Sureness Efficient Energy Technique for Cooperative Spectrum Sensing in Cognitive Radios

Mahmoud Khasawneh1, Anjali Agarwal1, Nishith Goel2, Marzia Zaman2, Saed Alrabaee1

1 Department of Electrical and Computer Engineering, Concordia University, Montreal, Canada {m_khasaw, aagarwal, s_alraba}@encs.concordia.ca

2 Cistel Technology Inc., Ottawa, Canada {ngoel, Marzia}@cistel.com

Abstract— Spectrum sensing is used to detect the unexploited sub-bands in the radio environment. In order to improve the accuracy of the spectrum sensing, the cooperative spectrum sensing method is assumed to be the best method to be used. However, it consumes power by exchanging the sensing results among the participating nodes. An efficient cooperative spectrum sensing method is proposed in this paper to save the power consumed in spectrum sensing results’ reporting. The clustering method is used to divide all the cognitive nodes in a specific number of clusters. Each cluster has a cluster head (CH) which is responsible for collecting data from different cognitive nodes in the same cluster and sending the cluster’s decision back to them. After each cluster head receives the clusters’ decisions, the final decision is made. Simulation results show that the proposed method helps in power saving in comparison to the traditional method.

Keywords-Spectrum sensing; cognitive nodes; clusters; cluster head;

I. INTRODUCTION Recently, the cognitive radio (CR) [1] technique became

one of the most common studied techniques in the wireless networks field. It allows more users to use the available spectrum. In CR networks, unlicensed users, which are referred to as secondary users (SUs), are allowed to dynamically access the frequency bands when licensed users which are referred to as primary users (PUs) are inactive. Because of the inefficient spectrum utilization of the licensed spectrum owners (primary users, PUs), and the increase in the spectrum demand, cognitive radio is proposed as a promising technology to improve spectrum utilization. Spectrum sensing is the first stage in cognitive radio in which unlicensed users, usually called secondary users (SUs), try to detect the presence or absence of the primary users (PUs) in their pre-reserved spectrum. Frustration in spectrum sensing results might cause substantial interference for those who use the spectrum. On the other hand, wrong results of the spectrum sensing lead to inefficient spectrum utilization. The probability of getting correct sensing results is low, if the spectrum sensing is made by each secondary user individually. If the cooperation concept is applied among the different secondary users, this probability will be increased. So, cooperative spectrum sensing helps in achieving a higher accurate correct decision ratio. It alleviates the negative impacts on performance caused by multipath fading and

shadowing [2]. It allows the secondary users to share their initial decisions about the vacant spectrum bands and then make their final decisions. Every participating user first detects the spectrum using any spectrum sensing method such as matched filter, energy detection, or cyclostationary feature detection [3], and then they exchange their detection decisions. Detecting spectrum holes in a fast method opens the doors for the researchers to develop new methods in spectrum sensing by taking different sensing situations when the conditions of the CR network are more dynamic. Collaboration concept is used to make the detection faster [4]. The issue of dynamic channels access as a partially observed Markov process is studied in [5]. Despite the advantages of the cooperative spectrum sensing, it leads to power consumption. Mostly, for battery-operated mobile terminals, the power resource is limited. Few researchers have suggested solutions for this problem. In order to decrease reporting power consumption, a censoring scheme is studied in [6] and [7] by ignoring uninformative test statistics or local decisions. In [8], a time scheduling scheme is shown to decrease the number of local decisions.

In the above works, all secondary users’ decisions are forwarded to a specific one receiver directly, but the distance between some SUs and that the receiver might be long which might be corrupted and become incorrect. On the other hand, their decisions are valuable to improve the spectrum sensing performance, since the increment in the number of participating SUs in the sensing will increase the accuracy of the sensing results. In order to guarantee correct transmission of their decisions to the receiver, more power is needed due to signal distortion and fluctuation with the communication distance increment. Moreover, if this receiver becomes inactive and the secondary users do not recognize that, they will continue sending their decisions without getting any reply which consumes more power and leads to low spectrum sensing efficiency.

