new directions in cognitive radio and spectrum sharing

New directions in cognitive radio and spectrumsharing

Anant Sahaipresenting joint work with:

Danijela Cabric Mubaraq Mishra Rahul TandraArash Parsa Amin Gohari Kristen Woyach

George Atia Saligrama Venkatesh

BWRC and Wireless Foundations CenterU.C. Berkeley

Major support from the National Science Foundation

IEEE Workshop on Networking Technologies for SDR Networks

Anant Sahai (UC Berkeley) Cognitive Sharing 6/15/2008 1 / 40

Spectrum, spectrum, everywhere, but . . .

Available spectrum looks scarce.

Measurements suggest the allocated spectrum is vastly underutilized.


Examine the problem from first principles

What is the deep reason for the existing waste?



What is the deep reason for the existing waste?3R’s: Rate,Reliability, andRobustness

◮ Rate and Reliability: our usual focus⋆ Fighting ergodic uncertainty⋆ Shannon capacity limits⋆ Error correcting codes





◮ Robustness: the rest⋆ “Outage” within a system⋆ Coexistence with other systems⋆ Traditional approach: static guard bands (“frequency plans”)






Separation of time and space scales◮ Years/decades: frequency planning◮ ms/minutes: actual use






Separation of time and space scales◮ Years/decades: frequency planning◮ ms/minutes: actual use

In the future, technical solutions must bridge these scales!Rethinkrobustness architecture to enable rate/reliability gains.


The basic policy alternatives for sharing

A new comprehensive commons — eliminate legacy users entirely.




Eliminate some legacy users and reallocate their spectrum.




Eliminate some legacy users and reallocate their spectrum.Preserve some priority for “primary users”

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

Secondary has permission Markets UWBSecondary must take care Denials Opportunistic

Current ultra-wideband: blanket permission◮ “Speak softly, but use a wideband”◮ Energy limited regime — works because most bands are not used







Current ultra-wideband: blanket permission◮ “Speak softly, but use a wideband”◮ Energy limited regime — works because most bands are not used◮ Not future-proof!







Current ultra-wideband: blanket permission◮ “Speak softly, but use a wideband”◮ Energy limited regime — works because most bands are not used◮ Not future-proof!

Even future “licensed” systems will likely haveopportunisticfeatures.


Layering revisited

PHY

MAC

Regulatory

Application

Networking


Outline

MotivationSpectrum sensing: uncertainty is key challenge

◮ Single-detector sensitivity◮ Overhead-oriented metrics◮ Cooperation and multiband sensing

Technical questions in regulation◮ A simple model ofa posteriori enforcement◮ Do you know who I am?

Conclusions


Sensing the primary’s presence

fc+W/2fc−W/2

UnknownActivity

UnknownActivityBand of Interest

Spectrum picture

Look for the primary in the ‘band of interest’Within band model:

◮ Primary signal:X(t)◮ Background and receiver noise:W(t)


Impact of uncertainty: energy detector

Actual noise power,σ

2a ∈ [ 1

ασ

2n, ασ

2n]

If

P + σ2a ≤ ασ

2n

⇒ P ≤α

2 − 1α

σ2n

Energy detector fails to detectthe signal

UncertaintyZone

Signalpresent

TargetSensitivity

σ 2nα

σ 2n1/α ��

��

}Impossible

Noise power

Test statistic


SNR wall for energy detector

−40 −35 −30 −25 −20 −15 −10 −5 00

2

4

6

8

10

12

14

Nominal SNR

log 10

N

x = 0.1 dBx = 0.001 dB x = 1 dB


SNR wall for energy detector

0 0.5 1 1.5 2 2.5 3−14

−12

−10

−8

−6

−4

−2

0

2Position of SNR wall for radiometer

Noise uncertainty x (in dB)

SN

Rw

all (

in d

B)

−3.3 dB


Primary structure vs environmental uncertainty

Primary Detector Key Uncertainty

Constellation any Noise distributionPilot coherent Phase-coherence timePilot any Noise color

