• No results found

A Cluster Formation Protocol for Cognitive Radio Ad hoc Networks


Academic year: 2021

Share "A Cluster Formation Protocol for Cognitive Radio Ad hoc Networks"


Loading.... (view fulltext now)

Full text


A Cluster Formation Protocol for Cognitive Radio

Ad hoc Networks

Alfred Asterjadhi

, Nicola Baldo

and Michele Zorzi

Department of Information Engineering – University of Padova, Italy

Centre Tecnol`ogic de Telecomunicacions de Catalunya – Barcelona, Spain



aasterja, zorzi


@dei.unipd.it, nbaldo@cttc.es

Abstract—In this paper we present a solution for the realization of large Cognitive Radio Networks. The solution features a spectrum-aware neighbor discovery and clustering scheme that works in conjunction with a network coded cognitive control channel in order to allow Cognitive Radio devices to opportunis-tically access the unused spectrum. We evaluate the performance of the proposed solution with respect to the characteristics of the formed clusters as well as the reliability of the dissemination of the control information within the cluster.


Cognitive Radio (CR) has enjoyed a strong interest by the research community in recent years. By leveraging on the possibility of CR devices (a.k.a. secondary users or unlicensed users) to access unused portions of the licensed spectrum, these strategies are expected to improve the utilization ef-ficiency of the spectrum itself, being thus able to provide additional and improved quality communication services with respect to existing wireless technologies.

Unfortunately, regardless of the huge research effort that has been devoted to this field, CRs are still far from being a reality. This is because there are numerous challenges that need to be addressed in order to identify suitable strategies for the implementation of CR networks. In particular, in this paper we focus our attention on the following challenges:

The control channel problem: CRs need to coordinate among themselves for spectrum access purposes, however they also need to communicate by accessing the spectrum in order to achieve coordination. A possible solution is to use statically allocated spectrum bands for this purpose, however it has been often argued that this solution is inappropriate for several reasons [1]–[3]. Therefore, the control channel problem is still mostly unsolved.

Channel allocation and multi-channel medium access:

many solutions have been proposed in the recent liter-ature, which address either one of these issues; how-ever, how to integrate these two aspects into a single CR solution has not been much investigated so far. In particular, most work on channel allocation [4], [5] does not consider how to perform medium access in practice, while most work on multi-channel medium access [6], [7] aims at enhancing the performance of existing MAC protocols like IEEE 802.11 by using more channels, without considering issues such as optimal frequency reuse and overall spectrum utilization efficiency.

Primary user detection: while in some cases a CR

network might rely on a centralized database to provide information on the spectrum availability at different times and locations, there are several scenarios (such as ad hoc, personal and sensor networks) in which this solution is not feasible due to the lack of Internet connectivity; as a consequence, a general-purpose CR solution is required to be able to rely on primary user detection only. Recently, it has been argued [8] that single-user detec-tion strategies do not perform well enough to limit the secondary interference to primary users in a satisfactory way. Cooperative detection strategies are more promising in this respect, however their integration in a complete CR solution is still an open challenge.

Scalability: many of the solutions mentioned earlier have been designed only for limited-size CR networks, for example due to the presence of centralized controllers. However, we would ideally like to be able to extend such a paradigm to virtually infinite CR networks.

We stress that many solutions have been proposed in the recent literature to address each of the above mentioned problems in isolation, however none of the CR architectures proposed so far addresses all of them in an efficient way. Well known architectures such as IEEE 802.22 [9] and CORVUS [10], based respectively on a centralized and decentralized approach, rely on global and/or local common control channels to share spectrum sensing information and to exchange the signaling information that is a pre-requisite for initiating communications.

This solution limits scalability in terms of devices and traffic density. As for medium access and resource allocation, IEEE 802.22 presents a centralized solution where the secondary base stations take care of resource management issues based on inputs collected during a distributed sensing phase carried out by CRs. However, coordination among multiple secondary base stations, especially in terms of mutual and aggregated interference that they may generate to primary receivers, is still an open issue [11]. On the other hand, [10] does not explicitly address the problem of multi channel and multiuser access in the absence of a centrally organized architecture.

