You are on page 1of 29

Chapter 1

Characterization of Signals and


Systems

1.1 Introduction
This chapter will characterize the signals and systems that are used to transmit digital
information over a communication channel. Specifically, this chapter

1. characterizes deterministic bandpass signals and systems

2. characterizes bandpass stationary stochastic processes

3. introduces the vector space representation of signals

4. introduces types of digital modulation and the spectral characteristics of these digital
signals.

First, some elementary signal types, which are used extensively in this chapter will be de-
fined. These signals types are baseband waveforms, bandpass waveforms (deterministic),
and narrowband bandpass waveforms.

Baseband Waveform - the spectral magnitude of the signal, m(t) is nonzero for frequen-
cies in the vicinity of the origin (f = 0) and negligible elsewhere, an example of which
is shown in Figure 1.1.

Bandpass Waveform - significant spectral magnitude components of the signal, s(t), are
clustered in a band that does not include f = 0. This band is concentrated about
f = ±fc , where fc  0 and fc is called the carrier frequency, an example of which is
shown in Figure 1.2.

Narrowband Bandpass Waveform - A bandpass signal which meets the condition 2W 


fc , where fc is near the center of the band and 2W is the bandwidth of the bandpass
signal as shown in Figure 1.2.

1
|M(f )|

M(f ) = F {m(t)}

−W W f

Figure 1.1: The Spectral Magnitude of an Example Baseband Waveform

|S(f )| S(f ) = F {s(t)}

−f c fc f
2W

Figure 1.2: The Spectral Magnitude of an Example Bandpass Waveform

2
1.2 Communication System
The general form of the digital communication systems discussed in this chapter is shown in
Figure 1.3. The various components and signals in Figure 1.3 are defined as follows:

Noise
n(t)
TRANSMITTER RECEIVER
Information
input Transmissionr(t) Carrier v(t) Signal
Signal u(t)Carrier s(t) m̃(t)
Processing circuits Medium circuits Processing To information
m(t) (channel)
sink (user)

Figure 1.3: General Digital Communication System

m(t) - The modulating signal, message information signal being sent. It is a baseband
signal (ie. the spectrum of m(t) is concentrated about frequency, f = 0).
Transmitter - It generates the modulated signal, s(t). The modulation process involves
transforming the message signal into a bandpass signal at the carrier frequency, fc .
u(t) is the complex lowpass equivalent (or the complex envelope) of the bandpass
signal (will be defined shortly). The signal processing block maps m(t) into u(t). The
carrier circuit block converts u(t) into the bandpass signal s(t).
Transmission channel - The modulated bandpass signal is passed through a noisy chan-
nel. Two general catagories of channels are hardwire and softwire. Some examples
of hardwire channels are telephone wires, coaxial cables, waveguides and fiber optic
cables. Examples of softwire channels are air, a vacuum and water. The channel mod-
ifies the signal s(t) if the channel frequency response is not ideal (ie if it does not have
a constant gain and linear phase) and corrupts the signal with channel noise, n(t).
Channel noise can be produced by natural electrical disturbances such as lightning
and manmade sources such as transmission lines, car ignition systems and computers.
The channel could also have multiple paths, which would produce multiple signals with
different travelling time delays. These multipath signals could destructively add at the
receiver causing fades (for example, this occurs in cellular radio or in-building radio).
The received corrupted signal is labelled r(t).
Receiver - The carrier circuit down converts the bandpass signal, r(t) into a lowpass equiv-
alent signal, v(t), which is a filtered and corrupted version of u(t). The signal process-
ing block then produces an estimate of the message signal, m̃(t), which is a corrupted
version of m(t).
Note that, in general, u(t) and v(t) are complex signals, and they will be used later to
simplify the analysis of communication systems. Since u(t) and v(t) are complex they are
not physically realizable.

3
1.3 Representation of Bandpass Signals
A bandpass signal can be represented by

s(t) = a(t) cos [ωc t + θ(t)] , (1.1)

where

a(t) is the time varying amplitude of the signal,

θ(t) is the time varying phase of the signal,

ωc = 2πfc and fc is called the carrier frequency of the signal.

This signal representation is typical of baseband digital or analog information signals which
are transmitted using some type of carrier modulation. If s(t) is a narrowband bandpass
1
waveform, then fc  2W and thus a(t) and θ(t) are slowly time varying envelope and phase,
respectively. An example of this for a deterministic narrowband bandpass signal s(t) is given
in Figure 1.4.

s(t)

1
≈ 2W

carrier with time varying


time varying phase envelope

Figure 1.4: An Example of a Deterministic Narrowband Bandpass Signal

An alternate representation for the above bandpass signal is given by

s(t) = <[u(t)ejωc t ] (1.2)

where

<[·] denotes the real part of [.],

and u(t) = a(t)ejθ(t) is referred to as the complex envelope of s(t).