In traditional broadcast scheme, each SU transmits its decision to all the SUs which has many disadvantages such as it consumes too much power in transmitting and receiving the spectrum sensing results, and takes more time too.

These two disadvantages motivate us to develop an approach that helps in reducing the power consumed and the time as well. A sureness cluster based cooperative approach is proposed in this paper. All secondary users SUs in the network

2012 International Conference on Telecommunications and Multimedia (TEMU)

978-1-4673-2781-7/12/$31.00 ©2012 IEEE 25

are grouped in clusters, and one of the cluster nodes is randomly chosen to be the cluster head. This cluster head is responsible for receiving sensing results from the cluster nodes (SUs), making the cluster decision, forwarding the cluster decision to the other clusters, collecting other clusters’ decisions and sending them back to its cluster nodes. Each cluster node should be sure of its initial decision before forwarding it to the cluster head. Meanwhile, the cluster head should be sure about the cluster’s decision before sending it to other clusters’ heads. Our proposed model has many advantages in comparison to the broadcast approach such as: the transmission distance between the SUs and the cluster heads is less than the distance to the common receiver which needs much less power in transmitting the sensing results. On the other hand, the sureness concept helps in increasing the accuracy of the sensing results. Analytical results demonstrate significant power can be conserved by using our proposed approach.

This paper is built on two main parts. First, the method of aggregation the spectrum sensing results from all the SUs in the network and how to make the final decision about the presence/absence of the PUs in their licensed spectrum. Second, showing how this approach reduces the consumed power in sharing the results of the spectrum sensing among the SUs and in making their final decision.

The rest of this paper is organized as follows: an overview of cognitive radio is explained in Section II. We demonstrate the system model of the proposed sureness cluster-based cooperative spectrum sensing, and the general system view is defined in Section III. Section IV presents simulation results showing the performance of the proposed model. The future works is shown in Section V. We conclude this paper in Section VI.

II. AN OVERVIEW OF COGNITIVE RADIO The principle of Cognitive Radio was firstly mentioned and

explained by Joseph Mitola [1]. Cognitive Radio could be defined as an efficient technology that allows more users to use the available spectrum. Spectrum sensing is assumed as the basic functionality in CR. Spectrum sensing aims to find the vacant spectrum holes for dynamic use. In general, there are two sensing modes, reactive sensing and proactive sensing [10]. Generally, the spectrum sensing techniques can be categorized as transmitter detection, cooperative detection, and interference-based detection [1]. In transmitter detection, the PU transmitter presence in its spectrum band is determined. Three schemes are generally used for the transmitter detection that are matched filter detection, energy detection and cyclostationary feature detection [1]. Matched filter detection is used if the secondary user has information about the primary user’s signal. While if not enough information about PU’s signal is available, energy detection is applied. In cyclostationary feature detection, modulated signals are coupled with other signals. In cooperative detection technique, cooperation concept between the SUs is applied in order to improve the sensing results. The last technique,

interference-based detection technique, has been introduced by the FCC in [11], wherein the interference temperature is measured and compared with statistical information to make the decision about the PU presence in its spectrum band.

Spectrum management is another important issue in CR. The objective of spectrum management is to share the spectrum between many users, PUs and SUs, in such a way that accomplishes their different goals and requirements.

