Prospects and challenges for spectrum sharing by cognitive ...sahai/Presentations/Harvard09.pdfAnant...

Prospects and challenges for spectrum sharing bycognitive radios

Anant Sahaipresenting joint work with students:

Mubaraq Mishra Rahul Tandra Kristen Woyachalong with my BU Collaborators:

George Atia Venkatesh Saligrama

BWRC and Wireless Foundations CenterU.C. Berkeley

Boston University

Support from the National Science Foundation, C2IT, and Sumitomo

Harvard EE Seminar

Anant Sahai (UC Berkeley) Cognitive Radio 02/19/2009 1 / 1

Spectrum, spectrum, everywhere, but . . .


Outline

How much usable white-space is there?How can we understand sensing?

Light-handed regulation: identity

Light-handed regulation: deterrence


How much white-space is there in a single band?


Consider channel 39 . . .


The pollution perspective: 15dB above noise


5dB above noise with -35dB spillover from next door


The protection perspective


The protection perspective: 4W


The protection perspective: 20W


The protection perspective: 100kW


The protection perspective: 1MW


How much to protect? 0.1dB margin


How much to protect? 1.0dB margin


How much to protect? 10dB margin


How much to protect? 1.0dB margin vs 5db pollution


How much to protect? 1dB margin with adjacent


Can we sense these holes?


Can we sense these holes? 90%


Can we sense these holes? 99%


Can we sense these holes? FCC rules


How much white-space is there across bands?

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of channels recovered

CC

DF

Actually available by area

Actually available by population

−114dBm

rule by area

−114dBm

rule by population


. . . If we account for adjacent-channel effects?

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


CC

DF



−114dBm

rule by area−114dB

m rule

by population


How much white-space is there across bands?

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


CC

DF



−114dBm

rule by area

−114dBm

rule by population

0 5 10 15 20 25 30 35 40 45 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


CC

DF



−114dBm

rule by area−114dB

m rule

by population

Detection Scheme/RuleBy Area By Population

LVHF HVHF LUHF HUHF LVHF HVHF LUHF HUHF2,5,6 7-13 14-51 52-69 2,5,6 7-13 14-51 52-69

Pollution (5dB,45dB adj.) 1.6 1.63 15.6 15.8 1.62 0.729 6.63 14.8Geolocation 1.52 2.86 22.3 16.2 1.69 2.09 14.8 15.8Geolocation with adj. 1.24 1.63 14.1 14.6 1.25 0.703 5.36 13.1Sense -114dBm 0.985 0.409 7.7 13.8 1.13 0.167 2.57 13.6-114dBm,-110dBm adj. 0.515 0.0635 2.63 9.83 0.576 0.008 0.284 8.87


. . . If we vary the allowed power?


What is the underlying public policy tradeoff?

.1 1 100

5

10

15

20

25

30

35

40

45

Margin (dB)

Ave

rage

num

ber

of c

hann

els

per

user

White space use: actual population density

White space use:

Broadcast use:

uniform population density

Broadcast use: actual population density

uniform population density

15dB, actual population density

15dB, uniform population density


10dB, actual population density


5dB, actual population densityWhite space use: protection & pollution


What is the underlying public policy tradeoff?

0.1 1 101

10

100

1000

Margin (dB)

Peo

ple

gain

ed v

ersu

s pe

rson

lost

Cumulative gain−loss, uniform population density

Cumulative gain−loss, actual population density

Instantaneous gain−loss, uniform population density

Instantaneous gain−loss, actual population density

ActualPop. Density15dB10dBfor 5dB pollution rule

Achievable margins

Achievable margins

for 5dB pollution rule 10dB 15dBUniformPop.Density


Outline

How much usable white-space is there?

How can we understand sensing?Light-handed regulation: identity



What are the right metrics for sensing?

WPAR =∫ ∞

rn

PFH(r) w(r) rdr

PFH(r): probability of finding a spectrum hole at distance r.

w(r): weighting function satisfying∫∞

rnw(r) r dr = 1.


What are the right metrics for sensing?

WPAR =∫ ∞

rn

PFH(r) w(r) rdr

FHI = sup0≤r≤rn

supFr∈Fr

PFr(D = 0|ractual = r)

where Fr is the uncertainty about the distribution Fr underlying algorithm D.Anant Sahai (UC Berkeley) Cognitive Radio 02/19/2009 11 / 1

Single-user sensing


Single-user sensing: finite samples


Single-user sensing: SNR Walls


Cooperation


Outline


How can we understand sensing?Light-handed regulation: identity

I Prior work:F Faulhaber ’05F Hall, Barbeau, Kranakis ’03F Brik, Banerjee, Gruteser, Oh ’08F Rasmussen and Capkun ’07F Rozovsky and Kumar ’01



Identity through taboos

Network IDUser ID

× Device IDTX Identity: Band 1

TX Identity: Band 2

TX Identity: Band 3

. . .Cannot transmit


Single secondary user case


Multiple Users: cooperation and/or “framing”


