𝜔 .
2𝜋 2𝜋

This is the inverse discrete-time Fourier transform, or the synthesis equation.

The main differences between the discrete-time and continuous-time Fourier transforms are the
following. (1) The discrete-time Fourier transform 𝑋(𝑒𝑗𝜔 ) is periodic in 𝜔 with period 2𝜋 ,
whereas the continuous-time Fourier transform is not necessarily periodic. (2) The synthesis
equation for the discrete-time Fourier transform only involves an integral over an interval of
length 2𝜋, whereas the one for the continuous-time Fourier transform is over the entire 𝜔 axis.
Both of these are due to the fact that 𝑒𝑗𝜔𝑛 is 2𝜋-periodic in 𝜔 , whereas the continuous-time
complex exponential is not.

Since the frequency spectrum of 𝑋(𝑒𝑗𝜔 ) is only uniquely specified over an interval of length
2𝜋, we have to be careful about what we mean by "high" and "low" frequencies. Recalling the
discussion of discrete-time complex exponentials, high-frequency signals in discrete-time have
frequencies close to odd multiples of 𝜋, whereas low-frequency signals have frequencies close
to even multiples of 𝜋.

Example 5.12. Consider the signal.

𝑥[𝑛] = 𝑎𝑛 𝑢[𝑛], |𝑎| < 1.

We have
∞ ∞
𝑗𝜔 ) −𝑗𝜔𝑛
𝑋(𝑒 = ∑ 𝑥[𝑛]𝑒 = ∑ 𝑎𝑛 𝑒 −𝑗𝜔𝑛
𝑛=−∞ 𝑛=0
∞
𝑛
= ∑ (𝑎𝑒 −𝑗𝜔 )
𝑛=0
1
= .
1 − 𝑎𝑒 −𝑗𝜔

If we plot the magnitude of 𝑋 (𝑒𝑗𝜔 ), we see an illustration of the "high" versus "low" frequency
effect. Specifically, if 𝑎 > 0 then the signal 𝑥[𝑛] does not have any oscillations and |𝑋(𝑒𝑗𝜔 )|
has its highest magnitude around even multiples of 𝜋. However, if 𝑎 < 0, then the signal 𝑥[𝑛]
oscillates between positive and negative values at each time-step; this "high-frequency"
behavior is captured by the fact that |𝑋(𝑒𝑗𝜔 )| has its largest magnitude near odd multiples of 𝜋.

Example 5.13. Consider the signal

32 | P a g e
 𝑥[𝑛] = 𝑎 |𝑛| , |𝑎| < 1.

We have
∞ ∞
𝑗𝜔 ) −𝑗𝜔𝑛
𝑋 (𝑒 = ∑ 𝑥[𝑛]𝑒 = ∑ 𝑎 |𝑛| 𝑒 −𝑗𝜔𝑛
𝑛=−∞ 𝑛=−∞
−1 ∞
−𝑛 −𝑗𝜔𝑛
= ∑ 𝑎 𝑒 + ∑ 𝑎𝑛 𝑒 −𝑗𝜔𝑛
𝑛=−∞ 𝑛=0
∞ ∞

= ∑ 𝑎𝑛 𝑒𝑗𝜔𝑛 + ∑ 𝑎𝑛 𝑒 −𝑗𝜔𝑛
𝑛=1 𝑛=0
𝑎𝑒𝑗𝜔 1
= +
1− 𝑎𝑒𝑗𝜔1 − 𝑎𝑒 −𝑗𝜔
1 − 𝑎2
= .
1 − 2𝑎cos (𝜔) + 𝑎2

5.6. The Fourier Transform of Discrete-time periodic Signals.

First, consider the signal

𝑥[𝑛] = 𝑒𝑗 𝜔0 𝑛 .