III. SYSTEM MODEL

A. System General Overview We present the general view of the system in this section.

There are two types of users in the network: primary users (PUs) {PU1…PUN}, and secondary users (SUs) {SU1…SUM}. While PUs have licenses to access the spectrum, SUs are trying to find vacant spectrum bands in order to use them for their data transmission. The M SUs are grouped in J clusters based on geometric locations, wherein each cluster contains K SUs such that ∑ K� = M

���� . In each cluster, one SU is randomly

chosen to be the cluster head (CH). The remaining SUs are named cluster nodes. Figure 1 illustrates an example of how to form clusters. We assume, in this example, that there are 9 SUs are grouped in three clusters which contain 3 SUs in each. These three clusters are in the same geographical region. SU3, SU6, and SU9 are randomly chosen as cluster heads at the beginning of the spectrum sensing phase. In each cluster, the cluster nodes send their sensing results to their CHs this messages exchange is represented in first step (a). The second step (b) is to forward the clusters’ decisions to the other clusters through the CHs communication. After that, each CH collects other clusters’ decisions as in step (c), and finally CH forwards all the clusters’ decisions to its clusters which is represented in (d). The energy detection method is used by all SUs to detect the presence or absence of the PU in its spectrum band, i.e. to make their initial decisions about the PUs’ bands, where two hypotheses are used to represent that as follows:

�� ∶ �� ���� �� ����

�� ∶ �� ���� �� ����

SUs measure the signal strengths in all channels, and by using the energy detection method SUs makes the initial decision about the presence/ absence of PU in its reserved channels. If the decision in a SU is z� then 0 will be stored at a spectrum sensing matrix (SSM) for the corresponding channel of that PU which is being checked otherwise 1 will be assigned. On each cluster node (SU) the spectrum sensing matrix, which contains the initial decisions of all channels of different PUs, has the following format:

�

��� ��� ⋯ ���

⋮ ⋱ ⋮

��� ⋯ ���

!

26

where, each row represents the SU initial decisions about all channels of each PU. The first row represents the initial decisions about all channels (from 1 to ) of PU number 1, while the second row represents the channels of PU number 2 and so on. represents the channel id where each channel has a unique id within the PUs.

Figure 1. An example of our scheme, where CH1, CH2, and CH3 are different clusters heads.

Each cluster head (CH) is responsible for collecting the cluster nodes’ decisions and making the cluster’s decision. We assume that each CH maintains a cluster decision matrix (CDM) that has the following format:

�

"#�� "#�� ⋯ "#��

⋮ ⋱ ⋮

"#�� ⋯ "#��

!

where, each row represents the cluster’s decision about all channels of each PU. The first row represents the cluster’s initial decisions about all channels (from 1 to ) of PU number 1, while the second row represents the channels of PU number 2 and so on.

We assume that there is a pre-reserved communication channel that is used for all information exchanges between all SUs, where this channel is capable of handling this traffic.

B. Sureness Cluster-Based Cooperative Spectrum Sensing In this section, we introduce the proposed sureness cluster

cooperative spectrum sensing scheme and its detailed algorithm is also proposed. In our scheme, sureness concept means that every SU has to be sure of its spectrum sensing decision before forwarding it. There are two levels of sureness, the first level is in each SU (i.e. each SU has to be sure of its initial decision before forwarding it to its cluster head), and while the second level is in each CH (i.e. each CH has to be sure of its cluster decision before forwarding it to other cluster heads). Spectrum sensing is made periodically which means at each sensing round, all SUs including the CHs re-sense the spectrum and update their matrices with the new sensing values. Cooperation concept is applied among the different SUs. It is when all SUs take other SUs’ decisions in consideration to make their final decisions.

At the first round of the sensing phase, all SUs and CHs sureness values are true. This means that the SUs will send their initial decision to their CHs and the CHs will send their decisions to other CHs. Each SU sets its sureness value as

follows: at each spectrum sensing round, the SU gets the decisions of all clusters via its CH, and then compares its decision with the other clusters’ decisions. If more than 50% of the clusters’ decisions are similar to its decision, then it sets its sureness value to be true, otherwise it is false. In other words, if the sureness value of a SU is true at the beginning of the sensing round and by end of the sensing round more than 50% of the clusters’ decisions are same as its decision, then its sureness value remains true, otherwise the SU loses it sureness so it cannot send their sensing results in the next round of sensing. Simultaneously, each CH sureness value is also set using the same way.

