[go: up one dir, main page]

0% found this document useful (0 votes)
20 views31 pages

확통1 LectureNote07 on Memoryless Random Processes

Uploaded by

jedem10224
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
20 views31 pages

확통1 LectureNote07 on Memoryless Random Processes

Uploaded by

jedem10224
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 31

Memoryless Random Processes

Memoryless Random Processes


Definition:
Sequence (or function) of random variables which has a special
memoryless property: {𝑋𝑛 , 𝑛 ≥ 1} or {𝑋 𝑡 , 𝑡 ≥ 0}

Examples:
• Arrival example: Number of arrivals to a bank at time 𝑡
• Queuing example: Length of a waiting line at a bank at time 𝑛

• Gambler’s ruin: The gambler’s fortune at time 𝑛 (Markov


Chains)
• Engineering example: Signal corrupted with noise at time 𝑡
2
Bernoulli Process
• A sequence of independent Bernoulli trials.
• At each trial:
− 𝐏 success = 𝐏 𝑋 = 1 = 𝑝
− 𝐏 failure = 𝐏 𝑋 = 0 = 1 − 𝑝
• Examples:
− Sequence of ups and downs of stock markets
− Sequence of lottery wins/losses
− Arrivals (each second) to a bank.

3
Number of successes S in n time slots

𝑛
• 𝐏 𝑆=𝑘 = 𝑘
𝑝𝑘 (1 − 𝑝)𝑛−𝑘 , 𝑘 = 0,1, ⋯ , 𝑛.
(Binomial)

• Mean: 𝐄 𝑆 = 𝑛𝑝

• Variance: 𝐕 𝑆 = 𝑛𝑝(1 − 𝑝)

4
Interarrival Times

• 𝑇1 : number of trials until first success or the first


interarrival time
• Interarrival times have a memoryless property

• 𝐏 𝑇1 = 𝑡 = 𝑝(1 − 𝑝)𝑡−1 , 𝑡 = 1,2, ⋯. (Geometric)

• Mean: 𝐄 𝑇1 = 1/𝑝

• Variance: 𝐕 𝑇1 = (1 − 𝑝)/𝑝2
5
Fresh Start and Memoryless Properties

• Fresh Start:
Given 𝑛, the process is said to fresh start if the future sequence
𝑋𝑛+1 , 𝑋𝑛+2 , ⋯ is also a Bernoulli process and is independent
of the past {𝑋𝑘 , 0 ≤ 𝑘 ≤ 𝑛}.

• Memoryless:
Suppose we observe the process for 𝑛 times and no success
occurred. Then the process is said to be memoryless if the
PMF of the remaining time 𝑇1 for the first arrival does not
depend on 𝑛:
𝐏 𝑇1 = 𝑛 + 𝑡 𝑇1 > 𝑛 = 𝐏(𝑇1 = 𝑡)

6
𝑡ℎ
Time of the 𝑘 Arrival (1)

• 𝑌𝑘 : number of trials until the 𝑘 𝑡ℎ success


• 𝑇𝑘 = 𝑌𝑘 − 𝑌𝑘−1 , 𝑘 = 2,3, ⋯ ; 𝑘 𝑡ℎ interarrival time

• It follows that:
𝑌𝑘 = 𝑇1 + 𝑇2 + ⋯ + 𝑇𝑘

7
Time of the 𝑘 𝑡ℎ Arrival (2)

𝑡−1
• 𝐏[𝑌𝑘 = 𝑡] = 𝑘−1
𝑝𝑘 (1 − 𝑝)𝑡−𝑘 , 𝑡 = 𝑘, 𝑘 + 1, ⋯ (Pascal)

• Mean: 𝐄 𝑌𝑘 = 𝑘/𝑝

• Variance: 𝐕 𝑌𝑘 = 𝑘(1 − 𝑝)/𝑝2

8
Bernoulli Process: Summary

• Discrete time process; success probability in each slot = 𝑝


• PMF of the number of arrivals in 𝑛 time slots: Binomial
• PMF of interarrival time: Geometric
• PMF of time to 𝑘 𝑡ℎ arrival: Pascal
• Memoryless
• Then, what about continuous arrival times? See next
− Example: no. of customer arrivals to a bank.
9
Poisson Process: Definition

• Let 𝐏 𝑘, 𝜏 the probability of 𝑘 arrivals (or events) in an


interval of duration 𝜏.
• Assumptions:
− The number of arrivals in disjoint
. time intervals are
independent (This is called independent increments)
− For very small 𝛿, we have
1 − 𝜆𝛿, if 𝑘 = 0
− 𝐏 𝑘, 𝛿 ≈ ቐ𝜆𝛿, if 𝑘 = 1
0, if 𝑘 ≥ 2
where 𝜆 = “arrival rate” of the process
− Note that the number of arrivals depends only on the
duration 𝛿 (This is called stationary increments) 10
From Bernoulli to Poisson (1)

