zhao swami mil com 08 tutorial
DESCRIPTION
SGSNTRANSCRIPT
Networking Cognitive Radios forDynamic Spectrum Access
Qing Zhao Ananthram SwamiQing [email protected] of California
Ananthram [email protected] Army Research Lab
Davis, CA Adelphi, MD
MILCOM 2008 Tutorial18 Nov 2008 San Diego
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
2
Tutorial Outline
Introduction Physical layer issues MAC layer issues Network layer issues Programs, Policies, StandardsConclusion
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
3
Tutorial Outline
Introduction Physical layer issues MAC layer issues Network layer issues Programs, Policies, StandardsConclusion
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Introduction
MotivationTaxonomy Taxonomy Technical challengesApplications
• Q. Zhao, A. Swami, “A Survey of Dynamic Spectrum Access: Signal Processing and Networking Perspectives,” IEEE ICASSP 2007.
Applications
g g p ,
• Q. Zhao, B.M. Sadler, “A Survey of Dynamic Spectrum Access,” IEEE Signal Processing Magazine, May, 2007.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
R l ti P i t 1927 O t AllRegulation Prior to 1927: Open to All
Herbert Hoover
The Secretary of Commerce...and The Secretary of Commerce...and Under-Secretary of Everything Else!
Agency: Department of Commerce.Service: AM radio broadcasting.
7
Service: AM radio broadcasting.Limited power: cannot deny license to anyone.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Sin 1927: Ti ht C ntr l b FCCSince 1927: Tight Control by FCC
Federal Communications Commission (FRC b. 1934).
8
All non-Federal Government use of the spectrum.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C t P li & S t S itCurrent Policy & Spectrum Scarcity
Fixed allocationRigid eq i ements on ho to se
Little Sharing
9
Rigid requirements on how to use
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Spectrum Opportunities in Space, Time, & Frequency
(Credit: DARPA XG) (Credit: ACSP Cornell)
10
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Di r Id C nf in T rmDiverse Ideas, Confusing Terms
Dynamic spectrum accessDynamic spectrum accessDynamic spectrum allocationSpectrum property rightsSpectrum property rightsSpectrum commonsOpportunistic spectrum accessSpectrum poolingSpectrum underlaySpectrum overlayCognitive radio
12
…
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
A T n m f DSAA Taxonomy of DSADynamic
Current Policy
Fixed allocation Little sharing Rigid requirement
ySpectrum
Access
Rigid requirementon how to use
13
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Thr DSA M d lThree DSA ModelsDynamic ySpectrum
AccessCurrent Policy
Fixed allocation Little sharing Rigid requirement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
g
14
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
E l i U M d lExclusive Use ModelDynamic ySpectrum
AccessCurrent Policy
Fixed allocationLittle sharing Rigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
g
15
Maintains the basic structure: license for exclusive use
Introduces flexibility in allocation and spectrum usage
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
E l i U M d lDynamic
Exclusive Use Modely
Spectrum Access
Current Policy
Fixed allocationLittle sharing Rigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
S t D i Spectrum Property
Rights
Dynamic Spectrum Allocation
16
Maintains the basic structure: license for exclusive use
Introduces flexibility in allocation and spectrum usage
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Sp tr m Pr p rt Ri htSpectrum Property RightsDynamic ySpectrum
AccessCurrent Policy
Fixed allocation Little sharing Rigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
S t D i Spectrum Property
Rights
Dynamic Spectrum Allocation
17
Allows selling and trading spectrum and freely choosing technology
Let economy & market determine the most profitable use of spectrum
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
N b l Priz Winnin IdNobel Prize Winning Idea
Ronald H. Coase
Nobel Prize Laureate in Economics (1991)
18
R. Coase, “The federal communications commission,” J. Law and Economics, pp. 1–40, 1959.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C Th r mCoase TheoremCoase Theorem: All government gallocations of a public good are equally efficient in the absence of transaction coststransaction costs.
Ronald H. Coase
Nobel Prize Laureate in Economics (1991)
19
R. Coase, “The federal communications commission,” J. Law and Economics, pp. 1–40, 1959.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C Th r mCoase TheoremCoase Theorem: All government ll i f bli d allocations of a public good are
equally efficient in the absence of transaction costs.
Ronald H. Coase
Nobel Prize Laureate in Economics (1991)
Milton Friedman George J Stigler
20
Milton Friedman
Nobel Prize Laureate in Economics (1976)
George J. Stigler
Nobel Prize Laureate in Economics (1982)
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C Th r mCoase TheoremCoase Theorem: All government allocations of a public good are equally efficient in the absence of transaction coststransaction costs.
Government Regulation: not to find the most efficient allocation, but ,to minimize transaction costs.
Spectrum Property Rights: Allow Ronald H. Coase
licensees to sell and trade spectrum and freely choose technology.
Nobel Prize Laureate in Economics (1991)
21
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Sp tr m Pr p rt Ri htSpectrum Property Rights
How To Define and Enforce?
Spatial, Spectral Spillover
How To Define and Enforce?
p , p pMeasured or computed?Tx or Rx’s responsibility?
How to detect trespassing
(D. Hatfield and P. Weiser, 2005 DySpan)( , y p )
Need Accurate Yet Simple Signal Propagation Models
22
Need Accurate Yet Simple Signal Propagation Models
And Filter Design For Effective Spillover Suppression.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
D n mi Sp tr m All ti nDynamic Spectrum AllocationDynamic ySpectrum
AccessCurrent Policy
Fixed allocationLittle sharing Rigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
S t D i Spectrum Property
Rights
Dynamic Spectrum Allocation
23
Dynamic spectrum assignment to different services
Exploiting spatial and temporal traffic statistics
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
D n mi Sp tr m All ti nDynamic Spectrum AllocationDemand Frequency
UMTSDVB TDVB-T
Time
DVB-T
Time
UMTS
Traffic statistical modeling, estimation, and predictionCent ali ed and dist ib ted spect m allocation
24
Centralized and distributed spectrum allocation(L. Xu, etal, 2000, P. Leaves, etal, 2004)
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Op n Sh rin M d lOpen Sharing ModelDynamic ySpectrum
AccessCurrent Policy
Fixed allocation Little sharingRigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
S t D i Spectrum Property
Rights
Dynamic Spectrum Allocation
25
Open sharing among peer users (spectrum commons)
Draws support from the success of unlicensed ISM bands
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Hi r r hi l A M d lHierarchical Access ModelDynamic ySpectrum
AccessCurrent Policy
Fixed allocation Little sharingRigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
S t D i Spectrum Property
Rights
Dynamic Spectrum Allocation
26
Hierarchical access with primary and secondary users
sharing with limited interference to primary users (licensees)
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Hi r r hi l A M d lHierarchical Access ModelDynamic ySpectrum
AccessCurrent Policy
Fixed allocation Little sharingRigid req irement
Exclusive Use Model
Open Sharing Model
Hierarchical Access Model
Rigid requirementon how to use
S t D i Spectrum SpectrumSpectrum Property
Rights
Dynamic Spectrum Allocation
SpectrumUnderlay(UWB)
SpectrumOverlay
(OSA, pooling)
27
Spectrum underlay: constraint on transmission power
Spectrum overlay: constraint on when and where to transmit
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Und rl O rlUnderlay vs. OverlaySpectrum Underlay (UWB) Spectrum Overlay (OSA)
PSD PSD
Primary Primary
28Secondary Secondaryf f
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
A T f DSAA Taxonomy of DSADynamic Dynamic Spectrum
Access
Exclusive Use Model
Open Sharing Model
Hierarchical Access ModelUse Model Sharing Model Access Model
Spectrum Property
Rights
Dynamic Spectrum Allocation
Spectrum Underlay(UWB)
Spectrum Overlay
(OSA, pooling)
29
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C niti R diCognitive RadioSoftware Defined RadioSo t a e e ed ad o
Promoted by Mitola in 1991A multiband radio supporting multiple air interfaces and reconfigurable through softwareinterfaces and reconfigurable through software
Cognitive RadioCognitive RadioPromoted by Mitola in 1998Built upon a software defined radio platformContext-aware, autonomous reconfigurableLearning from and adapting to environmentApplications not limited to DSA
30
Applications not limited to DSA
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C iti R di Th Ph i l Pl tfCognitive Radio: The Physical Platform
Dynamic Dynamic Spectrum
Access
Exclusive Use Model
Open Sharing Model
Hierarchical Access ModelUse Model Sharing Model Access Model
Spectrum Property
Rights
Dynamic Spectrum Allocation
Spectrum Underlay(UWB)
Spectrum Overlay
(OSA, pooling)
31Cognitive Radio
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
T d D i S t AToward Dynamic Spectrum Access
32Overlay(54-700M)
Underlay(3.1-10.6G)
Auction(A-TV,PCS,C)
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Sp tr m O rl : Appli ti nSpectrum Overlay: Applications
Opportunistic use based on hierarchical price structuresEmergency response and military applications
33
Integration of emerging applications such as sensor networks
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Sp tr m O rl : T hni l ISpectrum Overlay: Technical IssuesPhysical Layery y
Opportunity sensingInterference Aggregation
MAC LayerOpportunity tracking and learningOpportunity tracking and learningOpportunity exploitation with imperfect sensingOpportunity sharing
Network LayerP t l d ti
34
Power control and routing
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 1
Spectrum Sensing and Opportunity Identification
PHY Layer Issues
• Model and detection problem
• How should we sense?
• Cooperative Sensing
• Hardware Challenges
• Waveform Design & Modulation
• Interference Constraints
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 2
Channel Sensing Model: Slotted Primary Users
Opportunities
Channel 1
Channel N0 1 2 3 T
S1(1) = 0 S1(2) = 1 S1(3) = 0 S1(T ) = 0
SN(1) = 1 SN(2) = 0 SN(3) = 0 SN(T ) = 0
t
t
N independent channels, each with bandwidth Bi.
Secondary users search for opportunities independently.
Every primary tx interferes with all secondary users (symmetric).
