Introduction to

Telecommunication

The Energy Spectral Density and

Power Spectral Density

Asst. Prof. Dr. Ertuğrul SAATÇI

Overview
The Fourier Transform provides frequency domain information
for both energy signals and power signals.
The energy spectral density (ESD) provides information of the
energy of a signal at each frequency.
Applies only to energy signals.
The power spectral density (PSD) provides information of the
power of a signal at each frequency.
Applies only to power signals.
Can be applied to random as well as deterministic signals.
Today we will review the ESD and PSD and their relationship to
the autocorrelation function.
Reading
2.8, 2.9

1
Developing Energy Spectral Density
Recall the Convolution / Multiplication property:
x1 t x2 t X1 f X 2 f
Writing this as an equality

x1 x2 t d X1 f X 2 f e j 2 ft
df

Reversing in time and conjugating we have

x1 x2* t d X 1 f X 2* f e j 2 ft
df

Letting t=0 and setting x2(t)=x1(t)

x1 x1* d X 1 f X 1* f df

Developing Energy Spectral Density – cont.

x1 x1* d X 1 f X 1* f df

Rewriting
2 2
x1 d X1 f df

Energy

Since the integral of |X1(f)|2 over f provides the total energy,

we can call it the energy spectral density.
2
x f X1 f

The energy spectral density is a nonnegative real-valued quantity

for all f, even though the signal x1(t) may itself be complex valued.

2
Autocorrelation Function
Energy Signals
The autocorrelation function provides a measure of the similarity
between a signal and its delayed version.
The autocorrelation function of the energy signal x1(t) for a lag

Rx x1 t x1* t dt

The time lag plays the role of a scanning or searching variable.

Setting the delay =0 we arrive at the total energy

2
Rx 0 x1 t x1* t dt x1 t dt

Autocorrelation Function – cont.

Energy Signals
Starting from the correlation relationship we described earlier

x1 t x1* t dt X 1 f X 1* f e j 2 f
df
However, the first integral is simply the autocorrelation
function.

2
Rx X 1 f X 1* f e j 2 f
df X1 f e j2 f
df

Rx x f e j2 f
df

Thus, the autocorrelation function and the energy spectral

density are Fourier Transform pair.

3
Autocorrelation Function – cont.
Wiener-Khitchine Relations for Energy Signals

Given an energy signal x1(t), the autocorrelation function and

energy spectral density form a Fourier-transform pair.

j2 f
x f Rx e d FT

Rx x f e j2 f
df IFT

Note that the Fourier transformation given in the first equation

is performed with respect to the adjustable lag .

This equation pair constitutes the Wiener-Khitchine Relations

for Energy Signals.

Autocorrelation Function – cont.

Wiener-Khitchine Relations for Energy Signals

By setting f=0
The total area under the curve of the complex-valued
autocorrelation function of a complex-valued energy signal is
equal to the real-valued energy spectral at zero frequency.

x 0 Rx d

By setting =0
The total area under the curve of the real-valued energy
spectral density of an energy signal is equal to the total
energy of the signal.

Rx 0 x f df

4
Example 4.1
Autocorrelation Function of Sinc Pulse
Determine the autocorrelation function of a sinc pulse.
x t sinc t
We could use the definition directly, but it doesn’t appear to be
straightforward.

Rx x1 t x1* t dt

sinc t sinc t dt

How else might we do this problem?

Example 4.1 – cont.

Autocorrelation Function of Sinc Pulse
Since we know that the autocorrelation function is the inverse
Fourier Transform of the ESD, let us examine that approach. It
will be easier.
The ESD is simply
2
x f X f
2
rect f rect f
Taking the inverse Fourier Transform we have

Rx x f e j2 f
df sinc
Thus, the autocorrelation function of a sinc pulse is simply a
sinc pulse.

sinc t rect f

5
Filtering Energy Signals
We know from system theory that the Fourier Transform of the
output of a system can be written as

Y f H f X f
The ESD at the output of the system can be written as
2 2 2 2
y f Y f H f X f H f X f
2
H f x f
Which tells us that when an energy signal is transmitted
through a LTI filter,
the ESD at the output of the system equals the ESD of the
input multiplied by the squared magnitude of the transfer
function.

Filtering Energy Signals

An indirect method for evaluating the effect of LTI filtering on
the autocorrelation function of an energy signal:
1. Determine the Fourier transforms of x(t) and h(t), obtaining
X(f) and H(f), respectively.
2. Use y(f)=|H(f)|2 x(f) to determine the energy spectral
density y(f) of the output y(t).
3. Determine Ry( ) by applying the inverse Fourier transform
to y(f) obtained under point 2.
FT FT
1. x t X f , h t H f
2 2
2. x f X f , y f H f x f

3. Ry y f e j2 f
df

6
Example 4.2
Energy of Lowpass Filtered version of
Rectangular Pulse

A rectangular pulse x(t) of unit amplitude and unit duration is

passed through an ideal lowpass filter |H(f)| of bandwidth B as
shown in the figure below.
The rectangular pulse at the filter input
has unit energy. Evaluate the effect of
varying the bandwidth B on the energy
of the filter output.