• Bernoulli: Arrival probability in each time slot = 𝑝

• Poisson: Arrival probability in each 𝛿-interval = 𝜆𝛿

• Let 𝑛 = 𝑡/𝛿 and 𝑝 = 𝜆𝛿 = 𝜆𝑡/𝑛 :

Number of arrivals Number of successes


in a 𝑡-interval = in 𝑛 time slots
(Poisson) (Binomial)
11
From Bernoulli to Poisson (2)

• Number of arrivals in a 𝑡-interval as 𝑛 → ∞ :

12
PMF of the Number of Arrivals

• 𝑁 = the number of arrivals in a 𝜏-interval:


𝐏 𝑁 = 𝑘 = 𝐏 𝑘, 𝜏 = (𝜆𝜏)𝑘 𝑒 −𝜆𝜏 /𝑘!, 𝑘 = 0,1, ⋯ (Poisson)

• Mean: 𝐄[𝑁] = 𝜆𝜏

• Variance: 𝐕 𝑁 = 𝜆𝜏
𝑠)
• Transform: 𝑀𝑁 𝑠 = 𝑒 −𝜆𝜏(1−𝑒

13
Email Example

• You get email according to a Poisson process, at a rate of 0.4


messages per hour (𝜆 = 0.4, unit time = an hour). You check
your email every thirty minutes (𝜏 = half an hour).

0.4
− 𝜆𝜏 = = 0.2
2

(0.2)0 𝑒 −0.2
− P[no new messages] = = 𝑒 −0.2 = 0.819
0!

(0.2)1 𝑒 −0.2
− P[one new message] = = 0.2 ∙ 𝑒 −0.2 = 0.164
1!

14
Interarrival Time (1)

𝑡
• 𝑌1 : time of the first arrival

• “First order” interarrival time:


𝑓𝑌1 𝑡 = 𝜆𝑒 −𝜆𝑡 , 𝑡 ≥ 0. (Exponential)

• Why ?
𝐏 𝑌1 ≤ 𝑡 = 1 − 𝐏 0, 𝑡 = 1 − 𝑒 −𝜆𝑡

15
Interarrival Time (2)

• Fresh Start Property: The time of the next arrival is


independent from the past.

• Memoryless property: Suppose we observe the process


for 𝑇 seconds and no success occurred. Then the density of
the remaining time for arrival is exponential.

• Email Example: You start checking your email. How long


will you wait, in average, until you receive your next email?
1
𝐄 𝑌1 = = 2.5 hours
0.4
16
Time of the 𝑘 𝑡ℎ Arrival (1)

𝑌𝑘

• 𝑌𝑘 : time of the 𝑘 𝑡ℎ arrival

• 𝑇𝑘 = 𝑌𝑘 − 𝑌𝑘−1 = 𝑘 𝑡ℎ interarrival time, 𝑘 = 2,3, ⋯

• It follows that:
𝑌𝑘 = 𝑇1 + 𝑇2 + ⋯ + 𝑇𝑘

17
Time of the 𝑘 𝑡ℎ Arrival (2)
𝑘 𝑘+1

𝑌𝑘
𝑁𝑡 events

𝜆𝑡 𝑘−1 𝜆𝑒 −𝜆𝑡
• 𝑓𝑌𝑘 𝑡 = , 𝑡≥0 (Erlang of order 𝒌)
𝑘−1 !
∞ ∞
(𝜆𝑡)𝑛 𝑒 −𝜆𝑡
(∵)𝐹𝑌𝑘 𝑡 = 𝐏 𝑌𝑘 ≤ 𝑡 = 𝐏 𝑁𝑡 ≥ 𝑘 = ෍ 𝐏(𝑛, 𝑡) = ෍
𝑛!
𝑛=𝑘 𝑛=𝑘
∞ ∞
𝑛𝜆(𝜆𝑡)𝑛−1 −𝜆𝑡 𝜆𝑡 𝑛 −𝜆𝑡
𝑓𝑌𝑘 (𝑡) = ෍ 𝑒 −෍ λ𝑒
𝑛! 𝑛!
𝑛=𝑘 𝑛=𝑘
∞ (λ𝑡)𝑛−1 ∞ 𝜆𝑡 𝑛 −λ𝑡
= ෍ λ𝑒 −λ𝑡 − ෍ λ𝑒
𝑛=𝑘 𝑛 −1 ! 𝑛=𝑘 𝑛!
𝑘−1
(λ𝑡)
= λ𝑒 −λ𝑡
𝑘−1 !
18
Poisson Process: Summary

