1
24 June 2010 1
Fusion Strategies for Collabora3ve Spectrum Sensing
Kamran Arshad and Klaus Moessner Centre for Communica3on Systems Research (CCSR),
University of Surrey, United Kingdom Email: [email protected]
2010 ERRT Workshop Mainz Germany 2 24 June 2010
Outline
• Objective • Motivation • Spectrum Sensing • Collaborative Spectrum Sensing
• Hard decision based fusion optimisation • Genetic Algorithms based soft decision fusion optimisation
• Conclusions
3 24 June 2010
Outline
• Objective • Motivation • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
24 June 2010 4
Objective
2010 ERRT Workshop Mainz Germany
“To optimise performance of
collaborative spectrum sensing.”
2
5 24 June 2010
Outline
• Objective • Motivation • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
6 24 June 2010
Motivation for this research...
• People/users do not need spectrum, they need capacity and adequate quality communication means!
• Technology converts the limited spectrum into capacity
2010 ERRT Workshop Mainz Germany
Demand for Spectrum
Limited Resource
Crowded
Bandwidth hungry
Services
7 24 June 2010
Motivation – ctd.
• Possible solutions – Use higher spectrum (higher frequency bands)
• Do fundamental physical limits allows us to do so?
– Increase modulation efficiency or spectrum efficiency • Can we go beyond Shannon capacity?
– MIMO techniques • How much spectrum can be squeezed through more efficient antenna
techniques?
– Cognitive Radio exploiting White Spaces?
2010 ERRT Workshop Mainz Germany 8 24 June 2010
Motivation – ctd.
• What is a White Space (WS)? – Unoccupied spectrum bands at a particular location and time – Exists even in urban areas (depending on location and time)
• Where does WS come from? – Fixed and rigid spectrum allocations – Terrain signal blockage – Uneven demand for spectrum – TVWS emerge due to the digital switchover in Britain (Europe,
USA and some other countries as well)
• How to utilise them? – Cognitive Radio by performing spectrum sensing
2010 ERRT Workshop Mainz Germany
3
9 24 June 2010
Motivations - ctd
2010 ERRT Workshop Mainz Germany
Source: OFCOM /Dettmer R, ‘Up the revolution’, IEE Review, May 2005, p. 44
RURAL UK
SUBURBAN UK (near Heathrow
airport)
URBAN UK (Central London) Spectrum scarcity is mainly due to fixed
allocation policies
Unused
Heavily used
10 24 June 2010
Cognitive Radio as “enabler”
• Many definitions exist • In simple words:
– An intelligent radio that makes decision for its operating frequency, modulation scheme, transmitting power etc based on factors like:
• Current location • Policies at that particular location • Time of the day and available white
spaces • Negotiations with other opportunistic
devices
2010 ERRT Workshop Mainz Germany
11 24 June 2010
Outline
• Objective • Motivations • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
12 24 June 2010
Spectrum Sensing
• Goal is to reliably detect the presence of Primary (Licensed) User
• Three main approaches: – Match Filter detection – Energy Detection – Cyclostationary
Feature Detection
2010 ERRT Workshop Mainz Germany
Energy Detection • “Optimal” detector • Simple architecture • Easy to implement • Less complexity
Spectrum Sensing
Spectrum Decision
Channel Capacity
Primary User Detection
RF Stimuli
Spectrum Hole
Radio Environment
Spectrum Mobility
Decision Request
Transmitted Signal
Spectrum Sensing
Spectrum Sharing
4
13 24 June 2010
Local Spectrum Sensing - Results
2010 ERRT Workshop Mainz Germany
10-4 10-3 10-2 10-1 10010-4
10-3
10-2
10-1
100
Probability of false alarm, Pf
Prob
abilit
y of
mis
s de
tect
ion,
Pm
Shadowing (σdB = 6) - analysis
Shadowing (σdB = 6) - simulation
Shadowing (σdB = 2) - simulation
Shadowing (σdB = 2) - analysis
Shadowing (σdB = 12) - analysis
Shadowing (σdB = 12) - simulation
AWGN - simulationAWGN - analysisRayleigh Fading - analysisRayleigh Fading - simulation
Simulation Parameters Nodes = 1 SNR = 5dB BT product = 5 Channel(s) = AWGN,
Rayleigh, Shadowing
14 24 June 2010
Local Spectrum Sensing – Results (Ctd)
2010 ERRT Workshop Mainz Germany
-15 -10 -5 0 5 1010-4
10-3
10-2
10-1
100
SNR (in dB)
Prob
abilit
y of
mis
s de
tect
ion,
Pm
AWGNRayleigh FadingShadow Fading (σdB=6)
Simulation Parameters Nodes = 1 SNR = 5dB BT product = 5 Pfa = 10-1
Channel(s) = AWGN, Rayleigh, Shadowing
15 24 June 2010
Limitations of Local Spectrum Sensing
2010 ERRT Workshop Mainz Germany
Interference
Rayleigh Fading
Weak signals recvd. due to multi-path fading and may not detect PT
Interference
Cannot detect PT
Shadowing
CR can not detect PT due to large obstacles
PT
PU
CR
Collaborative Spectrum Sensing
16 24 June 2010
Outline
• Objective • Motivations • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
5
17 24 June 2010
Collaborative Spectrum Sensing (CSS)
• A central unit (fusion centre) collects sensing information, identifies the available spectrum, and broadcasts this information to other cognitive radios
• Use of control channels to share spectrum sensing result • Nodes may send 1-bit decision (Hard decision - HDC) or
observation (Soft decision - SDC) to fusion centre • Why collaboration?