4
This representation is equivalent to the previous cosine representation as shown below

s(t) = <[u(t)ejωc t ]
= <[a(t)ejθ(t) ejωc t ]
= <[a(t)ej(ωc t+θ(t)) ]
= <[a(t)(cos (ωc t + θ(t)) + j sin (ωc t + θ(t)))]
= a(t) cos (ωc t + θ(t)) (1.3)

Another representation for the bandpass signal can be obtained by expanding the cosine
term in
s(t) = a(t) cos (ωc t + θ(t)) (1.4)
using the identity
cos (x ± y) = cos x cos y ∓ sin x sin y (1.5)
as

s(t) = a(t) cos (θ(t)) cos (ωc t) − a(t) sin (θ(t)) sin (ωc t)
= x(t) cos (ωc t) − y(t) sin (ωc t) (1.6)

where

x(t) = a(t) cos (θ(t)) (1.7)


y(t) = a(t) cos (θ(t)). (1.8)

Note that x(t) and y(t) are referred to as the quadrature components of s(t), where x(t) is
the in-phase component and y(t) is the quadrature component. These are called quadrature
components because s(t) may be represented as two phasors with the second at an angle of
+90◦ with respect to the first as shown in

s(t) = x(t) cos (ωc t) + y(t) cos (ωc t + 90◦ ). (1.9)

In the above three representations of s(t), the waveforms u(t), x(t), y(t), a(t) and θ(t)
are all baseband or lowpass signals, except for u(t), which is complex. Also note that a(t)
is a non-negative signal. The lowpass signals from the three s(t) representations can be
compared using phasor diagrams. In terms of a rotating phasor diagram, s(t) is shown in
Figure 1.5. Suppressing the rotation term, ωC t, eliminates the carrier term, the resulting
phasor diagram shown in Figure 1.6, represents the lowpass equivalent system and gives the
relationships of the various lowpass signals.

1.3.1 Complex Number Representation


Complex numbers are an integral part of the lowpass equivalent representation, thus it is
worthwhile to briefly review complex number theory. A complex number, z, can be defined
as an ordered pair

z = (x, y)

5
Im

s(t)

a(t)
wc t + θ(t)

Re

Figure 1.5: The Bandpass Signal, s(t), Represented as a Rotating Phasor

Im

u(t)

a(t) y(t)
θ(t)

Re
x(t)

Figure 1.6: The Lowpass Equivalent Signal, u(t), Represented as a Phasor

6
of real numbers x and y, where x and y are known as the real and imaginary parts of z

x = <(z) (1.10)
y = =(z). (1.11)

Note that engineers typically represent the complex number z as

z = x + jy (1.12)

A geometric interpetation of this representation is shown in Figure 1.7. Basically, a complex

Im

(x, y)
x + jy

Re

Figure 1.7: A Geometric Representation of a Complex Number

number is just an ordered pair in a two dimensional space. This ordered pair can also be
represented in polar form using

x = r cos θ, (1.13)
y = r sin θ, (1.14)

with r and θ shown in Figure 1.8. It should be obvious from Figure 1.8 that

Im

(x, y)
x + jy = z
r
θ

Re

Figure 1.8: Polar Representation of a Complex Number

7
q
r = |z| = x2 + y 2, (1.15)
y
tan θ = , and (1.16)
x
z = r(cos θ + j sin θ) (1.17)

which is the polar form of z.


The third representation of a complex number is the exponential form, which can be
derived from the polar form using Euler’s formula

eθ = cos θ + j sin θ (1.18)

giving

z = r(cos θ + j sin θ)
= rejθ (1.19)

in the engineering field this geometrically represents a phasor. The exponential representa-
tion is shown in Figure 1.9.

Im

z = rejθ

r
θ

Re

Figure 1.9: Exponential Representation of a Complex Number

1.3.2 Lowpass Equivalent Summary


The low pass equivalent, u(t), of a signal

s(t) = a(t) cos (ωc t + θ(t)) = <{u(t)ejωc t } (1.20)

can be represented in three forms

u(t) = x(t) + jy(t), complex number form (1.21)


= a(t) cos θ(t) + ja(t) sin θ(t), polar form (1.22)
= a(t)ejθ(t) , exponential form (1.23)

8
where the interrelationships are defined by
q
a(t) = |u(t)| = x2 (t) + y 2 (t) (1.24)
y(t)
θ(t) = 6 u(t) = arg[u(t)] = arctan (1.25)
x(t)
x(t) = <{u(t)} = a(t) cos θ(t) (1.26)
y(t) = ={u(t)} = a(t) sin θ(t) (1.27)

These parameters are geometrically summarized in Figure 1.10.

Im

u(t)

a(t)
y(t)
θ(t)

x(t) Re

Figure 1.10: Geometric Relationships Between Lowpass Parameters

The lowpass equivalent (complex envelope) signal, u(t), is a baseband signal. In the block
diagram of Figure 1.11, shown previously, the carrier circuits basically involve frequency
translations of u(t). It is desirable to eliminate these frequency translations when analyzing

Noise
n(t)
TRANSMITTER RECEIVER
Information
input Transmissionr(t) Carrier v(t) Signal
Signal u(t)Carrier s(t) m̃(t)
Processing circuits Medium circuits Processing To information
m(t) (channel)
sink (user)

Figure 1.11: General Digital Communcation System

communication systems. This would simplify the performance analysis and also make the
analysis independent of carrier frequencies and channel frequency bands. This can be done
by reducing all bandpass signals and channels to equivalent lowpass signals and channels.

