88

Introduction to Random Processes

Definition: Let (,,P) be a probability space. Let X be the mapping from the
sample space to a space of functions called sample functions. Then X is called a
random process (r.p.) if at each time t
i
the mapping X is a random variable (r.v.),
i.e., X(t
i
, ) belongs to for each fixed t
i
, t
i
R (the real line).

Example: A continuous-time r.p. with a finite sample space ={
1
,
2

n
} is
illustrated on the next slide.
Observations:
1. For each fixed
i
, X(t,
i
) is an ordinary time function
2. For each fixed t
i
, X(t
i
,
i
) is an ordinary random variable.

As a consequence of these observations, at each t
i
the random process can be
described by either the probability density function (pdf), if exists, or by the
cumulative distribution function (cdf).
89














90
Let X(t,) be a random process (r.p.), then its mean, mean-square, and variance at
time t
i
are defined by





For the sake of conciseness, let X(t
i
) X(t
i
, ).
A measure of the statistical likeness (dependence) of the process from one instant
to another is the autocorrelation function. Mathematically, for arbitrary X(t)

or


{ } ) , ( ) (
i i X
t X E t =
{ }
2
2
) , ( ) (
i i
t X E t X =
| || | { }
2
2
2
) ( ) (
) , ( ) , ( ) , ( ) , ( ) (
i X i
i X i i X i i X
t t X
t t X t t X E t

=
=

{ }, ) ( ) ( ) , (
2
*
1 2 1
t X t X E t t R
X

( )
1
2
1 2 1 1 2 2 1 2
1 2
1 2 1 1 2 2
( , ; , ) , ( ) is a continuous-valued r. p.
( , )
( ) , ( ) , ( ) is a discrete-valued r. p.
x
R
X
R
x x f x t x t dx dx X t
R t t
x x P X t x X t x X t

=

= =

91
Let the ac component of X(t) be defined by

then, the covariance of X(t) is defined by




Definition: A random process X(t) is stationary if it has the same n
th
order
distribution function as X(t +T), T R, that is, for


The first-order probability density function of X(t), if it exists, is defined by


( ) ( ) ( ),
c i i X i
X t X t t
{ }
{ }
{ }
1 2 1 2
1 2 1 2 1 2 1 2
1 2 1 2
1 2 1 2
( , ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( , ) ( ) ( )
X c c
X X X X
X X
X X X
C t t E X t X t
E X t X t X t t t X t t t
E X t X t t t
R t t t t

= +
=
=
| |
1
( ) ( ) ,
T
n
X t X t = . X
1 1 1 1
( , ; ; , ) ( , ; ; , ), .
n n n n
F x t x t F x t T x t T t R = + + . .
X X
1 1 1 1 1 1 1
( , ) ( , ) ( ,0) ( ), for .
X X X X
f x t f x t T f x f x t T = + = = =
92
Clearly, the first-order pdf does not depend on time. Hence,

which are constants.
Example: The first-order probability density function of a wide-sense stationary
Gaussian r.p. is given by


The second-order density function of a stationary r.p. X(t) is given by


where
Hence,

{ }
1 1
( ) ( ) , and
X X
t E X t = =
( )
,
2
exp
2
1
) , (
2
2
1
1 1
(

=
X
X
X
X
x
t x f

2 2
1 1
2
} ) ) ( {( ) (
X X X
t X E t = =
1 1 2 2 1 1 2 2
1 2 2 1 1
1 2 2 1 1
( , ; , ) ( , ; , )
( ,0; , ),
( , ; ), ,
f x t x t f x t T x t T
f x x t t t T
f x x t t t T
= + +
= =
= =
X X
X
X
| |
1 2
( ) ( ) .
T
X t X t = X
{ } ) ( ) ( ) ( ) , (
1 2 2 1 2 1
t t R t X t X E t t R
X X
= =

93
A random process (r.p.) X(t) is wide sense stationary (WSS) if
1. Its mean is constant (does not depend on time)
2. Its autocorrelation depends on the time difference only.
Mathematically,

and for all t
1
and t
2
,


Remark Strict-sense stationary (SSS) implies WSS. However, the converse is not
always true.
Let t
2
=t
1
+, then if X(t) is WSS, we get




{ }
X
t X E = ) (
{ } ), ( ) ( ) (
1 2 2 1
t t R t X t X E
X
=

{ } { }
1 2 1 1
1 1
( ) ( ) ( ) ( )
( )
( ).
X
X
E X t X t E X t X t
R t t
R

= +
= +
=
94
Observation: If t
1
=t
2
, then

is equal to the mean square value (total average power) of X(t).
Furthermore, assuming X(t) is a real-valued random process, we get



This implies that

and
These two conditions imply

Equivalently

{ }
2
1
( ) (0) 0.
X
E X t R =
| | { } { }
( ) 0 ) ( ) 0 ( 2
) 0 ( ) ( 2 ) 0 (
) ( ) ( ) ( 2 ) ( ) ( ) (
2 2
2
=
+ =
+ + + = +

X X
X X X
R R
R R R
t X t X t X t X E t X t X E
) 0 ( ) (
X X
R R
). 0 ( ) (
X X
R R
), 0 ( ) ( ) 0 (
X X X
R R R
). 0 ( ) (
X X
R R
95
Let t
1
=t
2
+, then

Consequently,

Observation: The autocorrelation function is an even function of its argument .
Example: Consider the random binary signal shown below







where T
b
is the bit interval and t
d
is the value that the random delay T
d
takes on.

). ( ) ( ) , (
1 2 2 1
= =
X X X
R t t R t t R
( ) ( ), an even function of its argument.
X X
R R =
96
Assumptions:
1. During any bit interval where n is an arbitrary integer,

2. The time delay is uniformly distributed, namely,

Let us first compute the mean
x
(t).



To compute the autocorrelation function we must consider several situations. Let
t
1
>t
2
, and t
1
and t
2
lie in two different bit intervals, then X(t
1
) is statistically
independent of X(t
2
). Mathematically,

,
b
nT
{ } { }
bit bit 1/ 2 P A P A = = = =

=
otherwise , 0
0 , / 1
) (
b d b
d T
T t T
t f
d
{ } { } { }
. 0
2
1
2
1
bit ) ( bit ) ( ) (
=
=
= + = = =
A A
A P A A P A t X E t
X

{ } { } { } 0 0 0 ) ( ) ( ) ( ) ( ) , (
2 1 2 1 2 1
= = = = t X E t X E t X t X E t t R
X
97
Likewise, if t
2
>t
1
and t
1
and t
2
lie in two different bit intervals, the autocorrelation
of X(t) is equal to zero, i.e.,

In conclusion, if t
1
and t
2
lie in two different bit intervals,

On the other hand, if t
1
and t
2
occur in the same bit interval and then






This condition will hold when
1. For arbitrary n,
. 0 ) , (
2 1
= t t R
X
, 0 ) , (
2 1
= t t R
X
,
1 2 b
T t t <
{ } { } { }
( ) { } { } { }
{ }
{ }
2
1 2 1 2
2
1 2
2
1 2
2
1 2
2
1 2
( , ) and lieinthesamebitinterval ( )
and lieinthesamebitinterval ( )
1
and lieinthesamebitinterval
2
1
( ) and lieinthesamebitinterval
2
and lieinthesamebiti
X
R t t A P t t X t A
A P t t X t A
A P t t
A P t t
A P t t
= =
+ =
=
+
=
{ }
nterval
d b d b
t nT t t t T n + < +
2 1
) 1 (
98
Since n is arbitrary, let n =1, then

or



Thus,






2. t
1
and t
2
will also lie in the same interval if

or
d b d
t T t t t + <
2 1
.
1 2
t t T t
d b
<
{ } { }
| |
1
2
1 2 2 1
1 2
2 1
2 1.
and lieinthesamebitinterval
1
1
1 ,
b
b d
t
d
b t T
b
b
b
P t t P t T T t
dt
T
t T t
T
t t
t t
T

= <
=
= +

d b d
t T t t t + <
1 2
.
2 1
t t T t
d b
<
99
In this case,






Putting these two results together, we get



Example: Let a random process X(t) be described by

where
c
is a known constant and A is a zero-mean random variable.
{ } { }
| |
2
1
1 2 1 2
2 1
2 1
1 2.
and lieinthesamebitinterval
1
1
1 ,
b
b d
t
d
b t T
b
b
b
P t t P t T T t
dt
T
t t T
T
t t
t t
T

