chapter 4 random signals and systemstugnajk/elec3800_ch4_11s.pdf · random signals and systems...

87
Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer Engineering Auburn University

Upload: others

Post on 16-Mar-2020

15 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

Random

Signals and S

ystems

Chapter 4

JitendraK

Tugnait

James B

Davis P

rofessor

Departm

ent of Electrical &

Com

puter Engineering

Auburn U

niversity

Page 2: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 3: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 4: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 5: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 6: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

2A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Statistics

•P

robability and Random

Variables

»A

ssume w

e know how

the RV

behaves –either by

assumption or physics of the problem

.

•S

tatistics»

Analysis of data (specific values of an R

V)

Page 7: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

3A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Applications of S

tatistics

•S

ampling T

heory»

How

do we select sam

ples from a large population?

»E

xamples

–P

ublic opinion polls

–IC

manufacturing

•H

ypothesis Testing

»H

ow do w

e decide which of tw

o hypotheses are true?

»D

oes my electronic fuel injector really im

prove gas mileage?

•L

inear Regression

»D

erive a linear equation that describes the data.

»D

oes my autom

atic target detector work as w

ell as a manual

one?

Page 8: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

4A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Term

inology

•Population (N

Collection of data being studied–

All registered voters

–A

ll IC’s from

a manufacturing line

•S

ample (n)

»P

art of the population that is selected at random–

Voters contacted in a survey

–IC

’s selected for test

•N

ote:»

N=

population size»

n = sam

ple size

Page 9: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

5A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple –A

bin with 1000 R

esistors

0100

200300

400500

600700

800900

10000

200

400

600

800

1000

1200

1400

Resistor N

umber

Resistor Values ()

What is the m

ean of these resistors?

Random

ly select 3 resistors

We w

ant to estimate the population m

ean.T

he average of these three values is called the sample m

ean.W

e use the sample m

ean to estimate the population m

ean.

Page 10: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

6A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

•L

et Xbe the value of a resistor from

this bin of 1000 resistors

•S

elect 3 resistors at random

•T

he mean of this sam

ple (called the sample

mean) is:

3

11

11

11963

n

ii

ii

xx

xn

123

1133

1205

1250

xxx

Page 11: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

7A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

•H

ow m

any resistors do we need in our sam

ple to ensure that the sam

ple mean that is w

ithin ±1

of the true mean?

Page 12: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

8A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

More T

erminology / N

otation

Notation: (capital letters; R

V’s, low

er case: particular value)

random variable representing the population

an arbitrary sample taken from

the population

a particular value of

a particular sample taken from

the population

true (population)

ii

XXxX

xXm

ean (an unknown constant,

an RV

)

ˆ theoreticalsam

ple mean

actual sample m

ean (the mean of a particular sam

ple)

not

Xx

Page 13: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

9A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

•U

sually we are interested in the m

ean of an arbitrary set of sam

ples, Xi

»N

ote that is itself a random variable

•S

uppose we picked another set of 3 resistors from

the batch

•T

he mean of a particular sam

ple depends on which 3

resistors you pick.

1

n

ii

XX

n

123

1275

1198

1185

xxx

the mean of this sam

ple is 1219.33

(previousm

eanw

as1196) x

Page 14: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

10A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

010

2030

4050

6070

8090

1000

200

400

600

800

1000

1200

1400

Trial

Sample Mean ()S

ample m

eans for n=3

Page 15: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

11A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Tw

o Expressions for the S

ample M

ean

•N

ote we have tw

o expressions for the sample m

ean

•T

he mean of n

arbitrary samples

»A

random variable used for theoretical analysis

•T

he mean of n

actual samples

»A

number com

puted from the data

•is a particular value of

»Just like in random

variables, e.g. P(X

<x)

1

n

ii

XX

n

1

1n

ii

xx

n

X̂x

Page 16: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

12A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

•W

hat is the mean of the sam

ple mean ?

•O

n average, the sample m

ean equals the true m

ean»

This is called an unbiased

estimator

Mean of the S

ample M

ean

1

11

11

nn

n

ii

ii

i

EX

EX

EX

XX

nn

n

population mean (true m

ean)

where

is an arbitrary sample from

the population i

XX

Page 17: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

13A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

010

2030

4050

6070

8090

1000

200

400

600

800

1000

1200

1400

Trial

Sample Mean ()

Sam

ple means for n=

3

Exam

ple

Population m

ean =

1200

How

precisely can we estim

ate the mean?