9
1.3.3 Frequency Domain View of Bandpass and Lowpass Equiva-
lent Signals
A bandpass signal can be represented in the time domain by

s(t) = <{u(t)ejωc t } (1.28)

In the frequency domain a bandpass signal is defined by


(
if |f | < fc − W
S(f ) = 0 (1.29)
or |f | > fc + W
An example of a possible spectrum for a bandpass signal is shown in Figure 1.12. In Figure

S(f )
|S(f )|

−fc − W −fc −fc + W fc − W fc fc + W f

arg[S(f )]

Figure 1.12: Example Bandpass Spectrum

1.12, S(f ) = F {s(t)}, where s(t) is a real signal. Note that, if s(t) is a real signal, its
spectrum S(f ), will exhibit hermitian symmetry, which is defined as follows.
Hermitian Symmetry - If s(t) is a real signal then

S(−f ) = S ∗ (f ), (1.30)

where denotes complex conjugate. An eqivalent form is

|S(−f )| = |S(f )|,


arg[S(−f )] = −arg[S(f )],

which shows that S(f ) has even amplitude symmetry and odd phase symmetry. It is worth-
while to investigate the relationship between the lowpass equivalent spectrum, U(f ), and
the signal spectrum, S(f ). The relationship of S(f ) with respect to U(f ) can be determined
as follows. First take the Fourier transform of s(t) = <{u(t)ejωc t },
Z ∞
S(f ) = s(t)e−j2πf t dt
−∞
Z ∞
= <{u(t)ejωc t }e−j2πf t dt, (1.31)
−∞

10
and then substitute the identity <{ξ} = 12 (ξ + ξ ∗ ) into the above equation to give

1Z ∞
S(f ) = (u(t)ejωc t + (u(t)ejωc t )∗ )e−j2πf t dt
2 −∞
Z ∞ Z ∞ 
1
= {u(t)ejωc t }e−j2πf t + {(u(t)ejωct )∗ }e−j2πf t dt
2 −∞ −∞
1h i
= F {u(t)ejωct } + F {(u(t)ejωct )∗ }
2
1
= [F {z(t)} + F {(z ∗ (t)}] (1.32)
2
The Fourier transform of

z(t) = u(t)ej2πfc t

is determined using the Fourier transform identity

v(t)ej2πfc t ←→ V (f − fc )

to give

Z(f ) = F [z(t)] = U(f − fc )

The term F {(z(t)∗ } can be determined using the Fourier relationship

v ∗ (t) ←→ V ∗ (−f )

to give
F [z ∗ (t)] = Z ∗ (−f ) = U ∗ (−f − fc ) (1.33)
Substituting Equation (1.33) and (1.33) into (1.32) gives
1
S(f ) = [U(f − fc ) + U ∗ (−f − fc )] (1.34)
2
This equation demonstrates that the frequency content of u(t) must be concentrated about
f = 0, (ie. a lowpass signal), since the frequency content of s(t) is concentrated about fc . A
possible U(f ) that will satisfy (1.34) is given by

U(f ) = 2S(f + fc )us (f + fc ) (1.35)

where us (f ) is the unit step function defined by


(
1 f >0
us (f ) = (1.36)
0 otherwise

Equation (1.35) simply indicates that U(f ) equals the positive frequency portion of 2S(f )
translated down to the origin. An example of U(f ) being generated from (1.35) is shown in
Figure 1.13. Note that these plots do not show the phase of the spectrum. S(f ) is the top

11
S(f )

−fc 0 fc f

S(f + fc )

a
us (f + fc )
1

−2fc −fc 0 f

U(f )
U(f ) = 2S(f + fc )Us (f + fc )
2a

Figure 1.13: An Example of the Generation of U(f ) from S(f )

U(f )

−W 0 W f

Figure 1.14: U(f ) for the Spectrum in Figure 1.12

12
plot in Figure 1.13. The second plot shows S(f ) shifted by fc , S(f + fc ) and us (f ). The
third plot shows U(f ).
Using the approach in the above example, U(f ), shown in Figure 1.14, can be determined
for the spectrum S(f ) of Figure 1.12. Note that U(f ) in Figure 1.14 does not exhibit
hermitian symmetry (ie. even magnitude symmetry and odd phase symmetry). Since U(f )
does not exhibit hermitian symmetry, u(t), will be complex, thus it is called the complex
envelope representation.
Example
An amplitude modulated (AM) signal is given by

s(t) = [1 + m(t)] cos ωc t (1.37)

where m(t) is the baseband information signal. By inspection, the lowpass equivalent, u(t),
is given by
u(t) = 1 + m(t) (1.38)
and its Fourier transform is
U(f ) = δ(f ) + M(f ) (1.39)
The spectrum of s(t) is given by
1
S(f ) = [U(f − fc ) + U ∗ (−f − fc )]
2
1
= [δ(f − fc ) + M(f − fc ) + δ ∗ (−f − fc ) + M ∗ (−f − fc )] (1.40)
2
This can be simplified by noting that the delta function can be defined as an even function,

δ(f ) = δ(−f ), (1.41)

to give
1
S(f ) = [δ(f − fc ) + M(f − fc ) + δ(f + fc ) + M ∗ (−f − fc )]. (1.42)
2
Note that in this example, m(t) is real, so u(t) is also real. This is indicated by the hermitian
symmetry shown Figures 1.15 and 1.16. S(f ) also exhibits hermitian symmetry as shown
in Figure 1.17.