In our prior work [12], [13] we proposed the Network Coded

Cognitive Control Channel (NC4) for the realization of

multi-channel CR networks. NC4 addresses in a joint fashion the


Fig. 1. A typical clustered cognitive radio ad hoc network.

medium access and the efficient channel allocation, and as such stands as a very promising solution for CR networks. Our prior work, however, considered only networks having finite size; this is a significant limitation, since the ultimate CR architecture is expected to be utilized in very large networks, in order to provide support for coordinated spectrum access to a large number of CRs which can potentially belong to different operator networks.

In order to allow NC4 to scale up to networks of virtually

infinite size, an attractive approach is to divide the network

in clusters (see Figure 1), so that NC4 can be used on each

cluster independently. However, this approach faces additional challenges regarding how the clustering should be performed,

and how well NC4 will perform when run separately on each

cluster. In this paper we discuss how to effectively address these new challenges. In particular, the original contributions of this paper are: 1) the definition of a neighbor discovery strategy suitable for multi hop CR networks; 2) the definition of an enhanced clustering algorithm which enables the creation

of multi-hop clusters suitable for the operation of NC4; 3) the

adaptation of NC4 to work in cluster-based networks where

CRs are interested in receiving information from all the CRs of the same cluster while avoiding interfering with primary users and adjacent clusters communication; 4) the evaluation of the performance of the proposed scheme.


In this section, we first introduce the system model on which the rest of the paper is based. Subsequently, we describe the main components of the proposed scheme, namely the Neighbor Discovery Protocol, the Spectrum-aware Cluster Organization Protocol, and the cluster-aware Network Coded Cognitive Control Channel.

A. System model

We assume that the electromagnetic spectrum is divided

into a set of C non-overlapping channels assigned to the

primary users, that can be opportunistically accessed by CRs.

Additionally, CRs are equipped with a single RF module and can be tuned to only one channel at any given time. We adopt a protocol interference model both for the communications among CRs and for the interactions between primary users and cognitive radios. For CRs, we consider an interference range equal to the communication range, which is for sim-plicity set equal to the unit length. As for the primary users, they communicate using a Time Division Duplex mode, and consequently in any given channel the secondary users can detect primary receivers.

Let i∈ {1, . . . , N}denote a CR. We define:

• Ci ⊆C the set of channels which CRi has detected to

be available;

• Nik the set of CRs which arek-hop neighbors of i, i.e.,

which are at most k hops away from CR i (including

node iitself);

B. Discovery protocol fork-hop neighbors

In order to participate in the cluster formation process that we will describe in the next subsection, each CR needs to possess information on which other CRs are in its neighbor-hood and their local spectrum availability. We use the term

neighbor discovery to indicate the process of retrieving this information. In the following we describe possible approaches for the realization of neighbor discovery in a multi-channel CR network.

Neighbor discovery starts with a CR broadcasting control packets in order to solicit packet exchange from all its neigh-bors. The control packet exchange continues for a limited time, referred to as neighbor discovery delay, during which CRs exchange control packets until they receive replies from all CRs in their neighborhood.

We note that the discovery process in cognitive radio ad hoc networks is very challenging. This is because CRs have

to meet all their k-hop neighbors in a hop and

multi-channel network. Hence, these neighbors can be tuned on different channels, as their channel availability highly depends on the activity of primary users in the area.

To perform neighbor discovery in cognitive radio networks there have been several proposals in the literature [14]–[17]. In [14] the authors propose an asynchronous neighbor discovery and leader election algorithm for single hop cognitive radio networks. The time required to elect the leader and determine

its neighbors is O(N M2), where N is the number of nodes

andM is the maximum number of channels. In [15] neighbor

discovery relies on the identification of a single common

control channel which is rotated in the set of M channels.

Based on this deterministic and synchronization-relaxed com-mon control channel pattern, neighboring nodes can exchange

their mutual information after O(M) time slots for a generic


We note that the above described solutions can be adopted to implement the neighbor discovery which is required in our solution. However, these solutions provide every node with the

list of 1-hop neighbors only, whereas the clustering protocol


availability of information concerning k-hop neighbors. For this reason, we propose to modify these algorithms to provide

the list ofk-hop neighbors by allowing nodes to concatenate

and transmit multiple packets during the same time slot, in order to mutually exchange their list of neighbors until the

k-hop list is provided. This can be done by following the same

strategy as in [15] and additionally including in each packet a hop counter which allows nodes to discard packets that come

from neighbors which are more thank hops away.