How to detect whitespace?
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 3
802.22 Draft DFS Sensing Requirements
Parameter Digital TV Wireless Microphone
(Part 74)
Channel Detection Time ≤ 2 sec ≤ 2 sec
Channel Move Time 2 sec 2 sec
Detection Threshold - 116 dBM - 107 dBm
(required sensitivity) (over 6 MHz) (over 200 KHz)
Probability of detection 0.9 0.9
Probability of false alarm 0.1 0.1
SNR - 21 db - 12 dB
Low SNR regime
FCC ET Docket no. 03-122, November 18, 2003, Cordeiro et al, Ghosh et al, Shellhammer
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 4
Spectrum Sensor at PHY
Binary Hypotheses Test:
H0 (idle) vs. H1 (busy)
Two Types of Sensing Errors:
opportunity overlook: H0 → H1 ǫ∆= prob. of overlook
opportunity misidentification: H1 → H0 δ∆= prob. of misidentification
Receiver Operating Characteristics (ROC): 1 − δ vs. ǫ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm ε
Pro
babi
lity
of D
etec
tion
1 −
δ
δε
Which point δ to operate at?
overlook vs. misidentification
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 5
Spectrum Sensor at PHY
Binary Hypotheses Test:
H0 (idle) vs. H1 (busy)
Two Types of Sensing Errors:
opportunity overlook: H0 → H1 ǫ∆= prob. of overlook
opportunity misidentification: H1 → H0 δ∆= prob. of misidentification
Receiver Operating Characteristics (ROC): 1 − δ vs. ǫ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm ε
Pro
babi
lity
of D
etec
tion
1 −
δ
δε How to choose operating point δ?
Overlook vs. Misidentification
Which is worse:
false alarm or miss detection?
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 6
Spectrum Sensor at PHY: MAC performance
Binary Hypotheses Test:
H0 (idle) vs. H1 (busy)
O
ǫ
1 − δ
1 − ζ
δ > ζ δ < ζ
conservative aggressive
MAC Layer Performance
Probability of success PS
(throughput)
Probability of collision PC
Objective: max PS s.t. PC ≤ ζ
How to choose operating
point δ?
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 7
Channel Sensing Model: unslotted primary users
Acknowledgement
Secondary users
Slot (L)
Sensing Transmission
Channel 1
Channel N
Ls L − Ls
t
t
Slotted secondary usage, with sensing, data, and ACK periods
Problem: Given measurements during sensing time, detect the channel state
Problem: during transmission time.
Problem: Is a sensed idle channel an opportunity?
Challenge: Even with perfect sensing, opportunity detection is subject to
Challenge: errors.
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 8
Spectrum Sensor at PHY
Binary Hypotheses Test: (channel n in slot k)
H0 (On(k) = 1 : opportunity) vs. H1 (On(k) = 0 : no opportunity)
Let On(k) denote the sensing outcome.
Two Types of Sensing Errors:
false alarm: H0 → H1 ǫn(k)∆= PrOn(k) = 0|On(k) = 1
miss detection: H1 → H0 δn(k)∆= PrOn(k) = 0|On(k) = 1
Receiver Operating Characteristics (ROC): 1 − δ vs. ǫ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm ε
Pro
babi
lity
of D
etec
tion
1 −
δ
δε
How to choose the operating point (ǫn(k), δn(k))
for each channel at each slot?
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 9
Spectrum Sensing: Some key questions
How should we sense?
Choice of detectors
Tradeoff SU QoS with PU protection
Detecting spectrum opportunities
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 10
Energy Detection
Pros: easily implemented; minimal assumptions
Cons: poor performance with noise uncertainty
Cons: and with multiple secondary users
Cons: Performance ∼ 1/SNR2 at low SNR
H0 : (idle) y(n) = w(n), n = 1, ..., N, AWGN
H1 : (busy) y(n) = w(n) + s(n)
Decide H1 if z =1
N
N∑
n=1
|y(n)|2 > τ(N, σ2
w)
Under H0 : z ∼ N (σ2
w, σ4
w/N)
Under H1 : z ∼ N (σ2
y, σ4
y/N), σ2
y := σ2
w + σ2
s
µ1 − µ0 = σ0Q−1(PFA) − σ1Q
−1(PD)
√NSNR = Q−1(PFA) − (1 + SNR)Q−1(PD)
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 11
Choice of Detectors - Cyclic Detectors (2)
Exploit guard bands in frequency, known carriers, data rates, modulation type
Pros: fc, Ts easy to detect via square-law devices, or cyclic approaches
Pros: Cyclic approaches useful when σ2n is unknown (avoid SNR wall)
Test Statistic : S(f ; τ ) =1
N
∑
n
y(n)y(n + τ )e−j2πfn
Pros: Easily implemented via FFTs
Cons: Timing and frequency jitter can be detrimental
Cons: Requires long integration times
Cons: RF non-linearities; Spectral leakage (ACI).
Cabric et al, Asiloamr’04 Cabric-Brodersen, PIMRC’05
Ghozzi et al, Crowncom’06 da Silva-Choi-Kim,ITA Wkhsp ’07
Lee-Yoon-Kim,ICIPC’07 Kim et al, Dyspan’07
Ye et al, SPS Wkshp ’07 Tu et al, PIMRC’07
Sutton-Nolan-Doyle, JSAC’08
General references on cyclic detection: Giannakis; Gardener
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 12
Choice of Detectors: Matched Filter (3)
Exploit pilots or sync (PN) sequences in primary (WRAN 802.22)
Test Statistic : z =1
N
N∑
n=1
y(n)s(n)
Pros: Correlation detection is usually better than energy detection.
Pros: Performance ∼ 1/SNR at low SNR
Q−1(PFA) − Q−1(PD) =√
NSNR
Cons: fading may null pilot; need to cope with time and freq syncLi et al, JSAC’07 - Exploits pilots, for interference detection
Kundargi-Tewfik, ICASSP’08 - sequential tests with pilots
Yu-Sung-Lee, ICASSP’08 - exploit PU pilots
Other Detectors based on
Receiver leakage Wild-Ramachandran, Dyspan’05
Signal correlation Zeng et al, PIMRC’07
Fast fading Larson-Regnoli, CommLett’07
Multiple antennas Pandhripande-Linnartz, ICC’07
HMM classifier Kyouwoong et al, Dyspan’07
Wavelet-based Tian-Giannakis, CrownCom’06
Multi-resolution Neihart-Roy-Allstot, ISCAS’07
Compressed sensing Tian-Giannakis, ICASSP’07
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 13
Spectrum Sensing: Some key questions
How should we sense?
Choice of detectors
Tradeoff SU QoS with PU protection
Detecting spectrum opportunities
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 14
How long should the sensing time be
• Channel coherence
• Primary’s traffic patterns (e.g., fractional on-time)
• Interference constraints
• Primary and secondary user powers; noise power
• Fading, multipath, shadowing
• Multiple primaries? Spatial distribution
• Multiple secondary users? (aggregate interference)
• SU QoS (rate, reliability, latency) and constraints (power, cooperation)
• Can we exploit PU Modulation, pilots, sync signals,
• Complexity and specifics of algorithms
• Robustness
Detection problem cannot be solved in isolation
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 15
Optimizing sensing time for detection
N ≈ 2(SNR)−2[Q−1(PFA) − Q−1(PD)]2
What if we do not know the noise variance?
Could use sample estimate of noise variance,
σ2w ∈ [aσ2
w, bσ2w], a, b ∼ 1/
√N
To ensure desired performance with uncertainty, need
N ≈ 2(SNR − ∆)−2[bQ−1(PFA) − (SNR + a)Q−1(PD)]2
Energy Detector breaks down when SNR ≈ ∆ = b − a, uncertainty
Tandra and Sahai’s SNR wall, JSTSP, 2008
Example: 6 MHz BW, 1 sec. obs time, ∆ ≈ 0.0022, SNR threshold = −23dB,
close to operating SNR of -21 dB in the 802.22 standard
Robustness to model imperfections important at low SNR
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 16
Optimizing sensing time for throughput
Trade off sensing accuracy of throughput
Slot size of length N - devote n samples for sensing
Maximize throughput efficiency
η(n) :=N − n
N[1 − PFA(n)]
For specified PD (interference constraint)
PFA(n) = Q(
(1 + SNR)Q−1(PD) + SNR√
n)
n∗ and throughput increase as N ↑
n∗ ↓ and η ↑ as SNR ↑
n∗ ↑ and η ↓ as interference constraint ↓
Does this represent SU performance?
Wang et al, WCNC 2007
Kattepur et al, ICICSP 2007
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 17
Spectrum Sensing: Some key questions
How should we sense?
Choice of detectors
Tradeoff SU QoS with PU protection
Detecting spectrum opportunities
Choosing ‘sensing radius’ or threshold
Interaction with MAC
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 18
Whitespace Detection to Opportunity Detection
Is detecting primary signals = detecting spectrum opportunity?
How does PHY performance translate to MAC performance?
We want to detect primary receivers!
PU locations are unknown
1 − PMD
PFA0 1
1 − ζ
If PU is loud but SU is not
listening, is it interference?
SU-TX and SU-TX must
jointly detect opportunities
PS = (1 − PFA) Pr[H0]
PC = PMD
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 19
Spectrum Opportunity: Definition
A B
Interference
Primary Tx
Primary Rx
RI: interference range
RI: of primary users
rI: interference range
rI: of secondary users
A channel is an opportunity for A −→ B if
the transmission from A to B can succeed
the interference power to primary is below a prescribed level
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 20
Spectrum Opportunity: Definition
A B
Interference
RI
Primary Tx
Primary Rx
RI: interference range
RI: of primary users RI ∝ P1/αtx
rI: interference range
rI: of secondary users
rI: ∝ p1/αtx
A channel is an opportunity for A −→ B if
the transmission from A to B can succeed
the interference power to primary is below a prescribed level
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 21
Spectrum Opportunity: Definition
A B
Interference
rI
RI
Primary Tx
Primary Rx
RI: interference range
RI: of primary users RI ∝ P1/αtx
rI: interference range
rI: of secondary users rI ∝ p1/αtx
A channel is an opportunity for A −→ B if
the transmission from A to B can succeed
the interference power to primary is below a prescribed level
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 22
Spectrum Opportunity: Properties
A B
Interference
rI
RI
Primary Tx
Primary Rx
RI: interference range
RI: of primary users RI ∝ P1/αtx
rI: interference range
rI: of secondary users rI ∝ p1/αtx
determined by both transmitting and receiving activities of primary users.
