0% found this document useful (0 votes)
1 views

ch2_Probability_random_variables_and_random_signal_principles

Uploaded by

purimişga Korti
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
1 views

ch2_Probability_random_variables_and_random_signal_principles

Uploaded by

purimişga Korti
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 27

CHAPTER

TWO
THE RANDOM VARIABLE

2.0 INTRODUCTION

In the previous chapter we introduced the concept of an event to describe


characteristics of outcomes of an experiment. Events allowed us more flexib¬
ility in determining properties of an experiment than could be obtained by
considering only the outcomes themselves. An event could be almost any¬
thing from “descriptive,” such as “draw a spade,” to numerical, such as “the
outcome is 3.”
In this chapter, we introduce a new concept that will allow events to be
defined m~a more consistent manner; they will always be numerical. The new
concept is that of a random variable, and it will constitute a powerful tool in
the solution of practical probabilistic problems.

2.1 THE RANDOM VARIABLE CONCEPT

Definition of a Random Variable


We define a real random variablef as a real function of the elements of a
sample space S. We shall represent a random variable by a capital letter
(such as W, X, or Y) and any particular value of theTandom variable by a

t Complex random variables are considered in Chapter 5.

33
34 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Figure 2.1-1 A random variable mapping of a sample space.

lower case letter (such as w, x, or y). Thus, given an experiment defined by


a sample space S with elements s, we assign to every s a real number

X(s) (2.1-1)

according to some rule and call X(s) a random variable.


A random variable X can be considered to be a function that maps all
elements of the sample space into points on the real line or some parts
thereof. We illustrate, by two examples, the mapping of a random variable.

Example 2.1-1 An experiment consists of rolling a die and flipping a


coin. The applicable sample space is illustrated in Figure 2.1-1. Let the
random variable be a function X chosen such that (1) a coin head (H)
outcome corresponds to positive values of X that are equal to the num¬
bers that show up on the die, and (2) a coin tail (T) outcome corre¬
sponds to negative values of X that are equal in magnitude to twice the
number that shows on the die. Here X maps the sample space of 12
elements into 12 values of X from —12 to 6 as shown in Figure 2.1-1.

Example 2.1-2 Figure 2.1-2 illustrates an experiment where the pointer


on a wheel of chance is spun. The possible outcomes are the numbers
from 0 to 12 marked on the wheel. The sample space consists of the
numbers in the set {0 < s < 12}. We define a random variable by the
function

X - X(s) = s2

Points in S now map onto the real line as the set {0 < x < 144}.
THE RANDOM VARIABLE 35

12

As seen in these two examples, a random variable is a function that


maps each point in S into some point on the real line. It is not necessary that
the sample-space points map uniquely, however. More than one point in S
may map into a single value of X. For example, in the extreme case, we
might map all six points in the sample space for the experiment “throw a
die and observe the number that shows up” into the one point X = 2.

Conditions for a Function to be a Random Variable


Thus, a random variable may be almost any function we wish. We shall,
however, require that it not be multivalued. That is, every point in S must
correspond to only one value of the random variable.
Moreover, we shall require that two additional conditions be satisfied in
order that a function A be a random variable (Papoulis, 1965, p. 88). First,
the set {X < x} shall be an event for any real number x. The satisfaction of
this condition will be no trouble in practical problems. This set corresponds
to those points s in the sample space for which the random variable A(s)
does not exceed the number x. The probability of this event, denoted by
< x|, is equal to the sum of the probabilities of all the elementary events
corresponding to {X < x).
The second condition we require is that the probabilities of the events
{X = oo} and [X = — ool be Of
P{X= -oo) = 0 P{X= oo} = 0 (2.1-2)
This condition does not prevent X from being either - oo or oo for some
values of s; it only requires that the probability of the set of those s be zero.
36 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Discrete and Continuous Random Variables


A discrete random variable is one having only discrete values. Example 2.1-1
illustrated a discrete random variable. The sample space for a discrete
random variable can be discrete, continuous, or even a mixture of discrete
and continuous points. For example, the “wheel of chance” of Example
2.1-2 has a continuous sample space, but we could define a discrete random
variable as having the value 1 for the set of outcomes {0 < s < 6} and — 1 for
{6 < s < 12}. The result is a discrete random variable defined on a contin¬
uous sample space.
A continuous random variable is one having a continuous range of values.
It cannot be produced from a discrete sample space because of our require-
ment that all random variables be single-valued functions of all sample-
space points. Similarly, a purely continuous random variable cannot result
from a mixed sample space because of the presence of the discrete portion of
the sample space. The random variable of Example 2.1-2 is continuous.

Mixed Random Variable

A mixed random variable is one for which some of its values are discrete and
some are continuous. The mixed case is usually the least important type of
random variable, but it occurs in some problems of practical significance.