– Significantly decreases the probabilities of mis-detection and false alarm
– Helps solving hidden primary user problem – More effective when collaborating users observe independent
fading or shadowing
2010 ERRT Workshop Mainz Germany 18 24 June 2010
Performance of CSS - HDC
2010 ERRT Workshop Mainz Germany
10-4 10-3 10-2 10-1 10010-4
10-3
10-2
10-1
100
Probability of false alarm, Pf
Prob
abilit
y of
mis
s de
tect
ion,
Pm
2 users6 users10 users1 userAWGN
-15 -10 -5 0 5 1010-5
10-4
10-3
10-2
10-1
100
SNR (in dB)
Prob
abilit
y of
mis
s de
tect
ion,
Pm
1 user2 user4 user8 user20 userAWGN
Nodes = 5 SNR = 5 dB BT product = 5 Fusion rule = OR Channel(s) = AWGN, Shadowing
Nodes = 5 Pfa = 10-1
BT product = 5 Fusion rule = OR Channel(s) = AWGN, Shadowing
19 24 June 2010
Outline
• Objective • Motivations • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
20 24 June 2010
Optimum Fusion for HDC
• Is the “OR” fusion rule superior in all cases? • Three different scenarios are considered:
– Case 1 • All users have similar mean SNR
– Case 2 • Half of the users have higher mean SNR than other half
– Case 3 • When only one user has high mean SNR
• Decision Fusion Rule – Voting, OR, AND, 1-user rule
• Analytical formulation and Monte-carlo simulations were carried out to find optimal fusion in HDC
2010 ERRT Workshop Mainz Germany
6
24 June 2010 21
Analytical Formulation - HDC
2010 ERRT Workshop Mainz Germany
M = Total number of users Di = 1-bit decision of ith user
Global probability of detection and false alarm is given as,
Pd and Pf are local probabilities S0 = group of users decided signal is absent S1 = group of users decided signal is present
R(D) is decision fusion rule at fusion centre and defined as,
Where ith user is chosen as,
For 1-user rule
24 June 2010 22
Results - HDC
2010 ERRT Workshop Mainz Germany
AWG
N C
hann
el
Case 1 - All users have similar SNR Case 2 – Half of the users have high SNR Case 3 – Only one user have high SNR
Nodes = 5 BT product = 5 Channel(s) = AWGN
24 June 2010 23
Results - HDC
2010 ERRT Workshop Mainz Germany
Ray
leig
h Fa
ding
Cha
nnel
Nodes = 5 BT product = 5 Channel(s) = Rayleigh
Case 1 - All users have similar SNR Case 2 – Half of the users have high SNR Case 3 – Only one user have high SNR
24 June 2010 24
Results - HDC
2010 ERRT Workshop Mainz Germany
Sha
dow
ing Nodes = 5
BT product = 5 Channel(s) = Shadowing
Case 1 - All users have similar SNR Case 2 – Half of the users have high SNR Case 3 – Only one user have high SNR
7
25 24 June 2010
Outline
• Objective • Motivations • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
26 24 June 2010
Soft Decision Combining (SDC) Framework
2010 ERRT Workshop Mainz Germany
How to fuse local observations of cognitive radios at fusion centre to decide globally the existence of licensed user?
u1,γ1 Fusion Centre
Global Decision
g1
g2
gM
GA based Learning Module
Channel Estimator
u2,γ2
uM,γM
Fusion Module
Prob
lem
So
lutio
n A
ppro
ach For a given channel
conditions and targeted probability of false alarm, weights are assigned to the secondary user observations in such a way that it maximises global probability of detection. Optimum weights are calculated using genetic algorithms.