= <
=
= +

= +

<
|
|
.
|

\
|

=
otherwise , 0
interval bit same the in lie and , , 1
) , (
2 1 1 2
1 2 2
2 1
t t T t t
T
t t
A
t t R
b
b
X
t A t X
c
cos ) ( =
100
Then,



and


Define the mean square value of A by

then the autocorrelation of X(t) is found to be


Clearly, this r.p. is neither SSS nor WSS.
{ }
{ }
{ }

0
cos
cos
) ( ) (
0
=
=
=
=
t A E
t A E
t X E t
c
c
X

{ }
{ }
{ }
1 2 1 2
1 2
2
1 2
( , ) ( ) ( )
cos cos
cos cos
X
c c
c c
R t t E X t X t
E A t A t
E A t t


=
=
=
{ }
2 2
A E A
| | . , , ) ( cos ) ( cos
2
cos cos ) , (
2 1 1 2 2 1
2
2 1
2
2 1
t t t t t t
A
t t A t t R
c c c c X
+ + = =
101
Example: Let X(t) be a r.p. described by

with A a known constant amplitude,
c
a known constant frequency and be a
uniformly distributed r.v., i.e.,

Let A =1, then




But,




| | , ~ U
{ }
{ }
{ }
{ } { }
( ) ( )
cos( )
cos cos sin sin
cos cos sin sin .
X
c
c c
c c
t E X t
E t
E t t
t E t E

=
= +
=
=
{ } . 0 sin
2
1
2
1
cos cos = = =

d E
{ } . 0 cos
2
1
2
1
sin sin = = =

d E
), cos( ) ( + = t A t X
c

102
Hence,

Let us now compute its autocorrelation.






Again,


and

. , 0 ) ( t t
X X
= =
{ }
{ }
( ) ( )
( ) ( ) { ( )}
( ) { } ( ) { } ( )
1 2 1 2
1 2
1 2 1 2
1 2 1 2 1 2
1 2 1 2 1 2
( , ) ( ) ( )
cos( ) cos( )
1
cos ( ) 2 cos ( )
2
1
cos ( ) cos2 sin ( ) sin2 cos ( )
2
1
cos ( ) cos2 sin ( ) sin2 cos ( )
2
X
c c
c c
c c c
c c c
R t t E X t X t
E t t
E t t t t
E t t t t t t
t t E t t E t t




=
= + +

= + + + (
`

)
= + + +
= + + + (

{ }
1 1
cos2 cos2 sin2 0.
2 4
E d

= = =

{ }
1 1
sin2 sin2 cos2 0.
2 4
E d

= = =

103
Hence, the autocorrelation of X(t) is


Consequently, X(t) is WSS because the autocorrelation is a function of the time
difference.

Cross-Correlation Function
The cross-correlation function of two random processes X(t) and Y(t) is defined
by



If X(t) and Y(t) are jointly WSS, then


and

( ) R t t t t t t R
c X
=
2 1 1 2 2 1
, , ) ( cos
2
1
) , (
{ }
{ }
1 2 1 2
1 2 1 2
( , ) ( ) ( )
( , ) ( ) ( )
XY
YX
R t t E X t Y t
R t t E Y t X t

=
=
) ( ) , (
1 2 2 1
t t R t t R
XY XY
=
). ( ) (
1 2 2 1
t t R t t R
YX XY
=

104
Definition: A WSS r.p. X(t) is ergodic in the mean if the time average of its sample
functions converges to the ensemble (statistical) average E{X(t)} =
X
in the
mean-square sense, i.e.,


in the mean square sense, as T .

Observation: Note that M
X
(T ) is a random variable, thus we can compute its
statistical moments, i.e., its mean, variance, etc..
Let us begin with the calculation of its mean.



This means that M
X
(T ) is an unbiased estimate of
X
.

Suppose now that X(t) is a real-valued random process, then


( )
X
T
T
X
dt t X
T
T M

) (
2
1
( ) { } { }
1 1 1 1
( ) ( ) ( )
2 2 2 2
T T T T
X X X
T T T T
E M T E X t dt E X t dt E X t dt dt
T T T T



= = = = =
` `
) )

105






where C
X
(t
2
t
1
) is the auto covariance (ac component) of the process X(t).
If Var(M
X
(T)) is small, then the time average M
X
(T) is a good estimate of the
mean
X
. Furthermore, if this variance converges to zero as T , i.e.,

then M
X
(T) is a consistent estimate of
X
.

( ) { } ( ) ( ) { } ( ) { }
| | , ) (
4
1
) (
4
1
) (
4
1
) (
2
1
) (
2
1
2 1 1 2
2
2 1
2
1 2
2
2
2 1 1 2
2
2
2 2 1 1
2 2
2
dt dt t t C
T
dt dt t t R
T
dt dt t t R
T
dt t X
T
dt t X
T
E
T M E T M E T M Var
T
T
T
T
X
T
T
T
T
X X
T
T
X
T
T
X X
T
T
T
T
X X X X X




= =
=
)
`

=
= =

( ) { }
lim 0,
X
T
Var M T

=
106
Definition
A WSS r.p. X(t) is ergodic in correlation at the time shift if and only if


in the mean-square sense.
If the condition is true for all then X(t) is ergodic in correlation.

Remark: Although ergodicity holds in most practical cases, care must be
exercised to make sure the required conditions for ergodicity are met before
proceeding to use time averages instead of ensemble averages.
Let us consider what happens to the output of a continuous-time bounded-input
bounded-output (BIBO) stable linear and time invariant (LTI) system described by
the impulse response h(t), when its input is a WSS r.p. X(t). We know that


) ( ) ( ) (
2
1
lim
X
T
T
T
R dt t X t X
T
= +

. ) ( ) ( ) ( ) ( ) (

= = d h t X d X t h t Y
107
Thus, the mean value of the output is given by




where H() =F{h(t)}.
Observation: H(0) exists if and only if the system is BIBO stable.
Now, the autocorrelation of the output can be computed as follows:





{ } { }
), 0 ( ) ( ) (
) ( ) ( ) ( ) ( ) ( ) (
H d h d h
d h t X E d h t X E t Y E t
X X X
Y


= = =
=
)
`

= =

{ } { }
1 2 1 2 1 2
2 1 2 1
( , ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ).
Y
X Y
R t t E Y t Y t h h E X t X t d d
h h R t t d d R t t

= =
= + =


108
Likewise, the cross-correlation between X(t) and Y(t) is given by






Example: Consider the change detector T [], i.e, Y(t) =T [X(t)], such that
Y(t) =X(t) X(t-1), where X(t) is a real-valued WSS r.p. with zero mean.




Change detector.
{ }
{ }
). ( ) ( ) ( ) ( ) (
) ( ) ( ) ( ) ( ) ( ) (
) ( ) ( ) ( ) ( ) ( ) , (
1 2 1 2 1 2 1 2
2 1 2 1
2 1 2 1 2 1
t t R t t h t t R d t t R h
d t X t X E h d t X t X h E
d t X h t X E t Y t X E t t R
X XY X
XY
= = =
=
)
`

=
)
`

= =

109
Let us first compute the mean value of the output of the detector,


To detect a change, let us compute the autocorrelation of the output, i.e.,





For the detector to be effective, the variance of Y(t) should be large (the power of
the difference X(t) X(t-1) should be large).
Let the autocorrelation of the input be described by

{ }
{ }
{ } { }
( ) ( ) ( 1)
( )
( ) ( 1) 0.
Y
X X
t E E X t X t
Y t
E X t E X t

= =
= = =
{ } | || | { }
{ }
{ } { } { } { }
). ( ) 1 ( ) 1 ( ) ( 2
) ( ) 1 ( ) 1 ( ) (
) 1 ( ) 1 ( ) ( ) 1 ( ) 1 ( ) ( ) ( ) (
) 1 ( ) 1 ( ) ( ) 1 ( ) 1 ( ) ( ) ( ) (
) 1 ( ) ( ) 1 ( ) ( ) ( ) ( ) , (
2 1 1 2 1 2 1 2
1 2 1 2 1 2 1 2
2 1 2 1 2 1 2 1
2 1 2 1 2 1 2 1
2 2 1 1 2 1 2 1
t t R t t R t t R t t R
t t R t t R t t R t t R
t X t X E t X t X E t X t X E t X t X E
t X t X t X t X t X t X t X t X E
t X t X t X t X E t Y t Y E t t R
Y X X X
X X X X
Y
= + =
+ + =
+ =
+ =
= =
, 0 , ) (
1 2
2
2 1
> =


t t
X X
e t t R
110
Then, the autocorrelation of the output is


Since the mean of the output is zero, the autocovariance (ac power) of the output is
the same as its autocorrelation (total power), i.e.,