2.2 DISTRIBUTION FUNCTION

The probability P{X < x} is the probability of the event {X < x}. It is a
number that depends on x; that is, it is a function of x. We call this function, MdlUWMNI» unw* ...- - — —

denoted Fx(x), the cumulative probability distribution function of the random


variable X. Thus,

Fx(x)= P{X <x} (2.2-1)

We shall often call Fx(x) just the distribution function of X. The argument x
is any real number ranging from — oo to oo.
The distribution function has some specific properties derived from the
fact that Fx(x) is a probability. These are:f

(1) F*(-oo) = 0 (2.2-2 a)

(2) Fx(co) = 1 (2.2-2 b)

f We use the notation x+ to imply x + e where e > 0 is infinitesimally small; that is, £ ->• 0.
THE RANDOM VARIABLE 37

(3) 0 < Fx(x) < 1 (2.2-2c)

(4) Fx(xi) ^ Fx(x2) if x, < x2 (2.2-2d)

(5) P{Xi < X < x2} = Fx(x2) — Fx(xi) (2.2-2e)

(6) Fx(x + )=Fx(x) ~ (2.2-2/)

The first three of these properties are easy to justify, and the reader should
justify them as an exercise. The fourth states that Fx(x) is a nondecreasing
function of x. The fifth property states that the probability that X will have
values larger than some number xx but not exceeding another number x2 is
equal to the difference in Fx(x) evaluated at the two points. It is justified
from the fact that the events {X < xj and {xt < X < x2} are mutually
exclusive, so the probability of the event {I<x2} = {I <X(}u
{xi < X < x2} is the sum of the probabilities P{X < Xj} and
P{x1 < X < x2}. The sixth property states that Fx(x) is a function contin¬
uous from the right.
Properties 1, 2, 4, and 6 may be used as tests to determine if some
function, say Gx(x), could be a valid distribution function. If so, all four tests
must be passed.
If X is a discrete random variable, consideration of its distribution
function defined by (2.2-l jshows that"/ x(x) must have a stairstep form, such
as shown in Figure 2.2-lu. The amplitude of a step will equal the probability
of occurrence of the value of X where the step occurs. If the values of X are
denoted xf, we may write Fx(x) as

Fx{x) = y p{x = xMx


i = 1
- xi) (2.2-3)

where u(-) is the unit-step function defined by

x > 0
u (2.2-4)
x <0

and N may be infinite for some random variables. By introducing the


shortened notation

P(xt) = P{X = xj (2.2-5)

(2.2-3) can be written as

Fx(x) = £ P(xt)u(x - x;) (2.2-6)


1=1

We next consider an example that illustrates the distribution function of


a discrete random variable.
38 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

(a)

fx(x)

0.6 -

0.4
0.4
tk
0.2 0.2
. 0.2
1 1
i \
o.i 0.1

-1
L
0
,
1 2
I3 x
(b)

Figure 2.2-1 Distribution function (a) and density function (b) applicable to the discrete random
variable of Example 2.2-1. [Adapted from Peebles (1976) with permission of publishers Addison-
Wesley, Advanced Book Program.]

Example 2.2-1 Let X have the discrete values in the set {—1, —0.5, 0.7,
1.5, 3}. The corresponding probabilities are assumed to be {0.1, 0.2, 0.1,
0.4, 0.2}. Now P{X < — 1} = 0 because there are no sample space points
in the set {X < — 1}. Only when X — — 1 do we obtain one outcome.
Thus, there is an immediate jump in probability of 0.1 in the function
Fx(x) at the point x = — l.For—l<x< —0.5, there are no additional
sample space points so Fx(x) remains constant at the value 0.1. At
x = —0.5 there is another jump of 0.2 in Fx(x). This process continues
until all points are included. Fx(x) then equals 1.0 for all x above the last
point. Figure 2.2-la illustrates Fx(x) for this discrete random variable.

A continuous random variable will have a continuous distribution func¬


tion. We consider an example for which Fx(x) is the continuous function
shown in Figure 2.2-2a.
THE RANDOM VARIABLE 39

fxM

1
12 Figure 2.2-2 Distribution function (a) and
density function (b) applicable to the con¬
tinuous random variable of Example 2.2-2.
1 [Adapted from Peebles (1976) with per¬
0 6 12 jc mission of publishers Addison-Weslev,
(b) Advanced Book Program.]

Example 2.2-2 We return to the fair wheel-of-chance experiment. Let


the wheel be numbered from 0 to 12 as shown in Figure 2.1-2. Clearly
the probability of the event {X < 0} is 0 because there are no sample
space points in this set. For 0 < x < 12 the probability of {0 < X < x}
will increase linearly with x for a fair wheel. Thus, Fx(x) will behave as
shown in Figure 2.2-2a.

The distribution function of a mixed random variable will be a sum of


two parts, one of stairstep form, the other continuous.