We claim that the Fourier transform of this signal is

∞

𝑋( 𝑒𝑗𝜔 ) = ∑ 2𝜋𝛿(𝜔 − 𝜔0 − 2𝜋𝑙 )

𝑙=−∞

i.e., a set of impulse functions spaced 2𝜋 apart on the frequency axis. To verify this, note that
the inverse Fourier transform is given by

1
∫ 𝑋(𝑒𝑗𝜔 ) 𝑒𝑗𝜔𝑛 𝑑𝜔
2𝜋 2𝜋

The integral is only over an interval of length 2𝜋, and there is at most one impulse function
from 𝑋(𝑒𝑗𝜔 ) in any such interval. Let that impulse be located at 𝜔0 + 2𝜋𝑟 for some 𝑟 ∈ ℤ. Then
we have
1 1 (𝜔 0 +2𝜋𝑟)𝑛
∫ 𝑋(𝑒𝑗𝜔 )𝑒𝑗𝜔𝑛 𝑑𝜔 = ∫ 2𝜋𝛿 (𝜔 − 𝜔0 − 2𝜋𝑟)𝑒𝑗𝜔𝑛 𝑑𝜔 = 𝑒𝑗 = 𝑒𝑗𝜔0 𝑛 .
2𝜋 2𝜋 2𝜋 2𝜋

Thus consider a periodic discrete-time signal 𝑥[𝑛], with Fourier series

𝑁−1

𝑥[𝑛] = ∑ 𝑎𝑘 𝑒𝑗𝑘𝜔0 𝑛 = 𝑎0 + 𝑎1 𝑒𝑗𝜔0 𝑛 + 𝑎2 𝑒𝑗2 𝜔0 𝑛 + ⋯ + 𝑎𝑁−1 𝑒𝑗(𝑁−1)𝜔0 𝑛

𝑘=0

2𝜋
where 𝜔0 = . The Fourier transform of each term of the form 𝑎𝑘 𝑒𝑗𝑘𝜔0𝑛 is a set of impulses
𝑁
spaced 2𝜋 apart, with one located at 𝜔 = 𝑘𝜔0 . Furthermore each of these impulses is scaled by

33 | P a g e
𝑎𝑘 2𝜋 . Since 𝑎𝑘 = 𝑎𝑘+𝑁𝑙 for any 𝑙 (by the periodicity of the discrete-time Fourier series
coefficients), when we add up the Fourier transforms of all of the terms in the Fourier series
expansion of 𝑥[𝑛] , we obtain
∞
𝑗𝜔 )
2𝜋𝑘
𝑋( 𝑒 = ∑ 2𝜋𝑎𝑘 𝛿 (𝜔 − )
𝑁
𝑘=−∞

Thus, the Fourier transform of a discrete-time periodic signal is indeed a sequence of impulses
located at multiples of 𝜔0 , with a period of 2𝜋.

Example 5.14. Consider the impulse train

∞

𝑥[𝑛] = ∑ 𝛿[𝑛 − 𝑘𝑁]

𝑘=−∞

The Fourier series coefficients of 𝑥[𝑛] are given by

𝑁−1
1 1
𝑎𝑘 = ∑ 𝑥[𝑛]𝑒 −𝑗𝑘𝜔0 𝑛 =
𝑁 𝑁
𝑘=0

Thus the Fourier transform of 𝑥[𝑛] is

∞ ∞
2𝜋𝑘 2𝜋 2𝜋𝑘
𝑋( 𝑒𝑗𝜔 ) = ∑ 𝑎𝑘 2𝜋𝛿 (𝜔 − )= ∑ 𝛿 (𝜔 − ).
𝑁 𝑁 𝑁
𝑘=−∞ 𝑘=−∞

5.7. Properties of the Discrete-Time Fourier Transform

5.7.1. Periodicity

The Fourier transform of a signal 𝑥[𝑛] is periodic in frequency, with period 2𝜋 :

𝑋 (𝑒𝑗𝜔 ) = 𝑋( 𝑒𝑗(𝜔+2𝜋) )

This comes out of the fact that discrete-time complex exponentials are periodic in frequency
with period 2𝜋.

5.7.2. Linearity
It is easy to see that

ℱ{𝛼𝑥1 [𝑛] + 𝛽𝑥2 [𝑛] } = 𝛼𝑋1 (𝑒𝑗𝜔 ) + 𝛽𝑋2 (𝑒𝑗𝜔 )

34 | P a g e
5.7.3. Time and frequency Shifting

We have

ℱ{𝑥 [𝑛 − 𝑛0 ]} = 𝑒 −𝑗𝜔𝑛0 𝑋(𝑒𝑗𝜔 )

and
( )
ℱ {𝑒𝑗 𝜔0 𝑛 𝑥[𝑛]} = 𝑋( 𝑒𝑗 𝜔 −𝜔0 ).

The first property is easily proved using the inverse Fourier transform equation, and the
second property is proved using the Fourier transform equation.