• Number of arrivals in disjoint time intervals are independent


1 − 𝜆𝛿, if 𝑘 = 0
𝐏 𝑘, 𝛿 ≈ ቐ𝜆𝛿, if 𝑘 = 1 (for very small 𝛿)
0, if 𝑘 ≥ 2
𝐏 𝑘, 𝜏 = (𝜆𝜏)𝑘 𝑒 −𝜆𝜏 /𝑘! (Poisson)

• Interarrival times (𝑘 = 1):


𝑓𝑌1 𝑡 = 𝜆𝑒 −𝜆𝑡 , 𝑡 ≥ 0 (Exponential)

• Time to the 𝑘 𝑡ℎ arrival:


𝑓𝑌𝑘 (𝑡) = 𝜆(𝜆𝑡)𝑘−1 𝑒 −𝜆𝑡 / 𝑘 − 1 ! , 𝑡 ≥ 0 (Erlang)
19
Bernoulli vs. Poisson

Bernoulli Poisson
Arrival Times Discrete Continuous
Arrival Rate 𝑝/trial 𝜆/unit time
PMF of Arrival Numbers Binomial Poisson
PMF of Interarrival Time Geometric Exponential
PMF of the 𝑘th Arrival Time Pascal Erlang

20
Random Incidence Paradox (1)
𝑋𝑠
𝑡𝑛−2 𝑡𝑛−1 𝑡∗ 𝑡𝑛 𝑡𝑛+1
𝑋𝑛−1 𝑋𝑛+1
Elapsed time Chosen time instant Residual time
𝑋𝑒 = 𝑡 ∗ − 𝑡𝑛−1 𝑌 = 𝑡𝑛 − 𝑡 ∗
• Let us fix a time instance 𝑡 ∗ and consider the length 𝑋𝑠 of
the selected interarrival interval that contains 𝑡 ∗ .
• We argue that 𝑋𝑠 is the length of a typical interarrival time,
and is exponentially distributed, but this is false.
• Let [𝑡𝑛−1 , 𝑡𝑛 ] be the interval that contains 𝑡 ∗ , so that 𝑋𝑠 =
𝑡𝑛 − 𝑡𝑛−1 . We split 𝑋𝑠 into two parts,
𝑋𝑠 = 𝑋𝑒 + 𝑌 = 𝑡 ∗ − 𝑡𝑛−1 + (𝑡𝑛 − 𝑡 ∗ )
where 𝑋𝑒 = 𝑡 ∗ − 𝑡𝑛−1 is the elapsed time since the (𝑛 − 1)th
arrival, and 𝑌 = 𝑡𝑛 − 𝑡 ∗ is the residual time until the 𝑛th
arrival. 21
Random Incidence Paradox (2)
• By the independence property of the Poisson process, the two
parts are independent and are exponentially distributed rvs
• By the memoryless property, the Poisson process starts fresh
at time 𝑡 ∗ , and therefore the residual time 𝑌 is exponential
with parameter 𝜆. The elapsed time 𝑋𝑒 is also exponential
with parameter 𝜆, because
𝐏 𝑋𝑒 > 𝜏 = 𝐏 no arrival in 𝑡 ∗ − 𝜏, 𝑡 ∗ = 𝐏 0, 𝜏 .
• We therefore show that the selected interval 𝑋𝑠 is the sum of
two independent exponential rvs with parameter 𝜆, i. e.,
Erlang of order 2, with mean 2/𝜆.
• The key issue is that an observer who arrives at an arbitrary
time is more likely to fall in a large rather than a small
interarrival interval. Thus, the expected length seen by the
observer is not 1/𝜆 but a higher value, 2/𝜆.
22
Random Incidence Paradox (3)
• Let us derive the density 𝑓𝑠 𝑥 of the selected interval 𝑋𝑠 . Our
basic observation is that longer intervals between arrival points
occupy larger segment of time axis than do shorter intervals,
and therefore it is more likely that 𝑡 ∗ will fall in a long interval.
• Thus, for the selected interval 𝑋𝑠 , we have
𝑓𝑋𝑠 𝑥 𝑑𝑥 = 𝐾𝑥𝑓 𝑥 𝑑𝑥
where 𝑓(𝑥) is the pdf of common interarrival times {𝑋𝑛 }, the
left-hand side is 𝐏{𝑥 < 𝑋𝑠 ≤ 𝑥 + 𝑑𝑥} and the right-hand side is
the linear expression about interval length 𝑥 and the relative
occurrence of such intervals 𝐏 𝑥 < 𝑋𝑛 ≤ 𝑥 + 𝑑𝑥 = 𝑓 𝑥 𝑑𝑥.
Integrating both sides, we find 𝐾 = 1/𝑚 with 𝑚 = 𝐄(𝑋𝑛 ).
• Thus,𝑓𝑋𝑠 𝑥 is given in terms of the common density 𝑓 𝑥 of
interarrival times by
𝑥𝑓(𝑥)
𝑓𝑋𝑠 𝑥 = .
𝑚 23
Random Incidence Paradox (4)
• Let us proceed to find the density 𝑓𝑌 𝑦 of residual time 𝑌. If we
are told that 𝑋𝑠 = 𝑥, then the conditional CDF is given by
𝑦
𝐹𝑌|𝑋𝑠 =𝑥 𝑦 𝑥 = , 0 ≤ 𝑦 ≤ 𝑥.
𝑥
Then, for 0 ≤ 𝑦 ≤ 𝑥,
1
𝑓𝑌|𝑋𝑠 =𝑥 𝑦|𝑥 = ,
𝑥
𝑓 𝑥
𝑓𝑋𝑠 ,𝑌 𝑥, 𝑦 = 𝑓𝑋𝑠 𝑥 ∙ 𝑓𝑌|𝑋𝑠 =𝑥 𝑦|𝑥 = .
𝑚
• Integrating 𝑓𝑋𝑠 ,𝑌 𝑥, 𝑦 over 𝑥 we obtain 𝑓𝑌 𝑦 , namely