2.3 DENSITY FUNCTION

The probability density function, denoted by fx(x), is defined as the derivative


of the distribution function:

(2.3-1)
AW = ^
We often call fx(x) just the density function of the random variable X.
MMIMMMMMNMtMHlMMnrmietMnOT
40 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Existence
If the derivative of Fx(x) exists then fx(x) exists and is given by (2.3-1). There
may, however, be places where dFx(x)/dx is not defined. For example, a
continuous random variable will have a continuous distribution Fx(x), but
Fx(x) may have corners (points of abrupt change in slope). The distribution
shown in Figure 2.2-2a is such a function. For such cases, we plot fx(x) as a
function with step-type discontinuities (such as in Figure 2.2-2b). We shall
assume that the number of points where Fx(x) is not differentiable is
countable.
For discrete random variables having a stairstep form of distribution
functionaje..inXroduce-4hc concept of the unit-impulse function <5(x) to
describe the derivative of Fx(x) at its stairstep points. The unit-impulse
function and its properties are reviewed in Appendix A. It is shown there
that S(x) may be defined by its integral property

f00
</>(x0) = I 4>(x)S(x — x0) dx (2.3-2)
— 00

where (p(x) is any function continuous at the point x = x0 ; <3(x) can be in-
terpreted as a “ function ** with infinite amplitude, area of unity, and zero
duration. The unit-impulse and the umFstep functions are related by

<;»)
or

[ dq = u(x) (2.3-4)
* — 00

The more general impulse function is shown symbolically as a vertical arrow


occurring at the point x = x0 and having an amplitude equal to the ampli¬
tude of the step function for which it is the derivative.
We return to the case of a discrete random variable and differentiate
Fx(x), as given by (2.2-6), to obtain

fx{x) = X P(x,)S(x - x,) (2.3-5)


i= 1

Thus, the density function for a discrete random variable exists in the sense
that we use impulse functions to describe the derivative of Fx(x) at its
stairstep points. Figure 2.2-1 b is an example of the density function for the
random variable having the function of Figure 2.2-la as its distribution.
A physical interpretation of (2.3-5) is readily achieved. Clearly, the prob¬
ability of X having one of its particular values, say is a number P(x-). If
THE RANDOM VARIABLE 41

this probability is assigned to the point xt, then the density of probability is
infinite because a point has no “ width ” on the x axis. The infinite “ ampli¬
tude ” of the impulse function describes this infinite density. The “ size ” of
the density of probability at x = xf is accounted for by the scale factor P(x,)
giving P(Xi)S(x - xf) for the density at the point x = xt .

Properties of Density Functions

Several properties that fx(x) satisfies may be stated:


o
<

(1) all x
VI

(2.3-6 a)
x ;
t

. 00
(2) I fx(x) dx = 1 (2.3-6 b)
* — ao

(3) Fx(x) = 1 ’* fx(i) dq


* (2.3-6 c)
- 00

(4) p{xi < X < x2} — | fx{x) dx (2.3-6d)


'^i

Proofs of these properties are left to the reader as exercises. Properties 1 and
2 require that the density function be nonnegative and have an area of unity.
These two properties may also be used as tests to see if some function, say
9x(x), can be a valid probability density function. Both tests must be satisfied
for validity. Property 3 is just another way of writing (2.3-1) and serves as the
link between Fx(x) and /*(x). Property 4 relates the probability that X will
have values from xl to, and including, x2 to the density function.

Example 2.3-1 Let us test the function gx(x) shown in Figure 2.3-la to
see if it can be a valid density function. It obviously satisfies property 1
since it is nonnegative. Its area is acc which must equal unity to satisfy
property 2. Therefore a = 1/a is necessary if gx(x) is to be a density.
Suppose a = 1/a. To find the applicable distribution function we
first write

0 Xq a > x > Xq t a

1 ,
~~2 (x x0 T aj x0 — a < x < x0
9x(x)

1
(x - x0) X0 < x < Xq T a
a a^
42 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Figure 2.3-1 A possible probabil¬


ity density function (a) and a
distribution function (b) appli¬
cable to Example 2.3-1.

Next, by using (2.3-6c), we obtain


0 x0 — a > x

1
gx(x) dx = ^ (x ~ xo + a)2 x0 — a < x < x0
XQ-IX

1 rx 11
Gx(*) — - + | gx(x) dx = - + - (x - x0)
^ ' xo 2. (X

1
(x - x0y x0 < x < x0 + a
2a'

i. x0 + a < x
This function is plotted in Figure 2.3-15.