Power spectral density (PSD)
Definition: Let X(t) be a WSS r.p. with autocorrelation function R
X
( ). Then the
power spectral density S
X
() is the Fourier transform of R
X
( ), i.e.,


Moreover, under certain general conditions,


. 2 ) , (
1
2
1
2 2
2 1
1 2 1 2 1 2
+
=
t t
X
t t
X
t t
X Y
e e e t t R


( ). 1 2 2 ) 0 (
2 2 2 2 2


= = = e e e R Y
X X X X Y
{ }
( ) ( ) ( ) .
j
X X X
S R R e d

= =

{ }
1
1
( ) ( ) ( ) .
2
j
X X X
R S S e d

= =

111
Properties of the PSD:
1. S
X
() is real-valued since R
X
( ) is such that R
X
( ) =R
X
*
(-) .
2. If X(t) is real-valued, then S
X
() is an even function of since R
X
( ) is real and
even. Otherwise, S
X
() is not an even function of .
3. S
X
() 0, .
4.

Example: Let X(t) have autocorrelation function


What is the power spectral density of X(t)?



). (
) (
) 1 (
2
2
2

X
n
F
n
X
n
n
S
d
R d

. 0 , ) (
2
> =



e R
X X
{ }
2 2
( ) ( )
j j
X X X X
S R e e d e d





= = =

112
or







Practical approach to calculate S
X
()
Define the truncated (windowed) version of the WSS r.p. X(t) by



Let

2 2
2
2 2
2
2
0
0
2
0
0
2 2
2
) 1 0 (
1
) 0 1 (
1
1 1
) (

+
=
(

+
+ +
=
(

=
(
(

=
(

+ = =

X
X
X
j j
X
j j
X
j
X X
j j
j j
e
j
e
j
d e d e d e S
| |

elsewhere , 0
, ), (
) (
T T t t X
t X
T
{ }
( ) ( ) ( ) .
T
j t
T T
T
X t X t e dt

=

113
Then,

Since is a random variable, we can compute its ensemble average, i.e,




Let = t s and s = s, then


where



. ) ( ) ( ) (
) (
2



=
T
T
T
T
s t j
T
dtds e s X t X


2
) (
T
{ } { }
. ) (
) ( ) ( ) (
) (
) (
2






=
=
T
T
T
T
s t j
X
T
T
T
T
s t j
T
dtds e s t R
dtds e s X t X E E

=
) , (
) (
) ( ) , ( ) (
s
j
X
T
T
T
T
s t j
X
dsd e R s J dtds e s t R

1
1 0
1 1
=

=
s
s
t
s
s t
J

114
Graphically, the map is described by









Space transformation


115
Equivalently,






Therefore,


and


. ) (
2
1 2
) ( ) 2 ( ) ( ) 2 (
) ( ) ( ) ( ) , (
2
2
0
2
2
0
0
2
2
0 ) , (

|
|
.
|

\
|
=
+ + =
+ =
T
T
j
X
T
j
X
T
j
X
T
T
T
j
X
TT
T
j
X
s
j
X
d e R
T
T
d e R T d e R T
dsd e R dsd e R dsd e R s J

{ }

|
|
.
|

\
|
=
T
T
j
X T
d e R
T
E
T
2
2
2
) (
2
1 ) (
2
1

{ }
2 1
lim ( ) ( ) ( )
2
j
T X X
T
E R e d S
T

= =

116
where the Fourier transform of the truncated signal is

Therefore,

Hence S
X
() is real, nonnegative and is related to the average power at frequency
.

Example: Let X(t) have the autocorrelation function R
X
( ) =
2
(t). Then



Consider again the output of a BIBO stable LTI system excited by a WSS r.p. X(t)
Then

=
T
T
t j
T
dt e t X . ) ( ) (


{ }
2
) (
2
1
lim ) (
T
T
X
E
T
S

=
. , ) ( ) ( ) (
2 2


= = =

d e d e R S
j j
X X

+ =
+ = =
. ) ( ) ( ) (
) ( ) ( ) ( ) ( ) (

d d d e R h h
d e d d R h h d e R S
j
X
j
X
j
Y Y
117
Let = + - . With respect to the innermost integral, d = d and = + - .
Thus,




or




since


It is clear from the above result that

Average total power of Y(t)


|
|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
=
=

+




d e R d e h d e h
d d d e R h h S
j
X
j j
j
X Y
) ( ) ( ) (
) ( ) ( ) ( ) (
) (
), ( ) (
) ( ) ( ) ( ) (
2


X
X Y
S H
S H H S
=
=

*
( ) ( ) ( ) ( ), for real ( ).
j j j
h e d h e d h e d H h






| | | |
= = =
| |
\ . \ .

. , 0 ) (
Y
S
{ } ). 0 ( ) (
2
Y
R t Y E =
118
Now,

This implies that the total average power of Y(t) (mean square value of Y(t)) is



Example: Consider an LTI system with impulse response h(t) =2e
-5t
u(t), excited by
a r.p. X(t) with mean
X
and autocorrelation R
X
() =10(). What are the analytical
expressions of R
Y
() and S
Y
()?
We know that
Hence,

and


{ }
1
1
( ) ( ) ( )
2
j
Y Y Y
R S S e d

= =

{ }
2 2 2 1 1
( ) (0) ( ) ( ) ( ) ( ) ( ) .
2 2
Y Y X X
E Y t R S d H S d H f S f df



= = = =

{ }
2
5
2
( ) ( ) ( ) with ( ) 2 ( )
5
t
Y X
S H S H e u t
j

= = =
+

+
=
+
=
+
= ,
25
40
) 5 )( 5 (
40
10
5
2
) (
2
2
j j j
S
Y
1 1
2 2 2
1 1 40 40 2 5
( ) ( ) 4
2 2 25 25 25
j j
Y Y
R S e d e d

= = = =
` `
+ + +
) )

119
or

Gaussian processes
Let Z be a Gaussian random variable. Then its pdf has the form


Consider now the situation where
1) {X
1
, X
2
, ..., X
N
} is a sequence of statistically independent r.v.s.
2) The X
i
s have the same pdf with mean
X
and variance
If the X
i
s meet these two criteria, then they are independent and identically
distributed (iid).
Furthermore, if




=

, 4 ) (
5
e R
Y
. ,
2
) (
exp
2
1
) (
2
2
z
z
z f
Z
Z
Z
Z

(

.
2
X

=
|
|
.
|

\
|
=
N
i
X
X i
N
X
N
Y
1
,
1

120
Then, as N , Y
N
becomes a zero-mean Gaussian r.v. with unit variance, i.e.


This is the so-called central limit theorem.
It is important to note that if the X
i
s are iid, then the pdf of Y
N
is such that if



An n-fold convolution, where f
Z
(z) is the common pdf of the Z
i
s.