Each CH makes the cluster decision based on the majority voting mechanism. Each CH counts the number of the SUs whose sensing results are 1 (i.e. spectrum is busy) and simultaneously counts the number of the SUs whose sensing results are 0 (i.e. spectrum is idle). After that the CH makes the cluster’s decision where the number of SUs which have the same sensing results should be greater than 50% of the total number of SUs in that cluster. The same mechanism is applied to SUs in making their final decisions where more than 50% of clusters’ decisions should have the same decision as their initial decisions.

If the sureness value of a SU is false, it stops transmitting its sensing results but keeps sensing the spectrum and tracking the majority of all the SUs. As soon as it gets its sureness value back, it restarts sending its sensing results.

When a CH receives other CHs’ decisions, it maintains a decision matrix (D) which has the following format:

�

#�� #�� ⋯ #��

⋮ ⋱ ⋮

#�� ⋯ #��

!

where, first row represents the system decision that has been collected from all clusters about all the channels of PU number 1, and so on. D matrix is sent by each cluster head to its cluster nodes where each cluster node uses it to check its sureness about its decision in each round and to find the final decision in the final round of the sensing.

The disadvantage in our proposed scheme is the low accuracy rate of the spectrum sensing results when the number of cluster nodes that have the sureness in their decisions is small. In order to solve this problem, a rule is configured on each CH wherein every CH checks the number of cluster nodes that participates in the spectrum sensing round, at least more than half of them have to participate in order to take the cluster decision, otherwise, a notification message is sent to all cluster nodes to send their sensing results. The algorithm of the proposed scheme is summarized in Table I.

In our scheme, more power saving is achieved in comparison with the broadcast scheme. Simultaneously, it is an efficient scheme in comparison with other schemes. In other schemes [9], a central point takes care of sensing and making the decision of the spectrum sensing, so a long time is spent in making the final decision of the spectrum that results in a lower sensing accuracy rate, thus low efficiency.

27

TABLE I. SURENESS CLUSTER BASED ALGORITHM

The main difference between the initial decision function (IDF) and the second decision function (SDF) is that the IDF is made based on the local observations of the cluster node, while the SDF is made based on the other clusters’ decisions by using the D matrix.

Each SU works as transmitter and receiver simultaneously. So power is consumed in transmitting and receiving the spectrum sensing results.

The total consumed power P has two components: the consumed power in the transmission phase P$%&'%((*+*,-) and the consumed power in the receiving phase P(%/%�0%((*+*,-). The following equation is used to represent that:

P = P$%&'%((*+*,-) + P(%/%�0%((*+*,-) (1)

In the transmission period the power is consumed by exchanging the spectrum sensing results from the cluster nodes to its CH, CH to other CHs, and CH back to its cluster nodes which can be computed as follows:

P$%&'%((*+*,-) = ∑ 23�→"5+ 6(6 − 1)2"5→"5

6,9

:=1,�=1 +

∑ 2"5�→3;<��ℎ���

6

�=1 (2)

We assume that same power value is needed by each SU to transmit. In the receiving period, the power is consumed in receiving the spectrum sensing results. Most of the time the CHs are ON and consuming power while the cluster nodes consume a small level of power in a small time which can be neglected, so most of the power is consumed by receiving other clusters’ decisions which can be represented as follows: P(%/%�0%((*+*,-) = J(J − 1)P(%/%�0%( (3) where P(%/%�0%( represents the consumed power from the receiver.

IV. PERFORMANCE EVALUATION In this section, we demonstrate the performance of our

proposed scheme by showing some simulation results. In order to compare our scheme with other schemes, the traditional broadcast scheme in [9, 12] is simulated too. We compare the transmission and the receiving power consumption in our scheme with the broadcast scheme.