2.4 THE GAUSSIAN RANDOM VARIABLE

A random variable X is called gaussiani if its density function has the form

fx(x) = -Jr^e-(x-“x*ll°x2 (2.4-1)


V27t<Tx

t After the German mathematician Johann Friedrich Carl Gauss (1777-1855). The gaussian
density is often called the normal density.
THE RANDOM VARIABLE 43

Figure 2.4-1 Density (a) and. distribution (b) functions of a gaussian random variable.

where ox > 0 and — co<ax<co are real constants. This function is


sketched in Figure 2.4-iu. Its maximum value (2no\y^2 occurs at x = ax.
Its “ spread ” about the point x = ax is related to ox • The function decreases
to 0.607 times its maximum at x = ax + ax and x = ax — ox.
The gaussian density is the most important of all densities. It enters into
nearly all areas of engineering and science. We shall encounter the gaussian
random variable frequently in later work when we discuss some important
types of noise.
The distribution function is found from (2.3-6c) using (2.4-1). The inte¬
gral is

Fx(x) = - 2 =

V 27TCTX
I'
- oc
di (2.4-2)
44 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

This integral has no known closed-form solution and must be evaluated by


numerical methods. To make the results generally available, we could
develop a set of tables of Fx(x) for various x with ax and gx as parameters.
However, this approach has limited value because there is an infinite
number of possible combinations of ax and gx, which requires an infinite
number of tables. A better approach is possible where only one table of
Fx(x) is developed that corresponds to normalized (specific) values of ax and
gx . We then show that the one table can be used in the general case where ax

and ax can be arbitrary.


We start by first selecting the normalized case where ax = 0 and
gx= 1. Denote the corresponding distribution function by F(x). From

(2.4-2), F(x) is

(2.4-3)

which is a function of x only. This function is tabularized in Appendix B for


x > 0. For negative values of x we use the relationship

F( — x)= 1 — F(x) (2.4-4)

To show that the general distribution function Fx(x) of (2.4-2) can be


found in terms of F(x) of (2.4-3), we make the variable change

« = (i - ax)/ax (2.4-5)

in (2.4-2) to obtain
i Ax-ax)/<rx
Fx(x) = , — | e~u2/2 du (2.4-6)

From (2.4-3), this expression is clearly equivalent to

'*> -
Figure 2.4-lb depicts the behavior of Fx(x).
f(x r) (2.4-7)

We consider two examples to illustrate the application of (2.4-7).

Example 2.4-1 We find the probability of the event {X < 5.5} for a
gaussian random variable having ax = 3 and gx = 2.
Here (x — ax)/Gx = (5.5-3)/2 = 1.25. From (2.4-7) and the definition
of Fx(x)

p{X < 5.5} = Fx(5.5) = F(1.25)

By using the table in Appendix B

P{X <5.5} = F(1.25) = 0.8944


THE RANDOM VARIABLE 45

Example 2.4-2 Assume that the height of clouds above the ground at
some location is a gaussian random variable X with ax = 1830 m and
ox = 460 m. We find the probability that clouds will be higher than
2750 m (about 9000 ft). From (2.4-7) and Appendix B:
P{X > 2750} = 1 P{X < 2750} = 1 Fx(2750)
/2750 - 1830^
1 - F 1 - F(2.0)
\ 460
1 0.9772 = 0.0228
The probability that clouds are higher than 2750 m is therefore about
2.28 percent if their behavior is as assumed.

2.5 OTHER DISTRIBUTION AND DENSITY EXAMPLES S&p

Many distribution functions are important enough to have been given


names. We give three examples. The first is for a discrete random variable;
the remaining two are for continuous random variables. Other distributions
are listed by Peebles (1976).

Binomial
Let 0 < p < 1, and N = 1,2,..., then the function

fx(x) = (^)pk(l - pf^k5(x - k) (2.5-1)

is called the binomial density function. The quantity (*) is the binomial
coefficient defined in (1.7-4) as

IN\ N\
[ k / “ k'.(N - k)l *2'5"2)

The binomial density can be applied to the Bernoulli trial experiment of


Chapter 1. It applies to many games of chance, detection problems in radar
and sonar, and many experiments having only two possible outcomes on
any given trial.
By integration of (2.5-1), the binomial distribution function is found:

Fx(x) = Z ~ Pf~ku(x ~ k) (2.5-3)

Figure 2.5-1 illustrates the binomial density and distribution functions


for N = 6 and p = 0.25.
46 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

FxM

nq^94 0.9954 0.9998 1-0000


1.0
0.8306

0.5340
0.5

0.1780

_ _ I_I_1
0 12 3 4 5 6 *

C b)

Figure2.5-1 Binomial density (a) and distribution (b) functions for the case N = 6 and p = 0.25.

Exponential
The exponential density and distribution functions are:

1 e-(x-a)/b
x > a
fx(x) = 1 b (2.5-4)
0 x < a
\ _ e~(x~a)/b
x > a
Fx(x) = (2.5-5)
0 x < a

for real numbers — oo < a < oo and b > 0. These functions are plotted in
Figure 2.5-2.
The exponential density is useful in describing raindrop sizes when a
large number of rainstorm measurements are made. It is also known to
approximately describe the fluctuations in signal strength received by radar
from certain types of aircraft as illustrated by the following example.
THE RANDOM VARIABLE 47

Figure 2.5-2 Exponential density (a) and distribution (b) functions.