How

precisely can we estim

ate the mean?

Page 18: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

14A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Variance of the S

ample M

ean

•O

n average the sample m

ean equals the true m

ean

•H

ow m

uch does the sample m

ean vary about the true m

ean?»

How

precise is the sample m

ean?

•W

e need to compute the variance of the sam

ple m

ean

Page 19: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

15A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Variance of the S

ample M

ean

Assum

e N>

>n

(removing n

samples does not alter the pop. m

ean)

2

2

1

Var

n

ii

XE

XX

n

22

22

11

11

2

2

11

assuming

and are independent

for

for

nn

nn

ij

ij

ij

ij

ij

ij

EX

XX

EX

XX

nn

XX

Xi

jE

XX

Xi

j

22

22

2

Var

Xn

Xn

nX

Xn

22

2X

X

nn

Page 20: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

16A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

True variance =

σ2

= 50

2=

2500

22

2500ˆ

Var(

)833.33

3

ˆˆ

Std(

)V

ar()

28.87

Xn

XX

22

2500ˆ

Var(

)83.33

30

ˆˆ

Std(

)V

ar()

9.13

Xn

XX

010

2030

4050

6070

8090

1000

200

400

600

800

1000

1200

1400

Trial

Sample Mean ()

Sam

ple means for n=

30

010

2030

4050

6070

8090

1000

200

400

600

800

1000

1200

1400

Trial

Sample Mean ()

Sam

ple means for n=

3

Page 21: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

17A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

•T

he variance of the sample m

ean is

»T

he sample m

ean gets closer to the true mean as the

sample size increases

•T

he above formula w

orks when N

>>

n.»

If the population size is small, it also w

orks when

the samples can be returned to the population after

testing.–

Called sam

pling with replacem

ent.

Variance of the S

ample M

ean

2

ˆV

ar=

X

n

Page 22: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

18A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Sam

pling Without R

eplacement

•W

hen the population size is small and the

samples are not replaced, the m

ean of the rem

aining population may be different from

the original population. In this case

2

Var

=-1

Nn

Xn

N

Page 23: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

19A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:endless production line produces diodes random

ly tested forreverse current and forw

ard current. If

has truem

ean and variance of , how

many diodes m

ust betested to get a sam

ple mean w

ith 5% of the true m

ean?

1I

1I

1I

612

10&

10A

61

128

Var

0.0510

105

10

400 I

nn

,souse

Var(

)N

nX

n

Page 24: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

20A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:A

production line produces 30 diodesrandom

ly tested without

replacementfor reverse current I-1

and forward current I1 . If I-1

has true mean and variance of 10

-6and 10

-12A

, how m

any diodes m

ust be tested to get a sample m

ean with 5%

of the true mean?

61

128

0.0510

1030

510

301

27.9728

Var

I

n

n

n

issm

all,souse

Var(

)1

Nn

NX

nN

Note that

when N

=n, the

variance of the sam

ple m

ean is zero.

Page 25: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

21A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple –A

bin with 1000 R

esistors

0100

200300

400500

600700

800900

10000

200

400

600

800

1000

1200

1400

Resistor N

umber

Resistor Values ()

We know

how to estim

ate the mean.

What is the variance of these resistors?

Page 26: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

22A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Sam

ple Variance

221

1

n

ii

SX

Xn

•S

ince we can estim

ate the population mean, w

e can estimate the

population variance with the sam

ple variance

How

ever,

2

22

1

XX

nE

Sn

This estim

ate of the population variance is biased.

Instead, we use

22

211

11

n

ii

nS

SX

Xn

n

This is called the sam

ple variance.

2

2XE

S

This estim

ate of the sample variance is unbiased.

Page 27: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

23A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple (n=3)

010

2030

4050

6070

8090

1000

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

Trial

Sample Variance (2)

Sam

ple variances for n=3

Population

variance =

25002

Page 28: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

24A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple (n=30)

010

2030

4050

6070

8090

1000

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

Trial

Sample Variance (2)

Sam

ple variances for n=30

Population

variance =

25002

Page 29: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

25A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Variance of the S

ample V

ariance

•T

he variance of the sample variance is

»W

here μ4

is the fourth central mom

ent of the population.