Asymmetric (an opportunity for A −→ B may not be one for B −→ A).
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 23
Detection of Primary Receivers (LBT)
AXY
Primary TxPrimary Rx
rI
rD
Rp + rI
rI: interference range, Rp: primary tx range, rD: detection range
Detecting primary Rx within rI by detecting primary Tx within rD
False alarms and miss detections occur due to noise and fading
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 24
From Detecting Signal to Detecting Opportunity
A B
X
Y
rI
rD
RI
Prob. of Detection (1 − PMD)
Prob. of False Alarm0
1
1
rD ↓
rD ↑
H0: opportunity, H1: alternative.
Even with perfect ears, exposed Tx (X) ⇒ FA, hidden Rx (Y ) ⇒ MD.
Adjusting detection range rD leads to different operating points.
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 25
Miss Detection May not Lead to Collision
A B
rI
RI
Interference
Primary Tx
Primary Rx
There is no primary receiver around A
There are primary transmitters around B
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 26
Miss Detection May Lead to Success
A B
rI
RI
Primary Tx
Primary Rx
There are primary receivers around A
There is no primary transmitter around B
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 27
Correctly Identified Opportunity May Not Lead to Success
A B
rIRI
DATA
Primary Tx
Primary Rx
A B
rI
RI
ACK
Primary Tx
Primary Rx
Successful data transmission and failed ACK
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 28
Network Model
Primary users form a Poisson point process with density λ.
Each primary user transmits with probability p in a slot.
Primary receivers are uniformly distributed within Rp of their transmitters.
Rp
Rp
Rp
2 Analytical expressions for
PFA, PMD, PC, PS
2 For LBT and for RTS-CTS enabled
LBT
Zhao-Ren-Swami, Asilomar ’07
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 29
Summary of Opportunity Detection
Spectrum Opportunity
Determined by both transmitting and receiving activities of primary users
Asymmetric (an opportunity for A −→ B may not be one for B −→ A)
Equivalence of Detecting Signal and Opportunity
Inevitability of opportunity detection errors
A necessary and sufficient condition
Translation from PHY Performance to MAC Performance
Crucial for choosing optimal detector operating point
Complex dependency on the application type and MAC
Choice of sensor operating point cannot be decoupled from sensing and
accessing policies
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 30
Spectrum Sensing: Some key questions
How should we sense?
Choice of detectors
Tradeoff SU QoS with PU protection
Detecting spectrum opportunities
Cooperative sensing
Hardware challenges
Waveform Design & Modulation
Interference Constraints
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 31
Cooperative Schemes
Benefits: combat fading, shadowing, poor sensors
Overhead: control channel? broker?
Trust issues: - jammed links, malicious nodes
Fairness
Time scales - latency
Increased uncertainty due to aggregate interference
Many based on ‘distributed estimation detection’ ideas
Cabric-Mishra-Brodersen, Asilomar’04
Ghasemi-Sousa, Dyspan’05
Mishra-Sahai-Brodersen, ICC’06 - correlated fading; detecting malicious users
Qihang et al, PIMRC’06 - uses Dempster-Shafer theory
Yi et al, PIMRC’07 - relies on multi-hop cooperation
Gandetto-Regazzoni, JSAC’07 - distributed detection
Tahrepour et al, IET Commun, 07 - asymptotic theory, sequential detection
Ma-Li, Globecom’07 - extensions of MRC, EGC
Chen-Wang-Li, ISWCS’07 - learns local ROC parameters
Peh-Liang, WCNC’07 - select users for cooperation
Ganesan et al, TWC’07, JSAC’08- optimal pairing of SU’s to improve detection
Quan-Sayed,JSTSP’08 - linear combining
Unnikrishnan-Veeravalli, JSTSP’08 - linear-quadratic fusion of LLR’s
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 32
Hardware Challenges
Large bandwidth and sampling rates
PU load dictates scanning architecture
Dynamic range
Linearity of analog circuits (mixers, etc)
Adjacent channel interference
Adaptive notch filtering
Active interference cancellation
Cabric-Mishra-Brodersen, Asilomar’04
Cabric-Brodersen, PIMRC’05
Mayer et al, ECWT’07, PIMRC’07
Luu-Daneshrad, JSAC 2007
Jia-Zhang-Shen, JSAC 2008
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 33
Waveform Design / Modulation
OFDMA has emerged as natural standard
Time-frequency granularity well-suited to filling holes
ACI in sensing
Hardware non-linearities and ACI in transmit
Null subcarriers to protect PU from ACI
( 1440 data carriers out of 2048 in 802.22)
PAPR issues
Dictates pulse shape design
Symbol period dictated by channel and SO coherence times
SU transmit power dictated by allowed interference to PUWeiss-Jondral, Comm Mag, 2004
Berthold-Jondral, Dyspan 2005
Tang, Dyspan 2005
Wright, AccessNets 07
c©Q. Zhao, A. Swami, Tutorial at MILCOM 2008. 34
Interference Constraints
Policy Issues: What to impose, How to monitor?
How to impose?
Need to specify allowed probability of interference ζ
at prescribed interference level η to PU
[η, ζ] is a PU-protection / SU-QoS tradeoff
Conditional or joint probability of collision ?
Impose on per-slot basis, or on average ?
From aggregate to node-level parameters?
Requires knowledge of node location, traffic, and channel models
How to monitor?
Qing Zhao, ICASSP’07
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 1
MAC Issues in Opportunistic Spectrum Access
PSfrag replacements
Ch 1 Ch 2 Ch 3
NAK
NAKNAK
PSfrag replacements
Ch 1
Ch 2
Ch 3
NAK
ε
1 − δ
1 − ζ
δ > ζ δ < ζ
conservative aggressive
optimal (δ∗ = ζ)
References
[1 ] Y. Chen, Q. Zhao, and A. Swami, “Joint Design and Separation Principle for Opportunistic Spectrum
Access in the Presence of Sensing Errors,” IEEE Transactions on Information Theory, May 2008.
[2 ] Q. Zhao, B. Krishnamachari, and K. Liu, “On Myopic Sensing for Multi-Channel Opportunistic Access,”
to appear in IEEE Transactions on Wireless Communications.
[3 ] K. Liu, Q. Zhao, “A Restless Bandit Formulation of Opportunistic Access: Indexablity and Index Policy,”
Proc. of IEEE SDR Workshop, June, 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 2
Basic MAC Issues in OSA
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T
S1(1) = 0 S1(2) = 1 S1(3) = 0 S1(T ) = 0
SN(1) = 1 SN(2) = 0 SN(3) = 0 SN(T ) = 0
t
t
I Search for fast-varying opportunities in multiple channels.
I Limited Sensing: can only sense and access a subset of channels in each slot.
Which channels to sense and whether to access?
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 3
Outline
I Slotted independent Markov channels with perfect sensing
I (Correlated) channels with imperfect sensing
I Unslotted primary systems
I Energy-constrained opportunistic spectrum access in fading
I Distributed sensing for multiple users
I Opportunistic spectrum access in self-similar primary traffic
I Concluding remarks
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 4
Gilbert-Elliot Channel Model
I N independent Gilbert-Elliot channels with rate Bi (i = 1, · · · , N).
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T t
t
PSfrag replacements
0 1
(busy) (idle)
p(i)01
p(i)11p
(i)00
p(i)10
0E.N. Gilbert, “Capacity of burst-noise channels,” Bell Syst. Tech. J., vol. 39, pp. 1253-1265, Sept. 1960.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 5
Positive Memory vs. Negative Memory
PSfrag replacements
0 1(busy) (idle)
p0,1
p1,1p0,0
p1,0
Markov Channels with Positive Memory (p11 ≥ p01)
! ! ! ! ! ! ! ! ! ! !
! ! ! ! ! ! ! ! ! ! !
! ! ! ! ! ! ! ! ! ! !
! ! ! ! ! ! ! ! ! ! !
Markov Channel with Negative Memory (p11 < p01)" " "
" " "" " "
# # ## # #
# # #
$ $ $ $ $ $ $
$ $ $ $ $ $ $
$ $ $ $ $ $ $
% % % % % % %
% % % % % % %
% % % % % % %
& & & & &
& & & & &
& & & & &
' ' ' ' '
' ' ' ' '
' ' ' ' '
( ( (( ( (
( ( () ) )
) ) )) ) )
* * * * *
* * * * *
* * * * *
+ + + + +
+ + + + +
+ + + + +
, , , ,
, , , ,
, , , ,
- - - -
- - - -
- - - -
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 6
Sensing Policy for Opportunity Tracking
. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .
/ / / / / / / // / / / / / / // / / / / / / // / / / / / / /0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 11 1 1 1 1 1 1 1 11 1 1 1 1 1 1 1 11 1 1 1 1 1 1 1 1
2 2 2 2 2 2 2 22 2 2 2 2 2 2 22 2 2 2 2 2 2 22 2 2 2 2 2 2 2
3 3 3 3 3 3 3 33 3 3 3 3 3 3 33 3 3 3 3 3 3 33 3 3 3 3 3 3 3
4 4 4 4 4 4 4 44 4 4 4 4 4 4 44 4 4 4 4 4 4 44 4 4 4 4 4 4 4
5 5 5 5 5 5 5 55 5 5 5 5 5 5 55 5 5 5 5 5 5 55 5 5 5 5 5 5 5
6 6 6 6 6 6 6 66 6 6 6 6 6 6 66 6 6 6 6 6 6 66 6 6 6 6 6 6 6
7 7 7 7 7 7 7 77 7 7 7 7 7 7 77 7 7 7 7 7 7 77 7 7 7 7 7 7 7
8 8 8 8 8 8 8 8 88 8 8 8 8 8 8 8 88 8 8 8 8 8 8 8 88 8 8 8 8 8 8 8 8
9 9 9 9 9 9 9 9 99 9 9 9 9 9 9 9 99 9 9 9 9 9 9 9 99 9 9 9 9 9 9 9 9
: : : : : : : : :: : : : : : : : :: : : : : : : : :: : : : : : : : :
; ; ; ; ; ; ; ; ;; ; ; ; ; ; ; ; ;; ; ; ; ; ; ; ; ;; ; ; ; ; ; ; ; ;
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T t
t
I Sensing Policy πs
2 Choose M out of N channels to sense in each slot.
2 Without loss of generality, consider M = 1.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 7
Sensing Policy for Opportunity Tracking
< < < < < < < << < < < < < < << < < < < < < << < < < < < < <
= = = = = = = == = = = = = = == = = = = = = == = = = = = = => > > > > > > > >> > > > > > > > >> > > > > > > > >> > > > > > > > >
? ? ? ? ? ? ? ? ?? ? ? ? ? ? ? ? ?? ? ? ? ? ? ? ? ?? ? ? ? ? ? ? ? ?
@ @ @ @ @ @ @ @@ @ @ @ @ @ @ @@ @ @ @ @ @ @ @@ @ @ @ @ @ @ @
A A A A A A A AA A A A A A A AA A A A A A A AA A A A A A A A
B B B B B B B BB B B B B B B BB B B B B B B BB B B B B B B B
C C C C C C C CC C C C C C C CC C C C C C C CC C C C C C C C
D D D D D D D DD D D D D D D DD D D D D D D DD D D D D D D D
E E E E E E E EE E E E E E E EE E E E E E E EE E E E E E E E
F F F F F F F F FF F F F F F F F FF F F F F F F F FF F F F F F F F F
G G G G G G G G GG G G G G G G G GG G G G G G G G GG G G G G G G G G
H H H H H H H H HH H H H H H H H HH H H H H H H H HH H H H H H H H H
I I I I I I I I II I I I I I I I II I I I I I I I II I I I I I I I I
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T t
t
Immediate Reward
2 If the chosen channel i is idle, Bi units reward is accrued.
2 If the chosen channel i is busy, no reward; wait until the next slot.
Objective: choose sensing policy πs to
max E[throughput]
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 8
Optimal Sensing Policy for Opportunity Tracking
PSfrag replacements
0 t t + 1 TO(1) O(2) O(t − 1) O(t)
R(t) R(t + 1)
· · ·
· · ·
Use entire observation history
Max total remaining reward
I Learn from the observation history.
I Foresighted planning: maximize total remaining reward.
I Optimal action: Gaining immediate reward vs. Gaining spectrum information.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 9
Sensing Policy: Learn from Observations
0 5 10 15 20 25 30
0.5
0.52
0.54
0.56
0.58
0.6
0.62
0.64
0.66
0.68
0.7
Time (slot)
Thro
ughp
ut (b
its p
er s
lot)
Use only statistical info
Optimal Approach
I Cognition: improved performance by learning from accumulating observ.
I Difficulty: exponential complexity; sensitivity to model mismatch.
I Goal: structural policies that are simple, robust, and optimal.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 10
Slotted Independent Markov Channels: outline
I A restless multi-armed bandit formulation
I Indexability and index policies
I When optimality, simplicity, and robustness are achieved simultaneously
I Scaling behavior of the max. throughput w.r.t. the number N of channels
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 11
Multi-Armed Bandit
Multi-Armed Bandit Process
I A bandit with N independent arms.
I Fully observable states of all arms Zi(t)
I Activate arm i and get reward Ri(Zi(t)).
I Active arms change state (Markovian).
I Passive arms are frozen.
Objective: Decide which arm to activate in each slot for max long-term reward.
Optimal Policy: Gittins Index (1979)
I Compute an index for each state of each arm.
I Activate the arm whose current state has the largest index.
Advantage: Reduces an N-D problem to N independent 1-D problem.0J.C.Gittins, “Bandit Processes and Dynamic Allocation Indices,” in Journal of the Royal Sttistical Society, Series B (Methodological), Vol.41, No.2 (1979), 148-177.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 12
Restless Multi-Armed Bandit
Restless Multi-armed Bandit Problem
I Passive arms also change states.
I Can active M ≥ 1 arms simultaneously.
Structure of Optimal Policy
I Not yet known.
Complexity
I PSPACE-hard.
0P. Whittle, ”Restless bandits: Activity allocation in a changing world”, in Journal of Applied Probability, Volume 25, 1988.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 13
Restless Multi-Armed Bandit Formulation
J J J J J J J JJ J J J J J J JJ J J J J J J JJ J J J J J J J
K K K K K K K KK K K K K K K KK K K K K K K KK K K K K K K KL L L L L L L L LL L L L L L L L LL L L L L L L L LL L L L L L L L L
M M M M M M M M MM M M M M M M M MM M M M M M M M MM M M M M M M M M
N N N N N N N NN N N N N N N NN N N N N N N NN N N N N N N N
O O O O O O O OO O O O O O O OO O O O O O O OO O O O O O O O
P P P P P P P PP P P P P P P PP P P P P P P PP P P P P P P P
Q Q Q Q Q Q Q QQ Q Q Q Q Q Q QQ Q Q Q Q Q Q QQ Q Q Q Q Q Q Q
R R R R R R R RR R R R R R R RR R R R R R R RR R R R R R R R
S S S S S S S SS S S S S S S SS S S S S S S SS S S S S S S S
T T T T T T T T TT T T T T T T T TT T T T T T T T TT T T T T T T T T
U U U U U U U U UU U U U U U U U UU U U U U U U U UU U U U U U U U U
V V V V V V V V VV V V V V V V V VV V V V V V V V VV V V V V V V V V
W W W W W W W W WW W W W W W W W WW W W W W W W W WW W W W W W W W W
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T t
t
I Each channel is considered as an arm.
I If channel i is sensed, then it is “activated”.
I The channel states S1, ...SN are not observable
I ⇒ Cannot use the channel state as the state of each arm
0K. Liu, Q. Zhao, “A Restless Bandit Formulation of Opportunistic Access: Indexablity and Index Policy,” Proc. of IEEE SDR Workshop, June, 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 14
From Unobservable to Observable
I Information state: the state of each arm should be its observation history.
I Sufficient statistic: the a posterior distribution (belief vector) Ω(t) that
I exploits the entire observation history.
Ω(t) = [ω1(t), · · · , ωN (t)]
ωi(t) = Pr[channel i is idle in slot t | O(1), · · · , O(t − 1)︸ ︷︷ ︸
observations
]
I The state of arm i in slot t is ωi(t).
I The expected immediate reward obtained when activate arm i is ωi(t) Bi.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 15
Markovian Transition of the Belief StatePSfrag replacements
0 1
(busy) (idle)
p(i)0,1
p(i)1,1p
(i)0,0
p(i)1,0
I If channel i is activated in slot t:
ωi(t + 1) =
p(i)11 , if Si(t) = 1
p(i)01 , if Si(t) = 0
.
I If channel i is made passive in slot t:
ωi(t + 1) = ωi(t)p(i)11 + (1 − ωi(t))p
(i)01 .
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 16
Index Policies
Are there simple index policies with good performance?
Myopic policy: maximize immediate reward
I Index of channel i: Ii(t) = ωi(t) Bi
I Action: choose the channel with the largest index
1 2 3 4 5 60.65
0.7
0.75
0.8
0.85
0.9
Time Slot
Trou
ghpu
t(bits
per
slo
t)
Optimal policyMyopic policy
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 17
Whittle’s Index Policy
Whittle’s Index Policy
2 Subsidy for passivity: provide a subsidy ν when the arm is made passive.
2 Whittle’s index: the minimum subsidy ν that makes the passive action
optimal at the current state.
Performance
2 Optimal under relaxed constraint on the average number of active arms.
2 Asymptotically optimal (N → ∞ w. MN
fixed) under certain conditions.
2 Near optimal performance observed from extensive numerical examples.
Difficulties
2 Existence of Whittle’s index (indexability) is often difficult to establish.
2 High complexity to compute index for uncountable states of each arm.
0P. Whittle, ”Restless bandits: Activity allocation in a changing world”, in Journal of Applied Probability, Volume 25, 1988.0Richard R. Weber; Gideon Weiss, “On an Index Policy for Restless Bandits,” in Journal of Applied Probability, Vol.27, No.3. (Sep,. 1990), pp. 637-648.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 18
The Optimality of Threshold Policy and Indexability
Indexability: The optimal policy for a single-armed bandit is a threshold policy: there exists a
ω∗(ν) ∈ R such that it is optimal to activate the arm if the current belief ω > ω∗(ν); otherwise it is
optimal to make the arm passive. Furthermore, the threshold ω∗(ν) monotonically increases from
−∞ to ∞ as subsidy ν goes from −∞ to ∞, thus the bandit is indexable.