Pulse-shape cyclostationary Delay-coherence time



fc+W/2fc−W/2

UnknownActivity

UnknownActivity

Pilot tone

Spectrum picture

Band of Interest






UnknownActivity

Pilot

tone

−W/2fc f

c+W/2

Noise +Interferencelevel

MeasurementZone

UnknownActivity Band of Interest

}Primary Detector Key Uncertainty





X(t)

t






T (Y )1 T (Y )2 T (Y )3 T (Y )4 T (Y )

5

H ( f )1 H ( f )

3H ( f )4H ( f )2 H ( f )

5

t





Detector robustness with coherence time

100

101

102

103

104

105

106

−60

−50

−40

−30

−20

−10

0

10

20

Coherence time, Nc samples

SN

R (i

n dB

)Location of SNR wall for various detectors

Modified feature detectorEnergy detectorPilot detector, 10% pilot powerCompletely known signal


Are we just being paranoid?

Experimental validationUsed BEE2 and 2.4 GHz radio front-ends




Noise levels move around over a day.




The spectral correlation function shows spectral redundancy in a transformeddomain.




But this redundancy is blurred away by fast fading.


Outline




Conclusions


How to model the requirement of safety

ON ON

Consider time-domain



ONSense

Use Band ON


Can sense a whitespace and use it.



ONSense

Use BandInterference

safe again



Some interference unavoidable.



ONSense

Use BandInterference

safe again



Some interference unavoidable.

Otherwisefear of return makes it impossible to recover.


Consider strong primary transmitters

A


Define a protected radius

B


Mice can get close...

B


But keep the lions far away!

B


Fading

��

��

��

��

0 5 10 15 20 25 30−50

−40

−30

−20

−10

0

10

20

30

40

50Maximum power for secondary transmitter

SNR margin ψ at secondary receiver (dB)

Max

sec

onda

ry p

ower

(dB

W)

No shadowing10 dB shadowing

10 dB

If you hear a weak signal, are you far away, or just locally faded?

The possibility of 10 dB of fading results in a 10 dB shift of the requireddetection margin

How to choose X dB of fading margin?


The spatial equivalents�� (a)

(b)

Occupied Spectrum Hole Recovered Spectrum

time

TX 1

TX 2

TX 3

r p

r p

r p


What are we giving up?

��

��

��

��

Safe, but might be faded(fading uncertainty)

Will recover using diversity.



��

��

��

��



Lights on, but no one home(receiver uncertainty)

Could be recovered using denials.But not worth it.



��

��

��

��



Lights on, but no one home(receiver uncertainty)

Could be recovered using denials.But not worth it.

Safe, but not shadowed enough(symmetry uncertainty)Anant Sahai (UC Berkeley) Cognitive Sharing 6/15/2008 19 / 40

Example Distribution of Primary Users


Performance: Weighted Probability of Area Recovered

WPAR =

∫ ∞

rn

w(r)PFH(r) rdr

PFH(r) is the probability of finding a spectrum hole at distancer fromprimary.w(r) is a weighting function satisfying

∫ ∞

rnw(r) r dr = 1.



WPAR =

∫ ∞

rn

w(r)PFH(r) rdr


∫ ∞

rnw(r) r dr = 1.

◮ More people are likely to be close to city centers



WPAR =

∫ ∞

rn

w(r)PFH(r) rdr


∫ ∞

rnw(r) r dr = 1.

◮ More people are likely to be close to city centers◮ After enough distance, a new primary might exist.



WPAR =

∫ ∞

rn

w(r)PFH(r) rdr


∫ ∞

rnw(r) r dr = 1.

◮ More people are likely to be close to city centers◮ After enough distance, a new primary might exist.

e.g. exponential:w(r) = K exp(−κr)


What should the safety metric be?

Key issue: incentives and trust



Key issue: incentives and trust◮ Probability of interference depends on the model.