•T

his formula is valid only w

hen N>

>n

•A

measure of how

much the sam

ple variance fluctuates around the population variance

44

22

Var

1

nS

n

Page 30: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

26A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Higher-O

rder Central M

oments

•G

aussian

•U

niform»

Can assum

e, without loss of generality that:

–the range is -Δ

x/2 to Δx/2

–n

is even (odd central mom

ents are zero)

11

11

12

2

1

22

12

(1)

(1)

22

(1)2

(1)2

xx

nn

nn

nn

nn

nx

x

n

n

xx

xx

xE

XX

EX

xdx

xn

xn

xn

x

x

n

0 odd

13

5(

1) even

n

n

nE

XX

nn

Page 31: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

27A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Com

putational Formula for S

ample V

ariance

2

2

21

12

1

1

22

12

11

1w

here

Or,

ˆ

1

nn

ii

ni

ii

i

n

ii

n

ii

nx

x

sx

xn

nn

xx

n

xn

x

sn

N

ote sample m

ean, not Σx

i

Page 32: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

28A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:

207202184204206198197213191201

101

102

1

2

22

11

2

Sam

ple mean:

12003

200.310

10

401,825

Sam

ple variance:

11

11

110

401,825200.3

101

101

69.34 ii

ii

nn

ii

ii

xx

x

ns

xx

nn

n

Page 33: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

29A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:

207202184204206198197213191201

If the sample size is sm

all compared to the

population size, the true variance is 75, and the true 4

thcentral m

oment is 16875,

2

Variance of the sam

ple mean:

75ˆ

Var(

)7.5

10X

n

4

22

42

2

Variance of the sam

ple variance:

()

10(16,87575

)V

ar()

1,388.9(

1)(10

1)

nS

n

Page 34: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

30A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:S

uppose we sam

ple a random w

aveform of infinite duration

at equally spaced intervals.A

ssume the w

aveform has a true m

ean of 10 and true variance of9, and the sam

ples are independent.

What is the variance of the sam

ple mean ?

2

Var

Xn

n

29

100.02

0.04225

nn

How

many sam

ples do we need to estim

ate the mean w

ithin 2%of its true value?

Page 35: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

31A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•S

uppose we use n=

225, and compute a sam

ple m

ean. Are w

e guaranteed that our value is w

ithin 2% of the true m

ean?»

NO

!!

•W

hat is the probability that is within 2%

of the true m

ean?»

We need the P

DF

ofto answ

er this question

Page 36: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

32A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•In this case, the sam

ple is large (n=225) and the

central limit theorem

tells us that is Gaussian

regardless of what the P

DF’s of X

i are.»

Recall from

a previous slide that

10.210

9.810

ˆ9.8

10.210.2

9.80.2

0.2

11

21

12

0.84131

0.6826

PX

FF

2

ˆ10

Var(

)0.04

225

XXn

68% chance that the

sample m

ean is w

ithin 2% of the

true mean

Page 37: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 38: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 39: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

33A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•T

o answer questions like this, w

e need to know

the probability distributions of:»

The sam

ple mean,

»T

he normalized sam

ple mean:

ˆ (true variance know

n)X

XZ

n

ˆ (true variance unknow

n)X

XT

Sn

Recall that norm

alized random variables have zero m

ean and unit variance. We

normalize a random

variable by subtracting the mean and dividing

by the standard deviation.

Page 40: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

34A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•If the sam

ples are independent, σ2

known, and

»T

he sample size is large

»O

r, the Xi are G

aussian

•T

he sample m

ean is Gaussian w

ith

•T

he normalized

sample m

ean is also Gaussian

(30)

n

ˆV

arE

XX

Xn

X̂X

Zn

Page 41: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

35A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•W

hat if the sample size is large and σ

2

unknown?

»T

he sample m

ean is Gaussian

»C

an use as an estimate of σ

2

–W

hen nis large, should be sm

all

•T

he normalized sam

ple mean is approxim

ately G

aussian

2S

2V

ar()

S X̂X

ZS

n

(30)

n

Page 42: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

36A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:100 bipolar transistors

Want the m

ean value of the current gaintrue m

ean and variances are:

What is the probability that the sam

ple mean is betw

een 119 and 121?

2

120and

25A

S

ˆ119

120120

121120

ˆ(119

121)0.5

0.50.5

(2

2)

(2)(

2)

2(2)

1

2(0.9772)1

0.9544

XP

XPP

Z

Page 43: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

37A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•W

hat do we do w

hen the sample size is sm

all and σ

2is unknow

n?

•T

he normalized sam

ple mean

»is not G

aussian even if the Xi are G

aussian because one R

V, , is divided by another,

X̂X

TS

n

X̂2

S

Page 44: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

38A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•W

hat do we do w

hen the sample size is sm

all and σ

2is unknow

n?