The filter input is defined as

x t rect t
and its Fourier transform equals
X f sinc f

Example 4.2 – cont.

Energy of Lowpass Filtered version of
Rectangular Pulse
The ESD x(f) of the filter input x(t) can be given as
X f sinc f
2
x f X f sinc 2 f

This normalized ESD is

shown in the figure for
positive frequencies
only.

7
Example 4.2 – cont.
Energy of Lowpass Filtered version of
Rectangular Pulse
The ESD y(f) of the filter output y(t) can be given as
2
y f H f x f

x f B f B
0 otherwise
The energy of the filter output is therefore
B B B
Ey y f df x f df 2 x f df 2 sinc 2 f df
B 0 0

Since the filter input has unit energy, this result can be seen as
the ratio of the energy of the filter output to that of the filter
input. B
Energy of filter output Ey
Ey 2 sinc 2 f df
Energy of filter input Ex 0

Example 4.2 – cont.

Energy of Lowpass Filtered version of
Rectangular Pulse
B
Energy of filter output Ey
Ey 2 sinc 2 f df
Energy of filter input Ex 0
The rectangular pulse applied to the filter input has unit
duration; therefore, the variable f in above equation represents
a normalized frequency.

The figure for the ratio

shows that just over
90 percent of the total
energy of a rectangular
pulse lies inside the
main spectral lobe of
this pulse.

8
 Cross-Correlation Function
Energy Signals
The cross-correlation function provides a quantitative measure of
the similarity (or coherence) between one signal and the time
delayed version of a second signal.
If x(t) and y(t) are a pair of complex-valued energy signals:

Rxy x t y* t dt
If x(t) and y(t) are somewhat similar, then the cross-
correlation function Rxy( ) will be finite over some range of .

The energy signals x(t) and y(t) are said to be orthogonal over
the entire time domain
if Rxy(0)=0.
Rxy 0 x t y* t dt 0

Cross-Correlation Function
Energy Signals
The second cross-correlation function Ryx( ) is defined as

Ryx y t x* t dt
Relation between the cross-correlation functions Rxy( ) and Ryx( )
Rxy R*yx
Proof: Ryx y t x* t dt , R*yx y* t x t dt

R*yx y* t x t dt , t

R*yx y* x d Rxy

Unlike convolution, correlation is not in general commutative Rxy( )≠Ryx( ).

9
Cross-Correlation Function
Energy Signals
Cross-spectral densities of a pair of complex-valued energy
signals x(t) and y(t), are defined as the respective Fourier
transforms of the cross-correlation functions Rxy( ) and Ryx( ).
j2 f
xy f Rxy e d

j2 f
yx f Ryx e d
With the help of the correlation theorem

xy f X f Y* f

yx f Y f X* f

x1 t x2* t dt X 1 f X 2* f

Cross-Correlation Function
Energy Signals
Two properties of the cross-spectral density
1. Unlike the energy spectral density, cross-spectral density is
complex-valued in general.
2. xy(f)=
* (f) from which it follows that, in general,
yx

xy(f)≠ yx(f).

xy f X f Y* f xy f *
yx f
yx f Y f X *
f xy f yx f

10
Power Spectral Density
Recall the definition of the average power as the time averaged
energy of a signal x(t).
T
1 2
P lim x t dt
T 2T T
The signal x(t) is said to be a power signal (periodic) if P<∞.
Since power signals have infinite energy, they may not be
Fourier transformable. To overcome the problem,
consider a truncated version of the signal x(t). Define
t
xT t rect x t
2T
As long as the duration T is finite, the truncated signal xT(t)
has finite energy and therefore Fourier transformable.
xT t XT f

Power Spectral Density – cont.

Using truncated signal xT(t), we may rewrite the average power
P in terms of xT(t):
1 2
P lim xT t dt
T 2T
From the Rayleigh (Parseval) energy theorem
2 2
xT t dt XT f df

1 2
P lim XT f df
T 2T

11
Power Spectral Density – cont.

1 2
P lim XT f df
T 2T
The convergence of the integral in the limit as T approaches
infinity,
permits us to interchange the order of the limit and
integration operations

1 2
P lim XT f df S x f df
T 2T
where
1 2
Sx f lim XT f
T 2T
is termed the Power Spectral Density or Power Spectrum of the
power signal x(t).

Power Spectral Density – cont.

1 2
Sx f lim XT f
T 2T

The quantity |XT(f)|2/2T is called the periodogram of the signal.

PSD is a nonnegative real-valued quantity for all frequencies.

The total area under the curve of the PSD of a power signal is
equal to the average power of that signal.