Key issue: incentives and trust◮ Probability of interference depends on the model.◮ Primary does not andshould not trust full model for:

⋆ Noise distribution




⋆ Noise distribution⋆ Secondary deployment assumptions




⋆ Noise distribution⋆ Secondary deployment assumptions⋆ Fading distribution




⋆ Noise distribution⋆ Secondary deployment assumptions⋆ Fading distribution

This fear must by accounted for:

FHI = sup0≤r≤rn

supFr∈Fr

PFr(D = 0|ractual = r)

whereFr is the uncertain distribution underlying algorithmD.


The story so far in these metrics

10−4

10−3

10−2

10−1

100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1Single Detector Performance

Wei

gh

ted

Pro

bab

ility

of

Fal

se A

larm

(W

PA

R)

Fear of Harmful Interference (FHI

)

Perfect detector, Number of Samples (N) = ∞Complete knowledge, Number of Samples (N) = 100Single Quantile knowledge, Number of Samples (N) = 100



102

103

104

105

106

107

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Gains from increasing the number of samples (FHI

= .1)

Wei

gh

ted

Pro

bab

ility

of

Are

a R

cove

red

(W

PA

R)

Number of samples (N)

Perfect detectorComplete knowledgeSingle quantile knowledge



v

r n

Area never recoverd

Area recovered

No Noise uncertainty

v

r n

With Noise uncertainty



10−4

10−3

10−2

10−1

100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Fear of harmful interference (FHI

)

Wei

ghte

d P

roba

bilit

y of

Are

a R

ecov

ered

(W

PA

R)

Radiometer with noise uncertainty

No noise uncertainty (x=0)

1 dB noise uncertainty (x =1)

0.1 dB noise uncertainty (x=0.1)

0.01 dB noise uncertainty (x=0.01)


Outline




Conclusions


Spatial domain: How can cooperation help?

It lessens the required detector performanceRough Analogy: Deck of cards wherered

cards signify bad fades.

Probability that I get aredcard: VeryHigh (50%)!

Probability that all users getredcards: Very Low

��

��

��







��

��

��

Multipath varies significantly on the scale ofλ

4 (10cm at 800MHz).Shadowing varies significantly on the scale 20-500m







��

��

��



Close enough to be relevant, far enough to be independent.







��

��

��



Close enough to be relevant, far enough to be independent.

It also increases robustness to the fading model.


Does cooperation really work?

Experimental validation

Used BEE2 and 2.4 GHz radio front-ends


Time-domain: How can cooperation help?

Interference diversity!


Cooperation story in correct metrics

10−4

10−3

10−2

10−1

100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Wei

ghte

d P

roba

bilit

y of

Are

a R

ecov

ered

(W

PA

R)


)

ML cooperation

ML rule, 5 users (M=5)ML rule, 4 users (M=4)ML rule, 3 users (M=3)ML rule, 2 users (M=2)ML rule, 1 user (M=1)



100

101

102

103

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Wei

ghte

d P

roba

bilit

y of

Are

a R

ecov

ered

(W

PA

R)

Number of cooperating radios (M)

Multi−user cooperation: ML vs OR rule (FHI

= 10−2)

ML rule with complete knowledgeOR ruleOR rule with bounded single−quantile knowledgeML rule with single quantile knowledge



10−4

10−3

10−2

10−1

100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Wei

ghte

d P

roba

bilit

y of

Are

a R

ecov

ered

(W

PA

R)


)

Two user (M=10) ML detector with varying correlation uncertainty

ρ

max =0

ρmax

=0.5

ρmax

=0.8

ρmax

=1


Multiband detection: hope for the future

10−4

10−3

10−2

10−1

100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Target PHI

Wei

ghte

d P

FH

Impact of GPS

Single radioSingle radio with GPS



10−5

10−4

10−3

10−2

10−1

100

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Target PHI

Wei

ghte

d P

FH

MAP Cooperation with GPS

1 radio with GPS2 radios with GPS3 radios with GPS4 radios with GPS5 radios with GPS



Tower - A

Tower - B

r A

r B

r n

2r n + D



CR #2

CR #1 Tower A

Primary 1 Primary 2

Global Shadowing (S A )

Global Shadowing (S B )

Local Shadowing (L 2 )

Local Shadowing (L 1 )

Tower B Primary 3 Primary 4

Multipath (M 21 , M 22 , M 23 , M 24 )

