AbstractIn Cognitive Radio networks energy detection is one of the most common techniques used to detect primary user

activity. This is because its low cost and complexity compared to

other detection techniques. In order to enhance the performance

of energy detection, cognitive radio nodes can cooperate together

to decide whether the primary user is active or not.

In this paper, different fusion rules for cooperative sensing are

analyzed. In addition to analytical results, Monte-Carlo

simulations are carried out to verify the results and study effects

of different parameters like SNR and number of samples.

Index TermsCognitive Radio, Cooperative Sensing, Energy

Detection

I. INTRODUCTION

Radio spectrum is considered one of the most valuable

resources for all countries. Current communication systems can

utilize a limited portion of spectrum bands. Cognitive Radio

(CR) is a promising radio system introduced to enhance spectral

efficiency by allowing non-licensed users to utilize spectrum

holes. This will utilize the spectrum more efficiently because

the licensed users will not use the spectrum all the time in all

places. Currently, FCC allowed unlicensed users to use white

spaces in TV spectrum bands [3].

One of the main objectives of CR is to enhance spectrum

utilization without affecting the primary (licensed) user. To

achieve this, the secondary unlicensed user (SU) should sense

the medium to detect primary user (PU) activity. When the SU

detects that primary band is idle, it will decide to access the

medium. The detection of primary user activity is very critical

because collision will occur if the SU miss-detected the PU.

To improve the sensing process, many detectors could be

used like matched filters, energy detectors and cyclostationary

detectors. Moreover, cooperative sensing can enhance the

sensing efficiency by exploiting readings from different CR

nodes. This yields better accuracy than individual sensing.

In this paper, the performance of different fusion rules will

be analyzed through analytical and simulation results.

The paper is organized as follows: section II describes the

system model and Monte-Carlo simulation. Section III

discusses the performance of cooperative sensing fusion rules.

In section IV, the impact of SNR on cooperative sensing is

analyzed. The impact of number of samples is investigated in

section V. Section VI shows the impact of changing number of

cooperative nodes. Finally, different cooperative sensing

schemes are compared in section VII.

II. SYSTEM MODEL

In this paper, performance of cooperative spectrum sensing

techniques are evaluated using analytically. In addition, Monte-

Carlo simulations are used to verify the obtained analytical

results. We consider a system where Nu secondary users sense

the existence of a primary signal and exchange their sensing

information. Primary signal and noise are modeled as zero

mean Gaussian distribution with variance 2,

2 , respectively.

The SNR is assumed to be the same at all cooperative nodes.

Each of the cooperating nodes makes its own decision about the

existence of primary user by comparing the energy of samples of the received signal with certain threshold . As depicted in Fig. 1, these individual decisions are sent to a central

node called fusion center which receives individual decisions

and apply certain fusion rule to get the final decision. In this

paper, the following fusion rules are analyzed:

1) OR Fusion Rule:

In this rule, a PU is detected if at least one of

individual nodes detected the PU. 2) AND Fusion Rule:

In this rule, a PU is detected if all individual nodes

detected the PU. 3) Majority Fusion Rule:

In this rule, a PU is detected if the majority of

Performance Evaluation of Cooperative Sensing Schemes in Cognitive Radio Networks

Hazem Abdel Megeed, Hossam Taha, Mohamed Fouad, Nassr Ismail

Fig. 1 System model of cooperative sensing

individual nodes detected the PU. 4) M out of N Fusion Rule:

In this rule, a PU is detected if M out of N individual

nodes detected the PU. This is the most general case

where all other rules can be obtained at different values

for M. And fusion rule is obtained when having M =1

and Or fusion rule can be obtained by choosing M=N.

This rule can also yields Majority fusion rule when

M=N/2. For each fusion rule, analytical expression of probability of

false alarm and probability of detection is studied. These analytical results will be verified from the Monte-Carlo

simulation.

A. Analytical Model

As shown in [1], , of a single node preforming energy detection could be calculated as follows:

= (

2

24) = (

2

2)

= ( (

2 + 2)

24 + 422) =

(

2 (1 +

2

2)

2 + 42

2 )

Using the expression of , the threshold used for energy comparison is:

= 2(21() + )

According to fusion rule applied by fusion center, the

performance of , is determined. The following expressions model the performance of each fusion rule:

1) OR Fusion Rule:

_ = 1 (1 )

=1

_ = 1 (1 )

=1

2) AND Fusion Rule:

_ =()

=1

_ =()

=1

3) Majority Fusion Rule:

_ = (()

(1 ))

=2 +1

_ = (()

(1 ))

=2 +1

4) M out of N Fusion Rule:

_ = (()

(1 ))

=

_ = (()

(1 ))

=

Fig.4. ROC curve for Majority fusion rule (Ns=10, Nu=10, SNR=-5 dB)

Fig.5. ROC curve for M out of N fusion rule (Ns=10, Nu=10, M=3, SNR=-5 dB)

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

1M out of N Fusion Rule "M = 3, N = 10"

Probability of False AlarmP

robabili

ty o

f D

ete

ctio

n

Monte-Carlo

Theoretical

Fig.2 ROC curve for AND fusion rule (Ns=10, Nu=10, SNR=-5 dB)

Fig.3 ROC curve for OR fusion rule (Ns=10, Nu=10, SNR=-5 dB)

B. Monte-Carlo Simulation

Monte-Carlo simulations are performed using Matlab.