1.4 Lowpass Equivalent of a Bandpass System


A linear bandpass system which is described by its impulse response, h(t) or its frequency
response, H(f ), can also be defined in terms of an equivalent lowpass system with impulse
response, c(t), which is, in general, complex-valued. The Fourier transform of c(t) is C(f )
and this will be defined as
C(f ) = H(f + fc )us (f + fc ) (1.43)
This is the positive portion of H(f ) translated down to the origin. As previously discussed
for U(f ), C(f ), in general, lacks hermetian symmetry and thus c(t) is complex. Using (1.43)

13
M(f )

arg[M(f )] |M(f )|

−W W f

Figure 1.15: The Spectrum of the Signal m(t)

U(f )

δ(f )

−W W f

Figure 1.16: The Spectrum of u(t)

S(f )

lower upper
sideband sideband

−fc − W −fc −fc + W fc − W fc fc + W f

Figure 1.17: The Spectrum of s(t)

14
and the fact that h(t) is real, which means H(f ) is hermitian symmetric, H(f ) can be defined
as

H(f ) = C(f − fc ) + C ∗ (−f − fc ).

The inverse fourier transform of H(f ) gives

h(t) = F −1 {H(f )} = F −1 {C(f − fc ) + C ∗ (−f − fc )}


= c(t)ej2πfc t + c∗ (t)e−j2πfc t
= 2<{c(t)ej2πfc t }

It is obvious that h(t) is real, as it was originally specified.

1.5 Lowpass Equivalent System Response


Assume that a narrowband bandpass system, h(t), has as its input a narrowband bandpass
signal, s(t), as shown in Figure 1.18. The response of the system, r(t), is given by the

S(t) r(t)
h(t)

Figure 1.18: Narrowband Bandpass System

convolution integral Z ∞
r(t) = s(τ )h(t − τ )dτ (1.44)
−∞

The output of the bandpass system, r(t), is also a bandpass signal and it can be represented
by
r(t) = <{v(t)ej2πfc t }, (1.45)
Taking the Fourier transform of the convolution integral, the resulting system frequency
response is
R(f ) = S(f )H(f ) (1.46)
The previously derived expressions
1
S(f ) = [U(f − fc ) + U ∗ (−f − fc )] (1.47)
2
H(f ) = [C(f − fc ) + C ∗ (−f − fc )] (1.48)

can be substituted into R(f ), to give

R(f ) = S(f )H(f ) (1.49)


1
R(f ) = [U(f − fc ) + U ∗ (−f − fc )][C(f − fc ) + C ∗ (−f − fc )] (1.50)
2

15
Expanding this equation gives
1
R(f ) = [U(f − fc )C(f − fc ) + U ∗ (−f − fc )C ∗ (−f − fc )
2
+U(f − fc )C ∗ (−f − fc ) + U ∗ (−f − fc )C(f − fc )] (1.51)

The third and fourth terms in the above expression are zero since there is no spectral overlap
(assuming they are bandlimited). This can be demonstrated by taking the third term in
(1.51) and plotting its two components as shown in Figures 1.19 and 1.20. From these two

|U(f − fc )|

fc f

Figure 1.19: Magnitude of U(f − fc )

|C ∗ (−f − fc )|

−fc f

Figure 1.20: Magnitude of C ∗ (−f − fc )

figures it should be obvious that the product will be

U(f − fc )C ∗ (−f − fc ) = 0 (1.52)

Similarly
U ∗ (−f − fc )C(f − fc ) = 0 (1.53)
resulting in
1
R(f ) = [U(f − fc )C(f − fc ) + U ∗ (−f − fc )C ∗ (−f − fc )] (1.54)
2
Defining
V (f ) = U(f )C(f ) (1.55)

16
and substituting this into R(f ) gives
1
R(f ) = [U(f − fc )C(f − fc ) + U ∗ (−f − fc )C ∗ (−f − fc )]
2
1
= [V (f − fc ) + V ∗ (−f − fc )] (1.56)
2
V (f ) is the output spectrum of the equivalent lowpass system excited by the equivalent
lowpass signal. Taking the inverse fourier transform of V (f ) results in the time domain
convolution integral Z ∞
v(t) = u(τ )c(t − τ )dτ (1.57)
−∞
A block diagram of the lowpass system is shown in Figure 1.21. From the above discussion,

u(t) v(t)
c(t)

Figure 1.21: Lowpass System

it is apparent that linear bandpass signals and systems can be analyzed using equivalent
lowpass signals and systems.

