AE2235-II Instrumentation & Signals

Energy and Power

Max Mulder

Version 2019-2020
Friday May 1, 2020

Delft
University of
Technology

Challenge the future

 Energy/Power

Energy/Power Parseval Spectral Density

. 2 / 26
1-4 Energy and Power Signals
In our discussion about Fourier transforms we loosely discussed the
properties of signals of being ‘absolutely integrable’. In order to
properly classify signals, we use the definition of Ziemer Section 1-4:
• signals that have finite energy E, and
• signals that have finite average power P .
The definitions of ‘energy’ and ‘power’ of a signal stem from
electrical engineering. Suppose e(t) is the voltage across a resistor
R producing a current i(t).
i(t)

A B
R

Energy/Power Parseval Spectral Density

. 3 / 26
1-4 Energy and Power Signals (2)

Then the instantaneous power per Ohm is defined as:

e(t)i(t) 2
p(t) = R = i (t) (1-72)
Integrating over the interval |t| ≤ T , we define the total energy and
power “on a per Ohm basis” as:
RT
E = lim i2 (t)dt , in Joule, (1-73)
T →∞ −T

and:
T
1
P = lim 2T
R
i2 (t)dt , in Watt, (1-74)
T →∞ −T

respectively.
Energy/Power Parseval Spectral Density

. 4 / 26
1-4 Energy and Power Signals (3)

Note: In the book of Ziemer the signal energy and power are always (unless
specified otherwise) given on a ‘per Ohm basis’, i.e., normalized to unit resistance.
In doing so, one deviates from the common definition of power, which is simply
p(t) = e(t)i(t). So the power in (1-72) is indeed normalized with respect to R.
Now, we could have chosen to indicate it here with p̂(t) or something else,
distinguishing it from power p(t) = e(t)i(t), but we have chosen to stick to the
definition of Ziemer.
An important consequence is that also in the units of the power and energy
spectra (which are Watts and Joules, respectively) discussed in Ziemer, the
normalization has taken place. So when we discuss power in terms of Watts, we
are in fact talking about Watts per Ohm.
So, (1-73) is actually in Joule/Ω, (1-74) is actually Watt/Ω. And, later on, (1-75) is
actually in Joule/Ω (if x(t) is a current) or in JouleΩ (when x(t) is a voltage).
Similarly, (1-76) is actually in Watt/Ω (x(t) a current) or in WattΩ (x(t) a voltage).
Energy/Power Parseval Spectral Density

. 5 / 26
1-4 Energy and Power Signals (4) VIS
For an arbitrary signal x(t) the total energy normalized to unit
resistance is then defined as:

. RT
E = lim |x(t)|2 dt , Joules (1-75)
T →∞ −T

and the average power normalized to unit resistance is defined as:

. 1 RT
P = lim 2T |x(t)|2 dt , Watts (1-76)
T →∞ −T

Please note that as we are dealing with only real signals in this
lecture, the modulus signs may be removed from (1-75) and (1-76).

Energy/Power Parseval Spectral Density

. 6 / 26
1-4 Energy and Power Signals (5) VIS
Based on definitions (1-75) and (1-76) the following signal classes
are defined:

1. x(t) is an energy signal iaoi 0 < E < ∞ (i.e., P = 0)

2. x(t) is a power signal iaoi 0 < P < ∞ (i.e., E = ∞)
3. Signals that satisfy neither property are neither energy nor
power signals.
@home: Study Examples 1-11, 1-12 and 1-13 well.

Energy/Power Parseval Spectral Density

. 7 / 26
1-4 Energy and Power Signals (6)

Example 1-11 Consider signal x1 (t) = Ae−αt u(t), for positive α, i.e., a step
function multiplied with a decaying exponential function. Then use (1-75) to
compute its energy:
RT T
R
E = lim (Ae−αt u(t))(Ae−αt u(t))dt = A2 lim e−2αt dt
T →∞ −T T →∞ 0
T
2 e−2αt A2 A2
=A lim = lim (e −2αT
− 1) = 2α
T →∞ −2α 0 −2α T →∞

Hence, x1 (t) is an energy signal.