Example 5.15. Consider a discrete-time low-pass filter, whose Fourier transform 𝐻𝑙𝑝 (𝑒𝑗𝜔 ) is
a square pulse centered at even multiples of 𝜋.

Now consider the high pass filter 𝐻ℎ𝑝 (𝑒𝑗𝜔 ) which consists of square pulses centered at odd
multiples of 𝜋. We see that 𝐻ℎ𝑝 (𝑒𝑗𝜔 ) = 𝐻𝑙𝑝 (𝑒𝑗 (𝜔−𝜋 ) ). Thus we have

ℎℎ𝑝 [𝑛] = 𝑒𝑗𝜋𝑛 ℎ𝑙𝑝 [𝑛] = (−1)𝑛 ℎ𝑙𝑝 [𝑛].

5.7.4. First Order Differences

Consider the discrete-time analog of differentiation, which is to take the differences between
subsequent samples of the signal. By applying the linearity and time-shifting properties, we
have

ℱ{𝑥[𝑛] − 𝑥[𝑛 − 1]} = 𝑋(𝑒𝑗𝜔 ) − 𝑒 −𝑗𝜔 𝑋( 𝑒𝑗𝜔 ) = (1 − 𝑒 −𝑗𝜔 )𝑋( 𝑒𝑗𝜔 )

5.7.5. Conjugation

For any discrete-time signal (that has a Fourier transform), we have

ℱ {𝑥 ∗ [𝑛] } = 𝑋 ∗ (𝑒 −𝑗𝜔 ).

Furthermore, if 𝑥[𝑛] is real, we have 𝑥 ∗ [𝑛] = 𝑥[𝑛] and thus

𝑋(𝑒𝑗𝜔 ) = 𝑋 ∗ (𝑒 −𝑗𝜔 ) .

5.7.6. Time-Reversal

Consider the time-reversed signal 𝑥[−𝑛]. We have

35 | P a g e
 2𝜋 2𝜋
1 1
𝑥[−𝑛] = ∫ 𝑋 (𝑒𝑗𝜔 )𝑒 −𝑗𝜔𝑛 𝑑𝜔 = ∫ 𝑋(𝑒 −𝑗𝜔 ) 𝑒𝑗𝜔𝑛 𝑑𝜔
2𝜋 0 2𝜋 0

which is obtained by performing a change of variable 𝜔 → −𝜔 (note that the negative sign
introduced by this change is canceled out by the reversal of the bounds of integration that
arise because of the negation). Thus, we have

ℱ{𝑥[−𝑛]} = 𝑋 (𝑒 −𝑗𝜔 )

Together with the conjugation property, we see that for real-valued even signals (where
𝑥[𝑛] = 𝑥[−𝑛] ), we have

𝑋( 𝑒𝑗𝜔 ) = ℱ{𝑥[𝑛]} = ℱ{𝑥[−𝑛]} = 𝑋(𝑒 −𝑗𝜔 )

Thus, the Fourier transform of real, even signals is also real and even.

5.7.7. Time Expansion

Recall that for a continuous-time signal 𝑥(𝑡) and a scalar 𝑎 ≠ 0 , we had

1 𝑗𝜔
ℱ{𝑥(𝑎𝑡)} = 𝑋( )
|𝑎| 𝑎

Thus, an expansion in the time-domain led to a compression in the frequency domain and vice
versa.

In discrete-time, expansion and contraction of time-domain signals is not achieved simply by

scaling the time-variable. First, since the time index must be an integer, it does not make sense
to consider the signal 𝑥[𝑎𝑛] , where 𝑎 < 1 . Similarly, if we consider integer values of 𝑎 larger
than 1 , then the signal 𝑥[𝑎𝑛] only considers the values of 𝑥[𝑛] at integer multiples of 𝑎 , and all
information between those values is lost.