Properties of a Gaussian Processes
1) Let X(t) be a real Gaussian r.p. that is input to a BIBO stable LTI system with
impulse response h(t). Then Y(t) =h(t) X(t) is also a Gaussian r.p.
2) Let X(t) be a Gaussian r.p. and t
1
, t
2
be two known sampling instants of the
process X(t). Then the random variables X(t
1
) and X(t
2
) are jointly Gaussian
y e y f
y
Y
N


,
2
1
) (
2 /
2

( )
1
1
, / , 1, ,
N
N i i i X X
i
Y Z Z X i N
N

=
= = =

.
1 2
1times
( ) ( ) ( ) ( ) ( ) ( ) ( )
N N
Y Z Z Z Z Z Z
N
f z f z f z f z f z f z f z

= = . .
_
121
with pdf

where







3) Let X(t) be a Gaussian r.p. and t
1
, t
2
and be two known sampling instants. If the
r.v.s X(t
1
) and X(t
2
) are uncorrelated, i.e,

then the r.v.s X(t
1
) and X(t
2
) are independent.
( )
1 1
1
( ), ( )
1/2
1 1
exp ( ) ( ) ,
2 2
T
X t X t
f

(
=
(

x x x
(

=
(

=
2
1
2
1
) (
) (
X
X
t X
t X
X
( ) { }
( ) { }
{ }
1
1 1
2 2
2
( )
( )
E X t
X t
E E
X t
E X t

(

( (

= = = = (
`
( (

(
)

X
| |
)
`

=
2 2 1 1
2 2
1 1
) ( ) (
) (
) (

t X t X
t X
t X
E
{ } det =
( )( ) { } , 0 ) ( ) (
2 2 1 1
= t X t X E
122
Uncorrelatedness means that


and

Hence,


where

Thermal noise
Let V
TN
(t) be the thermal noise voltage that appears across the terminals of a
resistor due to the random motion of electrons. Then the mean square voltage
observed over a bandwidth of f can be approximated by
{ } , det
0
0
2
2
2
1
2
2
2
1

=
(

=
(

=
2
2
2
2 2
2
1
2
1 1
2 1
2
) (
2
) (
exp
2
1
) (

x x
f x
X
), ( ) (
2
) (
exp
2
1
2
) (
exp
2
1
) (
2 ) ( 1 ) (
2
2
2
2 2
2
2
1
2
1 1
1
2 1
x f x f
x x
f
t X t X
=
(

X
X
.
) (
) (
2
1
(

=
t X
t X
X
123


where k is Boltzmanns constant (1.38x10
-23
J oules/Kelvin), T is the absolute
temperature in degrees Kelvin, and R is the resistance in Ohms.
It should be clear that


We can think of V
TN
(t) as a r.p. that has a constant psd over the range of
frequencies


or


{ }
2 2
( ) 4 J oules / sec 4 volt ,
TN
E V t kTR f kTR f =
Watts or J oules/sec 4 ) (
1
2
f kT t V
R
E
TN

)
`

,
2
,
2
(

f f

elsewhere , 0
2
,
2
, /Hz volts 4
) (
2
f f
f kTR
f S
TN
V
124
Graphically,




PSD of thermal noise.
This is true because


White Noise
It is an extension of the concept of thermal noise, that is, if W(t) is a white noise
process, then


{ }
2
2
2
( ) (0) 4 4
TN
f
TN V
f
E V t R kTRdf kTR f

= = =

f
kT
f S
e
W
= ,
2
) (
125
where k is Boltzmanns constant and T
e
is the equivalent noise temperature of the
receiver.
Let N
0
=kT
e
, then and




White noise autocorrelation and its psd.

Observations
1. , are uncorrelated
2. If W(t) is Gaussian, then W(t
1
) and W(t
2
), t
1
t
2
, are also statistically independent.
3. The white noise model is good approximation whenever the bandwidth of the
noise is much larger that the bandwidth with an LTI system excited by such
a noise.
f
N
f S
W
= ,
2
) (
0
). (
2
) (
0

N
R
W
=
2 1 2 1
0
), ( and ) ( ) (
2
) ( t t t W t W
N
R
W
=
126
Example: Let an LTI system be described by



What are the autocorrelation and the psd of the output Y(t) when the input to the
system X(t) is white noise?

We already know that
Hence,



Now,

=
elsewhere , 0
, 1
) (
W
H

). ( ) ( ) (
2

x y
S H S =
0
2
0
,
( ) ( )
2
2
0, elsewhere
Y
N
W N
S H

= =

| | ( )
( )
( )

W
W N
W
W W N
W
N
e e
j
N
e
j
N
d e
N
R
jW jW
W
W
j
W
W
j
Y
sinc
2
sin
2
sin
2 4 4 2 2
1
) (
0 0
0 0 0 0
=
|
.
|

\
|
=
= = = =

=
=

127
Narrowband noise
In order to optimize communication system performance in the presence of noise,
filtering operations around the carrier frequency are performed at the front end of
the receiver. The result of these operations is that the receiver input wideband
noise is transformed into narrowband noise centered around the carrier frequency.

An example of a narrow-band noise signal is shown below







PSD of narrowband noise.

In-phase and quadrature representation: Let N
I
(t) and N
Q
(t) be the in-phase and
quadrature components, respectively, of the narrowband noise N(t).
128
Then the bandpass narrowband noise can be written as

Properties of N
I
(t) and N
Q
(t):
1)
2) If N(t) is Gaussian, then N
I
(t) and N
Q
(t) are jointly Gaussian.
3) If N(t) is stationary, then N
I
(t) and N
Q
(t) are jointly stationary.

4)
5)

6)

Example: Consider a receiver whose prefilter is described by


]. , [ ~ ), 2 sin( ) ( ) 2 cos( ) ( ) ( + + = U t f t N t f t N t N
c Q c I
{ } { } 0 ) ( ) ( = = t N E t N E
Q I

+ +
= =
elsewhere , 0
), ( ) (
) ( ) (
B f f f S f f S
f S f S
c N c N
N N
Q I
{ } { } { } ) ( ) ( ) ( t N Var t N Var t N Var
Q I
= =
| |

+
= =
elsewhere , 0
, ) ( ) (
) ( ) (
B f f f S f f S j
f S f S
c N c N
N N N N
I Q Q I

+
=
elsewhere , 0
, 1
) (
B f f B f
f H
c c
129
Let the received signal be described by

where s
T
(t) is the transmitted signal and w(t) is white Gaussian noise.
Let us consider the contribution of the white Gaussian noise to the output of the
filter only. In this case, if N(t) is the output due to the noise component, then


Moreover,



) ( ) ( ) ( t w t s t s
T R
+ =

+
=
elsewhere , 0
,
2
) (
0
c c
N
f B f B f
N
f S
{ }
1 2 2 2
0 0
0 c
( ) ( ) ( )
2 2
2 sinc(2B )cos(2 f ).
c c
c c
f B f B
j f j f j f
N N N
f B f B
N N
R S f S f e df e df e df
N B

+ +

= = = +
=

130
Finally,








PSD of in-phase and quadrature noise components.

< <
= =
elsewhere , 0
,
) ( ) (
0
B f B N
f S f S
Q I
N N
131
Noise performance of CW modulation

Let us now investigate the effects that channel imperfections have on the
performance of a CW modulated communication system.
We will first focus on the effect of noise. Consider the following receiver model





Generic receiver block diagram
where s
T
(t) is the transmitted signal
w(t) is white noise with psd N
0
/2, f
s
R
(t) =s
T
(t) +w(t) is the received signal
x(t) is the filtered signal available for demodulation
x
dem
(t) is the estimate of m(t)
132
At the output of the bandpass filter, we get

where and
If we assume that the bandpass filter is ideal with center frequency f
c
then the
noise is bandpass, i.e.,




Noise component psd at output of BPF

where the bandpass noise can be mathematically described by

n
I
(t) and n
Q
(t) are the in-phase and quadrature noise components.
) ( ) ( ) ( t n t s t x + =
) ( ) ( ) ( t s t h t s
T BPF
= ) ( ) ( ) ( t w t h t n
BPF
=
( ) ( )cos(2 ) ( )sin(2 )
I c Q c
n t n t f t n t f t = + +
133
Definition: The input signal-to-noise ratio (SNR)
i
is the ratio of the average power
of the filtered modulated signal s(t) to the average power of the filtered noise n(t).

Let P
s
Average power of s(t)
P
n
Average power of n(t)
Then,


At the modulator output, assuming the noise component is additive,


where s
dem
(t) contains information about the message signal m(t) and n
dem
(t) is the
noise contribution at the output of the demodulator.

Definition: The output signal-to-noise ratio (SNR)
o
is the ratio of the average power
of s
dem
(t) to the average power of the noise component n
dem
(t).