Multiple Users: noise-free non-strategic case

Min

imum

enf

orce

men

t ove

rhea

d

Number of users

Catch coalition of 4



0.1

0.2

0.3

0.4

0.5

02 73 65410 1010 101010

Time steps until conviction = 3000


Multiple Users: coalitions and overhead


Noisy-case: An equal-rate MAC-framework

Identifying Culprits MAC channelN Distinguishable Secondary Users N Distinct messages

K Coalition size K different usersTc Time-to-identification Codeword length Tc

γ Taboo-fraction γ average cost-constraint on codewordsUsers may/may-not cheat/interfere MAC channel model

limTc→∞

log NTc≤ min

k≤K

I(Xk1; Y|XK

k+1)k


Outline


How can we understand sensing?

Light-handed regulation: identityLight-handed regulation: deterrence

I Prior work:F Rose, Ulukus, Yates ’01F Popescu and Rose ’04F Etkin and Tse ’05F Huang, Berry, Honig ’04F Xu, Kamat, Trappe ’06


Single-band model

False Alarm

Legal TX

SecondaryTX No TX

No Cheat

Cheat

False Alarm

Legal TX

Primary

Cognitive

Jail

Pcatch

Ppen

p1

q1

Ptx = q/(q+p)


Single-band model

00.25

0.50.75

1

00.25

0.50.7510

0.25

0.5

0.75

1

pPtx

Pcheat

Always cheat

Never cheat

Pcatch = 1Ppen = 0.6


Single-band model

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

1

2

3

4

5

6

7

8

9

10

Ppen

β =

pai

n of

jail

Pcatch = 0.1

0.2

0.4

0.60.8

1


Multiple-bands: need to have something to lose



TX No TX

No Cheat

Cheat

False Alarm

Legal TX

q

SecondaryTX No TX

No Cheat

Cheat

False Alarm

Legal TX

Primary

Cognitive

Band 1Band 2

Band 3

Band B

Global Jail

Pcatch

Pcatch

Primary

Ppen

p1

q1

pN

qN

Ptx = q/(q+p)



0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

1

2

3

4

5

6

7

8

9

10

β =

hom

e ba

nds

requ

ired

to in

cent

iviz

e no

che

atin

g

Ppen

B = 1

B = 3

B = 5

B = 7

B = 9

Pcatch = 1


The problem of false convictions

TX No TX

No Cheat

Cheat

False Alarm

Legal TX

q

SecondaryTX No TX

No Cheat

Cheat

False Alarm

Legal TX

Primary

Cognitive

Band 1Band 2

Band 3

Band B

Global Jail

Pcatch

Pcatch

PrimaryPwrong

Pwrong

Ppen

p1

q1

pN

qN

Ptx = q/(q+p)



Pcatch = 1Pcatch = 0.5

Pcatch = 0.1Ppe

n

0

1

0.5Pwrong0.1 0.2 0.3 0.4

0.5

B = 3



Pcatch = 1Pcatch = 0.5

Pcatch = 0.1Ppe

n

0

1

0.5Pwrong0.1 0.2 0.3 0.4

0.5

B = 3

Ppe

n

0.5

10

1

01 2 6 84

Expansion

Pcatch = 0.1

Pcatch = 0.5

Pcatch = 1

Pwrong = 0.03


The “overhead” needed for bandwidth expansion

30

1

2

10 20

3

Expansion

0

0

1

0

0.5

Utility

Fraction of time in jail Ptx = 0.55Pcatch = 1

Pwrong = 0.03



Exp

ansi

on

0

40

20

30

10

Overhead0.1 0.2 0.3 0.4 0.5

Pwrong = 0.01

Pwrong = 0.06

Pwrong = 0.1

Pwrong = 0.035

Pwrong = .02

Pwrong = 0.001

MaximalExpansion

Ptx = 0.55Pcatch = 1



0 0.1 0.2 0.3 0.4 0.50

10

20

30

40

Overhead

Expansion

Pcatch = 1

0.8

0.6

0.4

0.2

0.1

Ptx = 0.55Pwrong = 0.02

MaximalExpansion


Conclusion: Freedom isn’t free

Interference management is Interference management notprimary’s responsibility primary’s responsibility

Secondary has permission Markets UWB/Spectrum MonitorsSecondary must take care Denials Opportunistic


References on www.eecs.berkeley.edu/∼sahai/

“How much white space is there?”

“What is a spectrum hole and what does it take to recognize one?”

“A technical perspective on light-handed regulation for cognitive radios”

“Cognitive Radios for Spectrum Sharing”


Prospects and challenges for spectrum sharing by cognitive ...sahai/Presentations/Harvard09.pdfAnant...

Documents

Transcript of Prospects and challenges for spectrum sharing by cognitive ...sahai/Presentations/Harvard09.pdfAnant...