Table II shows the simulation parameters, Table II Simulation Parameters

Parameter Value

Number of PUs Number of clusters Number of channels of each PU. Transmission power ?@ABACDA@

3 [10,100] 5 0.03 watt 0.1 watt

Transmission power represents the power consumed while the SUs are exchanging their spectrum sensing results. While the receiving power, Preceiver , represents the consumed power when the SUs receive the sensing results from the other SUs.

Initialization IDF= Initial Decision Function SSM=Spectrum Sensing Matrix CID=Initial Decision of the cluster CDM=Cluster Decision Matrix SDF=Second Decision Function which is any decision except the initial decision D=Decision matrix For round R 1: For SU m 2: Sense the spectrum 3: Do IDF 4: Update SSM 5: While (SUsureness) 6: {Forward SSM to CH} 7: If(SUM = = CH) 8: Check if more than half of SUs are sending 8: Receive SSM from cluster SUs 9: Do CID 10: Update CDM 11: While (CHsureness) 12: forward CDM to CHs 13: Receive CDM from CHs 14: Compute SDF 15: Update D 16: If ( SDF = =CID) 17: { CHsureness =TRUE 18: forward D to cluster SUs} 19: Else 20: CHsureness =FALSE 21: EndElse 22: EndIf 23: EndFor 24: For SU m 25: compute SDF 26: If(SDF==IDF) 27: SUsureness =TRUE 28: Else 29: SUsureness = FALSE 30: EndElse 31: EndFor 32: Endfor

28

Figure 2. Comparison of the consumed transmission power between the

broadcast scheme and our scheme

Figure 3. Transmission Power Performance for different clusters

Figure 4. Receiving Power Performance for different clusters

To simulate the two models i.e. sureness cluster-based and the broadcast models, we used MATLAB. The transmission and receiving power consumption of our proposed scheme are shown in Figures. 2 and 3. To demonstrate the power efficiency of our proposed scheme, we define the Transmission Power Consumption Ratio (TPCR) as follows:

TPCR =Our model transmission power

Broadcast transmission power

Also in same way, we define the Receiving Power Consumption Ratio (RPCR),

RPCR =Our model Receiving power

Broadcast Receiving power

In the traditional broadcast model, TPCR and RPCR values

are 1.

Figure 2 illustrates the difference between the conventional scheme in spectrum sensing which is the broadcast scheme and our proposed scheme. It is clear that our model helps in reducing the consumed power in transmitting the results of the spectrum sensing process in comparison with the broadcast model under the same network circumstances.

In Figure 3, the transmission power efficiency performance of our proposed method is simulated with different numbers of SUs. It can be observed that the transmission power consumption decreases with the increase of the number of SUs in the cluster. Simultaneously, the transmission power consumption ratio declines with the increase of clusters which means more additional cluster heads are needed.

Figure 4 shows the receiving power efficiency performance of our proposed method, where it is simulated with same system parameters. It can be seen that the receiving power consumption decreases with the increase of the number of SUs in comparison with the broadcast method. Simultaneously, the receiving power consumption ratio increases slightly with the increase of clusters which means more additional cluster heads are required; however this power consumption seems to be very small in comparison with the conventional broadcast method.

The maximum consumed power by a SU, if it is as a cluster node, depends on the size of spectrum sensing matrix (SSM). It increases when the number of the PUs and their channels to be detected increase. On the other hand if a SU works as a cluster head, the maximum consumed power depends on two factors that are the number of SUs which form the cluster and the total number of the clusters.

V. FUTURE WORK In our future work, we aim to use a more dynamic way of

forming the clusters which helps in reducing the consumed power of exchanging the sensing results among the different users and the exchanging time too. It is also important to study the different factors that might affect the decisions exchange

10 20 30 40 50 60 70 80 90 1000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Clusters