Suppose α → 0, then x2 (t) = Au(t). Clearly, the energy of x2 (t) is infinite. Its
average power, however, becomes:
1
RT 2 1
RT 2 1 A2
P = lim 2T (Au(t))(Au(t))dt = A lim 2T (1)dt = A lim 2T T = 2
T →∞ −T T →∞ 0 T →∞

Hence, x2 (t) is a power signal.

Energy/Power Parseval Spectral Density

. 8 / 26
1-4 Energy and Power Signals (7)

Example 1-12 Consider the periodic sinusoidal signal:

x(t) = A cos(2πf0 t + θ) (1-81)
The normalized average power of this signal is:
1 RT 2
P = lim 2T A cos2 (2πf0 t + θ)dt
T →∞ −T

Since cos2 (u) = 21 (1 + cos(2u)), we get:

2 RT
P = lim 4T A (1 + cos(4πf0 t + 2θ))dt
T →∞ −T
2  T 2
A
= lim 4T t + sin(4πf0 t+2θ) A
= 2 ,
T →∞ 4πf0
−T
since the last term on the right becomes zero. Sinusoidal signals are power
signals, they have an infinite energy.
Note that neither the phase θ, nor the frequency f0 of the sinusoidal signal has an
effect on the signal power. Only its amplitude affects power!!
Energy/Power Parseval Spectral Density

. 9 / 26
 Parseval

Energy/Power Parseval Spectral Density

. 10 / 26
3-6 Parseval’s Theorem (FS)
The average normalized power of a periodic waveform x(t) (period
T0 ) can be written as:
1 R
Pav = T T0 |x(t)|2 dt (3-53)
0

Since |x(t)|2 = x(t)x∗ (t) we can replace x∗ (t) with its Fourier series
representation (3-28):
 ∞ 
1 R
Pav = T T0 x(t)
P
Xn∗ e−j2πnf0 t dt
0 n=−∞
Z
∗ 1
P∞
= Xn x(t)e−j2πnf0 t dt, (3-54)
n=−∞ T 0 T0
| {z }
Xn
by interchanging the order of summation and integration.

Energy/Power Parseval Spectral Density

. 11 / 26
3-6 Parseval’s Theorem (FS) (2) VIS
Clearly, we may write:
∞
1 R 2
Pav = T T0 |x(t)| dt =
P
|Xn |2 (3-55a)
0 n=−∞
or:
∞
P
Pav = X02 +2 |Xn |2 (3-55b)
n=1

In other words, the average power of a periodic signal is simply the

sum of the powers in the phasor components of its Fourier series, or,
just the sum of its complex Fourier series coefficients squared.
Note that X0 is the signal average, which is often called the signal’s
‘DC’-component. The Xn ’s are its harmonic components.

Energy/Power Parseval Spectral Density

. 12 / 26
3-6 Parseval’s Theorem (FS) (3)

Example 3-8 The average power of the sine wave:

x(t) = 4 sin(50πt),
can be computed using the time-domain description above (using f0 = 50π/(2π) = 25
1
Hertz, so T0 = 25 s), by substitution in (3-53):
1/25 1/25
1 2
16 sin2 (50πt)dt = 16
R R
Pav = 1/25
(4 sin(50πt)) dt = 25 2 = 8W (3-56)
0 0
Its power can also be computed much more easily in the frequency domain, namely by
using its complex Fourier series coefficients, using (3-55a):

Since X1 = −2j, X−1 = X1∗ = 2j, so its power is 4 + 4 = 8W .

@home: For now read Example 3-9; study it when we have

discussed filtering.

Energy/Power Parseval Spectral Density

. 13 / 26
4-3 Parseval’s Theorem (FT)
For periodic signals x(t) we have just derived that the average power
in the time domain equals the average power in the frequency
domain:
∞
1 R
Pav = T T0 |x(t)|2 dt =
P
|Xn |2 , (3-55a)
0 n=−∞
a relation known as Parseval’s theorem (Section 3-6 of Ziemer).
It can be shown that a similar theorem (with the same name) exists
for the total energy in aperiodic signals.
We start with expressing the energy of a signal in the time and
frequency domains:
" #
. R∞ ∞
R ∞
R
E= |x(t)|2 dt = x∗ (t) X(f )ej2πf t df dt (4-12)
−∞ t=−∞ f =−∞

Energy/Power Parseval Spectral Density

. 14 / 26
4-3 Parseval’s Theorem (FT) (2) VIS
Reversal of the order of integration yields:
∞
 ∞ 
R R
E= X(f ) x∗ (t)ej2πf t dt df
f =−∞ t=−∞
 ∗
R∞ R∞
= X(f ) x(t)e−j2πf t dt df
f =−∞ t=−∞
R∞
= X(f )X ∗ (f )df (4-13)
f =−∞

We obtain Parseval’s Theorem for Fourier transforms:

R∞ ∞
R
E= |x(t)|2 dt = |X(f )|2 df (4-14)
t=−∞ f =−∞

or, the energy in time equals the energy in frequency.