C. A distributed cluster organization protocol

As we suggested earlier, a possible solution to address the issue of CR scalability up to virtually infinite CR networks is to divide the network itself into clusters, so that the main functionalities of a CR network (exchange of control information, channel allocation, medium access, primary user detection) can be addressed more efficiently. In this section, we discuss practical methods to divide the network into suitable clusters for CR operation. In this respect, several solutions have been proposed in the recent literature. In [18], the authors propose a Bargain Group Formation algorithm which relies on the availability of a common control channel. Similarly in [19] the authors propose a cluster based algorithm that divides the network in clusters taking into account the local spectrum availability; as part of this proposal, a neighbor discovery phase is also introduced. These algorithms form clusters by focusing on the constraint that there needs to be at least one channel which is free for all CRs in the same cluster. As a result, these schemes have the tendency to create clusters with a large number of members, but with a small (often equal

to1) number of free channels shared by all CRs. It is to be

noted that NC4 does not perform efficiently in this type of

scenario [20].

A different solution [15] aims at creating clusters that provide a good tradeoff between cluster size and maximum number of shared free channels in the cluster. The original objective in [15] is to make the execution of the cluster formation algorithm less frequent; in fact, if a primary user appears in one of the shared free channels, there will still be other channels suitable for the exchange of control information within the cluster. The algorithm in [15] is a maximum bi-clique algorithm which requires nodes to send 3 broadcast messages in order to partition the network into clusters which, as results show, tend to have reduced cluster size. While this algorithm is suitable to create spectrum-aware clusters, its execution time is higher, as nodes have to exchange 3 times more packets than the previous algorithms, making it less reactive to sudden spectrum changes. Moreover, we note that all the above mentioned algorithms partition the network in at

most1-hop clusters.

In order to allow NC4 to work efficiently, we need a

clustering algorithm that is able to create clusters that have

the highest number of CRs within a k-hop neighborhood,

while at the same time guaranteeing that CRs which are members of a cluster share a sufficient number of free channels in order to disseminate control packets and perform data

transmissions. The algorithms proposed so far in the literature only provide either a high number of CRs in a cluster [15] or a number of common free channels [18], [19]. In the following, we present a distributed algorithm (Combo) which partitions the CR network in non overlapping clusters based on local spectrum availability; in particular, the proposed algorithm aims at creating clusters of a given size (in number of hops) that takes into account the cardinality of the set of commonly available channels among CRs when making decisions.

The algorithm is inspired by [21], where the authors propose a clustering algorithm based on node IDs for the partitioning of the network in clusters. As we discussed in Section II-B, the neighbor discovery phase is able to provide to the CRs the

list of their k-hop neighbors, along with their corresponding

available channels. After the neighbor discovery phase, all CRs run the clustering algorithm independently, and base their decisions on the information stored in the ternary key

τj ={cj, dj,IDj}, where dj is the k-degree of connectivity

of CRsj, namely the cardinality of itsk-hop neighbors setNk


IDj is the cognitive radio ID, andcj is defined as follows:

cj= min

i∈Nk j


i.e.,cj is the minimum number of common channels that CR

j has with each of its neighbors. Based on this information

each CR calculates a weighted priority key ψj that will be

used during the cluster formation process to decide whether the CR will be a cluster head or join an existing cluster. A CR

j is elected as cluster head if its weighted priority key is the

highest among its neighbors, i.e., if the following condition is satisfied:

ψj = max

i∈Nk j


A cluster head CR initiates the clustering process by sending

a cluster formation request, broadcasting its ψj to its k-hop

neighbors following the same procedure as for the neighbor discovery phase [15], [19]. All nodes whose weighted priority key is the highest among the neighbors request the creation of a cluster with their ID as cluster ID. Nodes that overhear the request join the cluster if their priority is lower, otherwise, in case that they do not hear a broadcast message from any higher priority clusters, they elect themselves as cluster heads. The algorithm terminates once all nodes have made their choices and have been assigned uniquely to a cluster. Note that, even though all nodes only become members of a single cluster, it is possible to have some nodes (border nodes or gateway nodes) which store information about adjacent clusters. This information can be used to implement routing protocols and, most importantly, makes it possible to improve the dissemination performance as we discuss in the following.