Tra

nsm

issi

on P

ower

Con

sum

ptio

n R

atio

# of SUs per cluster=2

# of SUs per cluster=3# of SUs per cluster=4

In Broadcast scheme

10 20 30 40 50 60 70 80 90 1000.01

0.015

0.02

0.025

0.03

0.035

0.04

Number of Clusters

Tra

nsm

issi

on P

ower

Con

sum

ptio

n R

atio

# of SUs per cluster=8

# of SUs per cluster=9

# of SUs per cluster=10

10 20 30 40 50 60 70 80 90 1000.009

0.01

0.011

0.012

0.013

0.014

0.015

0.016

Number of Clusters

Rec

eivi

ng P

ower

Con

sum

ptio

n R

atio

# of SUs per cluster=8

# of SUs per cluster=9

# of SUs per cluster=10

29

between the SUs such as the fading and shadowing, and find solutions for that.

VI. CONCLUSIONS In this paper, an efficient spectrum sensing scheme was developed. This scheme helps in decreasing the power consumption in transmitting and receiving the results of spectrum sensing. Meanwhile, it increases the accuracy rate of the sensing results by allowing only the SUs who have sureness in its sensing results to participate in making the decision about unused frequency bands. The transmission and receiving power consumption of our proposed method has been derived and compared with that of the conventional broadcast method. Simulation results show significant decrement of transmission and receiving power consumption.

ACKNOWLEDGMENT The authors would like to acknowledge the financial support provided by MITACS – Accélération Québec funds.

REFERENCES [1] J. Mitola, “Cognitive radio for flexible multimedia communications,” in

Proc. MoMuC’99, pp. 3-10, 1999. [2] S. M. Mishra, D. Cabric, C. Chang, et al., "A real time cognitive radio

testbed for physical and link layer experiments," in Proc. IEEE DySPAN 2005, pp. 562–567, Nov 2005.

[3] I.F. Akyildiz, W.Y. Lee, M.C. Vuran, and S. Mohanty, “NeXt generation/dynamic spectrum access/cognitive radio wireless networks: A survey,” Computer Networks Journal, vol. 50, no. 13, pp. 2127-2159, Sep 2006.

[4] L. Lai, Y. Fan, and H. V. Poor, “Quickest detection in cognitive radio: A sequential change detection framework,” in Proc. IEEE GLOBECOM 2008, New Orleans, LA, pp. 1–5, Nov 2008.

[5] J. Unnikrishnan, and V. V. Veeravalli, “Algorithms for dynamic spectrum access with learning for cognitive radio,” IEEE Trans. Signal Processing, vol. 58, no. 2, pp. 750–760, Feb 2010.

[6] C. Sun, W. Zhang, and K. B. Letaief , “Cooperative spectrum sensing for cognitive radios under bandwidth constraints,” in Proc. IEEE WCNC 2007, pp.1-5, Mar 2007.

[7] J. Lunden, V. Koivunen, A. Huttunen, and H. V. Poor, “Censoring for Collaborative Spectrum Sensing in Cognitive Radios,” in Proc. ACSSC 2007, pp. 772-776, Nov 2007.

[8] A. Hoang, and Y. Liang, “Adaptive Scheduling of Spectrum Sensing Periods in Cognitive Radio Networks,” in Proc. IEEE GLOBECOM 2007, pp. 3128-3132, Nov 2007.

[9] T.A. Weiss, J. Hillenbrand, A. Krohn, and F.K. Jondral, “Efficient signaling of spectral resources in spectrum pooling systems,” Proc. 10th Symposium on Communications Vehicular Technology (SCVT), pp. 1 - 6 2003.

[10] D. Kakkar, A. Khosla, and M. Uddin, “Power Allocation with Random Removal Scheme in Cognitive Radio System,” Proceedings of the World Congress on Engineering (WCE ), pp.1796 – 1801, 2011.

[11] FCC, ET Docket No 03-237 Notice of inquiry and notice of proposed Rulemaking, November 2003. ET Docket No. 03-237.

[12] S. Shankar, N. C. Cordeiro, and K. Challapali, “Spectrum agile radios: utilization and sensing architectures,” Proc. IEEE DySPAN, pp. 160 - 169, 2005.

30