Energy/Power Parseval Spectral Density

. 15 / 26
 Spectral Density

Energy/Power Parseval Spectral Density

. 16 / 26
1-5 Energy and Power Spectra VIS
For some applications it is useful to define functions of frequency
that, when integrated over all frequencies, give total energy (for an
energy signal) or total power (for a power signal).
Denoting the energy spectral density with G(f ), we then have:
R∞
E= G(f )df , (1-87)
−∞
with E the signal’s total energy.
Denoting the power spectral density with S(f ), we then have:
R∞
P = S(f )df , (1-88)
−∞
with P the signal’s average power.

Energy/Power Parseval Spectral Density

. 17 / 26
1-5 Power Spectral Density
Example 1-12 (extended) Remember the sinusoidal signal:
x(t) = A cos(2πf0 t + θ), (1-81)
2
a power signal with average power A2 .
Now, since the power only depends on the signal’s amplitude, one may deduce
that a signal’s power spectral density looks very similar to the signal’s amplitude
spectrum (the two delta functions at (f − f0 ) and (f + f0 ), remember?).
A2
In fact, one can say that ‘half the signal power’ ( 4 )
can be associated to the
frequency +f0 , and the other half to the frequency −f0 . This is illustrated in Figure
1-18a (page 31).

Energy/Power Parseval Spectral Density

. 18 / 26
1-5 Power Spectral Density (2) VIS
Generalizing, for any signal possessing a two-sided line (amplitude)
spectrum (i.e., any periodic signal), we obtain the power spectral
density by simply taking each line of the amplitude spectrum,
squaring its value, and multiplying it by a delta function located at
that particular frequency.
Note that the power spectral density of any signal is an even
function of frequency, and that it possesses no information about the
phase of the signal.
@home: Study Example 1-14, Problem 1-45

Energy/Power Parseval Spectral Density

. 19 / 26
4-3 Energy Spectral Density VIS
When x(t) is a voltage, then the units of |X(f )|2 are (Volt second)2 .
Since E represents energy on a per Ohm basis we obtain for the unit
of |X(f )|2 (Watt second)/Hz = Joule/Hz.
In other words, the unit of |X(f )|2 is ‘energy density with frequency’,
and the energy spectral density G(f ) can be defined as:
.
G(f ) = |X(f )|2 (4-15)

Integrating (4-15) over all frequencies from −∞ to ∞ yields the total

(normalized) energy contained in the signal x(t).
Integrating (4-15) over frequency band [f1 , f2 ] yields the energy
contained in the signal within the range of frequencies.

Energy/Power Parseval Spectral Density

. 20 / 26
4-3 Energy Spectral Density (2)

So, when we have an energy spectral density that looks like:

2

1.5

|(X(f )|2
1

0.5

−4 −3 −2 −1 0 1 2 3 4
f , Hz

Integrate it from f = −∞ to f = ∞ to compute the total signal energy E:

2

1.5
|(X(f )|2

0.5

−4 −3 −2 −1 0 1 2 3 4
f , Hz

Energy/Power Parseval Spectral Density

. 21 / 26
4-3 Energy Spectral Density (3)

Integrate it from f = −f2 to f = −f1 and from f = f1 to f = f2 to compute the signal

energy for frequencies between f1 and f2 Hz (a bandwidth B = f2 − f1 ).
2

1.5
|(X(f )|2

0.5

0
−f2 −f1 f1 f2
−4 −3 −2 −1 0 1 2 3 4
f , Hz

Obviously, since |X(f )|2 is a real and even function, we can also integrate it between
f = f1 and f = f2 and multiply by 2!