Remark: Both s
dem
(t) and n
dem
(t) depend on the type of modulation scheme used.
. ) (
0 T
s
n
s
i
B N
P
P
P
SNR = =
), ( ) ( ) ( t n t s t x
dem dem dem
+ =
134
To establish a fair performance comparison between different modulation schemes
we shall assume the following:
1. s
T
(t) has the same average power, regardless of the modulation scheme
2. w(t) has the same power in the bandwidth W
m

Definition: The channel signal-to-noise ratio (SNR)
c
is the ratio of the average
power of the modulated signal to the average power of channel noise in the
message bandwidth W
m
Let the communication system performance figure of merit be

Consider first the performance of DSB using a coherent detector at the receiver , i.e,



. ) /( ) (
c o
SNR SNR
135
From the block diagram

where and w(t) is AWGN with psd
Let m(t) have zero mean and Then










), ( ) ( ) ( t w t s t s
DSB R
+ =
) ( ) cos( ) ( t m t A t s
c c DSB
+ = , 2 / ) (
0
N f S
w
=
. f
| |. 2 , 0 ~ U
{ } ( ) | | ( ) | | { }
( ) ( ) {
( )
( ) }
{ } ( ) ( ) { }
{ } ( ) { } { }
{ } { } ( ) { } { } ) ( ) ( cos ) ( ) (
) ( ) ( cos ) (
cos ) ( cos ) ( ) (
) ( ) ( cos ) ( ) (
) ( ) ( cos ) (
cos ) ( cos ) ( ) (
) ( cos ) ( ) ( ) ( cos ) ( ) ( ) (
0 0
0 0
2
2
t w t w E t E t m E t w E A
t w E t E t m E A
t t E t m t m E A
t w t w t t m t w A
t w t t m A
t t t m t m A E
t w t t m A t w t t m A E t s t s E
c c
c c
c c c
c c
c c
c c c
c c c c R R







+ + + + +
+ + + +
+ + + + =
+ + + + +
+ + + +
+ + + + =
+ + + + + + + = +
_ _
_ _
136
If m(t) is WSS, we get







Let the ideal bandpass filter have gain c in the frequency range

{ }
{ }
). ( ) (
2
) ( cos ) (
2
1
) ( ) ( cos ) (
2
1
) ( ) ( cos
2
1
) 2 2 ( cos
2
1
) (
), ( ) ( cos
2
1
) 2 2 ( cos
2
1
) ( ) ( ) (
0 2
2
0
2
2




R
s c m c
w c m c
w c c c m c
w c c c m c R R
R
N
R A
R R A
R t E R A
R t E R A t s t s E
= + =
+ =
+

+ + + =
+
)
`

+ + + = +
_
m c m c
W f f W f +
137
Then


Now,





Therefore,

+
= =
elsewhere , 0
), (
) ( ) ( ) (
2
2
m c m c s
s BPF X
W f f W f f S c
f S f H f S
R
R
{ }
{ } { }
| |
2
0
2
0
2 2
0
1
( ) ( ) ( )cos(2 ) ( )
2 2
1 1
( ) ( ) ( )
2 2 2
1 1
( ) ( ) .
4 4 2
R R
s s c m c
c m c c
c m c c m c
N
S f R A R f
N
A S f f f f f
N
A S f f A S f f


= = +
(
= + + +
(

= + + +

+ + + +
=
elsewhere , 0
,
2
) (
4
) (
4
) (
0
2 2 2 2 2
m c m c c m
c
c m
c
X
W f f W f
N c
f f S
A c
f f S
A c
f S
138
So, the psd of the message signal component s(t) is

,

whereas the psd of the noise component n(t) is

.

Let P
m
be the average power of m(t), i.e,


then the message signal component s(t) has average power



| |

+ + +
=
elsewhere , 0
, ) ( ) (
4
) (
2 2
m c m c c m c m
c
s
W f f W f f f S f f S
A c
f S

+
=
elsewhere , 0
,
2
) (
0
2
m c m c
n
W f f W f
N c
f S
, ) ( df f S P
m
m
W
W
m m

=
(
(

+ + =

+

+

m c
m c
m c
m c
W f
W f
c m
W f
W f
c m
c
s
df f f S df f f S
A c
P ) ( ) (
4
2 2
139
or

The average power of n(t) is



or

At the output of the mixer,

But, and
Hence,


.
2
2
4
) ( ) (
4
2 2 2 2 2 2
m c
m
c
W
W
m
W
W
m
c
s
P A c
P
A c
df f S df f S
A c
P
m
m
m
m
= =
(
(

+ =


. 2
2
) (
0
2 0
2
m
W f
W f
W f
W f
n n
W N c df df
N c
df f S P
m c
m c
m c
m c
=
(
(

+ = =

+

m
m c
m
m c
n
s
DSB i
W N
P A
W N c
P A c
P
P
SNR
0
2
0
2
2 2
,
4 2
2 /
) ( = = =
| | ). cos( ) ( ) ( ) cos( ) ( ) ( + + = + = t t n t s t t x t v
c c
) cos( ) ( ) ( + = t t m cA t s
c c
| |. ) sin( ) ( ) cos( ) ( ) ( + + = t t n t t n c t n
c Q c I
) cos( ) sin( ) ( ) ( cos ) ( ) ( cos ) ( ) (
2 2
+ + + + + = t t t cn t t cn t t m cA t v
c c Q c I c c
140
or


At the output of the lowpass filter, assuming an ideal response,


Since m(t) is uncorrelated (and therefore orthogonal) to n
I
(t) then


and

So, the average power of the component that contains the signal is



( ) | | ( ) | | ( ). ) ( 2 sin
2
) (
) ( 2 cos 1
2
) (
) ( 2 cos 1 ) (
2
) ( + + + + + + = t
t cn
t
t cn
t t m
cA
t v
c
Q
c
I
c
c
2
) (
) (
2
) (
t cn
t m
cA
t x
I c
dem
+ =
) (
4
) (
4
) (
2 2 2

I dem
n m
c
x
R
c
R
A c
R + =
) (
4
) (
4
) (
2 2 2

I dem
n m
c
x
S
c
f S
A c
f S + =
m
c
W
W
m
c
m
c
os
P
A c
df f S
A c
df f S
A c
P
m
m
4
) (
4
) (
4
2 2
2 2 2 2
=
= =

141
and the average power of the component that contains the noise is


Therefore, the output SNR is


Consequently the figure of merit (using the SNR at the output of the BPF) is


A ratio of 2 would mean that the demodulator gain is equal to 2, however, the pre-
detection bandpass filter has a bandwidth which is twice that of the message
bandwidth. Therefore, the noise bandwidth is doubled which implies that the noise
power is doubled at the demodulator input. The demodulator gain is necessary
to overcome this effect.
2 4
) (
4
0
2
0
2 2
N W c
df N
c
df f S
c
P
m
W
W
n on
m
m
I
= = =

0
2
0
2
2 2
,
2 2 /
4 /
) (
N W
P A
N W c
P A c
SNR
m
m c
m
m c
DSB o
= =
2
4 /
2 /
) (
) (
0
2
0
2
,
,
= =
N W P A
N W P A
SNR
SNR
m m c
m m c
DSB i
DSB o
142
Let us now compute the figure of merit in terms of the channel SNR.
The received signal power spectral density is still given by
2 2 0
1 1
( ) ( ) ( ) .
4 4 2
R
s c m c c m c
N
S f A S f f A S f f = + + +
Thus, the signal component average power is
2 2
2 2
( ) ( ) ( ) ( )
4 4
2
4 2
c m c m m m
c m c m m m
f W f W W W
c c
s m c m c m m
f W f W W W
c c m
m
A A
P S f f df S f f df S f df S f df
A A P
P
+ +

( (
= + + = + ( (
( (

= =


and the noise average power in the message signal bandwidth is
( )
0
0
2
2
m m
N
W N W = .
Hence, the channel SNR is given by
( )
2
2
,
0 0
2
2
c m
c m
c DSB
m m
A P
A P
SNR
N W N W
= = . The figure of merit is now
2
,
0
2
, 0
( )
/ 2
1
( ) / 2
o DSB
c m m
c DSB c m m
SNR
A P W N
SNR A P W N
= = .
143
Noise in AM receivers using Envelope detection

Recall that

where m(t) has average power P
m
and . Now

Therefore, the autocorrelation of the transmitted waveform is


Thus, the received signal power is


If the message bandwidth is W
m
then the noise power in the message bandwidth is


assuming AGWN with psd
( ) [1 ( )]cos( )
AM c a c
s t A k m t t = + +
~ [0 2 ] U ,
( ) cos( ) ( )cos( )
AM c c c a c
s t A t A k m t t = + + +
2 2 2
( ) cos( ) ( )cos( )
2 2
AM
c c a
s c m c
A A k
R R = +
2 2 2 2
2
(0) 1 , (0).
2 2 2
AM
c c a c
s m a m m m
A A k A
P R k P P R ( = + = + =