Example 2.5-1 The power reflected from an aircraft of complicated


shape that is received by a radar can be described by an exponential
random variable P. The density of P is therefore

1-p y1 0 p > 0
fp(p) = \ Po
(0 p < 0

where P0 is the average amount of received power. At some given time


P may have a value different from its average value and we ask: what
is the probability that the received power is larger than the power
received on the average?
We must find P{P > P0} = 1 - P{P < P0} = 1 - FP(P0). From
(2.5-5)
P{P > P0} = 1 _ (1 _ e-p°lpa) = e~1 % 0.368

In other words, the received power is larger than its average value about
36.8 percent of the time.
48 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Rayleigh
The Rayleighf density and distribution functions are:

17 (x — a)e (x a)2/b x > a


/*(*) = \b (2.5-6)
10 x < a

., 11 - e-{*-°wb x > a
F*(*)= (2.5-7)
|o x < a

for real constants — oo < a < oo and b > 0. These functions are plotted in
Figure 2.5-3.

t Named for the English physicist John William Strutt, Lord Rayleigh (1842-1919).

Figure 2.5-3 Rayleigh density (a) and distribution (b) functions.


THE RANDOM VARIABLE 49

The Rayleigh density describes the envelope of one type of noise when
passed through a bandpass filter. It also is important in analysis of errors in
various measurement systems.

2.6 CONDITIONAL DISTRIBUTION AND


DENSITY FUNCTIONS

The concept of conditional probability was introduced in Chapter 1. Recall


that, for two events A and B where P(B) f 0, the conditional probability of
A given B had occurred was

P(M B) = -p{B) 7 (2.6-1)

In this section we extend the conditional probability concept to include


random variables.

Conditional Distribution
Let A in (2.6-1) be identified as the event {X < x} for the random variable X.
The resulting probability P{X < x\B) is defined as the conditional distribu¬
tion function of X, which we denote Fx(x] B). Thus

P{X < x n £}
Fx{x | B) = P{X <x\B} = (2.6-2)
m
where we use the notation (X < t n B) to imply the joint event
AT < x} n B. This joint event consists of all outcomes s such that

X(s) x and s £ B (2.6-3)


The conditional distribution (2.6-2) applies to discrete, continuous, or
mixed random variables.

Properties of Conditional Distribution


All the properties of ordinary distributions apply to Fx(x \ B). In other
words, it has the following characteristics:

(1) Fx(-oo|B) = 0 (2.6-4 a)

(2) Fx(oo | B) = I (2.6-4 b)

(3) (2.6-4 c)
o
VI

VI

(4) Fx(xi \B) < Fx{x2\B) if Xi < x2 (2.6-4d)

(5) P{Xi <X<x2\B} = Fx(x2 | B) - Fx(Xl | B) (2.6-4e)

(6) Fx(x+ \B) = Fx(x | B) (2.6-4/)


50 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

These characteristics have the same general meanings as described earlier


following (2.2-2).

Conditional Density
In a manner similar to the ordinary density function, we define conditional
density function of the random variable X as the derivative of the con-
ditional distribution function. If we denote this density by fx(x\B), then

dFx(x 1B)
fx(x | B) = (2.6-5)
dx

If Fx(x\B) contains step discontinuities, as when X is a discrete or mixed


random variable, we assume that impulse functions are present in fx(x \ B) to
account for the derivatives at the discontinuities.

Properties of Conditional Density


Because conditional density is related to conditional distribution through
the derivative, it satisfies the same properties as the ordinary density func¬
tion. They are:

(1) fx(x | B) > 0 (2.6-6 a)

(2) | fx(x | B)dx= 1 (2.6-6 b)


— 00

(3) Fx{x | B) = '* fx(c\B)dz (2.6-6 c)


J
— 00

(4) P{Xl <X<x2\B} = f2fx(x \ B) dx (2.6-6d)


*i

We take an example to illustrate conditional density and distribution.

Example 2.6-1 Two boxes have red, green, and blue balls in them; the
number of balls of each color is given in Table 2.6-1. Our experiment
will be to select a box and then a ball from the selected box. One box
(number 2) is slightly larger than the other, causing it to be selected
more frequently. Let B2 be the event “select the larger box” while Bi is
the event “select the smaller box.” Assume P(B1) = %0 and P(B2) = %0-
(Bt and B2 are mutually exclusive and Bx u B2 is the certain event,
since some box must be selected; therefore, P^) + P{B2) must equal
unity.)
Now define a discrete random variable X to have values xx = 1,
THE RANDOM VARIABLE 51

Table 2.6-1 Numbers of colored balls in


two boxes

Box

Ball color 1 2 Totals

1 Red 5 80 85
2 Green 35 60 95
3 Blue 60 10 70

Totals 100 150 250

x2 = 2, and x3 = 3 when a red, green, or blue ball is selected, and let B


be an event equal to either Bl or B2. From Table 2.6-1:

80
P(X = l\B = Bl) P(X=1\B=B2) =
100 150

35 60
P(X = 2\B = B1) = P(X = 2\B=B2)
100 150

60 10
P(X = 3|B = Bl) = P(X = 3\B=B2) =
Too 150

The conditional probability density fx(x | Bx) becomes

By direct integration offx(x\ B{):

^(x|B1) = ~u(x- 1) + ~«(x-2) + -^M(x-3)

For comparison, we may find the density and distribution of X by


determining the probabilities P(X = 1), P(X = 2), and P(X = 3). These
are found from the total probability theorem embodied in (1.4-10):

P(X
P(X= 1) = P(X= l\Bl)P(B1)^-P(X= 1\B2)P(B2)

80 jf 8
5 I(2 + - 0.437
100'lio 150'HO

35 | 60 |f 8
P(X = 2) [2 + .390
100'110 150'.10

60 |(2
p|*-J>-T»(fo) + 7S>(fo)-om
100'Uo
52 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Thus
fx(x) = 0.437 S(x - 1) + 0.390 S(x - 2) + 0.173 S(x - 3)
and
Fx(x) — 0A31u(x — 1) + 0.390w(x — 2) + 0.173w(x — 3)
These distributions and densities are plotted in Figure 2.6-1.

Figure 2.6-1 Distributions (a) and densities (b) and (c) applicable to Example 2.6-1.
THE RANDOM VARIABLE 53

^Methods of Defining Conditioning Event


The preceding example illustrates how the conditioning event B can be
defined from some characteristic of the physical experiment. There are
several other ways of defining B (Cooper and McGillem, 1971, p. 61). We
shall consider two of these in detail.
In one method, event B is defined in terms of the random variable X. We
discuss this case further in the next paragraph. In another method, event B
may depend on some random variable other than X. We discuss this case
further in Chapter 4.
One way to define event B in terms of X is to let

B = {X <b) (2.6-7)
where b is some real number — oo < b < oo. After substituting (2.6-7) in
(2.6-2), we getf

Fx(x\X <b) = P{X <x\X <b} = ~ <2'6-8)

for all events {X < b] for which P{X < b\ F 0. Two cases must be con¬
sidered; one is where b < x; the second is where x < b. If b < x, the event
{X < b} is a subset of the event {X < x}, so {X < x} n {X < b} = {X < b}.
Equation (2.6-8) becomes

P{X < x n X < b} P{X < b}


Fx(x|X<6) = b <x (2.6-9)
P{X < b} P{X < b}

When x < b the event {X < x} is a subset of the event {X < b}, so {X < x} n
{X < b} = {X < x} and (2.6-8) becomes
P{X < x n X < b} P{X < x} FAx)
Fx(x\X<b) = x <b
P{X < 6} P{X < b} Fx(b)
(2.6-10)
By combining the last two expressions, we obtain

[xix)
x<b
Fx(x | X <b) = FAb) (2.6-11)
1 b <x
The conditional density function derives from the derivative of (2.6-11):

j fx{x) _ fx(x)
x<b
fx(x IX < b) = I Fx(b) f-oo fx(x) dx (2.6-12)
I 0 x > b.

f Notation used has allowed for deletion of some braces for convenience. Thus,
Fx(x\{X < b}) is written Fx(x\X < b) and P{{X < x} n {X < b}) becomes P{X < x n X < b).
54 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

Figure 2.6-2 Possible distribution functions (a) and density functions (b) applicable to a
conditioning event B = {X < b}.

Figure 2.6-2 sketches possible functions representing (2.6-11) and


(2.6-12).
From our assumption that the conditioning event has nonzero probabi¬
lity, we have 0 < Fx(b) < 1, so the expression of (2.6-11) shows that the
conditional distribution function is never smaller than the ordinary distribu¬
tion function:

Fx(x \X <b)> Fx(x) (2.6-13)

A similar statement holds for the conditional density function of (2.6-12)


wherever it is nonzero:

fx(x\X <b)>fx(x) x<b (2.6-14)

The principal results (2.6-11) and (2.6-12) can readily be extended to the
more general event B = {a < X < b} (see Problem 2-39).
THE RANDOM VARIABLE 55

PROBLEMS

2-1 The sample space for an experiment is S = {0, 1, 2.5, 6}. List all possible values of the
following random variables:
(a) X = 2s
(b) X = 5s2 - 1
(c) X = cos (ns)
(d) X = (1 - 3s)"1
2-2 Work Problem 2-1 for S = { — 2 < s < 5}.
2-3 Given that a random variable X has the following possible values, state if X is discrete,
continuous, or mixed.
(a) { — 20 < x < —5}
(b) {10, 12 < x < 14, 15, 17}
(c) {—10 for s > 2 and 5 for s < 2, where 1 < s < 6}
(d) (4, 3.1, 1, -2}
2-4 A random variable X is a function. So is probability P. Recall that the domain of a function
is the set of values its argument may take on while its range is the set of corresponding values of
the function. In terms of sets, events, and sample spaces, state the domain and range for X
and P.