Page 45: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

39A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Probability D

istribution of the Sam

ple Mean

•If »

nis sm

all (n<30)

»σ

2is unknow

n

»X

i are Gaussian

•T

he normalized sam

ple mean

has a student’s t distributionw

ith n-1 degrees of freedom

X̂X

TS

n

Page 46: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

40A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Student’s t D

istribution

12

2

12

12

v

T

vt

ft

vv

v

1

. gam

ma function

vn

12

1 1

2

As n

gets large, the tdistribution approaches the Gaussian distribution.

any 1

!integer

kk

kk

kk

Page 47: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

41A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:C

ompute student’s t density for t=

1 and 4 degrees of freedom

52

22.5

11

14

42

T f

2.5

1.51.5

1.50.5

0.5

1.5

0.51.3293

2.5

1.32931

1.250.2147

41

T f

Page 48: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 49: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

43A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

What does this have to do w

ith beer?

•T

he Student’s tdistribution was

discovered by William

Gossett for

use in statistical quality control

•G

ossett worked for G

uinness B

rewery, w

ho had a strict policy against em

ployees publishing under their ow

n names

•G

ossett published the paper under the pen nam

e “Student”

Page 50: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

44A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

PD

F of the N

ormalized S

ample M

ean

small

large

known

unknown

are Gaussian

iX

are not Gaussian

iX

small

large

known

unknown

ZZ

TZw

ith S w

ith S Z Z

**

*special cases

2

n2

n

Z: G

aussian distributionT

: Student’s tdistribution

Page 51: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

45A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Confidence Intervals

•S

uppose we are random

ly checking the values of resistors on a m

anufacturing line»

True m

ean is 100 and the true variance is 16

2

•If w

e randomly sam

ple a run of resistors and the sam

ple mean is 99

, is there a problem?

•T

o answer this question, w

e need to know the

expected range, or confidence interval, for the sam

ple mean

Page 52: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 53: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 54: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

46A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Confidence Intervals

Define the q-percent confidence intervalas the interval w

ithin w

hich the sample m

ean will lie w

ith q% probability.

The q%

confidence interval for the sample m

ean is:

kis a constant that depends on q

and density of .

ˆk

kX

XX

nn

confidence limits

ˆ100

kX

n

Xk

Xn

qf

xdx

Page 55: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

47A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:large population of resistors

Find 95%

confidence limits for n =

100 100

4

X

951.96 (T

able 4.1 in page 172 of the textbook)q

k

1.96

40.78

100ˆ

99.22100.78

k

n

X

A sam

ple mean of 99

could m

ean a problem w

ith the m

anufacturing line

Page 56: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

48A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:large population of resistors

Find 95%

confidence limits for n =

9

100

4

X

951.96 (T

able 4.1 in page 172 of the textbook)q

k

1.96

42.61

97.39102.61

k

n

X

A w

ider confidence interval corresponds to a poorer

estimate

of the mean

Page 57: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

49A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Confidence Intervals

•T

he confidence interval can also be expressed in terms of the

probability distribution function

•F

or small sam

ple size and unknown variance, use F

Tn-1

instead of Φ

100

ˆ100

ˆ

100

()

()

2(

)1

1(

)2

nn

nn

nn

n

q qP

Xk

XX

k

Xk

XX

kX

XX

qP

Pk

Zk

kk

k

k

Page 58: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 59: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

50A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

•W

hat is kfor a 90%

confidence interval when

n=100?

•W

hat is kfor a 90%

confidence interval when

n=9?

0.901

()

0.951.64

2k

k

8

0.901

()

0.951.86

2T

Fk

k

Page 60: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

51A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:random

ly sampled G

aussian waveform

The 95%

confidence interval is:

2

1094

Xsn

3.1829

3.1829

ˆ10

104

5.22714.773

XX

3

0.951

()

0.9753.182

2T

Fk

k

Page 61: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

52A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Hypothesis T

esting•

A particular car m

odel gets 26 mpg w

ith a standard deviation of 5 m

pg.

•A

new electronic fuel injection system

is thought to improve

gas mileage.

•36 gas m

ileage tests are performed w

ith mean 28.04 m

pg.»

Assum

e σis unaffected

•D

id the electronic fuel injector increase the gas mileage?

•T

o answer this and sim

ilar questions, we need to perform

a hypothesis test.