0
0
2 ,
2
n m m
N
P W W N
| |
= =
|
\ .
( )
0
2
N
W
S f =
144
The channel signal-to-noise ratio is


or


Consider the following receiver model




AM envelope detector receiver

2 2
1
2
0
( )
c a m
A k P
c AM
m
SNR
W N
(
+

,
=
2 2
0
1
( )
2
c a m
c AM
m
A k P
SNR
W N
,
( +

=
145
Let the bandpass filter be ideal, with unity gain and bandwidth 2W
m
. Then, at its
output,
or




where

and

Therefore, at the output of the envelope detector, we have


( ) ( ) ( )
AM
x t s t n t = +
| |
| |
{ }
( )
( ) ( )
( ) 1 ( ) cos( ) ( )cos( ) ( )sin( )
( ) ( ) cos( ) ( )sin( )
( ) , ( ) ( )
( )cos ( )
c
c a c I c Q c
c a c I c Q c
j t j t
c
x t A k m t t n t t n t t
A k A m t n t t n t t
Re x t e x t a t e
a t t t




+
= + + + + +
= + + + +
= =
= + +

| |
1
2 2
2
( ) ( ) ( ) ( )
c c a I Q
a t A A k m t n t n t
(
= + + +

1
( )
( ) tan
( ) ( )
Q
c c a I
n t
t
A A k m t n t

(
=
(
+ +

1
2 2
( )
( ) ( ) [ ( ) ( )] 1
( ) ( )
Q
c a c I
c a c I
n t
y t a t A k A m t n t
A k A m t n t
(
| |
( = = + + +
|
+ +
(
\ .

146
Remark: Phase information is irrelevant at the output of an envelope detector !

For an envelope detector to work properly, the carrier power must be much larger
than the average noise power P
n
. If this were the case, then


Note: The DC term does not contribute anything to the demodulation of m(t) and
can be removed by a DC blocking capacitor.

Let the output of the DC blocking capacitor be

Thus, the output signal-to-noise ratio is given by

, since the in-phase noise has its energy in the frequency

interval [-W
m
, W
m
].
Finally, assuming less than 100% of modulation, we get the following figure of
merit:
( ) ( ) ( ).
c a c I
y t A k A m t n t + +
noise signal + = + ) ( ) ( ) ( t n t m A k t y
I c a
2 2
0
( )
2
a c m
o AM
m
k A P
SNR
W N
,
=
147


Figure of merit

It is clear that the performance of AM with envelope detection is inferior to that of
DSB with coherent detection. This is due to explicit transmission of the carrier in
AM.

Example: Consider an AM system in AWGN with psd


Let m(t) have a bandwidth of 4 kHz and pdf


2 2
0
2 2
0
2
2
2
1
2
( )
1
( ) 1
a c m
m
c a m
m
k A P
W N
o AM
a m
A k P
c AM a m
W N
SNR
k P
SNR k P
,
(
+

,
= = = <
+
( ) triangle
2
m
x
f x
| |
=
|
\ .
. , 10
2
) (
12
0
f
Hz
Watts N
f S
W
= =

148
If the receiver signal is demodulated by an envelope detector and appropriate
postdetection filtering, assuming k
a
=0.9 find the minimum value of the carrier
amplitude A
c
that will yield (SNR)
0,AM
40 dB.

Solution: We know that

Now,




thus,



| |
( ) 1 09 ( ) cos( ),
AM c c
s t A m t t = + . +
~ [0 2 ] U ,
{ }
1 0 1
1 1 0
1 1
0 0
( ) tr ( 1) ( 1)
2
( 1) ( 1)
0, asexpected.
x
E m t x dx x x dx x x dx
x x dx x x dx

| |
= = + + +
|
\ .
= + + +
=


{ }
1 1
2 2 2
1 0
4 3
( ) 2 ( 1)
2
1 1 1 1
2 2 2
4 3 4 3 12 6
m
x
P E m t x tr dx x x dx
x X

| |
= = = +
|
\ .
(
( | |
= + = + = =
| (
(
\ .


149
So,


Now,




Consider now the case when the channel SNR is less than 1. At the input of the
envelope detector, we have (assuming =0 since envelope detector ignores phase)


where r(t) is the envelope of the noise.
( )
( )
2 2
2 2 1
6
12
0
2
(09)
( )
2
2(4000) 2 10
8437500
c
a c m
o AM
m
c
A
k A P
SNR
W N
x
A
,

.
= =
=
10
4
10
10log ( ) 40dB
log ( ) 4 ( ) 10
o AM
o AM o AM
SNR
SNR SNR
,
, ,

4
2 2
10
volts
8437500
0034 volts
c
c
A
A

.
| |
| | ( )
( ) 1 ( ) cos ( )
1 ( ) cos ( )cos ( ) ,
c a c
c a c c
x t A k m t t n t
A k m t t r t t t

= + +
= + + +
150
Now, means that almost all the time since r(t)
is a r.p.. Let us illustrate this using a time-varying phasor diagram.













( ) 1
c AM
SNR
,
<
| |
( ) 1 ( )
c a
r t A k m t > +
| |
( ) 1 ( ) 0 ( ) ( )
o
c a
x t A k m t r t t = + +
,
1
( )
( ) tan
( )
Q
I
n t
t
n t

(
=
(

151
It is clear from the picture that (using the direction of r(t) as the reference)
is equal to

and that


However,

Since

This means that the information has, on average, been lost and it makes no sense to
talk about any further or the possibility to demodulate m(t)!

( ) x t
,
| | | |
( ) ( ) 1 ( ) cos ( ) 1 ( ) sin ( )
c a c a
x t r t A k m t t jA k m t t = + + +
,
| |
( ) ( ) 1 ( ) cos ( ).
c a
y t r t A k m t t + +
{ }
( ) ~ [0 2 ] ( ) ( ) t U E y t r t ,
{ }
cos ( ) 0 E t =
( )
o AM
SNR
,
152
Noise in FM receivers

Consider the following FM receiver model






FM frequency discriminator receiver.

where the transmitted signal is



and w(t) is WGN with zero mean and psd N
0
/2, f

At the output of the bandpass filter,


0
( ) cos ( )
t
FM c c
s t A t k m d

| |
= +
|
\ .

( ) ( ) ( )
FM
x t s t n t = +
153
where the noise component is described by


Moreover, where
Thus,

Graphically, assuming (SNR)
c,FM
>>1







( )
( ) ( )cos( ) ( )sin( ) ( )cos ( )
I c Q c c
n t n t t n t t r t t t = = +
( )
( ) cos ( ) ,
FM c c
s t A t t = +
0
( ) ( )
t
t k m d

=

( ) ( )
( ) cos ( ) ( )cos ( )
c c c
x t A t t r t t t = + + +
( ) ( ) ( ) ( ).
c
x t A t r t t = +
,
154
From the above phasor diagram, using the signal component as the reference,


where


Now, a high (SNR)
c,FM
means that A
c
>>r(t) for almost all t. Therefore,




The job of the limiter is to remove the amplitude variations. Hence,

and
| | | |
( )
( ) ( )cos ( ) ( ) ( )sin ( ) ( )
( )
c
j t
x t A r t t t jr t t t
x t e

= + +
=
,
,
| |
| |
1
( )sin ( ) ( )
( ) ( ) tan
( )cos ( ) ( )
c
r t t t
t t
A r t t t

= +
(
+

( ) ( )
( )cos ( ) ( ) and ( ) ( )sin ( ) ( )
c c c
A r t t t A x t A j r t t t + +
,
| |
1
( )
( ) ( ) tan sin ( ) ( )
c
r t
t t t t
A

(
+
(

( )
c
l t A
| |
( )
( ) ( ) sin ( ) ( )
c
r t
t t t t
A
+
155
From our previous FM derivations, the output of an ideal discriminator is


where

is the noise contribution to the output of the discriminator.
At any given time t =t
0
,

since , the statistics of are determined by (t).
Hence,

The Fourier Transform of the last equation yields


| |
( )
1 1
2 2 2
( )
( ) ( ) sin ( ) ( ) ( ) ( ),
k d t
d dt
c
d r t
v t t t t m t n t
dt A