27 24 June 2010
Problem definition - SDC
• Maximise global probability of detection at the fusion centre, considering – Two scenarios (users with same mean SNR and with different
mean SNR values) – Noisy reporting channels with channel gains
• Global probability of detection can be defined as Qd = Q(f(w)), where f(w) is given by
2010 ERRT Workshop Mainz Germany
Problem
24 June 2010 28
Proposed weighted framework
2010 ERRT Workshop Mainz Germany
CR 1
CR 2
CR M
X
X
X
+
+
+
X
X
X
.
.
.
Fusion Rule
u 1
u 2
u M
g 1
g 2
g M
n 1
n 2
n M
w 1
w 2
w M
Sensing Environment
x 1
x 2
x M
Reporting Channels Fusion Centre
y 1
y 2
y M
y c Genetic Algorithm (GA)
8
29 24 June 2010
Why genetic algorithm?
• Very useful for complex and loosely defined problems.
• Quickly can scan a vast solution set.
• Global optimisation technique.
• Does not have to know any rules of the problem. – It works by its own internal
rules. • Supports parallel
processing. – Multiple solution capability
2010 ERRT Workshop Mainz Germany
Generation of initial
population
Start
Evaluate the current population
Is the reached quality
sufficient ?
Select the best
individual
StopMaximum number of iteration ?
Proportional Selection
Crossover
Mutation
Yes
No
Yes
No
Gen
erat
iona
l R
epla
cem
ent
Best solution
Search SpaceCoding/ Mapping
Decoding
24 June 2010 30
Simulation results - SDC
2010 ERRT Workshop Mainz Germany
10-3 10-2 10-1 10010-3
10-2
10-1
100
Probability of False Alarm, Qf
Prob
abili
ty o
f Mis
s D
etec
tion,
Qm
M=1, EGC (same SNR)M=3, EGC (same SNR)M=6, EGC (same SNR)M=6, EGC (different SNR)M=6, PC (different SNR)
Performance loss due to different SNR
PC – Proportional Combining EGC - Equal Gain Combining
24 June 2010 31
Simulation results - SDC
2010 ERRT Workshop Mainz Germany
10-3 10-2 10-1 10010-3
10-2
10-1
100
Probability of False Alarm, Qf
Prob
abilit
y of
Mis
s De
tect
ion,
Qm
Pf vs Pm in AWGN channel with different SNR
EGC, no channel gainsSC, no channel gainsEGC, with channel gainsSC, with channel gainsOPT, with channel gainsOPT, no channel gains
OPT = GA based optimal combining PC = Proportional Combining EGC = Equal Gain Combining
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.80
0.1
0.2
0.3
0.4
0.5
Probability of False Alarm, Qf
Pro
babi
lity
of M
iss
Det
ectio
n, Q
m
OPT (case 2)PC (case 2)EGC (case 3)PC (case 3)EGC (case 2)OPT (Case 3)EGC (case 1)PC (case 1)
AWGN Rayleigh
Case 1 - All users have good reporting channels Case 2 – All users have bad reporting channels Case 3 – Only two users have good reporting channels
32 24 June 2010
Outline
• Objective • Motivations • Spectrum Sensing • Collaborative Spectrum Sensing
– Hard decision based fusion optimisation – Genetic Algorithm based soft decision fusion optimisation
framework
• Conclusions
9
33 24 June 2010
Conclusions
• Collaborative spectrum sensing improves sensing performance significantly
• Different position (mean SNR) of users have significant effect on the performance of collaborative spectrum sensing
• Optimum fusion rule must consider mean SNR values of users in both cases i.e. HDC and SDC
• Proposed Genetic Algorithm based weighted collaborative spectrum sensing improves sensing performance
• Proposed scheme requires knowledge about SNR of each user as well as channel conditions – Larger reporting channel bandwidths are required – Topic of our current research
2010 ERRT Workshop Mainz Germany 34 24 June 2010
Acknowledgments
This work was performed in project E3 and QoSMOS which have received research funding from the Community’s Seventh Framework programme. This paper reflects only the author’s views and Community is not liable for any use that may be made of the information contained therein. The contributions of colleagues from E3 and QoSMOS consortium are hereby acknowledged.
2010 ERRT Workshop Mainz Germany