2-5 A man matches coin flips with a friend. He wins $2 if coins match and loses $2 if they do not
match. Sketch a sample space showing possible outcomes for this experiment and illustrate how
the points map onto the real line x that defines the values of the random variable X = “dollars
won on a trial.” Show a second mapping for a random variable Y = “dollars won by the friend
on a trial.”

2-6 Temperature in a given city varies randomly during any year from —21 to 49°C. A house in
the city has a thermostat that assumes only three positions: 1 represents “call for heat below
18.3°C,” 2 represents “dead or idle zone,” and 3 represents “call for air conditioning above
21.7°C.” Draw a sample space for this problem showing the mapping necessary to define a
random variable X — “ thermostat setting.”

2-7 A random voltage can have any value defined by the set S = [a < s < b). A quantizer divides
S into 6 equal-sized contiguous subsets and generates a voltage random variable X having values
( — 4, —2, 0, 4, 6}. Each value of X is equal to the midpoint of the subset of S from which it is
mapped.
(a) Sketch the sample space and the mapping to the line x that defines the values of X.
(b) Find a and b.
*2-8 A random signal can have any voltage value (at a given time) defined by the set
[a0 < s < aN}, where a0 and aN are real numbers and N is any integer N > 1. A voltage quan¬
tizer divides S into N equal-sized contiguous subsets and converts the signal level into one of
a set of discrete levels an, n = 1, 2, ..., N, that correspond to the “input” subsets
{a„-i < 5 < a„}. The set {al5 a2, ..., aN} can be taken as the discrete values of an “output”
random variable X of the quantizer. If the smallest “input” subset is defined by
A = ax — a0 and other subsets by an — a„_ t = 2n~1 A, determine A and the quantizer levels an in
/ terms of a0, aN, and N.

( 2-9 iAn honest coin is tossed three times.


(a) Sketch the applicable sample space S showing all possible elements. Let A be a
random variable that has values representing the number of heads obtained on any triple toss.
Sketch the mapping of S onto the real line defining X.
(b) Find the probabilities of the values of X.
2-10 Work Problem 2-9 for a biased coin for which P{head} = 0.6.
56 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

2-11 ^Resistor R2 in Figure P2-11 is randomly selected from a box of resistors containing 180-Q,
'47*0-0, 1000-Q, and 2200-0 resistors. All resistor values have the same likelihood of being
selected. The voltage E2 is a discrete random variable. Find the set of values E2 can have and
give their probabilities.

Rx = 820 O

Figure P2-11

2-12 A random variable X is called uniformly distributed if its density function is given by

r ( \ _ jV(b - a) a < x < b


x |0 elsewhere

for any real numbers a < b.


(a) Find the distribution function Fx(x) of X.
(b) Sketch /x(x) and Fx(x).

ind and sketch the density and distribution functions for the random variables of parts
and (c) in Problem 2-1 if the sample space elements have equal likelihoods of occurrence.

2-14 If temperature in Problem 2-6 is uniformly distributed (see Problem 2-12), sketch the
density and distribution functions of the random variable X.

2-15 For the random variable of Problem 2-12 find:


{a) P{0.9a + 0.1b < X < 0.7a + 0.3b}
(b) P{{a + b)/2 < X < b)

2-16 Determine which of the following are valid distribution functions:

x > 0
(a) Gx(x)
x < 0

x <0

(b) Gx(x) { 0.5 + 0.5 sin [7r(x — l)/2] 0 < x < 2

|1 x > 2

x
(c) Gx{x) - [u(x — a) — a(x — 2a)]
a

2-17 Determine the real constant a, for arbitrary real constants m and 0 < b, such that

/x(x) = ac~

is a valid density function (called the Laplacet density).

2-18 An intercom system master station provides music to six hospital rooms. The probability
that any one room will be switched on and draw power at any time is 0.4. When on, a room
draws 0.5 W.

t After the French mathematician Marquis Pierre Simon de Laplace (1749-1827).


THE RANDOM VARIABLE 57

(a) Find and plot the density and distribution functions for the random variable “power
delivered by the master station.”
(b) If the master-station amplifier is overloaded when more than 2 W is demanded, what
is its probability of overload?

2-19 The amplifier in the master station of Problem 2-18 is replaced by a 4-W unit that must
now supply 12 rooms. Is the probability of overload better than if two independent 2-W units
supplied six rooms each?

2-20 Justify that a distribution function Fx(x) satisfies (2.2-2a, b, c).


