Random Variables

INSTRUCTOR: SHAHROKH FARAHMAND

Basic Definitions
A RV is a mapping from the set of events to the real line

2
SLIDES BY: SHAHROKH FARAHMAND
Basic Definitions 2
Remember: We have certain non-measurable sets on the real line
 These sets are of mathematical importance only
 All sets of scientific and engineering importance are measurable
 They are of one of the forms

3
SLIDES BY: SHAHROKH FARAHMAND
Cumulative distribution function (CDF)

F(x) is nondecreasing

4
SLIDES BY: SHAHROKH FARAHMAND
Properties of CDF
CDF is continuous for continuous RVs, that is P[X=a]=0 for every a

5
SLIDES BY: SHAHROKH FARAHMAND
Example 1. Uniform distribution

6
SLIDES BY: SHAHROKH FARAHMAND
Example 2. Binomial distribution

7
SLIDES BY: SHAHROKH FARAHMAND
Example 3. Some binomial probabilities

8
SLIDES BY: SHAHROKH FARAHMAND
 Probability density function (PDF)

For continuous distributions

9
SLIDES BY: SHAHROKH FARAHMAND
Properties of PDF

10
SLIDES BY: SHAHROKH FARAHMAND
Gaussian distribution

Tail probability:

11
SLIDES BY: SHAHROKH FARAHMAND
Various probabilities for Gaussian

12
SLIDES BY: SHAHROKH FARAHMAND
 Rayleigh distribution

Example. Amplitude of NLOS fading channels

13
SLIDES BY: SHAHROKH FARAHMAND
 Exponential distribution

Example. Power of NLOS fading channels

14
SLIDES BY: SHAHROKH FARAHMAND
Uniform distribution

Used in cases that we do not have much info on

parameter X
Ex. Reception time of a random call on a [0,T]
Remember number of calls were modelled as
Poisson
Ex. Position of a cellular user in BS coverage area

15
SLIDES BY: SHAHROKH FARAHMAND
 Laplacian distribution

Main differences with Gaussian:

1) Heavier tail
2) Sharp raise in origin

Applications:
1) Modelling outliers
2) Prior for sparse recovery

16
SLIDES BY: SHAHROKH FARAHMAND
Chi-Squared distribution

Applications and properties

17
SLIDES BY: SHAHROKH FARAHMAND
Maximum ratio combining
Consider a single antenna Tx transmitting to a m antenna Base
Station (BS)
Channel gain between Tx and i’th antenna:
NLOS model (known at Rx):
Tx symbol: Received at antenna i:
Noise:

18
SLIDES BY: SHAHROKH FARAHMAND
MRC 2

Equivalent channel Equivalent noise

Equivalent channel is chi-squared with m degrees of freedom

MRC is the optimum combiner for fading AWGN channels in terms of
maximizing the SNR

19
SLIDES BY: SHAHROKH FARAHMAND
Probability of error in support recovery
Consider the following problem

Vector x has two zeros and a one, but we do not know which entry is
one. Want to find the one entry from knowledge of y,A. We apply the
following decision rule

Suppose WLOG that first element is nonzero

20
SLIDES BY: SHAHROKH FARAHMAND
Probability of error in support recovery 2

A common practice is to use random matrices A with IID entries

21
SLIDES BY: SHAHROKH FARAHMAND
 Discrete RVs
Examples. Bernoulli, Binomial, Poisson
 For Bernoulli: P[X=0]=P[f], P[X=1]=P[s]
 For Binomial

 For Poisson,

The PDF for discrete RVs is called

probability mass function (PMF)

Unified framework:

22
SLIDES BY: SHAHROKH FARAHMAND
Mixed RVs
A mix of continuous/discrete RVs.

23
SLIDES BY: SHAHROKH FARAHMAND
Geometric distribution
Repeat Bernoulli trials until first success is observed

Applications:
1) MAC layer Tx with ARQ (Remember Erasure channel)
2) Random Access: Slotted Aloha
3) End-t—End transport layer protocols (e.g., TCP)

24
SLIDES BY: SHAHROKH FARAHMAND
Example. Mixed RVs

25
SLIDES BY: SHAHROKH FARAHMAND
Example. Mixed RV 2

26
SLIDES BY: SHAHROKH FARAHMAND
Example. Mixed RV 3

27
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF Examples

28
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF Examples 2

29
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF Examples 3

30
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF Examples 4

31
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF Examples 5

32
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF Examples 6

33
SLIDES BY: SHAHROKH FARAHMAND
Joint CDFs/PDFs

Just integrate over the region that you want to compute its probability
How do we compute marginal from joint?

34
SLIDES BY: SHAHROKH FARAHMAND
Joint PDF/CDFs
Two ways to compute F(x):

For any set A

35
SLIDES BY: SHAHROKH FARAHMAND
Independent RVs
How do we obtain joint from marginal?
If X,Y are independent
 Subsequently

For example

36
SLIDES BY: SHAHROKH FARAHMAND
Examples for joint distribution 1

Question. Structure of PDF?

Question. Can we integrate separately?

37
SLIDES BY: SHAHROKH FARAHMAND
Example for joint distribution 2

38
SLIDES BY: SHAHROKH FARAHMAND
Examples for joint distribution 3

39
SLIDES BY: SHAHROKH FARAHMAND
Further examples 1

40
SLIDES BY: SHAHROKH FARAHMAND
Further examples 2

41
SLIDES BY: SHAHROKH FARAHMAND
Further examples 3

42
SLIDES BY: SHAHROKH FARAHMAND
Further examples 4

43
SLIDES BY: SHAHROKH FARAHMAND
Further examples 5

44
SLIDES BY: SHAHROKH FARAHMAND
Further examples 6

45
SLIDES BY: SHAHROKH FARAHMAND
Further examples 6

46
SLIDES BY: SHAHROKH FARAHMAND
Further examples 7

47
SLIDES BY: SHAHROKH FARAHMAND
Jointly Gaussian distribution

48
SLIDES BY: SHAHROKH FARAHMAND
Conditional PDF for continuous RVs

49
SLIDES BY: SHAHROKH FARAHMAND
Conditional JG distribution

50
SLIDES BY: SHAHROKH FARAHMAND
Conditional density of exponential

This conditional distribution is a shift exponential. It begins at t

instead of zero.
Interpretation. Suppose arrival time of a packet at a node has an
exponential distribution, provided that packet has not arrived till
time t, the arrival distribution remains exponential. This is referred to
as the memoryless property of exponential distribution.
51
SLIDES BY: SHAHROKH FARAHMAND
Thank You!

52