Simulation parameters are shown in Table 1. Noise and PU

signals are generated from Gaussian distribution with zero

mean and variance 2,

2 respectively. To simulate an AWGN

channel we generated the primary user signal as a vector of

length Ns where the elements of this vector are driven from

Gaussian distribution. Also, we generated AWGN vector of

length Ns driven from Gaussian distribution too.

At the receiver side SUi the received signal () can be expressed as:

() = () + () Where S(t) is the primary signal and W(t) is the AWGN.

Each SU compute the test statistic Energy of the received

signal as follows:

() = |()|2

=1

Now each user can calculate the energy detection threshold

from theoretical expression. Then, all nodes take their decisions as follows:

Decide primary user exists if:

() And it doesnt exist otherwise. To calculate the practical ,

() is compared against () = (), and the false alarm occurs whenever:

() Then each fusion rule is applied to get the final decision.

Because of the random nature of experiment, the decision

process will be simulated for several iterations. Finally, , will be calculated for each fusion rule.

III. PERFORMANCE OF COOPERATIVE SENSING

To measure the performance of different detectors

techniques, it is very common to use receiver operation

characteristic (ROC) curve. This curve shows the relation

between probability of detection and probability of false alarm.

In addition, the area under the ROC curve, denoted by AUC, is

also an important metric. If the detectors performance is no

other than flipping a coin, AUC will be 1

2, and it increases to

one as the detector performance improves [2]. Using theoretical

model and Monte-Carlo simulation, the performance of each

fusion rule is illustrated below.

A. OR Fusion Rule

The performance of hard cooperative sensing using OR

fusion rule is shown in Fig. 3. Theoretical results are compared

with Monte-Carlo simulations. It is observed that the theoretical

results and simulations have a good agreement with a slight

difference that decreases at very low and high probability of

false alarm.

Fig. 8. Comparison between theoretical ROC curves of different fusion rules

(Ns=10, Nu=10, M=7, SNR=-5 dB)

Fig. 9. Impact of changing M in M out of N on probability of detection (Ns=10,

SNR=-5 dB)

Fig. 6. Impact of SNR on single node ROC curve (Ns=10, Nu=10,

SNR=-5dB)

Fig. 7. Impact of number of samples on majority fusion rule ROC curve

(Ns=10, Nu=10, SNR=-5 dB)

TABLE I MONTE CARLO SIMULATION PARAMETERS

Symbol Quantity Value

Nu Number of users 10

Ns Number of samples 10

SNR Signal to noise ratio 10 dB Iterations Number of iterations 10000

B. AND Fusion Rule

Fig. 2 depicts ROC curve for hard cooperative sensing using

AND fusion rule. Theoretical results are compared with Monte-

Carlo simulations. It is observed that the proposed theoretical

results agree with the simulations with a slight difference that

decreases at very low and high probability of false alarm.

C. Majority Fusion Rule

As shown in Fig. 4 the difference between the theoretical and

Monte-Carlo simulation for Majority Fusion rule can be

elaborated as follows: the simulation based performance

approaches the theoretical when probability of false alarm

exceeds 0.3, which means that Majority Fusion rule in practice

can reach theoretical limits for relatively large values of

detection threshold.

D. M Out of N Fusion Rule

In Fig. 5 as shown in the graph the theoretical curve

represents better performance for low values of probability of

false alarm whereas for larger values both the theoretical and

the Monte-Carlo simulation based performance are nearly

IV. IMPACT OF SNR ON COOPERATIVE SENSING PERFORMANCE

As shown in Fig. 6, the impact of SNR on the performance

of energy detector is illustrated for single node non-

cooperative cognitive radio network. As shown in the figure

the probability of detection improves greatly with increasing

the SNR values, at SNR value of 10 dB the area under the ROC

curve approaches 1 which indicates that the energy detector in

this case operating efficiently and accurately, this is a result of

higher power levels of the sent signal being received and the

detector can easily differentiate between noise and primary

signal, hence we can conclude that increasing the SNR value

yields a near optimum detection.

V. IMPACT OF NUMBER OF SAMPLES ON COOPERATIVE SENSING

Fig. 7 depicts the performance of majority fusion rule at

different number of samples (NS). As expected, the probability

of detecting the primary user increases with increasing the

number of samples. As the number of samples increases, more

samples are used in sensing the existence of a primary user

which in turns increases probability of its detection. The same

thing is noticed in case of single node as shown in Fig. 10.

VI. IMPACT OF NUMBER OF COOPERATIVE NODES

As shown in Fig. 12, OR fusion rule needs only one node with

detection decision to decide that primary user exists, hence,

with large number of nodes, the probability that any node takes

a detection decision is larger, thus for different values of

Probability of false alarm, the increase in number of nodes

results in higher probability of detection. If the system accepts

larger value for probability of false alarm, then the system will

have larger value for probability of detection compared to lower

values for probability of false alarm at same number of nodes.

