This presentation shows performance Optimization of Hybrid Fusion Cluster-based Cooperative Spectrum Sensing in Cognitive Radio Networks. For more details, send an email to ayman.elsaleh@gmail.com

1. Performance Optimization of Hybrid
Fusion Cluster-based Cooperative
Spectrum Sensing in
Cognitive Radio Networks
Presented by :
Name : Thong Wing Yew
Student ID : 1061103246
Course : Telecommunications
Supervisor : Mr. Ayman Abd El-Saleh
Moderator : Mr. Aaras Y. Kraidi
1

2. Presentation Outline
 Objectives
 Project Overview & Recap of FYP Part I
 Performance Criteria
 Simulation Outcomes for Neyman-Pearson and
Minimax Criteria
 Conclusion
 Recommendations
2

3. Objectives
 Part I
 Derivation of mathematical model of the soft-hard fusion for cognitive
radio network using Neyman-Pearson criterion.
 Compare the effects of different channel’s parameters on the
performance of the system.
 Evaluate the impact of different number of users of the system on the
performance of the system.
 Part II
 Derivation of mathematical model of the soft-hard fusion for cognitive
radio network using Minimax criterion.
 Evaluation of Threshold Analysis by simulation and mathematical
derivation.
 Evaluate the similar parameters and effect of users of the system
using the framework of Minimax.
3

4. Project
Overview
4

5. Project Overview
Performance Optimization of Hybrid Fusion Cluster-based
Cooperative Spectrum Sensing in Cognitive Radio Networks
• Spectrum Under-utilization Cognitive Radio
• Detect the presence of licensed Spectrum Sensing
PU
• Destructive channel effects Cooperative Spectrum Sensing
• Data Fusion
• Soft Decision Fusion (SDF)
Hybrid Fusion Scheme
• Hard Decision Fusion (HDF)
• Implementing Hybrid Fusion Scheme Cluster-based CSS
• Evaluate other schemes and
parameters that give the best result Performance Optimization
5

6. Spectrum Sensing
Spectrum Underutilization
Some portions of the
frequency band are unused
most of the time  CR
Hidden Terminal Problem
The accuracy of spectrum
sensing is reduced
 Cooperative SS

7. Cluster-based Cooperative Spectrum
Sensing
Cluster 1
Cluster 2
Cluster Header
(CH)
Base Station
Primary User (BS)
(PU)
Cluster 3
Secondary Users
(SU)
7

8. Hard Decision Fusion Vs Soft Decision
Fusion
Soft Decision Fusion (SDF)
0 = PU absent
Hard Decision Fusion (HDF)
1 = PU present
Cluster Header Base Station
(CH) (BS)
Fusion Detection Overhead Traffic
Techniques Performance
Hard Decision
Fusion (HDF)
Soft Decision
Fusion (SDF)
8

9. Probabilities Definition
Pdf
β H1
Pcr Pd
Energy (T)
Pmd Pfa
Pd = 1 – Pmd = P ( T > β | H1 )  Desired
Pfa = 1 – Pcr = P ( T > β | H0 )  Undesired
Pmd = 1 – Pd = P ( T < β | H1 )  Undesired
Pcr = 1 – Pfa = P ( T < β | H0 )  Desired
9

10. Performance Criteria
Neyman-Pearson
Criterion
&
Minimax Criterion

11. Neyman-Pearson Vs Minimax
 Neyman-Pearson Criterion (FYP Part I)
 Minimal interference caused to PU
 Maximize Pd for a given Pf
 The threshold is fixed
 Minimax Criterion (FYP Part II)
 Higher chances of interfering PU (more aggressive)
 Minimize the total Pe = Pf + Pm
 The threshold is adjusted dynamically

12. Neyman-Pearson Criterion
For SDF
Pd depends on a fixed value of Pf
as well as weighting coefficient, ω

13. Soft Decision Fusion Schemes
How to search for the best ω in SDF ?
Conventional Schemes Proposed Schemes
Equal Gain Combination (EGC) Normal Deflection Coefficient (NDC)
Weight assigned to M SUs is equally ∑ H 0 covariance matrix under hypothesis H
0
distributed *   −1

ω opt , NDC = ω opt , NDC / || ω opt , NDC ||= ∑ H 0 θ
ωi = 1
M where θ i = K PRi | g i | 2 | hi | 2 σ S
2
Maximal-Ratio Combining (MRC) Modified Deflection Coefficient (MDC)
Weight assigned is dependent on the ∑ H1 covariance matrix under hypothesis H1
PU SNR value at the SU
*   
SNRi ω opt ,MDC = ωopt ,MDC / || ω opt ,MDC ||= ∑ H 1 θ
−1
||ω|| = 1 ωi =
SNRT
13

14. The ROC Curve
Receiver operating characteristic (ROC) as performance evaluation for
different simulations.
Po a ilit o D t cio , Q
1
r b b y f ee t n
09
.
08
.
d
07
.
06
.
05
.
04
.
03
.
02
.
E c ll n
x ee t
01
. Go
o d
W rh s
o t le s
0
0 02
. 04
. 06
. 08
. 1
Po a il yo F l eAa m Q
r b b it f as l r ,
f
Area of 1 = Perfect Test
Area of 0.5 = Worthless Test
14