(
= + = +
(

( )
1
2
( ) ( )sin ( ) ( )
c
d
d A dt
n t r t t t

= (

( ) { }
0 0 0
sin ( ) ( ) ( ) 0 E t t t , =
0
( ) ~ [0 2 ] t U , | |
sin ( ) ( ) t t
( )
1 1
( ) [ ( )sin ( )] .
2 2
Q
d
c c
dn t
d
n t r t t
A dt A dt

=
) ( ) ( ) ( ) ( 2
2
1
) (
2
1
) ( f N f H f N
A
f
j f fN j
A dt
t dn
F
A
f N
Q eq Q
c
Q
c
Q
c
d
= = =
)
`

=

156
This is equivalent to passing the noise component through the filter

So, at the output of the discriminator, we have


Let P
m
be the average message power. Then the power of the message component
at the output of the discriminator is k
f
2
P
m
. The psd of the noise component, on the
other hand, is


This means that the power of the noise component is



( )
c
f
eq A
H f j =
( ) ( ) ( ) ( ) ( ) ( ) ( )
2
eq Q f eq Q
k
v t m t h t n t k m t h t n t

+ = +
( ) ( )
2
2
0
2
,
2
0, elsewhere
Q
T
eq N c
f
B
N f
H f S f A

, ,

2
2
2
0
2
.
B
T
B
T
noise
c
N
P f df
A

=

157
However, in general, and the average noise power can be decreased by
passing v(t) through an ideal lowpass filter with bandwidth W
m
, i.e.


At the output of the lowpass filter,


The average power of the signal component is still k
f
2
P
m
.The psd of the noise
component is, however,




2
T
B
m
W >
( )
1,
0, elsewhere
m
lpf
f W
H f
, ,

( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
f lpf lpf eq Q
f lpf eq Q
y t k m t h t h t h t n t
k m t h t h t n t
+
= +
( ) ( ) ( )
2 2
Q
lpf eq N
H f H f S f =
2
0
2
,
0, elsewhere
m
c
N
f f W
A

, ,

158
and the average power of the noise component at the output of the lowpass filter is


so,

At the input of the bandpass filter, and the channel SNR in the
message bandwidth W
m
is given by


So, the figure of merit for FM is



2 3
0 0
2 2
2
3
m
m
W
m
c c W
N N
f df W
A A

3
0
2
2 2 2
3
2
0
3
3
( )
2
m
c
f m c f m
o FM
N W
m
A
k P A k P
SNR
N W
,
= =
( ) ( ) ( )
FM
s t s t w t = +
2
2
2
0 0
( )
2
c
A
c
c FM
m m
A
SNR
W N W N
,
= =
2 2
3
0
2
0
3
2
2
2
2
2
2
3 ( )
( )
c f m
m
c
m
A k P
W N f m o FM
A
c FM m m
W N
k P SNR
f
D
SNR W W
,
,
| |

= = =
|
\ .
159
Since increased B
T
results in an improved figure of merit ! This happens
with FM only!
Now, recall that


This means that the high frequency components of the noise contribution get
emphasized. This effect can be reduced by first emphasizing the high frequency
components of the message signal before modulation at the transmitter and then de-
emphasizing both message and noise at the output of the receiver, i.e.,




,
f
f k
2
0
2
,
( ) 2
0, elsewhere
d
T
c N
N
B
f f
A S f

, ,

160
where the de-emphasis filter is described by


In principle, the average power of the message at the output of the receiver is not
affected by the pre-emphasis-de-emphasis operation. The noise component, on the
other hand, will have its spectral characteristics modified by the de-emphasis filter,
i.e.,


The average power of the noise component at the output of the post-detection
lowpass filter is given by


( )
( )
1
de
pe
H f
H f
=
m
f W
( ) ( )
( )
2
2
0
2
2
,
0, elsewhere
d
de m
c de N
N
f H f f W
A H f S f

, ,

( )
2
2
0
,
2
.
m
m
W
de n de
c W
N
f H f df P
A

161
Clearly, H
pe
(f) must have highpass characteristics which implies that H
de
(f) has
lowpass chartacteristics.

Example: Let the pre-emphasis filter be described by


and








( ) ( )
1
1
1
pe de
H f jaf H f
jaf
= + =
+
( )
( )
2
0
,
2 2 2
0
2 2 2 2 2
1
0
2 2
1
0
2 2
1
1 1 1
1
1
tan
2
2 tan
m
m
m
m
m
m
W
n de
c W
W
c W
W
W c
m m
c
N f
P df
A a f
N
df
A a a a f
N
f af
A a a
N
W aW
a A a

=
+
(
(
| |
=
(
( |
+
\ . (

(
=
(

(
=
(

162
So,





or


The improvement with de-emphasis is therefore



0
2 2
2
2 1
1
( )
( )
c
f m
o FM de
N
m m a
a A
k P
SNR
W tan aW
, ,

=
(

( )
( )
3 2
1
2
2
3 3
tan
1
3
3 tan
m f m
m m
f m
m
a W k P
aW aW
de m
k P
m m
W
F M a W
F M
aW aW

. .
= =
. . (

0
2
1
0
2 2 3
2
2
for merit of Figure
N W
A
) aW ( tan aW N
P k A a
FM
m
c
m m
m f c
de
] [

=
) aW ( tan aW
P k W a
FM
m m
m f m
de
1
2 3
for merit of Figure

=
163
Channel Propagation Losses

Simple Channel Model

where
( )
T
s t is the transmitted signal, is the attenuation coefficient,
( )
0 1, w t < < is
WGN and
( )
r t is the received signal.
164
Free space line-of-sight propagation:
Let d be the separation distance between transmitter and receiver. Then the received
signal power ( )
r
P d is described by the Friis free space equation
( )
2
,
4
T r
r T
G G
P d P
d L

| | | |
=
| |
\ . \ .

where
T
P is the transmitted power,
T
G and
r
G are the transmitter and receiver antenna
gains, respectively, L is the system loss factor (not related to propagation, 1 L ), and
is the wavelength of the carrier in meters.
Now,
2
4
,
e
e
A
G A

= is the effective aperture of the antenna, namely, the ratio of the

available power at the terminals of the receiving antenna to the power flux density of a
plane wave incident on the antenna from that direction, and
0
/ , c f = is the speed of
light in meter/s and
0
f is the carrier frequency in Hz.
The loss factor L is due to transmission line attenuation, filter losses, and antenna
losses. Clearly, 1 L = a lossless system.
165
Friis propagation model is valid for d in the far-field (Fraunhofer region) of the
transmitting antenna.
The Fraunhofer distance is given by
2
2
,
f
D
d

= where D is the largest physical

linear dimension of the antenna. Furthermore, to be in the far-filed region,
f
d D > and
.
f
d >
Since does not hold for 0, d = large-scale propagation models use a close-in
distance
0 f
d d as a known received power reference point.
Hence,
( ) ( )
2
0
0 0
, .
r r f
d
P d P d d d d
d
| |
=
|
\ .

The dynamic range of
r
P can change by many orders of magnitude over a typical
coverage area of several square kilometers. As a result, dBm or dBW units are used to
express the received power levels, e.g.,
( )
( )
0
0
10 10
in 10log 20log ,
1
r
r
P d
d
P d dBm
mW d
| |
| |
= +
|
|
\ .
\ .

( )
0 r
P d is in mW units.
166
For practical systems that use low-gain antennas in the 1-2 GHz region,
0
1m d = for
indoor environments and
0
100m d = or
0
1km d = for outdoor environments.
Example: Find
f
d for an antenna with maximum dimension of 1 m and operating
frequency 900 MHz.
Solution: D =1 m, f =900 MHz
8
6
3x10 m/s 1
0.333m
900x10 Hz 3
c
f
= = = =
( ) ( )
2 2
2 2x 1
6m
0.333
f
D
d

= = =
Example: If the transmitted signal power is 5 Watts, express this power in units of dBm
and dBW. If power is applied to a unity-gain antenna with a 900 MHz carrier frequency,
find the received power in dBm at a free space distance of 100 m from the antenna.
Whats
( )
10km
r
P ? Assume unity gain for the receiver antenna.
167
Solution:
3
10 10
10 10
inmW 5x10 mW
indBm 10log 10log 37dBm
1mW 1mW
inW 5W
indBW 10log 10log 7dBW
1W 1W
t
t
t
t
P
P
P
P
( (
= = =
( (

( (
= = =
( (



Assuming a lossless system and unity-gain antennas,
( )
( )
( )( )( )
( ) ( )
2
2
4
2 2
5 1 1 1/ 3
100 3.5x10
4 400 1
t T r
r
PG G
P mW
d L

= = =
( )
( )
4
10
10 10
4
3.5x10 mW
indBm 10log 34.5dBm
1mW
100m inmW
100
10km indBm 10log 20log
1mW 10
34.5 40 74.5dBm
r
r
r
P
P
P

(
= =
(

(
| |
= +
( |
\ .