A different expansion of signals that preserves all their values is as follows. For a given signal
𝑥[𝑛] and positive integer 𝑘 , define the signal

Thus, the signal 𝑥(𝑘) [𝑛] is obtained by spreading the points of 𝑥[𝑛] apart by 𝑘 samples and
placing zeros between the samples. We have
∞ ∞
−𝑗𝜔𝑛
ℱ{ 𝑥(𝑘) [𝑛]} = ∑ 𝑥(𝑘) [𝑛]𝑒 = ∑ 𝑥(𝑘) [𝑟𝑘]𝑒 −𝑗𝜔𝑟𝑘
𝑛=−∞ 𝑟=−∞

since 𝑥(𝑘) [𝑛] is nonzero only at integer multiples of 𝑘 . Since 𝑥(𝑘) [𝑟𝑘] = 𝑥[𝑟], we have

36 | P a g e
 ∞

ℱ {𝑥(𝑘) [𝑛]} = ∑ 𝑥[𝑟]𝑒 −𝑗𝜔𝑟𝑘 = 𝑋( 𝑒𝑗𝑘𝜔 )

𝑟=−∞

Example 5.16. Consider the signal

1 𝑛 ∈ {0,2,4}
𝑥[𝑛] = {2 𝑛 ∈ {1,3,5}
0 otherwise

We note that we can write 𝑥[𝑛] as

𝑥[𝑛] = 𝑔[𝑛] + 2𝑔[𝑛 − 1]

where

1 𝑛 ∈ {0,2,4}
𝑔[𝑛] = {
0 otherwise

This 𝑔[𝑛] can be viewed as an expansion of the signal

1 0≤𝑛≤2
ℎ[𝑛] = { .
0 otherwise

Specifically, 𝑔[𝑛] = ℎ(2) [𝑛]. The Fourier transform of ℎ[𝑛] is given by

2 3
1 − 𝑒 −3𝑗𝜔 sin ( 𝜔)
𝐻 (𝑒 𝑗𝜔 )
=∑ 𝑒 −𝑗𝜔𝑛
= =𝑒 −𝑗𝜔 2
1 − 𝑒 −𝑗𝜔 𝜔
𝑛=0 sin ( )
2

Thus, the Fourier transform of 𝑔[𝑛] is given by

sin (3𝜔)
𝐺 (𝑒𝑗𝜔 ) = 𝐻 (𝑒𝑗2𝜔 ) = 𝑒 −2𝑗𝜔
sin (𝜔)

Finally,

sin (3𝜔)
𝑋 (𝑒𝑗𝜔 ) = 𝐺 (𝑒𝑗𝜔 ) + 2𝑒 −𝑗𝜔 𝐺 (𝑒𝑗𝜔 ) = 𝑒 −2𝑗𝜔 (1 + 2𝑒 −𝑗𝜔 )
sin (𝜔)

5.7.8. Differentiation in Frequency

Suppose 𝑥[𝑛] has Fourier transform 𝑋( 𝑒𝑗𝜔 ). Then we have

∞ ∞
𝑑 𝑗𝜔 )
𝑑 −𝑗𝜔𝑛
𝑋 (𝑒 = ∑ 𝑥[𝑛]𝑒 = ∑ (−𝑗𝑛)𝑥[𝑛]𝑒 −𝑗𝜔𝑛
𝑑𝜔 𝑑𝜔
𝑛=−∞ 𝑛=−∞

Thus

𝑑𝑋 ( 𝑒𝑗𝜔 )
ℱ{𝑛𝑥[𝑛]} = 𝑗
𝑑𝜔

37 | P a g e
5.7.9. Parseval’s Theorem

We have
∞
1 2
∑ |𝑥[𝑛]|2 = ∫ |𝑋(𝑒𝑗𝜔 )| 𝑑𝜔
2𝜋 2𝜋
𝑛=−∞

5.7.10. Convolution

Just as in continuous time, the discrete time signal 𝑒𝑗𝜔𝑛 is an eigenfunction of discrete-time
LTI systems. Specifically, if 𝑒𝑗𝜔𝑛 is applied to a (stable) LTI system with impulse response
ℎ[𝑛], the output of the system will be 𝐻 (𝑒𝑗𝜔 )𝑒𝑗𝜔𝑛 .

Thus consider a signal 𝑥[𝑛] written in terms of its Fourier transform as

1
𝑥[𝑛] = ∫ 𝑋(𝑒𝑗𝜔 ) 𝑒𝑗𝜔𝑛 𝑑𝜔
2𝜋 2𝜋

This is a linear combination of complex exponentials (where the scaling factor on the
1
complex exponential 𝑒𝑗𝜔𝑛 is 2𝜋 𝑋( 𝑒𝑗𝜔 ). By the LTI property, we thus have

1 1
𝑥[𝑛] = ∫ 𝑋 (𝑒𝑗𝜔 )𝑒𝑗𝜔𝑛 𝑑𝜔 → ∫ 𝑋( 𝑒𝑗𝜔 )𝐻 (𝑒𝑗𝜔 )𝑒𝑗𝜔𝑛 𝑑𝜔 = 𝑦[𝑛].
2𝜋 2𝜋 2𝜋 2𝜋

The expression on the right hand side is the output 𝑦[𝑛] of the system when the input is 𝑥[𝑛] .
Thus we have

𝑌(𝑒𝑗𝜔 ) = 𝐻( 𝑒𝑗𝜔 )𝑋(𝑒𝑗𝜔 ).