1.6 An Alternate Introduction to Lowpass Equivalent


This section introduces the lowpass equivalent by considering single single sided spectrums
and the Hilbert transform (see page 36 of Simulation of Communication Systems, by Jeruchim,
M.C., Balaban, P., Shannigan, K.S.)
The spectrum of any real valued signal exhibits hermitian symmetry about f = 0.
(S(−f ) = S ∗ (f )). This suggests that there is some redundancy between the positive and
negative portions of the frequency spectrum. This indeed is the case, the information of
a real-valued time signal is completely contained within its positive frequency spectrum.
In certain applications the negative frequency spectrum is discarded and only the positive
frequency spectrum is processed. An example of this is single sideband modulation. If the
negative frequency spectrum is discarded, the hermitian condition is not met and thus the
time signal, given by the inverse Fourier transform must be complex-valued. Discarding the
negative frequency spectrum results in a single sided or one-sided spectrum. To investigate
the implications of one-sided spectra, first define a real function of time, m(t), having the
spectrum, M(f ) which are shown in the Figures 1.22 and 1.23, respectively. The double-
sided spectrum, M(f ), can be converted to a one-sided spectrum by passing the real signal
through an ideal linear filter with the frequency response
(
2 if f > 0
L(f ) = (1.58)
0 if f < 0

17
m(t)

Figure 1.22: Example m(t)

|M(f )|

Figure 1.23: Example Spectrum for m(t)

18
which is shown in Figure 1.24. The magnitude of the one-sided spectrum, Z(f ), is shown
in Figure 1.25. Note that in generating the one-sided spectrum the negative portion is

L(f )

Figure 1.24: One Sided Filter L(f )

|Z(f )|

2a

Figure 1.25: One Sided Spectrum Z(f )

removed and piled on top of the positive portion, resulting in the area under the one-sided
magnitude spectrum being the same as the double-sided magnitude spectrum. The impulse
response of the filter, L(f ), can be determined using Fourier transform tables to give
j
l (t) = δ(t) + , (1.59)
πt
where the imaginary part of l(t) is taken to be zero at t = 0. The output time function, z(t),
which is the Fourier inverse of Z(f ), is given by the convolution
 
j
z(t) = m(t) ∗ δ(t) + . (1.60)
πt
Performing this convolution results in

z(t) = m(t) + j m̂(t) (1.61)

19
where Z
1 ∞ m(τ ) 1
m̂(t) = dτ = m(t) ∗ (1.62)
π −∞ t − τ πt
is called the Hilbert transform of m(t). The complex waveform, z(t), whose imaginary part
is the Hilbert transform of the real part is referred to as a complex analytic signal. Note that
the Fourier transform of a complex analytic signal has a zero value for negative frequencies,
ie. it is a one-sided frequency spectrum. The Fourier transform of the Hilbert transform
1
m̂(t) = m(t) ∗ (1.63)
πt
is given by
F {m̂(t)} ≡ M̂ (f ) = −jM(f )sgnf (1.64)
where 
 −j
 if f > 0
−jsgnf = 0 if f = 0 (1.65)


+j if f < 0
Thus it is theoretically possible to generate the Hilbert transform of a signal by passing it
through a linear circuit that provides a 90◦ phase lag at all positive frequencies (shifts the
phase by −90◦ ) and a 90◦ phase lead at all negative frequencies (shifts the phase by +90◦ ).
Some properties of the Hilbert transform and complex analytic signals are illustrated in the
following examples.
1) Given the signal

g(t) = cos ωo t

the Hilbert transform is


∞ cos ω τ Z
ˆ = 1
g(t)
o

π −∞ t − τ
1 Z ∞ cos ωo (τ − t + t)
= dτ
π −∞ t−τ
using cos (A − B) = cos A cos B + sin A sin B gives
Z Z
ˆ = cos ωo t
∞ cos ωo (t − τ ) sin ωo t ∞ sin ωo (t − τ )
g(t) dτ + dτ.
π −∞ t−τ π −∞ t−τ
This can be simplified by noting that cos ωo (t − τ ) is an even function, 1/(t − τ ) is an odd
function, and the product of an even function and an odd function is odd. Integrating this
odd function results in
Z ∞ cos ωo (t − τ )
dτ = 0.
−∞ t−τ
Using definite integral tables
Z ∞ sin ω0 (t − τ )
dτ = π if ω0 > 0. (1.66)
−∞ t−τ

20
and thus
ĝ(t) = sin ω0 t (1.67)
and the complex analytic signal corresponding to cos ω0 t is

z(t) = g(t) + jĝ(t)


= cos ω0 t + j sin ω0 t
= ejω0 t (1.68)

Note that z(t) = ejω0 t is also referred to as the pre-envelope of cos ω0 t. Thus the complex
analytic signal or pre-envelope corresponding to a cosine is just the complex phasor used in
circuit theory. Also, as discussed before, a real signal has a double-sided spectrum and a
complex analytic signal has a one-sided spectrum. This is apparent in this example, where
the spectrum of the cosine consists of two delta function at ±ω0 , each having a weight of 12 .
The spectrum of the complex analytic signal, ejω0 t , is a single delta function at ω0 having a
weight of 1, which is a one-sided spectrum.
2) In bandpass communication systems, signals of the form a(t) cos ωc t are often used
where the spectrum of a(t) contains only low frequencies and ωc is a relatively high frequency.
The complex analytic form of this type of signal is very important, as will be shown shortly.
We will consider a more general version of this problem by analyzing the signal

s(t) = a(t)g(t), (1.69)

where a(t) is a low-frequency signal and g(t) is a high-frequency signal, and the Fourier
transforms are defined by