15. Minimax Criterion
Assuming Pf = Pm, where Pm= 1-Pd
β
Consider Pe = Pf + Pm

16. Pe Vs SNR Curve
• Similar to BER Vs SNR plot
• Best to have the lowest possible Pe for a low SNR value

17. Simulation
Outcomes
&
Results

18. Parameters To Be Evaluated
 Sensing Bandwidth, B
 Sensing Time of Secondary Users, Ts
 Number of SU per Cluster, M
 Number of Clusters, N
 Probability of Reporting Error, Pe
 Different Combinations of MN
 Different Spectrum Sensing Schemes
 Threhold Analysis for Minimax
18

19. Sensed Bandwidth, B
1
0.9
0.8
d
0.7
Probability of Detection, Q
0.6
0.5
0.4
0.3
8MHz
0.2
6MHz
0.1 4MHz
2MHz
0
0 0.2 0.4 0.6 0.8 1
Probability of False Alarm, Q
f
Higher Sensed Bandwidth is preferred but …. K = 2.B.Ts
19

20. Sensing Time, Ts
1
0.9
0.8
d
0.7
Probability of Detection, Q
0.6
0.5
0.4
0.3
50us
0.2
25us
0.1 10us
1us
0
0 0.2 0.4 0.6 0.8 1
Probability of False Alarm, Q
f
Ts Tx
Longer Sensing Time is good but …. Access Period
20

21. Number of SU per Cluster, M
1
0.9
0.8
Probability of Detection, Q d
0.7
0.6
0.5
0.4
0.3
M=15
0.2
M=10
0.1 M=5
M=1
0
0 0.2 0.4 0.6 0.8 1
Probability of False Alarm Q
,
f
Higher M gives better results!
21

22. Number of Clusters, N
1
P b b i y fDt c o ,Q
0.9
r ai t o e t n
0.8
e i
d
0.7
0.6
0.5
l
0.4
o
0.3 N1
=0
N8
=
0.2
N6
=
0.1 N4
=
N2
=
0
0 0.2 0.4 0.6 0.8 1
P b bi y fF l e l r ,Q
r ai t o a A m
o l s a
f
Higher N gives better results!
22

23. MN Combination (Neyman-Pearson)
1
0.9
0.8
d
0.7
Probability of Detection, Q
0.6
0.5
0.4 MN = 15 MN = 4
0.3
M=15, N=1
0.2
M=5, N=3
0.1 M=3, N=5
M=1, N=15
0
0 0.2 0.4 0.6 0.8 1
Probability of False Alarm, Q
f
When M increases More SDF involved Better Performance
23

24. Probability of Reporting Error, Pe
1
0.99
0.98
Probability of Detection, Qd
0.97
0.96
0.95
0.94
0.93
Pe = 0
0.92
P = 0.15
e
0.91 Pe = 0.3
0.9
0 0.2 0.4 0.6 0.8 1
Probability of False Alarm Q
,
f
CH BS

25. Threshold Analysis (SNR=10dB)

26. Threshold Analysis (SNR=5dB)

27. Single Link Sensing Schemes
• SDF has better performance than HDF
• Proposed SDF schemes are better than conventional SDF schemes

28. Double Link Sensing Schemes (Neyman Pearson)
1
0.9
0.8
0.7
d
Probability of Detection, Q
0.6
0.5
0.4
0.3
SDF-SDF(NDC-NDC)
SDF-HDF(NDC-OR)
0.2 SDF-SDF(MRC-MRC)
SDF-HDF(MRC-OR)
0.1 SDF-SDF(EGC-EGC)
SDF-HDF(EGC-OR)
HDF-HDF (OR-OR)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Probability of False Alarm, Q
f
Spectrum SDF SDF
Sensing HDF HDF
Primary User (PU) Secondary Users (SU) Cluster Header (CH) Base Station (BS)

29. Double Link Sensing Schemes (Minimax)

30. Conclusion
 Cognitive radio is a way to maximize spectrum
utilization
 Hard Fusion – Less Overhead but Poorer Performance
Soft Fusion - Better Performance but Higher Overhead
 Employing Soft-Hard Fusion to get the best of both
methods (Hybrid Fusion)
 Higher Sensing Time and Bandwidth yields better
detection performance
 The proposed SDF schemes (NDC & MDC) outperform
the conventional SDF ones (EGC & MRC)
30

31. Recommendation for Future Works
• Explore the possibilities and effect of introducing the
weighting coefficients at different stages or links of the
network.
• Determine the best number of SU per cluster that gives
the best detection performance.
• Efficient way of selecting CH, either from an ordinary SU
or a dedicated BS.
• Develop an algorithm that minimize the sensing time of
a SU.
31

32. Thank You
Q & A Session
32