Multipath (M 11, M 12 , M 13 , M 14 )



0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.65

0.7

0.75

0.8

0.85

0.9

0.95

1

Pro

babi

lity

of F

indi

ng a

Hol

e (P

FH

)

Normalized distance from tower A (r/rn)

Multiband, 1 radioSingleband, 1 radioMultiband, 2 radiosSingleband, 2 radios


Outline




Conclusions


The regulatory challenge

Part-15-style device-certification would be great◮ Easy to enforce◮ Easy to generalize to single-user sensing



Part-15-style device-certification would be great◮ Easy to enforce◮ Easy to generalize to single-user sensing◮ Terrible for QoS




Could attempta priori certification of cooperative algorithms




Could attempta priori certification of cooperative algorithms◮ Undecidability? (Need a proof of cooperative correctness)




Could attempta priori certification of cooperative algorithms◮ Undecidability? (Need a proof of cooperative correctness)◮ Pick “one true algorithm” and approve it: evolvability problems




Could attempta priori certification of cooperative algorithms◮ Undecidability? (Need a proof of cooperative correctness)◮ Pick “one true algorithm” and approve it: evolvability problems◮ Squeezes out innovation: return of “beauty contest”




Could attempta priori certification of cooperative algorithms◮ Undecidability? (Need a proof of cooperative correctness)◮ Pick “one true algorithm” and approve it: evolvability problems◮ Squeezes out innovation: return of “beauty contest”◮ Draconian Digital Restrictions Management





Shift to somea posteriori enforcement





Shift to somea posteriori enforcement◮ Ideas already present in Coase ’59 and de Vany ’69.◮ Users incur liability for harmful interference and are punished


A toy model

TX No TX

S2: Wait S

1: FA

S3: Illegal TX S

0: Legal TX

S4: Pen. Box S

5: Pen. Box

Primary Use

Secondary Use

p

q

S0

S0

S1 S

1S

2S

2

S3

S3


A toy model

00.25

0.50.75

1

00.25

0.50.75

10

0.25

0.5

0.75

1

pP(Pri TX)

Uto

tal


A toy model

00.25

0.50.75

1

00.25

0.50.75

10

0.25

0.5

0.75

1

pP(Pri TX)

p chea

t


A toy model

00.25

0.50.75

1

00.25

0.50.75

10

0.25

0.5

0.75

1

pP(Pri TX)

Uco

llide


A toy model

00.25

0.50.75

1

00.25

0.50.75

10

0.250.5

0.751

pP(Pri TX)

p chea

t


A toy model

TX No TX

S2: Wait S

1: FA

S3: Illegal TX S

0: Legal TX

S4: Pen. Box S

5: Pen. Box

Primary Use

Secondary Use

p

q

S0

S0

S1 S

1S

2S

2

S3

S3


A toy model

00.25

0.50.75

1

0

0.25

0.5

0.75

10

0.25

0.5

0.75

1

pP(Pri TX)

p chea

t


A toy model

0 0.25 0.5 0.75 10

0.25

0.5

0.75

1P

(Pri

TX

)

p

β = 1.5β = 1.4β = 1.3β = 1.2β = 1.1A

B


A toy model

0 0.2 0.4 0.6 0.8 10

1

2

3

4

5

ppen

β

pcatch

= .2

pcatch

= .4

pcatch

= .6

pcatch

= 1


A toy model

00.25

0.50.75

1

00.25

0.50.75

10

0.25

0.5

0.75

1

pP(Pri TX)

Uto

tal


A toy model

TX No TX

S2: Wait S

1: FA

S3: Illegal TX S

0: Legal TX

S4: Pen. Box S

5: Pen. Box

Primary Use

Secondary Use

p

q

S0

S0

S1 S

1S

2S

2

S3

S3


A toy model

0 0.2 0.4 0.6 0.8 10

2

4

6

8

10

12

14

pwrong

β

pcatch

= 0.2

pcatch

= 0.4

pcatch

= 0.6

pcatch

= 0.8

pcatch

= 1


Outline




Conclusions


Wireless + anonymity: a serious problem

Faulhaber’s “Hit and run radios” vs Hatfield’s “Spectrum trolls”



Faulhaber’s “Hit and run radios” vs Hatfield’s “Spectrum trolls”What can we do about identity?