10
PSfrag replacements
Total remainingreward if active
Total remainingreward if passive
ω∗i (ν)
Passive Activeω < ω∗
i (ν) ω > ω∗i (ν)
ω
0K. Liu, Q. Zhao, “A Restless Bandit Formulation of Opportunistic Access: Indexablity and Index Policy,” Proc. of IEEE SDR Workshop, June, 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 19
Whittle’s Index In Closed-Form
I Positive memory (p11 ≥ p01)
Ii(ω) =
ωBi, if ω ≤ p01 or ω ≥ p11;
ω1−βp11+βω
Bi, if ωo ≤ ω < p11;
ω−βT (ω)+C(1−β)(β(1−βp11)−β(ω−βT (ω)))1−βp11−A(1−β)(β(1−βp11)−β(ω−βT (ω))) Bi, if p01 < ω < ωo;
I Negative memory (p11 < p01)
Ii(ω) =
ωBi, if ω ≤ p11 or ω ≥ p01;
βp01+ω(1−β)1+β(p01−ω) Bi, if T (p11) ≤ ω < p01;
(1−β)(1+βE)(βp01+ω(1−β))1−β(1−p01)−D(1−β)(β2p01+βω−β2ω)
Bi, if ωo ≤ ω < T (p11);
(1−β)(βp01+ω−βT (ω))−E(1−β)β(βT (ω)−βp01−ω)1−β(1−p01)+D(1−β)β(βT (ω)−βp01−ω) Bi, if p11 < ω < ωo;
,
0K. Liu, Q. Zhao, “A Restless Bandit Formulation of Opportunistic Access: Indexablity and Index Policy,” Proc. of IEEE SDR Workshop, June, 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 20
Whittle’s Index in Closed-Form
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Belief ω
Whi
ttle’
s in
dex
W(ω
)
p11
=0.4, p01
=0.8, β=0.9
p11
=0.9, p01
=0.1, β=0.9
I Whittle’s index is an increasing function of the belief state.
I Whittle’s index policy is equivalent to myopic policy for identical channels.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 21
The Performance of Whittle’s Index Policy
1 2 3 4 5 60.65
0.7
0.75
0.8
0.85
0.9
Time Slot
Trou
ghpu
t(bits
per
slo
t)
Optimal policyWhittle’s index policyMyopic policy
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 22
The Optimality of Whittle’s Index Policy
For stochastically identical channels,
I Whittle’s index policy = myopic policy
I Myopic policy is proven to be optimal when
2 channels have positive memory (p11 ≥ p01)
2 N = 2 and N = 3 channels with negative memory (p11 < p01)
0Q. Zhao, B. Krishnamachari, K. Liu, “On Myopic Sensing for Multi-Channel Opportunistic Access: Structure, Optimality, and Performance,” to appear in IEEE Trans. Wireless Communications.0T. Javidi, B. Krishnamachari, Q. Zhao, and M. Liu, “Optimality of Myopic Sensing in Multi-Channel Opportunistic Access,” ICC 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 23
Structure of Whittle’s Index Policy: Positive Memory Case
X X X X X X
X X X X X X
X X X X X X
X X X X X X
Y Y Y Y Y Y
Y Y Y Y Y Y
Y Y Y Y Y Y
Y Y Y Y Y Y
Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z
Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z
Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z
Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z Z
[ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [
[ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [
[ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [
[ [ [ [ [ [ [ [ [ [ [ [ [ [ [ [
\ \ \ \ \ \ \ \ \ \ \
\ \ \ \ \ \ \ \ \ \ \
\ \ \ \ \ \ \ \ \ \ \
\ \ \ \ \ \ \ \ \ \ \
] ] ] ] ] ] ] ] ] ] ]
] ] ] ] ] ] ] ] ] ] ]
] ] ] ] ] ] ] ] ] ] ]
] ] ] ] ] ] ] ] ] ] ]I Stay with idle channels and leave busy ones to the end of the queue.
1
2
3
4
2
1
3
4
N
N
PSfrag replacements
Sense“idle”
“idle”
“busy”
t = 1 t = 2
0Q. Zhao, B. Krishnamachari, K. Liu, “On Myopic Sensing for Multi-Channel Opportunistic Access: Structure, Optimality, and Performance,” to appear in IEEE Trans. Wireless Communications.0K. Liu and Q. Zhao, “Channel Probing for Opportunistic Access with Multi-channel Sensing,” IEEE Asilomar Conference on Signals, Systems, and Computers, Oct., 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 24
Structure of Whittle’s Index Policy: Negative Memory Case
^ ^ ^^ ^ ^
^ ^ ^^ ^ ^
_ _ __ _ _
_ _ __ _ _
` ` ` ` ` ` `
` ` ` ` ` ` `
` ` ` ` ` ` `
` ` ` ` ` ` `
a a a a a a a
a a a a a a a
a a a a a a a
a a a a a a a
b b b b b
b b b b b
b b b b b
b b b b b
c c c c c
c c c c c
c c c c c
c c c c c
d d dd d d
d d dd d d
e e ee e e
e e ee e e
f f f f f
f f f f f
f f f f f
f f f f f
g g g g g
g g g g g
g g g g g
g g g g g
h h h h
h h h h
h h h h
h h h h
i i i i
i i i i
i i i i
i i i i
I Stay with busy channels and leave idle ones to the end of the queue.
I Reverse the order of unobserved channels.
4
1
2
3
4
3
2
N
1
N
PSfrag replacements
Sense
“idle”
“idle”
“busy”
t = 1 t = 20Q. Zhao, B. Krishnamachari, K. Liu, “On Myopic Sensing for Multi-Channel Opportunistic Access: Structure, Optimality, and Performance,” to appear in IEEE Trans. Wireless Communications.0K. Liu and Q. Zhao, “Channel Probing for Opportunistic Access with Multi-channel Sensing,” IEEE Asilomar Conference on Signals, Systems, and Computers, Oct., 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 25
Robustness of Whittle’s Index Policy
I No need to know the transition probabilities except the order of p11 and p01.
I Automatically tracks model variations.
1 2 3 4 5 6 7 8 9 100.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
Time slot (T)
Thro
ughp
ut
p11
=0.6, p01
=0,1 (T<=5); p11
=0.9, p01
=0,4 (T>5)
Model Variation
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 26
Scaling Behavior of Maximum Throughput with N
I The upper bound is independent of N .
I The lower bound approaches to the upper bound at geometric rate with N .
I Throughput w. single-channel sensing saturates at geometric rate with N .
5 10 15 20 25 300.52
0.54
0.56
0.58
0.6
0.62
Number of channels
Low
er a
nd u
pper
bou
nds
p11
=0.8, p01
=0.1
The upper bound of the throughput limitThe lower bound of the throughput limit
0Q. Zhao, B. Krishnamachari, K. Liu, “On Myopic Sensing for Multi-Channel Opportunistic Access: Structure, Optimality, and Performance,” to appear in IEEE Trans. on Wireless Comm..
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 27
Slotted Independent Markov Channels: Summary
I A restless multi-armed bandit formulation
I Indexability and Whittle’s index policy in closed-forms
I When channels are stochastically identical
2 Whittle’s index policy = myopic policy.
2 A semi-universal structure ⇒ optimality + simplicity + robustness.
2 Scaling behavior: max throughput saturates at geometric rate under
limited sensing.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 28
Outline
I Slotted independent Markov channels with perfect sensing
I (Correlated) channels with imperfect sensing
2 A constrained POMDP formulation for joint PHY-MAC design
2 A separation principle
I Unslotted primary systems
I Energy-constrained opportunistic spectrum access in fading
I Distributed sensing for multiple users
I Opportunistic spectrum access in self-similar primary traffic
I Concluding remarks
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 29
Correlated Channels
j j j j j j j j
j j j j j j j j
j j j j j j j j
j j j j j j j j
k k k k k k k k
k k k k k k k k
k k k k k k k k
k k k k k k k kl l l l l l l l
l l l l l l l l
l l l l l l l l
l l l l l l l l
m m m m m m m m
m m m m m m m m
m m m m m m m m
m m m m m m m m
n n n n n n n n
n n n n n n n n
n n n n n n n n
n n n n n n n n
o o o o o o o o
o o o o o o o o
o o o o o o o o
o o o o o o o o
p p p p p p p p
p p p p p p p p
p p p p p p p p
p p p p p p p p
q q q q q q q q
q q q q q q q q
q q q q q q q q
q q q q q q q q
r r r r r r r r
r r r r r r r r
r r r r r r r r
r r r r r r r r
s s s s s s s s
s s s s s s s s
s s s s s s s s
s s s s s s s s
t t t t t t t t
t t t t t t t t
t t t t t t t t
t t t t t t t t
u u u u u u u u
u u u u u u u u
u u u u u u u u
u u u u u u u u
v v v v v v v v
v v v v v v v v
v v v v v v v v
v v v v v v v v
w w w w w w w w
w w w w w w w w
w w w w w w w w
w w w w w w w w
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T t
t
PSfrag replacements
(0, 0)
(1, 0)
(0, 1)
(1, 1)
I Potentially correlated channels.
I Markov chain with 2N states.
I Imperfect sensing.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 30
Sensing and Access Policies
x x x x x x x xx x x x x x x xx x x x x x x xx x x x x x x x
y y y y y y y yy y y y y y y yy y y y y y y yy y y y y y y yz z z z z z z z zz z z z z z z z zz z z z z z z z zz z z z z z z z z
| | | | | | | || | | | | | | || | | | | | | || | | | | | | |
~ ~ ~ ~ ~ ~ ~ ~~ ~ ~ ~ ~ ~ ~ ~~ ~ ~ ~ ~ ~ ~ ~~ ~ ~ ~ ~ ~ ~ ~
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T
S1(1) = 0 S1(2) = 1 S1(3) = 0 S1(T ) = 0
SN(1) = 1 SN(2) = 0 SN(3) = 0 SN(T ) = 0
t
t
Sensing Policy πs
2 Deterministic: choose which channel to sense in each slot
2 Randomized: choose the probability of sensing each channel
Access Policy πc
2 Deterministic: whether to transmit based on the sensing outcome
2 Randomized: transmission probability based on the sensing outcome
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 31
Sensing and Access Policies
PSfrag replacements
Opportunities
Channel 1
Channel N0 1 2 3 T
S1(1) = 0 S1(2) = 1 S1(3) = 0 S1(T ) = 0
SN(1) = 1 SN(2) = 0 SN(3) = 0 SN(T ) = 0
t
t
Reward and Collision
2 A reward R(t) = Bi is accrued when access an idle channel i.
2 A collision with primary users occurs when access a busy channel.
2 A successful transmission is acknowledged at the end of the slot.
Objective:
max E[throughput] s.t. collision probability Pc ≤ ζ
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 32
Sensing Policy: How to Achieve Synchronous Hopping?