𝑓 𝑥 𝑑𝑥
𝑓𝑌 𝑦 = න .
𝑥=𝑦 𝑚
This immediately gives the final result:
1 − 𝐹(𝑦)
𝑓𝑌 𝑦 = .
𝑚
• It gives the density of residual time in terms of the common
distribution 𝐹(∙) of interarrival times and its mean 𝑚. 24
Random Incidence of Non-Poisson Arrival
• Assume that buses arrive at a station deterministically, on the
hour, and 5 minutes after the hour. Then the interarrival times
alternate between 5 and 55 minutes. The average interarrival
time is 30 minutes.
• A person shows up at the bus station at a random time.
Question: What is the expected time until the next bus arrival?

• Answer: Such a person falls into an interarrival interval of


length 5 with probability 1/12, and an interarrival interval of
length 55 with probability 11/12. The expected length is
1 11
5∙ + 55 ∙ = 50.83,
12 12
which is considerably larger than 30, the average interarrival
time.
25
Merging of Poisson Processes

• Sum of independent Poisson random variables is Poisson.


• Sum of independent Poisson processes is Poisson.
• What is the probability that the next arrival comes from the
first process?
𝜆1 𝛿 𝜆1
=
𝜆1 𝛿 + 𝜆2 𝛿 𝜆1 + 𝜆2

26
Splitting of Poisson Processes

• Each message is routed along the first stream with


probability 𝑝, and along the second stream with
probability 1 − 𝑝.

− Routing of different messages are independent.


− Each output stream is Poisson.

27
Example: Email Filter (1)

• You have incoming email from two sources: valid email, and
spam. We assume both to be Poisson.
• You receive, on average, two valid emails per hour, and one
spam email every 5 hours: 𝜆1 = 2, 𝜆2 = 0.2
• Total incoming email rate 𝜆 = 𝜆1 + 𝜆2 = 2.2 emails per
hour
• 𝐏{a received email is spam} = 𝜆2ൗ𝜆1 +𝜆2 = 0.2Τ2.2 ≈ 0.09

28
Example: Email Filter (2)

• You install a spam filter, that filters out spam email correctly
80% of the time, but also identifies a valid email as spam 5%
of the time: 𝑝 = 0.95, 𝑞 = 0.2

• Inbox email rate = 𝑝𝜆1 + 𝑞𝜆2 = 0.95 × 2 + 0.2 × 0.2 = 1.94

• Spam folder email rate = 2.2 − 1.94 = 0.26

29
Example: Email Filter (3)

• 𝐏{an email in the inbox is spam}


𝑞𝜆2 (0.2 × 0.2)ൗ
= ൗ(𝑝𝜆 + 𝑞𝜆 ) = 1.94 ≈ 0.02
1 2

• 𝐏{an email in the spam folder is valid}


= (1−𝑝)𝜆1ൗ((1−𝑝)𝜆1 +(1−𝑞)𝜆2 ) = (0.05×2)ൗ0.26 ≈ 0.38

• Every how often should you check your spam folder, to find
one valid email, on average?
1
𝐄 𝑁 = 𝜆1 1 − 𝑝 𝜏 = 1 ⇒ 𝜏 = 0.05∙2 = 10 hours
30
Homework #7
Textbook “Introduction to Probability”, 2nd Edition, D. Bertsekas and J. Tsitsiklis
Chapter6 p.326-p.338, Problems 2, 3, 9, 10, 11, 12, 16
Due date: 아주BB 과제출제 확인

31

You might also like