Page 62: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

53A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Hypothesis T

esting

•W

e make tw

o hypotheses:

•W

e assume that H

0is true (that is w

hy it needs equality), and test if the data is consistent w

ith this assum

ption.

•If not consistent, w

e reject hypothesis H0

and accept hypothesis H

1 .

01

: 26 (called the null hypothesis)

: 26 (called the alternative or research hypothesis)

HX

HX

Page 63: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

54A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

.05

Hypothesis T

esting

•If w

e assume H

0is true, then the sam

ple mean has a

Gaussian distribution w

ith mean 26.

•W

e compute a critical value x

c such that

•It is easier to w

ork with the norm

alized sample m

ean.

ˆ0.95

cP

Xx

5From

table, 1.64

2627.36

36c

cc

zx

z

ˆ26

536

0.95c

XZP

Zz

Page 64: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 65: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

56A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

.05

Hypothesis T

esting

•S

o if the gas mileage is unchanged, 95%

of the sam

ple means w

ill be ≤27.36 m

pg

•T

wo possibilities:

»W

e get lucky (i.e. 28.04 is one of the 5%)

»T

he true mean is higher than 26

•W

e choose the second possibility»

We reject H

0 and accept H1 (the gas m

ileage im

proved)

Page 66: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

57A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Types of T

ests

Right-tailed

test(L

eft-sided test)H

1 : mean >

value

Left-tailed

test(R

ight-sided test)H

1 : mean <

value

Tw

o-tailed test

(Tw

o-sided test)H

1 : mean ≠

value

Page 67: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

58A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Norm

alized Sam

ple Mean

•In the previous exam

ple, we com

puted xc ,

which w

e compared to the sam

ple mean,

•In practice, how

ever, it is usually easier to norm

alize

and then compare z

to zc

•T

he normalized sam

ple mean is called a test

statistic.

x

xx

Xz

n

If

is unknown, use

.

If the sample size is sm

all, the

normalized sam

ple mean w

ill

have a student's t distribution.

s

Page 68: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

59A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Hypothesis T

esting Procedure

•C

onstruct two hypotheses: H

0 and H1

»H

1should be the negation of H

0

»E

quality should be with H

0

•C

ompute test statistic (norm

alized sample m

ean) , assum

ing H0

is true

•C

ompute critical value of test statistic.

•If

(Gaussian)

xX

zn

(Student's

)

xX

ts

nt

~G

aussian

for large

xX

zs

nn

111

accept

(right-tailed)

accept

(left-tailed)

or <

accept (tw

o-tailed)

cc

cc

zz

H

zz

H

zz

zz

H

Page 69: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

60A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Com

puting Critical V

alues

1)R

ight-tailed

2)L

eft-tailed

3)T

wo-tailed

100c

qP

Zz

100c

qz

100

11

1100

100

c

cc

qP

Zz

qq

zz

100c

qz

100

1100

100

cc

cc

cc

qP

zZ

z

qq

zz

zz

21

100c

qz

1

1002c

q

z

*For S

tudent’s t, replace Φw

ith Fγ and z

c with t

γ

Page 70: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

61A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Com

puting Critical V

alues cont’d

Right-tailed

Left-tailed

Tw

o-tailed

ZT

100c

qz

100c

qz

1

1002c

q

z

100T

c

qF

t

100

Tc

qF

t

1

1002T

c

q

Ft

degrees of freedom

Page 71: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

62A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:A

manufacturer claim

s the breakdown voltage of his capacitors is

300 V or greater.

100 capacitors tested:Is this claim

valid at the 99% confidence level?

290Vx

40Vs

01

Left-tailed test

:300 (claim

valid)

:300 (claim

invalid)

Sam

ple size is large (>

30) use w

ith =

290300

2.540

10

0.99

10.99

11

0.99

0.99

2.332.33

2.5

c

cc

c

cc

c

HX

HX

nZ

s

zPZ

zzz

z

zz

zz

0reject

(claim invalid)

H

cz

0accept

H1

acceptH

Page 72: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 73: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

64A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:S

ame as before but w

ith n =9

Sam

ple size is small and σ

2is unknow

n so use:

8

8

8

8

8 290300

0.75(t distribution w

ith 8 degrees of freedom)

409

0.99

10.99

11

0.99

0.992.896

2.896

0.75accept hypothesis (claim

invalid)

c

TcT

c

Tc

cc

c

tPT

t

FtF

t

Ft

tt

zz

cz

0accept

H1

acceptH

Page 74: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 75: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