PSfrag replacements
0 t t + 1 TK(1) K(2) K(t − 1) K(t)
R(t) R(t + 1)
· · ·
· · ·
Use entire observ. history (ACK/NAK)
Max total remaining rewarda(t)
I Use common observ. (ACK/NAK) to ensure Tx-Rx synchronous hopping.
I Sufficient statistic: the a posterior distribution (belief vector) Λ(t) that
I exploits the entire observation history.
Λ(t) = [λ1, · · · , λ2N ]
λi = Pr[state is i | K(1), · · · , K(t − 1)︸ ︷︷ ︸
common observ.
]
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 33
Access Policy: How to Deal with Sensing Errors?
max E[throughput] s.t. collision probability Pc ≤ ζ
Consequences of trusting spectrum sensor:
2 idle sensed as busy ⇒ missed opportunity
2 busy sensed as idle ⇒ collision
Access Policy: when and how much to trust the sensor
tx probability =
p0 if idle
p1 if busy
p0 < 1 : conservative
p1 > 0 : aggressive
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm ε
Pro
babi
lity
of D
etec
tion
1 −
δ
δε
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 34
Joint PHY-MAC Design: A Constrained POMDP
π∗δ , π
∗s , π
∗c = arg max E[
T∑
t=1
R(t)], subject to Pc ≤ ζ
I Belief vector Λ(t) (conditional distribution)
I is a sufficient statistic
I πδ: Λ(t) → sensor operating point or its PDF
I πs: Λ(t) → sensing action or its PMF
I πc: Λ(t),sensing outcome → access action
I πc: Λ(t),sensing outcome or tx probability 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm ε
Pro
babi
lity
of D
etec
tion
1 −
δ
δε
A constrained POMDP often requires randomized policy for optimality.
0Y. Chen, Q. Zhao, and A. Swami, “Joint Design and Separation Principle for Opportunistic Spectrum Access in the Presence of Sensing Errors,” IEEE Trans. on Info. Theory, May 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 35
Separation Principle
π∗δ , π
∗s , π
∗c = arg max E[
T∑
t=1
R(t)], subject to Pc ≤ ζ
Separation principle: πδ and πc can be decoupled from πs
I Choose sensor operating policy πδ and access policy πc
I to maximize immediate reward R(t) and ensure constraint Pc = ζ.
I =⇒ Static optimization problem
I =⇒ Deterministic policy δ∗, π∗c in closed form.
I Choose sensing policy πs to maximize total reward E
[∑T
t=1 R(t)]
.
I =⇒ An unconstrained POMDP
I =⇒ Optimality achieved with deterministic policies.
0Y. Chen, Q. Zhao, and A. Swami, “Joint Design and Separation Principle for Opportunistic Spectrum Access in the Presence of Sensing Errors,” IEEE Trans. on Info. Theory, May 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 36
Separation Principle
π∗δ , π
∗s , π
∗c = arg max E[
T∑
t=1
R(t)], subject to Pc ≤ ζ
Separation principle: πδ and πc can be decoupled from πs
I Choose sensor operating policy πδ and access policy πc
I to maximize immediate reward R(t) and ensure constraint Pc = ζ.
I =⇒ A static optimization problem
I =⇒ Optimal policies π∗δ, π
∗c with a universal structure in closed form.
I Choose sensing policy πs to maximize total reward E
[∑T
t=1 R(t)]
.
I =⇒ An unconstrained POMDP
I =⇒ Optimality achieved with deterministic policies.
I =⇒ For i.i.d. channels, myopic has the same semi-universal structure and optimality.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 37
The Optimal Sensor Operating and Access Policies
I when δ > ζ (conservative access)
tx probability =
0 if busyζδ
if idle
I when δ < ζ (aggressive access)
tx probability =
ζ−δ1−δ
if busy
1 if idle
I when δ = ζ (optimal joint design)
tx probability =
0 if busy
1 if idle
PSfrag replacements
ε
1 − δ
1 − ζ
δ > ζ δ < ζ
conservative aggressive
optimal (δ∗ = ζ)
Optimal policies are deterministic, stationary, and model independent:
δ∗ = ζ, π∗c = trust the sensor.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 38
Separation Principle
π∗δ , π
∗s , π
∗c = arg max E[
T∑
t=1
R(t)], subject to Pc ≤ ζ
Separation principle: πδ and πc can be decoupled from πs
I Choose sensor operating policy πδ and access policy πc
I to maximize immediate reward R(t) and ensure constraint Pc = ζ.
I =⇒ Static optimization problem
I =⇒ Optimal policies π∗δ, π
∗c with a universal structure in closed form.
I Choose sensing policy πs to maximize total reward E
[∑T
t=1 R(t)]
.
I =⇒ An unconstrained POMDP
I =⇒ Optimality achieved with deterministic policies.
I =⇒ For i.i.d. channels, myopic has the same semi-universal structure and optimality
0Q. Zhao and B. Krishnamachari, “Structure and Optimality of Myopic Policy in Opportunistic Access with Noisy Observations,” submitted to IEEE Transactions on Automatic Control.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 39
Correlated Channels with Imperfect Sensing: Summary
I A constrained POMDP formulation for joint PHY-MAC design
I The separation principle
I Step 1: Design the spectrum sensor and the access policy
2 Being myopic is optimal.
2 A universal structure ⇒ optimality + simplicity + robustness.
I Step 2: Design the sensing policy
2 An unconstrained POMDP.
2 A semi-universal structure ⇒ optimality + simplicity + robustness for
i.i.d. Markov channels.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 40
Outline
I Slotted independent Markov channels with perfect sensing
I (Correlated) channels with imperfect sensing
I Unslotted primary systems
I Energy-constrained opportunistic spectrum access in fading
I Distributed sensing for multiple users
I Opportunistic spectrum access in self-similar primary traffic
I Concluding remarks
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 41
Unslotted Primary Systems
PSfrag replacements
White space
Channel 1
Channel N
0
1
2
3
t
t
I N channels, each with bandwidth Bi.
I Channel i: two-state continuous Markov process with transition rates µi, λi.
PSfrag replacements
0 1(busy) (idle)
λi
µi
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 42
Reduce to OSA in Slotted Primary Systems
Acknowledgement
¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡
¢ ¢ ¢ ¢ ¢ ¢ ¢ ¢¢ ¢ ¢ ¢ ¢ ¢ ¢ ¢¢ ¢ ¢ ¢ ¢ ¢ ¢ ¢¢ ¢ ¢ ¢ ¢ ¢ ¢ ¢
£ £ £ £ £ £ £ ££ £ £ £ £ £ £ ££ £ £ £ £ £ £ ££ £ £ £ £ £ £ £
¤ ¤ ¤ ¤ ¤ ¤ ¤ ¤¤ ¤ ¤ ¤ ¤ ¤ ¤ ¤¤ ¤ ¤ ¤ ¤ ¤ ¤ ¤¤ ¤ ¤ ¤ ¤ ¤ ¤ ¤
¥ ¥ ¥ ¥ ¥ ¥ ¥ ¥¥ ¥ ¥ ¥ ¥ ¥ ¥ ¥¥ ¥ ¥ ¥ ¥ ¥ ¥ ¥¥ ¥ ¥ ¥ ¥ ¥ ¥ ¥
¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦
§ § § § § § § § § § § §§ § § § § § § § § § § §§ § § § § § § § § § § §§ § § § § § § § § § § §
¨ ¨ ¨ ¨ ¨ ¨¨ ¨ ¨ ¨ ¨ ¨¨ ¨ ¨ ¨ ¨ ¨¨ ¨ ¨ ¨ ¨ ¨
© © © © © ©© © © © © ©© © © © © ©© © © © © ©ª ª ª ª ª ª ª ª ª ªª ª ª ª ª ª ª ª ª ªª ª ª ª ª ª ª ª ª ªª ª ª ª ª ª ª ª ª ª
« « « « « « « « « «« « « « « « « « « «« « « « « « « « « «« « « « « « « « « «
¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬ ¬
® ® ®® ® ®® ® ®® ® ®¯ ¯ ¯¯ ¯ ¯¯ ¯ ¯¯ ¯ ¯
Secondary users
Slot (L)
Sensing Transmission
PSfrag replacements
Opportunities
Channel 1
Channel N
0
1
2
3
Ls L − Ls
t
t
I Secondary users adopt a slotted transmission structure with slot length L.
I A slot is partitioned into sensing time (Ls) and transmission time (L − Ls).
I The chosen channel is an opportunity if it stays idle during the tx period.
I Unslotted tx of primary users absorbed by sensing errors.
I The problem can be reduced to that in a slotted primary system.
0Q. Zhao and K. Liu, “Detecting, Tracking, and Exploiting Spectrum Opportunities in Unslotted Primary Systems,” RWS 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 43
Outline
I Slotted independent Markov channels with perfect sensing
I (Correlated) channels with imperfect sensing
I Unslotted primary systems
I Energy-constrained opportunistic spectrum access in fading
I Distributed sensing for multiple users
I Opportunistic spectrum access in self-similar primary traffic
I Concluding remarks
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 44
Energy-Constrained OSA in Fading
Energy Constraint
I Both sensing and access cost energy
I Finite initial energy
Optimal Sensing and Access Policies
I may choose not to sense when the belief vector indicates all channels are
unlikely to be idle
I may choose not to access when channels are in deep fade
0Y. Chen, Q. Zhao, and A. Swami, “Distributed Spectrum Sensing and Access in Cognitive Radio Networks with Energy Constraint” to appear in IEEE Trans. Signal Processing.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 45
Outline
I Slotted independent Markov channels with perfect sensing
I (Correlated) channels with imperfect sensing
I Unslotted primary systems
I Energy-constrained opportunistic spectrum access in fading
I Distributed sensing for multiple users
I Opportunistic spectrum access in self-similar primary traffic
I Concluding remarks
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 46
Sharing among Competing Distributed Users
2 4 6 8 10 12 14 16 18 200
0.5
1
1.5
2
2.5
Time (slot)
Nor
mal
ized
net
wor
k th
roug
hput
(Bit
unit
per s
lot)
Multi−user approach in multi−user settingSingle−user approach in multi−user settingSingle−user approach in single−user setting
Tradeoff: choosing the best channel vs. avoiding competing secondary users.
