Module

2
Random Processes

Version 2, ECE IIT, Kharagpur

 Lesson
6
Some useful
Distributions

Version 2, ECE IIT, Kharagpur

After reading this lesson, you will learn about

¾ Uniform Distribution
¾ Binomial Distribution
¾ Poison Distribution
¾ Gaussian Distribution
¾ Central Limit Theorem
¾ Generation of Gaussian distributed random numbers using computer
¾ Error function

Uniform (or Rectangular) Distribution

The pdf f(x) and the cdf F(x) are defined below:
f(x) = 0 for x < a and
for x > (a + b ).

However, let, f(x) = 1/b for a < x < (a + b). It is easy to see that,

+∞ a +b
⎛1⎞
∫ f ( x ) dx = ∫ ⎜ ⎟dx = 1
⎝b⎠
2.7.1
−∞ a
⎧0 for x < a ;
x ⎪
⎪x−a
The cdf F ( x ) = ∫ f ( x ) dx = ⎨ for a < x < ( a + b ) ; 2.7.2
−∞ ⎪ b
⎪⎩1 for x > ( a + b ) ;

Binomial Distribution
This distribution is associated with discrete random variables. Let ‘p’ is the
probability of an event (say, ‘S’, denoting success) in a statistical experiment. Then, the
probability that this event does not occur (i.e. failure or ‘F’ occurs) is (1 – p) and, for
convenience, let, q = 1 – p, i.e. p + q = 1. Now, let this experiment be conducted
repeatedly (without any fatigue or biasness or partiality) ‘n’ times. Binomial distribution
tells us the probability of exactly ‘x’ successes in ‘n’ trials of the experiment (x ≤ n). The
corresponding binomial probability distribution is:
⎛ ⎞ ⎡ n! ⎤ x n− x
f ( x ) = ⎜ n ⎟ p x q n− x = ⎢ ⎥ p (1 − p ) 2.7.3
⎝ x⎠ ⎣ x !( n − x )!⎦

n n
⎛ ⎞
Note that, ∑ f ( x ) = ∑ ⎜ n ⎟ p x q n− x = 1
x =0 ⎝ ⎠
x =0
x

Version 2, ECE IIT, Kharagpur

 The following general binomial expression and its derivative with respect to a
dummy variable ‘τ‘are useful in verifying the above comment and also for finding
expectations of ‘S’:
n
⎛ ⎞
( pτ + q )n = ∑ ⎜ nx ⎟ ( pτ ) x q n− x
x =0 ⎝ ⎠
Differentiating this expression once w.r.t. ‘τ‘, we get,
n−1
n
⎛ ⎞
n ( pτ + q ) p = ∑ ⎜ n ⎟ p x q n− x xτ x−1
x =0 ⎝ ⎠
x
n n
⎛ ⎞
The mean number of successes, E ( s ) = ∑ ( ) ∑ ⎜⎝ nx ⎟⎠ p x q n− x x
x f x =
x =0 x =0

A simple form of the last term can be obtained by putting τ = 1 in the previous
expression resulting in a fairly easy-to-remember mean of the random variable:
The mean number of successes, E (S) = np

Following a similar approach it can be shown that, the variance of the number of
successes,
σ2 = npq.

An approximation

The following approximation of the binomial distribution of 2.7.3, known as Laplace

approximation, is good to use when ‘n’ is large, i.e. n Æ ∞:
f ( x ) = {(2π npq)−1/ 2 }.exp[−{( x − np)2 }/(2npq)] 2.7.4

Poisson Distribution
A formal discussion on Poisson distribution usually follows a description of
binomial distribution because the Poisson distribution can be approximately viewed as a
limiting case of binomial distribution.

The approximate Poisson distribution, fp(x) is:

f p ( x ) = ⎡{.λ x }/ x !⎤ .e −λ 2.7.5
⎣ ⎦

‘λ‘ in the above expression is the mean of the distribution. A special feature of this
distribution is that the variance of this distribution is approximately the same as its mean
(λ).

Poisson distribution plays an important role in traffic modeling and analysis in data
networks.

Version 2, ECE IIT, Kharagpur

Gaussian (Normal) Distribution:
This is a commonly applied distribution often exhibited by continuous random
variables in nature. A relevant example is the weak electrical noise (voltage or
equivalently current) generated because of random (Brownian) movement of carriers in
electronic devices and components (such as resistor, diode, transistor etc.).

A Gaussian or, normally distributed random variable X has the following

probability density function:
fN(x) = [ 1/{σ. (2π)1/2}]. exp[ - {(x – m)2}/ (2σ2) ] 2.7.6
m = E[X] : Mean or expected value of the distribution and
σ2 = E[{X – E[X]}2] : the variance of the distribution.

Fig. 2.7.1 shows two Gaussian pdf curves with means 0.0 and 1.5 but same
variance 1.0. The particular curve with m = 0.0 and σ2 = 1.0 is known as the normalized
Gaussian pdf. It may be noted that a pdf curve is symmetric about its mean value. The
mean may be positive or negative. Further, a change in the mean of a Gaussian pdf curve
only shifts the curve horizontally without any change in shape of the curve. A smaller
variance, however, increases the sharpness of the peak value of the pdf which always
occurs at the average value of the random variable. This explains the significance of
‘variance’ of a distribution. A smaller variance means that a random variable mostly
assumes values close to its expected or mean value.