D. NC4 in a clustered network

In order to design an effective dynamic spectrum access scheme we need to solve the following two problems: 1) how to assign the available spectrum resources to the users, and 2) how to make the users coordinate among themselves for


medium access. The key concept of NC4 is to tackle both problems simultaneously. We argue that it is intuitive that the most efficient way of doing spectrum allocation and transmis-sion scheduling is to use a centralized approach which relies on the knowledge of both the spectrum availability and the communication needs (e.g., intended set of receivers, as well as possible Quality of Service requirements) of all the users. We

call this knowledge the complete control information, which

is made up of the control information pieces (control packets)

of each user in the cluster. In the literature, most of the times in which the complete control information is used for resource allocation purposes, a centralized scheme is assumed, i.e., there is a centralized controller which gathers all the control packets generated within the cluster and determines the resource allocation and transmission schedule for all the users in the cluster.

The NC4solution is different in that it allows the

implemen-tation of a scheme requiring the complete control information in a distributed fashion. We first proposed this solution in [20] for single-hop multi-channel networks, and in this paper we consider its extension to clustered multi-hop CR networks. The crucial aspect is that each user gathers the complete control information from the other users in the cluster, and independently determines the resource allocation for the whole cluster. If the same control information is successfully dissem-inated to all users, and if the resource allocation algorithm is deterministic, then each user will be able to determine the same resource allocation, without any further interactions.

For our scheme, we assume that time is divided inallocation

periods with duration Tall. Each allocation period is divided

into S slots of duration Tslot = Tall/S. In every slot, each

user will tune to one of the available channels according to a pre-determined channel switch pattern, and transmit exactly

one control packet. Tall andS are to be chosen with respect

to the expected secondary user density per channel, so that the transmission of the control packet occupies only a fraction of the slot duration.

Each user generates its own control packet at the beginning of each allocation period. Subsequently, in every slot, each user will tune to a new channel and exchange control packets with whatever other user he meets in that channel. The objective of this process is to disseminate to all the users within a cluster the complete control information. We require the channel switch pattern to have pseudo-random properties, so that the dissemination of the control information is possible to all nodes with high probability after a sufficient number of slots. At the end of an allocation period, each user will have retrieved the control information, and will use it to run any deterministic scheduling algorithm. We note that this algorithm will also determine the channel switch pattern to be followed in the next allocation period.

In order for this scheme to work properly, it is important that each secondary user retrieves the complete control information from all the secondary users that are in the same cluster with high probability. In fact, users that fail to retrieve the control information at the end of an allocation period (misinformed

users) can potentially cause a spectrum collision in the next

allocation period. This is because they have not gathered the complete control information. Hence, when running the scheduling algorithm for the next allocation period, they will determine a wrong channel switch pattern and transmission schedule, possibly interfering with other users within the cluster.

For this reason, we prefer those dissemination schemes

which offer a highretrieval probability(Pretr), defined as the

probability that a generic user correctly retrieves the control information from every other user in the same cluster. In our prior work [20] on single-hop multi-channel networks, we proposed the use of Network Coding for the dissemination of the control information, since it vastly outperforms other state-of-the-art dissemination strategies. In this paper, we propose to adapt this scheme for clustered multi-hop cognitive radio networks. Clearly, in this scenario the dissemination is more critical because of the multi-hop nature of the cluster and because of the differences in the channel allocation scheme. In the following we briefly describe the additional functionalities

that make NC4 suitable for operation in multi-hop clustered


As discussed previously, the cluster formation protocol assigns CRs with similar spectrum availability to the same cluster, which is identified by a cluster ID. Consider two

nearby clusters∆ andΓ. Each node δ∈∆ reserves memory

for a master buffer β∆ which is used to store all the control

packets generated within cluster ∆. When a node has to

transmit a control packet, it generates a linear combination

over GF(21) of the packets in β

∆, and broadcasts it to all

other nodes which happen to be on the same channel. The control packets include in their header the coefficients of the linear combination and the cluster ID. This ID also identifies the generation set, i.e., the buffer where the packets have to be stored, making it possible not to mix packets from different

clusters. This way it is possible for each nodeδ∈∆to retrieve

the control information generated by all the cluster members in sufficient slots. In Section III-B we evaluate the dissemination performance of this scheme for different network settings.