66A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:(tw

o-tailed test)Z

ener diodes with m

ean breakdown voltage of 10V

. Deviations on

either side are undesirable because they are used as voltage regulators.T

est 100 diodes:Is claim

valid at 95% confidence level?10.3V

x

1.2Vs

01

: 10V

(claim valid)

: 10V

(claim invalid)10.3

10large sam

ple sizeuse

2.51.2

100

compute

such thatc

HX

HX

z

z

0.95c

cP

zZ

z

0.95c

cz

z

1

0.95c

cz

z

21

0.951.95

20.975

1.96c

cc

zz

z

Since 2.5 does not lie in the rangeclaim

not valid. -1.96

z1.96

0accept

H

1accept

H

Page 76: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

67A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

Ex:sam

e as before but with n

=9

8

8

8

10.310

0.751.2

9

0.95

21

0.95

0.9752.306

cc

Tc

Tc

c

tPt

Tt

Ft

Ft

t

Since t =

0.75 lies in the intervalclaim

valid.2.306

2.306t

0accept

H

1accept

H

Page 77: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer
Page 78: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

69A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Curve F

itting and Linear R

egression

Ex:F

our light bulbs are tested to determine the relationship

between lifetim

e and operation voltage.

iV

Hrs

1105

1400

2110

1200

3115

1120

4120

950

ix

iy

Page 79: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

70A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Curve F

itting and Linear R

egression

•W

e would like a m

athematical relationship

between the operation voltage and lifetim

e so w

e could estimate lifetim

e for other operation voltages.»

For exam

ple, 90V, 112V

, 130V

•O

ne way to do this is to fit a straight line to the

data. This is called linear regression.

»W

e can fit other curves as well polynom

ials, F

ourier series.

Page 80: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

71A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Curve F

itting and Linear R

egression

•O

ur data model is y

i = m

xi +

b

•O

nly need two points to solve for m

and b, but we w

ould like to use all the data available.

•D

efine di =

yi –

mx

i –b

•C

hoose mand b to m

inimize the sum

of squared differences

2

1

2

11

1

2

11

1

0

0

n

ii

i

nn

n

ii

ii

ii

i

nn

n

ii

ii

ii

i

SSDy

mx

bm

m

xy

mx

bx

xy

mx

bx

2

1

11

11

0

0

n

ii

i

nn

ii

ii

nn

ii

ii

SSDy

mx

bb

b

ym

xbn

ym

xnb

22

11

nn

ii

ii

i

SSDd

ym

xb

Page 81: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

72A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

11

12

2

11

11

11

nn

n

ii

ii

ii

i

nn

ii

ii

n

iin

ii

nx

yx

y

m

nx

x

by

mx

xx

n

yy

n

Curve F

itting and Linear R

egression

Solving for m

and byields:

slope

y-intercept

Page 82: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

73A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Curve F

itting and Linear R

egression cont’d

For our exam

ple:

111

4

450

4670

521800

n

iin

iin

ii

i n

xyxy

112.5

1167.5

xy

2

1

2

1

50750

5556900

n

iin

ii

xy

2

4521800

4504670

28.64

50750450

1167.528.6

112.54385

28.64385

mbyx

90V-28.6

904385

1811 hrs

112V-28.6

1124385

1181.8 hrs

130V-28.6

1304385

667 hrs

Page 83: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

74A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Correlation C

oefficient

Can com

pute correlation coefficient for dataH

ow correlated is operation voltage w

ith lifetime?

11

1

22

22

11

11

nn

n

ii

ii

ii

i

nn

nn

ii

ii

ii

ii

nx

yx

y

r

nx

xn

yy

Com

pare to the theoretical formula:

For our example:

X

Y

EX

YX

Y

2

2

4521800

4504670

0.98834

50750450

4556900

4670r

highly negatively correlated

Page 84: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

75A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

0.05r0.83

r

0.50r

0.80r

From

Denney Jr., T

S, et al, M

agnetic Resonance in M

edicine, 2003

Page 85: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

76A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

0.09r0.91

r

0.63r

0.86r

Page 86: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

77A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple

0.08r0.74

r

0.58r

0.60r

Page 87: Chapter 4 Random Signals and Systemstugnajk/ELEC3800_ch4_11s.pdf · Random Signals and Systems Chapter 4 Jitendra K Tugnait James B Davis Professor Department of Electrical & Computer

78A

U E

LE

CT

RIC

AL

AN

D C

OM

PU

TE

R E

NG

INE

ER

ING

Exam

ple