2-21 Use the definition of the impulse function to evaluate the following integrals.
(Hint: Refer to Appendix A.)
,4
(a) j (3x2 + 2x — 4)c>(x — 3.2) dx
‘ 3

00
(b) ) cos (67rx)<5(x — 1) dx
— OO

r00 24S(x-2)dx
•L a, x4 + 3x2 + 2

r°°

(d) f | <5(x — x0)e }mx dx


- 00

,3
(e) j u(x — 3)<5(x — 3) dx
' -3

2-22 Show that the properties of a density function/^(x), as given by (2.3-6), are valid.
2-23 For the random variable defined in Example 2.3-1, find:
(a) P{x0 — 0.6a < X < x0 + 0.3a}
(b) P{X = x0}

2-24 A random variable X is gaussian with ax = 0 and ax = 1.


(a) What is the probability that | X | >2?
(b) What is the probability that X > 2?

2-25 Work Problem 2-24 if ax = 4 and ox = 2.


2-26 For the gaussian density function of (2.4-1), show that
00

xfx(x) dx = ax
— 00

2-27 For the gaussian density function of (2.4-1), show that

| (x ~ ax)2fx(x) dx =
- 00

2-28 )a production line manufactures 1000-ft resistors that must satisfy a 10% tolerance.
(a) If resistance is adequately described by a gaussian random variable X for which
ax — 1000 Q and ox — 40 Q, what fraction of the resistors are expected to be rejected?
(b) If a machine is not properly adjusted, the product resistances change to the case where
ax = 1050 Q (5% shift). What fraction is now rejected?

f The quantity j is the unit-imaginary; that is, j = 1.


58 PROBABILITY, RANDOM VARIABLES, AND RANDOM SIGNAL PRINCIPLES

2-29 Cannon shell impact position, as measured along the line of fire from the target point, can
be described by a gaussian random variable X. It is found that 15.15% of shells fall 11.2 m or
farther from the target in a direction toward the cannon, while 5.05% fall farther than 95.6 m
beyond the target. What are ax and ax for X?
2-30 (a) Use the exponential density of (2.5-4) and solve for I2 defined by

r00
12 = j x2fx{x) dx
— 00

(b) Solve for 7X defined by

fi = j xfx(x)dx
— 00

(c) Verify that % and /2 satisfy the equation J2 — l\ = b2.


2-31 Verify that the maximum value offx(x) for the Rayleigh density function of (2.5-6) occurs
at x = a + yjb/2 and is equal to j2/b exp ( — Y2) ~ 0.607 j2/b. This value of x is called the
mode of the random variable. (In general, a random variable may have more than one such
value—explain.)
2-32 Find the value x = x0 of a Rayleigh random variable for which P{X < x0} = P{*„ < x}.
This value of x is called the median of the random variable.
2-33 The lifetime of a system expressed in weeks is a Rayleigh random variable X for which

, , |(x/200k~*2/4O° 0 < x
Ux) = v ’
*v ; |0 x <0

(a) What is the probability that the system will not last a full week?
(b) What is the probability the system lifetime will exceed one year?
2-34 The Cauchyt random variable has the probability density function

b/n
fx(x)
V+ (x - a)2

for real numbers 0 < b and — oo < a < oo. Show that the distribution function of X is

Fx{x) = \ + - tan 1 (

I
2 tc \

2-35 The Log-Normal density function is given by

exp { — [In (x — b) — ax ]2/241


x > b
Jin ax{x - b)

0 x < b

for real constants 0 < ox, — oo < ax < oo, and — oo < b < oo, where In (x) denotes the natural
logarithm of x. Show that the corresponding distribution function is

In (x — b) — ax
x > b
Fx(x) =

x < b

where F( ) is given by (2.4-3).

t After the French mathematician Augustin Louis Cauchy (1789-1857).


THE RANDOM VARIABLE 59

' 2-36yA discrete random variable X is said to have a Poisson densityt if fx(x) is of the form
* bke-b

fxix) = Z —rr~ s(x ~ k)


fc =o K-

for any real constant 0 < b. Find the distribution function and sketch it for the special case

The number of cars arriving at a certain bank drive-in window during any 10-min period
is a Poisson random variable X with b = 2 (see Problem 2-36). Find:
(a) The probability that more than 3 cars will arrive during any 10-min period.
(b) The probability that no cars will arrive.
2-38 Rework Example 2.6-1 to find fx(x \ B2) and Fx(x \ B2). Sketch the two functions.
*2-39 Extend the analysis of the text, that leads to (2.6-11) and (2.6-12), to the more general
event B = {a < X < b}. Specifically, show that now

0 x < a

Fx(x | a < X < b) = <


Fx(x) ~ Fx(a) a <x < b
Fx(b) - Fx(a)

1 b < x

and

0 x < a

fx{x) fxix)
fx(x | a< X <b) = a < x < b
Fx(b) - Fx(a) Safx(X) dx

0 b < x

*2-40 Consider the system having a lifetime defined by the random variable X in Problem 2-33.
Given that the system will survive beyond 20 weeks, find the probability that it will survive
beyond 26 weeks.

t After the French mathematician Simeon Denis Poisson (1781-1840).

You might also like