0K. Liu, Q. Zhao, and Y. Chen, “Distributed Sensing and Access in Cognitive Radio Networks,” ISSSTA 2008.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 47
Spatial Opportunity Sharing
° °° °° °± ±± ±± ±
² ²² ²² ²³ ³³ ³³ ³
´ ´´ ´´ ´µ µµ µµ µ
¶ ¶¶ ¶¶ ¶¶ ¶· ·· ·· ·· ·
¸ ¸¸ ¸¸ ¸¹ ¹¹ ¹¹ ¹
PSfrag replacements
CH1 CH2
CH3A(1, 2)
C(3)
B(2)
D(1, 2)
E(1)
I Secondary users cannot use the channels assigned to nearby primary users.
I Neighboring secondary users interfere.
I How to allocate available channels to optimize a certain network utility.
I Tools: graph coloring and game theory.
0H. Zheng and C. Peng, “Collaboration and Fairness in Opportunistic Spectrum Access,” ICC 2005.0W. Wang and X. Liu, “List-coloring based channel allocation for open-spectrum wireless networks,” VTC 2005.
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 48
Outline
I Slotted independent Markov channels with perfect sensing
I (Correlated) channels with imperfect sensing
I Unslotted primary systems
I Energy-constrained opportunistic spectrum access in fading
I Distributed sensing for multiple users
I Opportunity spectrum access in self-similar primary traffic
2 K. Liu, X. Xiao, and Q. Zhao, “Opportunistic Spectrum Access in Self Similar Primary
Traffic,” IEEE Military Communication Conference (MILCOM), Nov., 2008.
I Concluding remarks
c©Q. Zhao, A. Swami, tutorial at MILCOM 2008. 49
Conclusion
Basic Components:
I Spectrum sensor: opportunity identification (PHY)
I Sensing policy: where in the spectrum to sense (MAC)
I Access policy: whether to tx given that sensing errors may occur (MAC)
Fundamental Tradeoffs
I Spectrum sensor: false alarm vs. miss detection
I Sensing policy: gaining immediate reward vs. gaining spectrum information
I Access policy: conservative vs. aggressive
I Spectrum sharing: choosing the best channel vs. avoiding competing users
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 1
Network Layer Issues in Opportunistic Spectrum Access
References
[1 ] W. Ren, Q. Zhao, and A. Swami, “Power Control in Spectrum Overlay Networks: How to Cross AMulti-Lane Highway,” ICASSP 2008.
[2 ] Q. Zhao, W. Ren, and A. Swami, “Spectrum Opportunity Detection: How Good Is Listen-Before-Talk?”Proc. of IEEE Asilomar Conference on Signals, Systems, and Computers, November, 2007.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 2
Long Hop vs. Relaying: We Know This
Whispering: Spatial Reuse Shouting: Fewer Hops
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 3
How to Cross A Multi-Lane Highway?
One lane at a time or dash through?
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 4
How to Cross A Multi-Lane Highway?
Detecting traffic in multiple lanes is more difficult.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 5
Unique Tradeoffs in Spectrum Overlay
I Transmission power affects how often opportunities occur.
2 Pr[opportunity] decreases exponentially with tx range squared (p2/αtx ).
I Transmission power affects the reliability of opportunity detection.
2 Reliable detection achieved when ptx/Ptx → 0 or ptx/Ptx → ∞.
I Optimal transmission power for transport throughput.
2
0W. Ren, Q. Zhao, and A. Swami, “Power Control in Spectrum Overlay Networks: How to Cross A Multi-Lane Highway,” ICASSP 2008.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 6
Quantification for Poisson Primary Networks
I Transmission power affects how often opportunities occur.
2 Pr[opportunity] decreases exponentially with tx range squared (p2/αtx ).
I Transmission power affects the reliability of opportunity detection.
2 Reliable detection achieved when ptx/Ptx → 0 or ptx/Ptx → ∞.
I Optimal transmission power for transport throughput.
2 p∗tx decreases with the traffic load of primary network.
0W. Ren, Q. Zhao, and A. Swami, “Power Control in Spectrum Overlay Networks: How to Cross A Multi-Lane Highway,” ICASSP 2008.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 7
Network Model
I Primary users form a Poisson point process with density λ.
I Each primary user transmits with probability p in a slot.
I Primary receivers are uniformly distributed within Rp of their transmitters.
PSfrag replacementsRp
Rp
Rp
2 Thinning Thm ⇒ Txs are Poisson.
2 Displacement Thm ⇒ Rxs are Poisson.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 8
Spectrum Opportunity: Definition
PSfrag replacements
A B
Interference
rI
RI
Primary Tx
Primary Rx
I RI: interference rangeI RI: of primary usersI RI: RI ∝ P
1/αtx
I rI: interference rangeI rI: of secondary usersI rI: rI ∝ p
1/αtx
A channel is an opportunity for A −→ B if
I the transmission from A to B can succeed
I the interference power to primary is below a prescribed level
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 9
Spectrum Opportunity: Properties
PSfrag replacements
A B
Interference
rI
RI
Primary Tx
Primary Rx
I RI: interference rangeI RI: of primary usersI RI: RI ∝ P
1/αtx
I rI: interference rangeI rI: of secondary usersI rI: rI ∝ p
1/αtx
I Detecting primary signals 6= detecting spectrum opportunity.
I Asymmetric (an opportunity for A −→ B may not be one for B −→ A).
0Q. Zhao, W. Ren, and A. Swami, “Spectrum Opportunity Detection: How Good Is Listen-Before-Talk?” Proc. of IEEE Asilomar Conference on Signals, Systems, and Computers, Nov., 2007.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 10
Probability of Spectrum Opportunity
Pr[opportunity] = Prno rx ≤ rI of A ∩ no tx ≤ RI of B
= exp
−pλ
∫∫
S0(rI+Rp,RI)
SI(r, Rp, rI)
πR2p
rdrdθ + πR2I
PSfrag replacements
A B
rI
RI
d
rI + Rp
S0(rI + Rp, RI)
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 11
Impact of Transmission Power on Pr[opportunity]
I Asymptotically Achievable Lower and Upper Bounds:
exp[−pλπ(r2I + R2
I)] < Pr[ opportunity ] ≤ exp(−pλπr2I)
I Pr[opportunity] decreases exponentially with r2I ∝ p
2/αtx .
100 200 300 400 500 6000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
rI
Pr[H
0]
Pr[H0]
exp[−pλπ(rI2+R
I2)]
exp(−pλπrI2)
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 12
Quantification for Poisson Primary Networks
I Transmission power affects how often opportunities occur.
2 Pr[opportunity] decreases exponentially with tx range squared (p2/αtx ).
I Transmission power affects the reliability of opportunity detection.
2 Reliable detection achieved when ptx/Ptx → 0 or ptx/Ptx → ∞.
I Optimal transmission power for transport throughput.
2 p∗tx decreases with the traffic load of primary network.
0W. Ren, Q. Zhao, and A. Swami, “Power Control in Spectrum Overlay Networks: How to Cross A Multi-Lane Highway,” ICASSP 2008.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 13
Spectrum Opportunity Detection
PSfrag replacements
A B
Interference
rI
RI
Primary Tx
Primary Rx
I RI: interference rangeI RI: of primary usersI RI: RI ∝ P
1/αtx
I rI: interference rangeI rI: of secondary usersI rI: rI ∝ p
1/αtx
Detecting primary signals 6= detecting spectrum opportunity.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 14
Detecting Primary Signals
PSfrag replacements
A
X
Y B
Primary TxPrimary Rx
rI
rD
Rp + rI
I rD: detection range.
I H0: no primary Tx within rD, H1: alternative.
I False alarms and miss detections occur due to noise and fading.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 15
From Detecting Signal to Detecting Opportunity
! !! !
""" ###
$$ %%
PSfrag replacements
A B
X
YPrimary TxPrimary RxrI
rD
RI
Rp + rI
PSfrag replacements
Prob. of Detection (1 − PMD)
Prob. of False Alarm0
1
1
rD ↓
rD ↑
I H0: opportunity, H1: alternative.
I Even with perfect ears, exposed Tx (X) ⇒ FA, hidden Rx (Y ) ⇒ MD.
I Adjusting detection range rD leads to different operating points.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 16
Detection Performance with Perfect Ears
I False Alarm Probability
PF = 1 − exp
−pλ
πr2
D − SI(d, rD, RI) −
∫∫
SA2
SI(r, Rp, rI)
πR2p
rdrdθ
I Miss Detection Probability
PMD =
exp(−pλπr2D) − exp
[
−pλ
(
π(r2D + R2
I) − SI(d, rD, RI) +∫∫
S0(rI+Rp,RI)−SA2
SI(r,Rp,rI)
πR2p
rdrdθ
)]
1 − exp
[
−pλ
(
∫∫
S0(rI+Rp,RI)
SI(r,Rp,rI)
πR2p
rdrdθ + πR2I
)]
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 17
Asymptotic Properties of ROC
I Reliable opportunity detection is achieved in two extreme regimes:
2 The point (PF (rD = RI), PD(rD = RI)) → (0, 1) when ptx/Ptx → 0.
2 The point (PF (rD = rI − RI), PD(rD = rI − RI)) → (0, 1) when ptx/Ptx → ∞.
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PF
PD
LBT (rI = 50)
LBT (rI = 300)
LBT (rI = 800)
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 18
Quantification for Poisson Primary Networks
I Transmission power affects how often opportunities occur.