Energy/Power Parseval Spectral Density

. 22 / 26
4-3 Energy Spectral Density (4)

Example Let’s consider the energy in a unit pulse x(t) (width τ = 1 seconds),
illustrated in the figure below, together with its energy spectral density function
G(f ) = |X(f )|2 , with X(f ) = τ sinc(τ f )|τ =1 .
1.5
The energy can be computed easily in time (using (4-12)) to be:
1
R∞ τ /2
R
2
E = |x(t)| dt = (1)2 dt = τ = 1 Joule.
x(t)

0.5
t=−∞ −τ /2
0
In frequency we get: (using (4-14)):
−0.5 R∞ R∞
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
2
t, s E = |X(f )| df = |τ sinc(τ f )|2 df
f =−∞ −∞
∞ R
1 sin(πτ f ) 2
2
0.8
=τ ( (πτ f )
) df (*)
|X(f )|2

0.6 −∞
0.4
R∞ sin(u) 2
Note that (
u
) du = π (Ziemer integrals, App. F,
0.2 −∞
0 page 383), so when substituting for u, πτ f and for du, πτ df we
1.
find that the integral on the right-hand side of (*) equals τ
−0.2
−3 −2 −1 0 1 2 3
f , Hz So, we get that E = τ , and with τ = 1 we get 1 Joule.

Energy/Power Parseval Spectral Density

. 23 / 26
4-3 Energy Spectral Density (5)

Example (continued) Clearly, we see that the energy of a unit pulse with width
τ equals τ Joules. Given the shape of the energy spectral density function G(f ), one may
wonder how much of this energy in the signal lies within the main lobe, in our example
between f ± 1 Hz (width τ = 1 second), or in general between f ± τ1 Hz.
Obviously, we can only do this in the frequency domain, by integrating the energy spectral
density function G(f ) = |X(f )|2 from f = − τ1 to f = τ1 Hz (the ‘null-to-null bandwidth’
B0 = τ1 ):
1 1
Rτ Rτ
Ef ∈[− 1 , 1 ] = |X(f )|2 df = τ 2 ( sin(πτ f) 2
(πτ f )
) df (*)
τ τ 1 1
f =− τ −τ

The integral (*) turns out to be a rather nasty one to calculate analytically, numerical
calculation yields that it equals approximately 0.9028 Joule.
Generally, one can show that 90% of the energy in the unit pulse (width τ seconds) lies
between f = −1/τ and f = 1/τ Hz.

Energy/Power Parseval Spectral Density

. 24 / 26
4-3 Energy Spectral Density (6)

Example 4-2 In Example 1-10 we have shown that the energy in the signal
−αt 1
x(t) = e u(t) (with α > 0) equals 2α . The FT of this signal can be derived as:
R∞ −αt −j2πf t
R∞ −αt −j2πf t R∞ −(α+j2πf )t
X(f ) = (e u(t))e dt = e e dt = e dt
−∞ 0 0
−(α+j2πf )t ∞
= e−(α+j2πf ) = 1 (0 − 1) = 1 (4-16)
0
−(α + j2πf ) α + j2πf

The energy spectral density G(f ) then equals |X(f )|2 which can be computed as follows:
G(f ) = |X(f )|2 = ( (Re(X(f )))2 + (Im(X(f )))2 )2
p
 2  2
α −2πf
= 2 2 + 2 2 = 2 1 2 (4-17)
α + (2πf ) α + (2πf ) α + (2πf )
The energy EB contained in the frequency range −B < f < B is then:
RB 2πB/α
EB = 1 df
v=2πf /α
= 1 R 1 dv = 1 arctan( 2πB ) (4-18)
α + (2πf )2
2 πα 1 + v2 πα α
−B 0
1
Concluding, when B → ∞ this integral becomes 2α
.

Energy/Power Parseval Spectral Density

. 25 / 26
Recommended home work
• Study 1-4, 1-5, 3-6 and 4-3
• Read the Summary points 4-3, 4-4
• Derive the equations yourself, get a grip on the mathematics!
• Make Exercises 1-33, 1-36, 1-37, 1-38, 1-39, 1-40, 1-41, 1-43,
1-45, 4-11, 4-12

Note that in the reader on Signal Modulation and Detection, by

C.C.J.M. Tiberius (part II) the Power Spectral Density function (PSD)
will be discussed in more detail (Section 1.3).

Energy/Power Parseval Spectral Density

. 26 / 26