Now, focus on a border node which, based on the clustering formation algorithm, has decided to be a member of cluster

Γ but at the same time can overhear packets sent by all

those nodes belonging to ∆ that are in its reception range.

Basically, due to the fact that the node is situated within the cluster border, it is able to receive control packets from

both clusters (the one the node is a member of, Γ, and the

adjacent one, ∆), and also to calculate the channel switch

pattern and transmission schedule of both clusters. This is done by allocating memory for one more slave buffer where

the node stores packets of the adjacent cluster ∆.1 This way

it is possible for border nodes to avoid transmitting in those slots in which they would collide with the transmissions of other CRs in the adjacent clusters. We note that if the CRs in the border of the adjacent cluster adopt the same strategy,

1Overall, the number of slave buffers depends on the number of adjacent clusters from which the border node can receive control packets.


they would also defer from transmission. Hence, the solution that we just proposed is conservative, and its efficiency could be improved by identifying a suitable strategy for selecting which border node will refrain from transmitting; this topic is left for future research.

E. Primary User Detection

Since the purpose of our scheme is to support the unlicensed reuse by secondary users of the spectrum resources which are unused or underused by their licensed holders, we need suitable methods to identify unused channels. Clearly, if all users have access to a database providing information on the spectrum availability at each particular location, this problem is solved. Unfortunately, this situation is not expected to be encountered very often in practice: for example, not all secondary users might have an Internet connection available to query a centralized spectrum database managed by the spectrum regulator or spectrum broker. For this reason, we consider the case in which every secondary user is equipped with an independent primary user detection system, such as an Energy Detector or a Cyclostationary Feature Detector.

We distinguish two different use cases for the determination of the available spectrum resources. The first use case is clus-tering: as per the algorithm that we described in Section II-C, every user is required to know the set of unused channels at its own location in order to perform the clustering. We note that, in this context, the consequence of a misdetection is that the formation of clusters will be affected; however, a misdetection does not necessarily cause interference to primary users, since whether the secondary users actually use a channel for transmission or not is determined afterwards

during the operation of NC4. As a consequence, we argue

that for the purpose of clustering it is satisfactory to have each secondary user determine the set of available channels at its location based solely on its own primary user detection data.

The second use case for the determination of available

spectrum resources is NC4. In this case, the accuracy of the

primary user detection process is critical, since those channels which are identified as free will be allocated to secondary users for data and control communications. We note, however, that the primary user detection data gathered by each secondary user as a result of its own sensing activity can be disseminated to all other users within the other control information. As a result, a Cooperative Detection (CD) strategy can be adopted to identify with greater accuracy and faster response times those spectrum resources which can be reused by secondary users. In [22] we discussed possibles way of implementing

CD on top of NC4, and showed that this approach is effective

in achieving a low probability of interference to primary users while at the same time yielding a good spectrum reuse efficiency for secondary users.


In this section we evaluate the performance of the proposed scheme with respect to the characteristics of the obtained

clusters and the intra-cluster control information dissemination reliability.

A. Cluster organization protocol performance

We first focus on the performance of the cluster organi-zation algorithm. We show, by means of simulations, that the algorithm organizes efficiently CRs in clusters based on the variations of their channel availability. To perform our evaluation, we randomly deployed CRs in a 50x50 square area with different numbers of nodes. Furthermore, for each scenario we took into account different numbers of primary

users P, each of them transmitting over a fixed number cp

of channels which are randomly selected from the set C of

all channels. Transmission ranges are set to 1 for CRs and

to 1.5 for primary users; the total number of channels is 10.

Simulation results are averaged over 100 different randomly deployed topologies.

We compare our protocol (Combo) with the following protocols: a) the lowest id algorithm (Lowest ID) [21], and b) the distributed clustering algorithm (ConID) proposed in [23], where the weight is set to the degree of nodes connectivity.