2 Pr[opportunity] decreases exponentially with tx range squared (p2/αtx ).
I Transmission power affects the reliability of opportunity detection.
2 Reliable detection achieved when ptx/Ptx → 0 or ptx/Ptx → ∞.
I Optimal transmission power for transport throughput.
2 p∗tx decreases with the traffic load of primary network.
0W. Ren, Q. Zhao, and A. Swami, “Power Control in Spectrum Overlay Networks: How to Cross A Multi-Lane Highway,” ICASSP 2008.
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 19
Optimal Power for Transport Throughput
p∗tx = arg max d(ptx) Pr[ success | ptx] s.t. Pr[ collision | ptx] ≤ ζ
& && &' '' '
( (( () )) )*+*,+,
-+- .+.
PSfrag replacements
A B
Interference
rI
RI
Tx
Rx
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PF
PD
LBT (rI = 50)
LBT (rI = 300)
LBT (rI = 800)
PSfrag replacements
A
B
InterferencerI
RI
TxRx
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 20
Numerical Examples
0 0.5 1 1.5 2 2.5 3 3.50
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Normalized Hop Length
Nor
mal
ized
Tra
nspo
rt C
apac
ity
low primary traffic loadhigh primary traffic load
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10.5
1
1.5
2
2.5
3
p (λ = 10/2002)
r* I/RI
c©Q. Zhao and A. Swami, tutorial at MILCOM 2008. 21
Conclusion
I Transmission power affects how often opportunities occur.
2 Pr[opportunity] decreases exponentially with tx range squared (p2/αtx ).
I Transmission power affects the reliability of opportunity detection.
2 Reliable detection achieved when ptx/Ptx → 0 or ptx/Ptx → ∞.
I Optimal transmission power for transport throughput.
2 p∗tx decreases with the traffic load of primary network.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Programs, Policies, Standards
Programs Programs • Policies
St d d • Standards
1
CN Programs
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
CN Programs
• NSF Programmable Wireless Networks (ProWiN)www.nsf.gov/cise/cns/prowin.jsp
• European research initiative: end-to-end reconfigurability (E2R) e2r.motlabs.com/front-page
• WINNER: Wireless world initiative new radio www.ist-winner.org• U.S. National Institute of Justice
www.ojp.usdoj.gov/nij/topics/technology/communication/research-priorities.htm
• U.S. Army Spectrum Exploitation Program• DARPA XG and related programs• CERDEC CN StudyCERDEC CN Study • FCC www.fcc.gov/oet/spectrum• ITU www.itu.int/publications/main_pub/frequency_html• …
2
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
DARPA Programs related to DSA / CR
3Source: Chris Ramming’s presentation at WAND Proposer day. Feb 2006, online atwww.darpa.mil/STO/Solicitations/WAND/pdf/CBMANET_WAND_proposer_day_briefing.pdf
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
DARPA XG Program
Develop Technology and System Concepts for DoD to Dynamically Access All Available SpectrumDynamically Access All Available Spectrum
• Complexity of planning large-scale dynamic networks• Spectrum regulation policies and processes vary by
country region even by DoD unit country, region, even by DoD unit – Military: Purple problem; Coalition planning, with NGO’s
• Interference prevention and coexistence with QoS and cost constraintscost constraints
Basic Components: • Real-time low-power wideband spectrum sensing• Rapid characterization of signals and waveforms• React and Adapt: rapid formulation of best course of action
Credit: Preston Marhsall, XG Briefings, e.g., SWANS Conf., April 2005
4
http://www.daml.org/meetings/2005/04/pi/DARPA_XG.pdf
Cognitive Network Study at CERDEC
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Cognitive Network Study at CERDEC
Dynamic Spectrum Utilization
CNetwork ControlSoftware Radio
Evolution Network Operations
Sources: David Jimenez Director (A)
Network OperationsNetwork Learning /
Reasoning
David Jimenez, Director (A), CERDEC STCD, “STCD Future Efforts & Direction”, 1 AFCEA Luncheon,7 Jan 2008. Derek S. Morris, CERDEC STCD, “Cognitive Networking for Tactical Army Applications” Tekes
5
Army Applications , Tekes Workshop, Mar 2008.
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Programs, Policies, Standards
• Programs • Programs Policy – is a deliberate plan of action to guide decisions and achieve to guide decisions and achieve rational outcome(s) St d d • Standards
6
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Examples of Policyp y
802.22 90% detection within 2 seconds at prescribed power level
UWBFCC t l kFCC spectral mask
WNaN WNaN Maintain 3dB SNR margin at protected receiversVacate channel within 500 ms
7
Vacate channel within 500 ms
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Many sources of policy
Regulatory policy: interference avoidance; protection to emergency services incumbents and licenseesto emergency services, incumbents and licensees
Military policy: Mission needs; C2 priorities; CSI avoidanceavoidance
Radio policy: control radio parameters
Policy implementations may differ e g back off Policy implementations may differ e.g., back-off window durations; reactive-proactive routing …
Policy Reasoning Engine to verify policy, detect Policy Reasoning Engine to verify policy, detect conflicts, negotiate and resolve conflicts
Yes, No, possibly with additional constraints
8
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Typical Policy Life CycleTypical Policy Life CycleHigh Level
Specification of Operational
Policy System Translates OP
Policy System Analyzes OPOperational
Policies (OP) I eInto Machine-ReadableOperational Policies
y
for conflicts/errors
Policy System RefinesOP Into
Policy System Validates
Compliance
Graphic Adapted from ITA briefing by Dinesh Verma, Dakshi Agrawal et al
OP Into Deployable Policies
In Post-Mortem Some CR Policy Languages:OWL, CoRaL
Policy System Distributes
Deployable PoliciesTo MANET Devices
MANET Devices EnforcePolicies
Policy Systemupdates
Devices to PoliciesEffective Post Operation
9
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Policy Representation
Issues:Must provide rich interfaceExpressiveness vs. applicability to CRInteroperability and testability V&V Interoperability and testability – V&V Security issues with dynamic policy updates
Examples: Examples: UML, XTM, RDF, RDFS, …. OWL – Web-based Ontology Language (BBN in XG) CoRaL – Cognitive Radio Language (SRI in XG)
Berlemann et al, DySPAN 05Chapin-Sicker, IEEE Comm Mag 06; Wilkins et al IEEE Wireless Comm 07
10
Wilkins et al, IEEE Wireless Comm 07Kokar, The Role of Ontologies in Cognitive Radioin Cognitive Radio Technology, ed., B. Fette, 2006.ITA: www.usukita.org
XG Policy Reasoner
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
XG Policy Reasoner
Source: P Marshall and T Martin “XG Communications Program
11
Source: P. Marshall and T. Martin, XG Communications ProgramOverview”, at WAND Industry Day, 27 Feb 2007, available atwww.darpa.mil/STO/Solicitations/WAND/pdf/XG_overview_for_WAND.pdf
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Programs, Policies, Standards
• Programs • Programs • Policy
St d dStandards
12
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Commercial Standards
• 802 22 www ieee802/22• 802.22 www.ieee802/22• Also various 802.11x and 802.16x
• IEEE P1900 and SCC41
13
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
802.22
Wireless Regional Area Networks (WRAN)U TV b d 54 t 862 MH• Uses TV bands 54 to 862 MHz
• Provides 6 – 7 – 8 MHz bands• 4 W EIRP default • Explicit requirements :
– Avoid interference with incumbents – Channel sensing and measurements g– Coexist with DTV and wireless microphones
• Explicit sub-slots to signal p g`Urgent coexistence’ (PU detected) and `Self-coexistence’ (another WRAN cells detected)
14
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
802.22 PHY/MAC Draft
Largely based on 802.16
• Modulation – OFDMA (IEEE 802.16d/e – WiBro) – QPSK, 16-QAM, 64-QAM– Channel bonding and aggregation optional– Adaptive sub-carrier allocation– Adaptive pilot insertion– Channel coding: LDPC, STBC, Turbo – 2048 carriers: 1440 modulated
• MAC also based on 802.16 standard
15
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
Other standards with `CR’ featuresOther standards with CR features
802.11h - Dynamic Frequency Selection (DFS) and transmit power control (TPC) to avoid interference with radar and satellite control (TPC) to avoid interference with radar and satellite
• Quiet channels to test for presence of radar• Test channels for radar before and during use
http://standards.ieee.org/getieee802/download/802.11h-2003.pdf
802.11y – 3.65-3.7 GHz band (fixed satellite / radar band)• Provides for spatial exclusion zones• Location-based policy (e.g., near borders)• Must protect incumbent users • Must protect incumbent users
802.16 h – Explicit CR protocol, with co-existence and interference avoidance
802.11j - 802.11 designed specially for to conform to local policiesin Japan; operates in 4.9-5 GHz band.
16
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C l i D i S t AConclusion: Dynamic Spectrum Access
Dynamic Dynamic Spectrum
Access
Exclusive Use Model
Open Sharing Model
Hierarchical Access ModelUse Model Sharing Model Access Model
Spectrum Property
Rights
Dynamic Spectrum Allocation
Spectrum Underlay(UWB)
Spectrum Overlay
(OSA, pooling)
18Cognitive Radio
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
C n l i n: Sp tr m O rlConclusion: Spectrum OverlayPhysical Layer
Opportunity sensingInterference Aggregation
MAC LayerOpportunity tracking and learningOpportunity exploitation with imperfect sensingOpportunity exploitation with imperfect sensingOpportunity sharing
Network LayerNetwork LayerPower control and routing
R l t P li19
Regulatory Policy
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
21
Google TrendsCR:South Korea ------------------------Taiwan -----India -----China --USA -
SDR: DenmarkS. KoreaCzech RepSwedenRomania
© Q. Zhao, A. Swami, Tutorial at MILCOM 2008
S Op ISome Open IssuesCo-existence issuesHardware Issues: PHY – RF Security and privacy Models and measurements Increased cross-layer interactions Multicast …Policy translators yAll the usual radio issues with a twist True MANET & co-existence still far away ?
20
y