As in the continuous-time case, convolution in the time-domain is given by multiplication in

the frequency domain.

Example 5.17. Consider the system shown in Fig. 5.18a in OW. Let us analyze the
relationship between 𝑦[𝑛] and 𝑥[𝑛] for that system.

First, we have 𝑤1 [𝑛] = (−1)𝑛 𝑥[𝑛] = 𝑒𝑗𝜋𝑛 𝑥[𝑛] . By the frequency shifting property, we see that
𝑊1 (𝑒𝑗𝜔 ) = 𝑋 (𝑒𝑗 (𝜔−𝜋) ). Next, we have

𝑊2 (𝑒𝑗𝜔 ) = 𝐻𝑙𝑝 ( 𝑒𝑗𝜔 )𝑊1 (𝑒𝑗𝜔 ) = 𝐻𝑙𝑝 (𝑒𝑗𝜔 )𝑋( 𝑒𝑗(𝜔−𝜋) ).

The signal 𝑤3 [𝑛] is given by 𝑤3 [𝑛] = (−1)𝑛𝑤2 [𝑛] , and thus 𝑊3 (𝑒𝑗𝜔 ) = 𝑊2 (𝑒𝑗 (𝜔 −𝜋) ). Putting
this together with the expression for 𝑊2 (𝑒𝑗𝜔 ), we obtain

𝑊3 (𝑒𝑗𝜔 ) = 𝐻𝑙𝑝 (𝑒𝑗(𝜔−𝜋) )𝑋 (𝑒𝑗 (𝜔−2𝜋) ) = 𝐻𝑙𝑝 ( 𝑒𝑗(𝜔−𝜋) )𝑋( 𝑒𝑗𝜔 ).

38 | P a g e
From the bottom path, we have 𝑊4 (𝑒𝑗𝜔 ) = 𝐻𝑙𝑝 ( 𝑒𝑗𝜔 )𝑋(𝑒𝑗𝜔 ) . Thus, we have

𝑌 (𝑒𝑗𝜔 ) = 𝑊3 ( 𝑒𝑗𝜔 ) + 𝑊4 ( 𝑒𝑗𝜔 ) = (𝐻𝑙𝑝 (𝑒𝑗𝜔 ) + 𝐻𝑙𝑝 (𝑒𝑗 (𝜔−𝜋) )) 𝑋(𝑒𝑗𝜔 ).

Recall that 𝐻𝑙𝑝 ( 𝑒𝑗(𝜔−𝜋) ) is a high-pass filter centered at 𝜋. Thus, this system acts as a
bandstop filter, blocking all frequencies in a certain range and letting all of the low and high
frequency signals through.

5.7.11. Multiplication

Consider two signals 𝑥1[𝑛] and 𝑥2 [𝑛] , and define 𝑔[𝑛] = 𝑥1[𝑛]𝑥2 [𝑛] . The discretetime Fourier
transform of 𝑔[𝑛] is given by
∞
𝑗𝜔 )
𝐺( 𝑒 = ∑ 𝑥1[𝑛]𝑥2 [𝑛]𝑒 −𝑗𝜔𝑛
𝑛=−∞

Replacing 𝑥1 [𝑛] by the synthesis equation

1
𝑥1 [𝑛] = ∫ 𝑋1 (𝑒𝑗𝜃 )𝑒𝑗𝜃𝑛 𝑑𝜃
2𝜋 2𝜋

where we simply replaced the dummy variable 𝜔 with 𝜃 to avoid confusion, we obtain
∞
𝑗𝜔 )
1
𝐺 (𝑒 = ∑ ∫ 𝑋1 (𝑒𝑗𝜃 )𝑒𝑗𝜃𝑛 𝑑𝜃𝑥2 [𝑛]𝑒 −𝑗𝜔𝑛
2𝜋 2𝜋
𝑛=−∞
∞
1
= ∫ 𝑋 (𝑒𝑗𝜃 ) ∑ 𝑥2 [𝑛]𝑒 −𝑗(𝜔−𝜃)𝑛 𝑑𝜃
2𝜋 2𝜋 1
𝑛=−∞
1
= ∫ 𝑋 (𝑒𝑗𝜃 )𝑋2 (𝑒𝑗 (𝜔−𝜃) )𝑑𝜃
2𝜋 2𝜋 1