P S x f df

12
 Example 4.3
Modulated Wave
Consider the modulated wave
x t g t cos 2 f c t
where g(t) is a power signal that is band limited to B Hertz.
Since the amplitude of the sinusoidal carrier of frequency fc is
varied linearly with signal g(t), we refer to x(t) as a modulated
wave. Find the PSD of x(t) in terms of PSD of g(t), given that
the frequency fc is larger than the bandwidth B.
Let gT(t) denote the truncated version of g(t).
t
gT t rect g t
2T
Correspondingly, the truncated version of x(t) can be given as

xT t gT t cos 2 f c t

Example 4.3 – cont.

Modulated Wave
Since
1 j2 fct j 2 fc t
xT t gT t cos 2 f ct gT t e e
2
using the frequency shifting property of the Fourier transform
1
XT f GT f fc GT f fc
2
where GT(f) is the Fourier transform of gT(t).
Given that fc>B, we find that GT(f-fc) and GT(f+fc) represent
nonoverlapping spectra; therefore their product is zero.
If we evaluate squared magnitude of XT(f) we get
2
2 1 1 2 2
XT f GT f fc GT f fc GT f fc GT f fc
2 4

13
 Example 4.3 – cont.
Modulated Wave
Applying the definition for the PSD of the power signal g(t)
1 2 1 2
Sx f lim XT f , Sg f lim GT f
T 2T T 2T
1 2 1 1 2 1 2
lim XT f lim GT f f c lim GT f f c
T 2T 4 T 2T T 2T
1
Sx f S g f fc S g f fc
4
Except for the scaling factor 1/4, the PSD of the modulated
wave x(t) is equal to the sum of
the PSD Sg(f) shifted to the right by fc and the PSD Sg(f)
shifted to the left by the same amount fc.

Autocorrelation Function
Power Signals
The autocorrelation function of a power signal is
T
1
Rx lim x t x* t dt
T 2T T

Further, just as with energy signals we can show that the

autocorrelation function of power signals and the power
spectral density are a Fourier Transform pair.

Rx Sx f

14
Power Spectral Density
Periodic Signals
While the previous definitions are helpful for general power
signals, for the important subclass of power signals,
periodic signals we can find a more direct expression for
the power spectral density.

Recall that we can write a periodic signal in terms of the

Fourier Series:

x t cn e jn 2 f0 t

Further, the autocorrelation function is

T
1
Rx lim x t x* t dt
T 2T T

Power Spectral Density – cont.

Periodic Signals
Substituting the Fourier Series representation into the
autocorrelation function definition:
T
1 jm 2 f0 t
Rx lim cn e jn 2 f0 t
cm* e dt
T 2T T n m
T
1 jm 2 f 0 t
lim cn e jn 2 f0 t *
m c e dt
T 2T T n m

T
1 jm 2 f 0 t
cn cm* lim e jn 2 f0 t
e dt
n m
T 2T T

T
1 j2 n m f0 t
cn cm* e jm 2 f0
lim e dt
n m
T 2T T

15
Power Spectral Density – cont.
Periodic Signals
T
1
Rx cn cm* e jm 2 f0
lim e j2 n m f0 t
dt
n m
T 2T T

cn cm* e jm 2 f0
n m cn cn*e jn 2 f0

n m n

2
Rx cn e jn 2 f0

The Power Spectral Density can then be found using the Fourier
Transform.

Power Spectral Density – cont.

Periodic Signals
We can find the PSD using the Fourier Transform.
2
Rx cn e jn 2 f0

Sx f F Rx
2
F cn e jn 2 f0

2
cn F e j 2 nf 0

2
Sx f cn f nf 0
n

16
Example 4.4
PSD of Periodic Signals

Find the PSD of the pulse train x(t).

t nT0
x t rect
n T

Example 4.4 – cont.

PSD of Periodic Signals

We know that the PSD of a periodic signal can be written as

2
Sx f cn f nf 0
n
Further, we can determine the Fourier Series coefficients as
t0 T T /2
1 jn 2 f0 t 1 jn 2 f 0 t
cn x t e dt e dt
T t0
T0 T /2

jn f 0 T
1 e e jn f0 T
1 sin nf 0 T T
sinc nf 0 T
T0 jn2 f 0 T0 nf 0 T0
Thus, we have 2
2 T
Sx f cn f nf 0 sinc nf 0 T f nf 0
n n T0

17
Filtering Power Signals
The Fourier Transform of the output of a system when the
input is periodic can be written as
Y f H f X f

H f cn f nf 0
n

H nf 0 cn f nf 0
n
FS coefficients
of output y ( t )

The PSD of the output y(t) can be written as

2 2 2
Sy f H nf 0 cn f nf 0 H nf 0 cn f nf 0
n n

Summary
In this lecture we have briefly reviewed the concepts of Energy
Spectral Density and Power Spectral Density.

The ESD or PSD of the output of a LTI system can be readily

found
using the ESD or PSD of the input and the transfer function
of the system.

The autocorrelation function and the power spectral density

form a Fourier Transform pair for power signals.

The autocorrelation function and the energy spectral density

form a Fourier Transform pair for energy signals.

18