The metrics we focus on are: in Figure 2 the number of clusters in the network, in Figure 3 the average cluster size, and in Figure 4 the ratio of the average number of common free channels in a cluster to the total number of free channels. As we can see all algorithms behave quite similarly in terms of the number of clusters and the cluster size for different numbers of primary users operating in the area. However, the Combo algorithm provides a higher ratio of common free channels in all cases, thus making it possible to considerably improve the dissemination performance, as we will discuss in the next subsection.

100 200 300 400 500 600 700 800 900 1000 0 5 10 15 20 25 number of clusters secondary users LowestID, P = 5 ConID, P = 5 Combo, P = 5 LowestID, P = 9 ConID, P = 9 Combo, P = 9

Fig. 2. Average number of clusters versus number of secondary users for different cluster protocols and primary users activity.

B. Intra-cluster control information dissemination

In this section we focus on the dissemination performance of control packets within a cluster, assuming that there are no packet losses due to intra-cluster interference. Once the clusters have been created, we have a disjoint set of clusters


100 200 300 400 500 600 700 800 900 1000 0 10 20 30 40 50 60 cluster size secondary users LowestID, P = 5 ConID, P = 5 Combo, P = 5 LowestID, P = 9 ConID, P = 9 Combo, P = 9

Fig. 3. Average cluster size versus number of secondary users for different cluster protocols and primary users activity.

100 200 300 400 500 600 700 800 900 1000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Ratio of ccc over free channels

secondary users LowestID, P = 5 ConID, P = 5 Combo, P = 5 LowestID, P = 9 ConID, P = 9 Combo, P = 9

Fig. 4. Average ratio of common control channels over free channels versus number of secondary users for different cluster protocols and primary users activity.

with different numbers of nodes interested in mutual exchange of control information. As discussed in Section II-D, we propose to disseminate the control information using the

NC4 scheme. An alternative approach that we consider for

performance evaluation purposes is a baseline scheme based on random message selection (RMS). Based on it, each time a node is scheduled for transmission it forwards one packet randomly selected among all the control packets it has received so far from its neighbor nodes. By doing so it can assure packet dissemination over multiple hops.

In Figure 5 we focus on the performance of control infor-mation dissemination in a 1-hop cluster topology for CRNs with different node densities and number of common free

channels. As evident from the figure, the use of NC4 reduces

considerably the number of slots required to assure a high retrieval probability for the control information. The perfor-mance gain with respect to RMS depends on the node density as well as on the number of free common control channels (ccc). More precisely, for low values of the node density

NC4 is approximately 3 times faster than RMS to deliver

all the control packets with Pretr = 0.97; the gain becomes

more substantial as the node density increases, reaching a 60

times reduction for λ= 30. This is due to the capability of

network coding to increase the rate of innovative information per packet. Under the RMS scheme CRs pick randomly a packet from their buffer and forward it every time they have to transmit, making it less likely that the packet will be useful to any other CR that happens to be in the same channel. On the other side, using network coding, CRs send linear combinations of their buffer’s content providing to their neighbors packets that are most likely going to increase their

decoding matrices rank. Hence, when NC4 is used, the CRs

are able to decode the control information earlier.

It is noteworthy that for NC4the number of slots required to

assure a givenPretrdecreases with increasingλ. This decrease

is associated to the fact that network coding performance improves with increasing number of CRs per channel, as the degree of connectivity of the CR network is higher. In the case of RMS, the benefit of a higher connectivity degree is overwhelmed by the overall increase of nodes per cluster which reduces significantly the probability of forwarding an innovative packet at the end of the dissemination phase. This is also confirmed by the fact that, for high densities, RMS requires almost the same number of slots to assure

Pretr = 0.97 independently of the number of free common control channels.

This behavior is emphasized when the number of channels

cp used by each primary user increases i.e., the number of

common control channels decreases. Limiting the number of free channels forces the nodes to access the same channel,

leading to a reduced number of slots to assure the samePretr

as before. The results suggest that in terms of promptness of the dissemination of control information it is better to have the fewest possible available channels and small clusters sizes. However, this implies that the control dissemination overhead would increase drastically when the number of channels is reduced, leading to very low resources for data transmission. Furthermore, small cluster sizes would lead to increased inter-cluster interference.