Efficiency of this system appears at large number of nodes. At

low values for probability of false alarm, reasonable values for

probability of detection are obtained.

On the contrary, AND fusion rule needs all nodes to have

detection decision to decide that primary user exists, hence,

with small number of nodes, the probability that all nodes take

a detection decision is high. But, with larger number of nodes,

this probability decreases dramatically and tends to zero. As

shown in Fig. 13, at the same number of nodes, if the system

accepts larger value for probability of false alarm, then the

system will go to zero probability of detection slower than the

case with lower values for probability of false alarm. This

implies that this rule is not efficient for large number of nodes

as this may give wrong decisions if only one node miss-detected

the primary user.

Majority fusion rule could give accepted values for

probability of detection as shown in Fig. 14, but this requires a

system that accepts slightly high values for probability of false

alarm. For each value of probability of false alarm, probability

of detection increases with increasing number of users then

saturates at a specific value. This value depends on probability

of false alarm design value. After saturation, probability of

detection curve gradually fall back to zero. The reason for this

behavior is that at the large number of nodes, the probability

that

2 + 1 nodes can detect the primary user decreases. The

point at which the curves go down is different according to the

accepted probability of false alarm in the system. This approach

achieves very good performance for different values of

probability of false alarm at SNR values > -2 dB.

In case of M out of N fusion rule shown in Fig. 15, value of

M is the important factor. Small values of M make the

Fig.10. Impact of Number of Samples (Ns) on single node ROC curve

(Ns=10, Nu=1, SNR=-5 dB)

Fig.11. Impact of SNR on single node ROC curve

(Ns=10, Nu=10, SNR=-5 dB)

probability of detection curve goes to one early. The probability

to have small number M nodes that have a detection decision is

high, especially for large values of total number of nodes N.

When M value increases, the probability to have M nodes with

detection decision at the same time decreases. Thus, the

probability of detection curve rises to high values and reaches

to one late. However, this rule is suitable for different systems

due to its sort of flexibility by changing M value.

VII. PERFORMANCE COMPARISON OF DIFFERENT FUSION RULES

Fig. 11 depicts the theoretical ROC curve of energy detection

algorithm for different fusion rules in co-operative Cognitive

Radio networks. The fusion rules performance impact can be

interpreted and analyzed by comparing these rules against each

other and against the non-cooperative case Single Node curve

as follows:

OR Fusion Rule: In this rule the primary user is detected

by the overall cognitive network whenever at least one

secondary user detects the primary user, the same applies

for the false alarm, hence the rapid increase in the value

of probability of detection compared to the single node

curve and the other fusion rules curves as well.

AND Fusion Rule: In this rule in order to detect the

presence of the primary user, all the secondary users In

the CR network must detect the primary user at the same

time, so as shown in the figure the AND fusion rule

yields the lowest probability of detection even compared

to the single node curve because in the AND rule any

faulty decision taken by any secondary user will affect

the overall system.

M out of N Rule: This is the most general rule where it

introduces a trade-off between the high probability of

detection of the OR rule and the low probability of False

alarm of AND rule. In our simulations M=7 is chosen.

As expected, its ROC curve lies between the two

extremes (AND and OR curves).

Majority Rule: In this rule for CR network to detect the

primary user, at least more than half the number of

secondary users must detect the primary user at the same

time. This rule yields an intermediate performance

amongst the other 3 rules and the single user too.

VIII. CONCLUSION

. Cognitive Radio is a promising system that utilizes the

spectrum and boost user experience. Sensing the primary users

is considered one of the main challenges of cognitive radio

systems. In this paper, different hard combining cooperative

sensing schemes are investigated analytically and using Monte-

Carlo simulations. OR rule offers the highest probability of

detection. However, this comes on the expense of having high

probability of false alarm. In contrast, and rule introduces the

lowest probability of false alarm, yet with low probability of

detection. Majority and M out of N rules introduce a tradeoff

between the AND and OR rules. Simulation results verify the

obtained analytical results.

REFERENCES

[1] Spyros Kyperountas, Neiyer Correal and Qicai Shi, A Comparison of Fusion Rules for Cooperative Spectrum Sensing in Fading Channels, EMS Research, Motorola

[2] Jiaqi Duan and Yong Li, "Performance analysis of cooperative spectrum sensing in different fading channels," in Proc. IEEE International

Fig. 14. Impact of number of nodes on PD of Majority fusion rule

Fig. 15. Impact of number of nodes on PD of M out of N fusion rule

Fig. 12. Impact of number of nodes on PD of OR fusion rule

Fig. 13. Impact of number of nodes on PD of AND fusion rule

conference on Computer Engineering and Technology (ICCET'10), pp.

v3-64-v3-68, June 2010. [3] J. Sachs, I. Maric, and A. Goldsmith, Cognitive Cellular Systems within

the TV Spectrum, Proc. IEEE Symp. New Frontiers in Dynamic Spectrum (DySPAN 10), Apr. 2010