This resembles the typical convolution of the signals 𝑋1 (𝑒𝑗𝜔 ) and 𝑋2 (𝑒𝑗𝜔 ) , except that the
integral is over only an interval of length 2𝜋 as opposed over the entire frequency axis. This is
called the periodic convolution of the two signals. Recall that we saw the same thing when we
considered the discrete-time Fourier series of the product of two periodic discrete-time signals.

Example 5.17. Consider the two signals

𝜋 𝜋
sin ( 𝑛) sin ( 𝑛)
𝑥1 [𝑛] = 2 , 𝑥2 [𝑛] = 4 .
𝜋𝑛 𝜋𝑛

The Fourier transforms of these signals are square pulses, where the pulse centered at 0
𝜋 𝜋 𝜋 𝜋
extend from − 2 to 2 (for 𝑋1 (𝑒𝑗𝜔 ) ) and from − 4 to 4 (for 𝑋2 (𝑒𝑗𝜔 ) ). The Fourier transform of
𝑔[𝑛] = 𝑥1 [𝑛]𝑥2 [𝑛] is given by

39 | P a g e
 1
𝐺 (𝑒𝑗𝜔 ) = ∫ 𝑋1 (𝑒𝑗𝜃 )𝑋2 (𝑒𝑗(𝜔−𝜃) )𝑑𝜃
2𝜋 2𝜋

Since we can choose any interval of length 2𝜋 to integrate over, let's choose the interval [−𝜋, 𝜋)
for convenience. We also only need to determine the values of the Fourier transform for values
of 𝜔 between −𝜋 and 𝜋, since the transform is periodic. Depending on the value of 𝜔 , there are
different cases that occur:
3𝜋
• If −𝜋 ≤ 𝜔 < − 4 , then there is no overlap in the signals 𝑋1 (𝑒𝑗𝜃 ) and 𝑋2 (𝑒𝑗(𝜔−𝜃) ), and
thus 𝐺(𝑗𝜔 ) is zero.
3𝜋 𝜋
• If − ≤ 𝜔 < − , then there is partial overlap in the signals; the product is a rectangle
4 4
𝜋 𝜋 1 3𝜋
with support from − 2 to 𝜔 + 4 , and thus 𝐺(𝑗𝜔) evaluates to 2𝜋 (𝜔 + ).
4
𝜋 𝜋 1 𝜋 1
• If − ≤ 𝜔 < , there is full overlap and 𝐺(𝑗𝜔) is ( )= .
4 4 2𝜋 2 4

𝜋 3𝜋 1 3𝜋
• If 4 ≤ 𝜔 < , there is partial overlap and 𝐺(𝑗𝜔) is 2𝜋 ( 4 − 𝜔).
4
3𝜋
• If ≤ 𝜔 < 𝜋, there is no overlap and 𝐺(𝑗𝜔) is zero.
4

Note that since we are only integrating over 𝜃 between −𝜋 and 𝜋, the values of 𝑋( 𝑒𝑗𝜃 )
ˆ
outside of that interval does not matter. Thus, we could also create a new signal 𝑋1 (𝑒𝑗𝜃 )
which is equal to 𝑋1 (𝑒𝑗𝜃 ) over the interval [−𝜋, 𝜋) and zero everywhere else. The Fourier
transform can then be written as
∞ ˆ
1 1
𝐺 ( 𝑒𝑗𝜔 ) = ∫ 𝑋1 (𝑒𝑗𝜃 )𝑋2 (𝑒𝑗 (𝜔−𝜃) )𝑑𝜃 = ∫ 𝑋1 (𝑒𝑗𝜃 )𝑋2 (𝑒𝑗(𝜔−𝜃) )𝑑𝜃,
2𝜋 2𝜋 2𝜋 −∞

ˆ
i.e., it is the usual convolution of the signals 𝑋1 (𝑒𝑗𝜔 ) and 𝑋2 (𝑒𝑗𝜔 ).

40 | P a g e