Fig.2.7.1. Gaussian pdf with mean = 0.0 and variance = 1.0 and Gaussian pdf with
mean = 1.5 and variance = 1.0

Version 2, ECE IIT, Kharagpur

 It can be shown with some approximation that about 68% of the values (assumed
by a Gaussian distributed random variable over a large number of unbiased trials) lie
within ± σ around the mean value and about 95% of the values lie within ± 2σ around the
mean value.

The sum of mutually independent random variables, each of which is Gaussian

distributed, is a Gaussian distributed random variable.

Central Limit Theorem

Let X1, X2, X3, ……, XN denote N mutually independent random variables
whose individual distributions are not known and they are not necessarily Gaussian
distributed. If mi , i = 1, 2, ….,N, indicates the mean of the i-th random variable and σi , i
= 1, 2, ….,N, indicates the variance of the i-th random variable, the central limit theorem,
under a few subtle conditions, results in a very powerful and significant inference. The
theorem establishes that the sum of the N random variables (say, X) is a random variable
which tends to follow a Gaussian (or, normal) distribution as N Æ ∞. Further, a) the
mean of the sum random variable X is the sum of the mean values of the constituent
random variables and b) the variance of the sum random variable X is the sum of the
variance values of the constituent random variables X1, X2, X3, ……, XN.

The Central Limit Theorem is very useful in modeling and analyzing several
situations in the study of electrical communications. However, one necessary condition to
look for before invoking Central Limit theorem is that no single random variable should
have significant contribution to the sum random variable.

Generation of Gaussian distributed random numbers using

computers
Gaussian distributed random numbers can be generated by generating random
numbers having uniform distribution. A simple (but not very good always) method of
doing this is by applying central limit theorem. A set of ‘n’ (usually n ≥ 12) independent
uniformly distributed random numbers are generated and they are summed up and scaled
appropriately to result in a Gaussian distributed random variable.

A faster and more precise way for generating Gaussian random variables is
known as the Box-Müller method. Only two uniform distributed random variables say,
X1 and X2, ensure generation of two (almost) independent and identically distributed
Gaussian random variables (N1 and N2). If x1 and x2 are two uncorrelated values assigned
to X1 and X2 respectively, the two independent Gaussian distributed values n1 and n2 are
obtained from the following relations:
n1 = √[ -2.(lnx1).cos2πx2 ]
n2 = √[ -2.(lnx2).sin2πx1 ]
2.7.7

Version 2, ECE IIT, Kharagpur

Error Functions
A few frequently used functions that are closely related to Gaussian distribution
are summarized below. The reader may refer back to this sub-section as necessary.

The error function, erf(x) is defined as:

x
2
erf ( x ) = ∫
2
e − z dz 2.7.8
π 0

It may be noted that the above definite integral is not easy to calculate and tables
containing approximate values for the argument ‘x’ are readily available. A few
properties of the error function erf(x) are noted below:

a) erf(-x) = - erf(x) [Symmetry Property]

∞
2
∫
2
b) as x → + ∞, erf(x) → 1, since e − z dz = 1
π 0
c) if ‘X’ is a Gaussian random variable with mean mx and variance σx2 ,
the probability that X lies in the interval (mx – a, mx +a) is:
a
2σ x
⎛ a ⎞ 2
P ( mx − a < X ≤ mx + a ) = erf ⎜⎜ ∫
2

⎟⎟ = e − z dz 2.7.9
⎝ 2σ x ⎠ π 0

The complementary error function is directly related to the error function erf(x):
∞ ∞
2 2
erfc ( u ) = ∫ exp ( − z ) dz = ∫e dz = erfc ( u ) = 1 − erf ( u )
2 − z2

π u π u

2.7.10

A useful bound on the complementary error function erfc(v):

For large positive v,

exp ( −v 2 ) ⎛ 1 ⎞ exp ( −v 2 )
⎜ −
1 ⎟< erfc ( )<
v
π ⋅ v ⎝ 2v 2 ⎠ π ⋅v
2.7.11

The Q- function or the Marcum function (Fig.2.7.2) is another frequently occurring

function. Considering a standardized Gaussian random variable X with mx = 0 & σx2 = 1,
the probability that an observed value of x will be greater than v is given by the following
Q – function:
∞
1 ⎛ x2 ⎞
Q (v) = ∫ ⎜⎝ − 2 ⎟⎠ dx
exp 2.7.12
2π v

Version 2, ECE IIT, Kharagpur

 f(x)

Fig.2.7.2 The shaded area is measured by the Q-function

We may note that,

1 ⎛ v ⎞
Q ( v ) = erfc ⎜ ⎟ 2.7.13
2 ⎝ 2⎠

And conversely, putting u = v

2
, erfc ( u ) = 2Q ( 2u ) 2.7.14

Problems
Q2.7.1) If a fair coin is tossed ten times, determine the probability that exactly two
heads occur.

Q2.7.2) Mention an example/situation where Poisson distribution may be applied.

Q2.7.3) Explain why Gaussian distribution is also called as Normal distribution.

Version 2, ECE IIT, Kharagpur