A(f ) = 0 , |f | > W (1.70)


G(f ) = 0 , |f | < W. (1.71)

The non overlapping spectra of A(f ) and G(f ) are shown in Figure 1.26.

|A(f )|
|G(f )|

−W W f

Figure 1.26: Non-Overlapping Spectra

The Hilbert transform of s(t) can be represented by

ŝ(t) = a(t)ĝ(t) (1.72)

21
if a(t) is a low-frequency signal and g(f ) is a high-frequency signal and their spectrum’s
do not overlap. This can be demonstrated by observing the spectrums of each. Taking the
Fourier transform of the left hand side of (1.72) gives
d
F [ŝ(t)] = F [a(t)g(t)]
= −jsgnf F [a(t)g(t)]
= −jsgnf A(f ) ∗ G(f ), (1.73)

and taking the Fourier transform of the right hand side of (1.72) gives

F [a(t)ĝ(t)] = A(f ) ∗ F [ĝ(t)]


= A(f ) ∗ (−jsgnf G(f )
= −jA(f ) ∗ (sgnf G(f )) (1.74)

Assume A(f ) and G(f ) are given by the plots in Figures 1.27 and 1.28, respectively. Note

|A(f )|

Figure 1.27: Assumed Spectrum for A(f )

G(f )

−fc fc f

Figure 1.28: Assumed Spectrum for G(f )

that the explanation will be easier to understand if it is assumed G(f ) consists of delta
functions (ie. cos ωc t) as shown in Figure 1.28. For this case the convolution of A(f ) ∗ G(f )

22
will result in A(f ) being shifted to ±fc . Lets ignore the −j term and compare sgnf A(f )∗G(f )
with A(f ) ∗ (sgnf G(f )). A plot of sgnf A(f ) ∗ G(f ) from F {ŝ(t)} (1.73) is given in Figure
1.29. Similarly, A(f ) ∗ (sgnf G(f )) from F [a(t)ĝ(t)] (1.74) can easily be visualized. First a

sgnf A(f ) ∗ G(f )

−fc fc f

Figure 1.29: sgnf A(f ) ∗ G(f )

plot of sgnf G(f ) is given in Figure 1.30. The convolution of sgnf G(f ) with A(f ) will result
in a plot exactly the same as the one shown in Figure 1.29.
The previous discussion demonstrated that
d
ŝ(t) = a(t)g(t) = a(t)ĝ(t) (1.75)

where it is assumed that a(t) is a low frequency signal and g(t) is a high frequency signal
and they do not overlap (in frequency). The important result, here, is that only the high-
frequency component in the signal a(t)g(t) is Hilbert transformed when the Hilbert transform
of the product is computed. Thus for a signal of the form

s(t) = a(t) cos (ωc t + θ(t)), (1.76)

where a(t) and θ(t) are low-frequency time functions, the complex analytic form is given by

c(t) = a(t)(cos (ωc t + θ(t)) + j cd


os(ωc t + θ(t))) (a(t) is low frequency) (1.77)

Using
cos (ωc t + θ(t)) = cos (ωc t) cos (θ(t)) − sin (ωc t) sin (θ(t)), (1.78)
the imaginary term in (1.77) is given by

cd
os(ωc t + θ(t)) = cd c
os(ωc t) cos (θ(t)) − sin(ω c t) sin (θ(t)) (1.79)

23
sgnf G(f )

fc
−fc f

Figure 1.30: sgnf G(f )

since θ(t) is a low frequency term. Taking the Hilbert transform of the two terms in the
RHS of (1.79) gives

cd
os(ωc t + θ(t)) = sin (ωc t) cos (θ(t)) + cos (ωc t) sin (θ(t))
= sin (ωc t + θ(t)) (1.80)

Thus the complex analytic form of

s(t) = a(t) cos (ωc t + θ(t)) (1.81)

is obtained by substituting (1.80) into (1.77) to give

c(t) = a(t)(cos (ωc t + θ(t)) + j sin (ωc t + θ(t)))


= a(t)ej(ωc t+θ(t))
= a(t)ejθ(t) ejωc t
= u(t)ejωc t (1.82)

where, as before, u(t) = a(t)ejθ(t) is called the complex envelope or the low pass equivalent
signal. Taking the Fourier transform of c(t) gives

C(f ) = U(f − fc ). (1.83)

u(t) is generally not complex analytic, thus U(f ) will not generally be zero for negative
frequency. But if c(t) is to be complex analytic C(f ) must be zero for negative frequency.
This requires that U(f ) be zero for f < −fc . This condition

U(f ) = 0 for f < −fc (1.84)

is approximately satisfied if s(t) is a narrowband bandpass signal.