5 10 15 20 25 30




slots required to disseminate with P

retr = 0.97 node density, λ RMS, ccc = 3 RMS, ccc = 6 RMS, ccc = 9 GF(21), ccc = 3 GF(21), ccc = 6 GF(21), ccc = 9

Fig. 5. Number of slots required to assurePretr= 0.97as a function of

the node density,λfor different numbers of free common control channels.


1 1.5 2 2.5 3 3.5 4 4.5 5





slots required to disseminate with P

retr = 0.97 number of hops, k RMS, λ = 3 RMS, λ = 5 GF(21), λ = 3 GF(21), λ = 5

Fig. 6. Number of slots required to assurePretr = 0.97as a function of the cluster radius for different values of the secondary user density and the number of free common control channels.

scenario in which nodes are distributed randomly with

differ-ent node densitiesλ∈ {3,5}. For each such topology nodes

disseminate control packets on the free channels determined by the clustering algorithm. We consider the cases where

the nodes in the cluster have 3 free channels, which reflects

the presence of several primary users P transmitting in that

location. We plot the number of slots required by NC4 to

assure aPretr= 0.97for different values of the cluster radius

k. From the figure, we note that the number of slots required

to disseminate control information increases with the cluster radius. This is because the cluster includes nodes that are farther away (in terms of hops) from each other, requiring more slots to disseminate packets to nodes that are located at opposite sides of the cluster. The number of slots diminishes slightly when the node density increases as the scheme is able to disseminate the packets faster, thanks to the larger number of nodes that are able to mix the packets.


In this paper we presented our solution for large Cogni-tive Radio networks which features a channel multi-hop Neighbor Discovery scheme, a distributed Clustering Algorithm, and a Network Coded Cognitive Control Channel which allows CRs to coordinate within each cluster to avoid interfering with primary users as well as with other CRs in adjacent clusters. Our performance evaluation showed that the proposed solution offers several advantages with respect to state-of-the-art approaches. Future research directions include the development of fast neighbor discovery strategies, as well as the investigation of efficient channel allocation techniques which leverage on the disseminated control information.


This work was supported in part by the European Commis-sion in the framework of the FP7 Network of Excellence in Wireless COMmunications NEWCOM++ (contract n. 216715) and the ARAGORN project (FP7 ICT-216856).


[1] J. Mo, H.-S. So, and J. Walrand, “Comparison of Multichannel MAC Protocols,”IEEE Transactions on Mobile Computing, vol. 7, no. 1, pp. 50–65, 2008.

[2] P. Bahl, R. Chandra, and J. Dunagan, “ SSCH: Slotted Seeded Channel Hopping for Capacity Improvement in IEEE 802.11 Ad-Hoc Wireless Networks,” inProc. of ACM MobiCom, Sept. 2004.

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

[4] L. Narayanan, “Channel Assignment and Graph Multicoloring,” Hand-book of wireless networks and mobile computing, pp. 71–94, 2002. [5] N. Nie and C. Comaniciu, “Adaptive Channel Allocation Spectrum

Etiquette for Cognitive Radio Networks,”Proc. of IEEE DySPAN, Nov. 2005.

[6] N. Choi, M. Patel, and S. Venkatesan, “A Full Duplex Multi-channel MAC Protocol for Multi-hop Cognitive Radio Networks,” inProc. of CrownCom, Jun. 2006.

[7] C. Cordeiro and K. Challapali, “C-MAC: A Cognitive MAC Protocol for Multi-channel Wireless Networks,” inProc. of IEEE DySPAN, Apr. 2007.

[8] OFCOM, “Digital Dividend: Cognitive Access Consultation on Licence-exempting Cognitive Devices using Interleaved Spectrum,” Feb. 2009. [Online]. Available: http://www.ofcom.org.uk/consult/condocs/ cognitive/cognitive.pdf

[9] C. Cordeiro, K. Challapali, D. Birru, and S. Shankar, “IEEE 802.22: The First Worldwide Wireless Standard Based on Cognitive Radios,” in

Proc. of IEEE DySPAN, Nov. 2005.

[10] D. Cabric, S. Mishra, D. Willkomm, R. Brodersen, and A. Wolisz, “CORVUS: A Cognitive Radio Approach for Usage of Virtual Unli-censed Spectrum,”IST Mobile and Wireless Communications Summit, June 2005.