◮ PHY-Layer Identity




◮ PHY-Layer Identity⋆ Easy to think about: just transmit your name⋆ Obvious overhead: power in identity




◮ PHY-Layer Identity⋆ Easy to think about: just transmit your name⋆ Obvious overhead: power in identity⋆ Removes flexibility




◮ PHY-Layer Identity⋆ Easy to think about: just transmit your name⋆ Obvious overhead: power in identity⋆ Removes flexibility⋆ Nonobvious catastrophic performance degradation for multiuser techniques.




◮ PHY-Layer Identity⋆ Easy to think about: just transmit your name⋆ Obvious overhead: power in identity⋆ Removes flexibility⋆ Nonobvious catastrophic performance degradation for multiuser techniques.⋆ Doesn’t help with Trolls





◮ MAC-Layer Identity





◮ MAC-Layer Identity⋆ A signature that shows up in the pattern of interference





◮ MAC-Layer Identity⋆ A signature that shows up in the pattern of interference⋆ What is the overhead?


Trollsbane: No harm, no foul


Trollsbane: No harm, no foul

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.30.75

0.8

0.85

0.9

0.95

1

∆=θ1−θ

0

Util

izat

ion

(1−

γ)

PF=15%, P

M=10%, T=280 slots


Identity through superimposed codes



0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.910

1

102

103

104

105

106

Utilization (p)

Tim

e to

Con

vict

ion

(Tc)

in s

lots

Information Theoretic LBRandom Coding UB found from UD codes propertiesRandom Coding UB found from the bipartite graph



102

103

104

105

106

107

108

109

102

103

104

105

106

Total Number of Systems (N)

Tim

e to

Con

vict

ion

T c in s

lots

Random Coding Upper BoundInformation Theoretic Lower Bound


Outline




Conclusions


Decouple sensor network and communication network


Secondary has permission Markets Spectrum MonitorsSecondary must take care Denials Opportunistic

Purely opportunistic useis harder than it looks.

Cooperation is just another word for “infrastructure”






Cooperation is just another word for “infrastructure”Dedicatedsensor-networkinfrastructure

◮ Assume network nodes know where they are◮ Assume large spatial extent beyond primary user scale◮ Construct primary usage map on slower time-scale






Cooperation is just another word for “infrastructure”Dedicatedsensor-networkinfrastructure

◮ Assume network nodes know where they are◮ Assume large spatial extent beyond primary user scale◮ Construct primary usage map on slower time-scale

Coordinate secondary radios by giving explicit permission.


“Disneyland” vs “Yosemite”

Owner controls access topreserve QoS

“Band-managers” own the bandand leases it out to users.

Monopoly





Monopoly

Public owns and sets broadguidelines for use

Unlicensed users are on theirown.

Competition





Monopoly

Public owns and sets broadguidelines for use

Unlicensed users are on theirown.

Competition

“Spectrum tour guide” can coordinate users without owning bands.


Speculation: who benefits from cognitive radio?

Conventional answer: new entrants and other small players denied accessto spectrum by the existing regime.




Sounds familiar: Rhetoric around the Internet boom in the 90s. . .




Sounds familiar: Rhetoric around the Internet boom in the 90s. . .But what actually happened?





◮ Many smaller niche applications were enabled.





◮ Many smaller niche applications were enabled.◮ But big players got bigger — used the new technology tocut costs and

further exploit economies of scale.




Observation:carriers are users of spectrum, not mere holders.




Observation:carriers are users of spectrum, not mere holders.Why pay for spectrum when you could just take it?

◮ Achieve ubiquitous coverage through opportunistic use!◮ Pay only when truly scarce.






Carriers can build and update the sensor network infrastructure.






Carriers can build and update the sensor network infrastructure.

Cheap devices will win out over more expensive ones.


new directions in cognitive radio and spectrum sharing

Documents