24
The complex analytic signal, c(t), is a bandpass signal with a single-sided spectrum. The
lowpass equivalent signal, u(t), equals a frequency shifted version of the complex analytic
signal, c(t), as shown by
c(t)e−jωc t = u(t)ejωc t e−jωc t
= u(t)

1.7 Bandpass Stationary Stochastic Processes


The previous description of lowpass equivalent signals applied to deterministic signals. In
this section the lowpass equivalent concept is extended to bandpass stationary stochastic
processes. Consider n(t), which is a wide-sense stationary stochastic process with zero mean
and power density spectrum Φnn (f ). The stochastic process, n(t), is referred to as a narrow-
band bandpass process if the width of the power spectral density is much smaller than fc ,
the carrier frequency. A narrowband bandpass process can be represented in the following
three forms.
n(t) = a(t) cos (ωc t + θ(t))
= x(t) cos (ωc t) − y(t) sin (ωc t)
= <[z(t)ejωc t ] (1.85)
where a(t) is the envelope of n(t), θ(t) is the phase of n(t), x(t) and y(t) are the quadrature
components of n(t), and z(t) is the complex envelope of n(t), z(t) = x(t) + jy(t). n(t) is a
wide-sense stationary random process which is characterized by its mean and autocorrelation
function (power density spectrum). Thus the above three forms are really only useful in terms
of determining these random process characteristics.
The autocorrelation function of n(t) can be determined using the quadrature represen-
tation for n(t).
ϕnn (τ ) = E[n(t)n(t − τ )] = E[n(t)n(t + τ )]
= E[(x(t) cos (ωc t) − y(t) sin (ωc t))
×(x(t + τ ) cos (ωc (t + τ )) − y(t + τ ) sin (ωc (t + τ )))]
= E[(x(t)x(t + τ ) cos (ωc t) cos (ωc (t + τ ))
+y(t)y(t + τ ) sin (ωc t) sin (ωc (t + τ ))
−y(t)x(t + τ ) sin (ωc t) cos (ωc (t + τ ))
−x(t)y(t + τ ) cos (ωc t) sin (ωc (t + τ ))]
= ϕxx (τ ) cos (ωc t) cos (ωc (t + τ ))
+ϕyy (τ ) sin (ωc t) sin (ωc (t + τ ))
−ϕxy (τ ) sin (ωc t) cos (ωc (t + τ ))
−ϕyx (τ ) cos (ωc t) sin (ωc (t + τ )) (1.86)
Substituting in the necessary trigometric identities and using the assumption that n(t) is
wide-sense stationary (see Proakis (page 154 2nd ed., page 160 3rd ed.) results in
ϕnn (τ ) = ϕxx (τ ) cos (ωc τ ) − ϕyx sin (ωc τ ) (1.87)

25
In deriving this expression, stationarity of n(t) is satisfied only if
ϕxx (τ ) = ϕyy (τ )
ϕxy (τ ) = −ϕyx (τ ) (1.88)
An alternate representation of ϕnn (t) can be derived from the quadrature representation.
ϕnn (τ ) = ϕxx (τ ) cos (ωc τ ) − ϕyx (τ ) sin (ωc τ )
= <{(ϕxx (τ ) + jϕyx (τ )) (cos (ωc τ ) + j sin (ωc τ )}
= <{ϕzz (τ )ejωc τ } (1.89)
where
ϕzz (τ ) = ϕxx (τ ) + jϕyx (τ ) (1.90)
ϕzz (τ ) is the autocorrelation function of the equivalent lowpass process z(t) = x(t) + jy(t).
ϕzz (τ ) can be determined directly using
1 1
ϕzz (τ ) = E[z(t)z ∗ (t − τ )] = E[z ∗ (t)z(t + τ )] (1.91)
2 2
Note the 12 term, is required for complex processes. (See Proakis, eqn (1.2.12) 2nd ed. or
eqn (2.2.12) 3rd ed.). Expanding (1.91) gives
1
ϕzz (τ ) = E[z ∗ (t)z(t + τ )]
2
1
= E[(x(t) − jy(t)) (x(t + τ ) + jy(t + τ ))]
2
1
= E[x(t)x(t + τ ) + y(t)y(t + τ ) − jy(t)x(t + τ ) + jx(t)y(t + τ )]
2
1
= [ϕxx (τ ) + ϕyy (τ ) − jϕxy (τ ) + jϕyx (τ )] (1.92)
2
The stationarity conditions for n(t), which were previously derived are
ϕxx (τ ) = ϕyy (τ )
ϕxy (τ ) = −ϕyx (τ ) (1.93)
Substituting these into ϕzz (τ ) gives
1
ϕzz (τ ) = [2ϕxx (τ ) + j2ϕyx (τ )]
2
= ϕxx (τ ) + jϕyx (τ ). (1.94)
Note that
ϕnn (τ ) = <{ϕzz (τ )ejωc τ } (1.95)
indicates that the autocorrelation function of the bandpass stochastic process, ϕnn (τ ), is
uniquely determined from the autocorrelation function ϕzz (τ ) of the equivalent lowpass pro-
cess, z(t), and the carrier frequency turn ejωc t . The power density spectrum, Φnn (f ) can be
determined using the Fourier transform,
Φnn (f ) = F {ϕnn (τ )}
= F {<(ϕzz (τ )ejωc τ )}, (1.96)