[11] S. Shankar and C. Cordeiro, “Analysis of Aggregated Interference at DTV Receivers in TV Bands,” inProc. of CrownCom, May 2008. [12] A. Asterjadhi, N. Baldo, and M. Zorzi, “A Distributed Network Coded

Control Channel for Multi-Hop Cognitive Radio Networks,” IEEE Network, vol. 23, no. 4, 2009.

[13] N. Baldo, A. Asterjadhi, L. Giupponi, and M. Zorzi, “A Scalable Dynamic Spectrum Access Solution for Large Wireless Networks,” in

Proc. of IEEE ISWPC, May 2010, submitted for publication. [14] C. Arachchige, S. Venkatesan, and N. Mittal, “An Asynchronous

Neigh-bor Discovery Algorithm for Cognitive Radio Networks,” inProc. of IEEE DySPAN, Oct. 2008.

[15] L. Lazos, S. Liu, and M. Krunz, “Spectrum Opportunity-Based Control Channel Assignment in Cognitive Radio Networks,” inProc. of IEEE SECON, June 2009.

[16] S. Krishnamurthy, M. Thoppian, S. Kuppa, R. Chandrasekaran, N. Mit-tal, S. Venkatesan, and R. Prakash, “Time-efficient distributed layer-2 auto-configuration for cognitive radio networks,”Elsevier Computer Networks, vol. 52, no. 4, pp. 831–849, 2008.

[17] S. Krishnamurthy, M. Thoppian, S. Venkatesan, and R. Prakash, “Con-trol Channel Based MAC-layer Configuration, Routing and Situation Awareness for Cognitive Radio Networks,” inProc. of IEEE MILCOM, October 2005.

[18] L. Cao and H. Zheng, “Distributed Spectrum Allocation via Local Bargaining,”Proc. of IEEE SECON, Sept. 2005.

[19] T. Chen, H. Zhang, G. Maggio, and I. Chlamtac, “CogMesh: A Cluster-based Cognitive Radio Network,” inProc. of IEEE DySPAN, Apr. 2007. [20] N. Baldo, A. Asterjadhi, and M. Zorzi, “Distributed Dynamic Spectrum Access using a Virtual Coded Control Channel,” in Proc. of ACM IWCMC, June 2009.

[21] C. R. Lin and M. Gerla, “Adaptive Clustering for Mobile Wireless Networks,”IEEE Journal on Selected Areas in Communications, vol. 15, no. 7, pp. 1265–1275, 1997.

[22] N. Baldo, A. Asterjadhi, and M. Zorzi, “Cooperative Detection and Spectrum Reuse using a Network Coded Cognitive Control Channel,” inProc. of IEEE SECON, June 2009.

[23] S. Basagni, “Distributed clustering for ad hoc networks,” in Proc. of I-SPAN, Washington, DC, USA, June 1999.


Fig. 1. A typical clustered cognitive radio ad hoc network.
Fig. 2. Average number of clusters versus number of secondary users for different cluster protocols and primary users activity.
Fig. 3. Average cluster size versus number of secondary users for different cluster protocols and primary users activity.
Fig. 6. Number of slots required to assure P retr = 0.97 as a function of the cluster radius for different values of the secondary user density and the number of free common control channels.


Related documents

Small overload region and large fatigue region associated with multiple ratchet marks on the fracture surface indicates that the load was light, but there were high

Note that the AR(1)-term remains significant (at the 5% level) in all estimations. The set of available 15-second time intervals reveals a significant reaction for all cases

Comparing the performance levels of the entire hydrograph revealed that the prediction ability of the two-segment statistical models was superior to that of the

First, I model activism as a sequential decision process consisting of three consecutive stages of demand negotiations, board representation, and proxy contest, and de…ne the

The Gonds, together with the other Scheduled Caste (SC) communities, are among the many populations of India that are poor in both land and income who perform irregular, insecure,

The service was launched in 2008 and is governed by Korea Institute of Science and Technology Information (KISTI), a national research institute to establish national

Fi- nally, the sceptical universities (15%) are observed to be lagging behind the rest in almost every respect. They are characterised by a limited use of digital services such

grants came to Germany. Overall, over the period 1996 to 2001, 714,265 ethnic German immigrants came to Germany, which corresponds to an average inflow rate relative to the