= =

168
Effective Noise Temperature and Noise Figure
A thermal noise source has power spectral density
( ) | |
S Watts Hz , .
2
n
kT
f f =
When constrained to a frequency bandwidth of B Hz, the noise power is
x2 Watts.
2
n
kT
P B kTB = =
Consider a thermal noise source connected to an amplifier and a load, i.e.,
Thermal noise
source
Amplifier
H(f)
Load

The power spectral density at the output of the amplifier is
( ) ( ) ( )
2
.
x n
S f H f S f =
169
The noise power at the output of the amplifier is
( ) ( ) ( )
2 2
2
x n
kT
P H f S f df H f df


= =


Define the noise equivalent bandwidth by
( )
2 1
,
2
neq
B H f df
G

where
( )
{ }
2
max ,
f
G H f then
x neq
P GkTB =
Now, a practical amplifier generates additional noise internally, i.e.,
0
.
i i
i
x x x neq x
x
neq
neq
P P P GkTB P
P
GkB T
GkB
= + = +
| |
= +
|
|
\ .

170
Define the noise equivalent temperature of the amplifier by ,
i
x
e
neq
P
T
GkB
then
( )
0
1
e
x neq e neq
T
P GkB T T GkTB
T
| |
= + = +
|
\ .

Define the noise figure as 1
e
T
F
T
+ , then
0
.
x neq
P GkTB F =
But, an input signal
( )
i
s t with power
i
s
P to an amplifier with power gain G produces an
output signal
( )
0
s t with power
0
.
i
s s
P GP =
Hence, the output power SNR is given by
0
0
0
1
i i
s s s
i x neq n
P GP P
S S
N P GkTB F P F F N
| | | |
= = = =
| |
\ . \ .

and ( )
10 10 10
0
10log 10log 10log ,
i
S S
F
N N
| | | |
=
| |
\ . \ .

171
where
10
10log ( ) F is the SNR loss due to the additional noise introduced by the
amplifier.
Example: Obtain the noise figure of 2 amplifiers in cascade with gains G
1
and G
2
,
respectively.
Solution: The noise output powers of the two amplifiers are described by
( )
( )
01
2
02 01 2 2
1 1
2 1 2 1 1 2 1
1 2
i
i i
x neq e
x
x x x neq e x neq e
neq
P G kB T T
P
P G P P GG kB T T P GG kB T T
GG kB
= +
| |
= + = + + = + +
|
|
\ .

Let
2
2
2
,
i
x
e
neq
P
T
G kB
= then
02
2 1 2
1 2 1 1 2
1 1
1 2
1
1
,
e e e
x neq e neq
neq Total
T T T
P GG kB T T GG kTB
G T G T
GG kTB F
| | | |
| |
= + + = + +
| |
|
\ .
\ . \ .
=

172
where
1 2
1
1
1
e e
Total
T T
F
T G T
| |
= + + =
|
\ .
( )
1 2
1 2
1 1
1 1
1 1 1 1.
e e
T T
F F
T G T G
| |
+ + + = +
|
\ .

and
2
, 1
1
e
e Total e
T
T T
G
= +
In general, for M amplifiers in cascade,
3 2
1
1 1 2 1 2 1
2
, 1
1 1 2 1
1 1 1
M
Total
M
e eM
e Total e
M
F F F
F F
G GG GG G
T T
T T
G GG G

= + + + +
= + + +

173
Propagation Losses:
The signal undergoes attenuation because of the physical medium, operating
frequency and the separation between transmitter and receiver.
Define the dimensionless propagation loss as
T
r
P
P
L or in dB as
( ) ( )
10 10
10log 10log .
dB T r
P P = L
For wireline transmission, the propagation loss is specified in dB per unit length, e.g.,
dB/km. For line-of-sight radio communications,
2
4 d

| |

|
\ .
L
174
Example: A signal is transmitted through a 100 km coaxial cable with 1 dB/km loss.
T
P
in dBW is -20 dBW (equivalent to 10 mW). What is
r
P and the output power of a LNA
with 25dB?
dB
G =
Solution:
dB
= L 100 km x 1 dB/km =100 dBW
, , r dB T dB dB
P P = = L -20-100 =-120 dBW
0 0, , r dB dB r dB
P GP P G P = = + = 25-120 =-95 dBW
Analog Repeaters:
Consider the following communication system with a repeater:
Transmitter

175
At the output of the repeater,
0
.
T
P
P G =
L
If
0
, .
T
G P P = = L
Let
i
S
N
| |
|
\ .
be the input SNR of the amplifier, then
0
1 1 1 1
.
r T T
i a a neq a neq a neq
P P P S S
N F N F kTB F kTB F kTB
| | | | | |
| | | |
= = = =
| | |
| |
| | |
\ . \ .
\ . \ . \ .
L L

This system can be viewed as the cascade of networks with noise figures L and .
a
F
Hence, the overall noise figure is
1
.
a
Total
F
F
G

= + L
Now, if
1
G =
L
, then
1
.
1
a
Total a
F
F F

= + = L L
L

176
Link Budget:
We already know that in free-space propagation
2
0
,
4
T T R
R f
P G G
P d d d
d L

| |
=
|
\ .
. The
product
T T
P G is also referred to as the equivalent isotropic radiated power (EIRP) and the
term
2
10 10
10log 20log
4 4 d d


| | | |
=
| |
\ . \ .
is the free-space loss in decibels.
In dBs,
, 10 ,
20log
4
R dBW dBW R dB dB
P EIRP G L
d

| |
= + +
|
\ .
.
Example: Consider a communication system that uses a transmitter with directional
antenna with gain 30
T
G dB = and transmits at the power level of 100 Watts, and a
receiver whose antenna gain is 3
R
G dB = . If the transmit frequency is 100 MHz, the
antennas are separated by 100 km and the system loss is 5dB, calculate the receive
power in dBW.
177
Solution: The wavelength is
8
8
3 10
3
10
c
f

= = meters
10
5
3
20log 112.4
4 10
free space
L dB

| |
= =
|

\ .
,
10
30 10log (100) 30 20 50 EIRP dBW = + = + =
,
112.4 50 3 5 64.4
R dbW
P dBW = + + = .
The noise power seen by the receiver antenna is described by ( )
n ant e
P k T T kTB = + = ,
where
23
1.38 10 / k Joules Kelvin

= ,
ant e
T T T = + , and B is the bandwidth of interest.
The antenna temperature
ant
T is not a physical temperature, but depends on where the
antenna is pointed and the frequency band of the received signal. For example, if the
antenna is mounted on a satellite and pointed towards the earth, a typical value is
300
ant
T K = . Furthermore,
0
( 1)
e
T T F = , where
0
290 T K = . Hence, the received
signal-to-noise (SNR) power ratio in the frequency band B is given by
2
0
,
4
R T T R
f
n
P P G G
d d d
P d LkTB

| |
=
|
\ .
.
178
Example: Consider again the communication system of the previous example. Suppose
now that the receiver noise figure is 5 3.162 F dB = = , the antenna temperature is
50
ant
T K = and the receiver bandwidth 30 kHz. What is the receiver SNR?
Solution: 290(3.162 1) 627.1
e
T K = = , 50 627.1 677.1
ant e
T T T K = + = + = , and
23 4 16
1.38 10 (677.1)(3 10 ) 2.8 10
n
P Watts

= = .
Now,
16
, 10
2.8 10
10log 155.5
1
n dBW
Watts
P dBW
Watts

| |
= =
|
\ .
and the SNR in dBW is
, ,
64.4 ( 155.5) 91.1
R
R dBW n dBW
n
dBW
P
P P dBW
P
| |
= = =
|
\ .
.