26
using <(ξ) = 12 (ξ + ξ ∗ )
1
Φnn (f ) = F [ϕzz (τ )ejωc τ + ϕ∗zz (τ )e−jωc τ ]. (1.97)
2
Applying the property (see eqn 1.2.13 2nd ed. or eqn 2.2.13 3rd ed. of Proakis)

ϕzz (τ ) = ϕ∗zz (−τ ) (ϕzz (τ ) obeys hermitian symmetry) (1.98)

gives
1
Φnn (f ) = F [ϕzz (τ )ejωc τ + ϕzz (−τ )e−jωc τ ]
2
1
= [Φzz (f − fc ) + Φzz (−f − fc )] (1.99)
2
Note that Φzz (f ) is a real valued function since ϕzz (t) obeys hermitian symmetry. The
ϕyx (τ ) term in ϕzz (τ ) = ϕxx (τ ) + jϕyx (τ ) is a cross correlation function which satisfies the
property (true for any cross correlation),

ϕyx (τ ) = ϕxy (−τ ). (1.100)

Substituting the stationarity condition of n(t)

ϕxy (τ ) = −ϕyx (τ ), (1.101)

into (1.100) gives


ϕxy (τ ) = −ϕxy (−τ ) (1.102)
This demonstrates that ϕxy (τ ) is an odd function. and thus ϕxy (0) = 0 (x(t) and y(t) are
uncorrelated at τ = 0). Also, since ϕxy (τ ) is odd, Φxy (f ) will be purely imaginary.
If x(t) and y(t) are uncorrelated for all τ , ϕxy (τ ) = 0 ∀τ , then ϕzz (τ ) is real and

Φzz (f ) = Φzz (−f ) (an even function). (1.103)

It also follows that if Φzz (f ) is an even function then ϕxy (τ ) = 0 ∀ τ .


For the special case where n(t) is gaussian, the quadrature components are jointly gaus-
sian. At τ = 0 they are uncorrelated since ϕxy (0) = 0, and thus they are statistically
independent (at τ = 0). This results in their joint probability density function being
1 − (x2 +2y2 )
p(x, y) = e 2σ (1.104)
2πσ 2
where
σ 2 = ϕxx (0) = ϕyy (0) = ϕnn (0) (1.105)
Example 1.1
n(t) is a narrowband bandpass process with PDS Φnn (f ) shown in Figure 1.31. Note
that Φzz (f ) is a real function. Thus Φnn (f ) is also real.
1. What should fc be such that the quadrature components are uncorrelated?

27
Φnn (f )

−5 −4 −3 3 4 5 f

Figure 1.31: Φnn (f ) for Example 1.1

Φzz (f )

−1 1 f

Figure 1.32: Φzz (f ) for Example 1.1

2. What should fc be such that the quadrature components are correlated?

Solution 1.1

1. Choose fc = 4Hz, which results in the Φzz (f ) given in Figure 1.32. You can see
that Φzz (f ) = Φzz (−f ) and thus x(t) and y(t) are uncorrelated, ϕxy (τ ) = 0. Thus
Φxx (f ) = Φzz (f )

2. Choose fc = 3Hz, which gives the Φzz (f ) shown in Figure 1.33. You can see that

Φzz (f )

1 2 f

Figure 1.33: Φzz (f ) for Example 1.12)

Φzz (f ) 6= Φzz (−f ), thus x(t) and y(t) are correlated, so ϕxy (τ ) 6= 0. But Φzz (f ) is still

28
real so
Φzz (f ) = Φxx (f ) + jΦyx (f ) (1.106)
indicates that Φyx (f ) must be purely imaginary.

1.8 Signal Space Representations


Everyone is familiar with traditional vector spaces, the most common being the three-
dimensional vector space of the physical world. Vector spaces are very useful, since they
can provide a geometrical interpretation which may give a valuable insight into a problem.
This geometrical representation can be used in digital communication systems by abstractly
representing the signals as vectors in a vector (linear) space.
The most familiar example of a linear space is a Euclidean space. In a Euclidean space, a
vector is represented by its coordinates, an n-dimensional space requires n coordinates. The
vector X is represented as [x1 , x2 , · · · , xn ]T . In a linear space the two operations permitted
are addition of vectors and multiplication by a scalar. For a Euclidean space the following
properties hold: If X, Y and Z are vectors then

X + Y = Y + X commutative law
X + (Y + Z) = (X + Y ) + Z associative law

A linear space must have a zero vector 0, and every vector must have an additive inverse,
denoted −X, such that

0+X = X
X + (−X) = 0

Multiplying a vector X by a scalar, α , produces a new vector, αX, which must be in the
linear space. For vectors X and Y , and constants α and β,

α(βX) = (αβ)X associative law (multiplication)


1×X = X
0×X = 0
α(X + Y ) = αX + αY distributive law
(α + β)X = αX + βX distributive law
(1.107)

Another important definition for the Euclidean space is the inner product (dot product)
of two vectors given by
X
n
X ·Y = xi yi∗ (1.108)
i=1

The inner product can also be evaluated using

X ·Y = XT Y ∗ (1.109)
X ·Y = k X k k Y k cos (θ) (1.110)

29

You might also like