Ad Hoc Networks 9 (2011) 1127–1139

Contents lists available at ScienceDirect

Ad Hoc Networks

journal homepage: www.elsevier .com/locate /adhoc

Adaptive power-controlled MAC protocols for improved throughputin hardware-constrained cognitive radio networks

Haythem Bany Salameh a,⇑, Marwan Krunz b

a Department of Telecommunication Engineering, Hijjawi Faculty for Engineering Technology, Yarmouk University, Irbid 21163, Jordanb Department of Electrical and Computer Engineering, University of Arizona, Tucson, AZ 85721, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 18 January 2010Received in revised form 31 August 2010Accepted 29 December 2010Available online 7 January 2011

Keywords:Bipartite matchingSingle-transceiverOpportunistic access

1570-8705/$ - see front matter � 2011 Elsevier B.Vdoi:10.1016/j.adhoc.2010.12.003

⇑ Corresponding author. Tel.: +962797708510; faE-mail addresses: haythem@yu.edu.jo (H.B. S

arizona.edu (M. Krunz).

Cognitive radios (CRs) are emerging as a promising technology to enhance spectrum utili-zation through opportunistic on-demand access. Many MAC protocols for cognitive radionetworks (CRNs) have been designed assuming multiple transceivers per CR user. How-ever, in practice, such an assumption comes at the cost of extra hardware. In this paper,we address the problem of assigning channels to CR transmissions in single-hop andmulti-hop CRNs, assuming one transceiver per CR. The primary goal of our design is tomaximize the number of feasible concurrent CR transmissions, and conserve energy as asecondary objective, with respect to both spectrum assignment and transmission powersubject to interference constraint and user rate demands. The problem is formulated underboth binary-level and multi-level spectrum opportunity frameworks. Our formulationapplies to any power-rate relationship. For single-hop CRNs, a centralized polynomial-timealgorithm based on bipartite matching that computes the optimal channel assignment isdeveloped. We then integrate this algorithm into distributed MAC protocols that preservefairness. For multi-hop ad hoc CRNs, we propose a novel distributed MAC protocol (WFC-MAC) that attempts to maximize the CRN throughput, assuming single transceiver radiosbut with ‘‘dual-receive’’ capability. WFC-MAC uses a cooperative assignment that reliesonly on information provided by the two communicating users. The main novelty inWFC-MAC lies in requiring no active coordination with licensed users and exploiting thedual-receive capability of radios, thus alleviating various channel access problems thatare common to multi-channel designs. We conduct theoretical analysis of our MAC proto-cols, and study their performance via simulations. The results indicate that compared withCSMA/CA variants, our protocols significantly decrease the blocking rate of CR transmis-sions, and hence improve network throughput.

� 2011 Elsevier B.V. All rights reserved.

1. Introduction

Heavy traffic load over the unlicensed portion of theradio spectrum (a.k.a., ISM bands) along with the ineffi-cient static allocation of the licensed spectrum have trig-gered the need for a new paradigm in spectrum

. All rights reserved.

x: +96227211129.alameh), krunz@ece.

allocation, whose main purpose is to improve spectrumefficiency through opportunistic spectrum access. Recentradio measurements conducted by the FCC and other agen-cies revealed vast temporal and geographical variations inthe utilization of the licensed spectrum, which can be aslow as 15% [1]. To overcome spectrum scarcity, cognitiveradios (which are based on programmable-radio plat-forms) have been proposed to allow opportunistic on de-mand access to the spectrum [1,2]. CR technology offerssuch opportunistic capability without affecting licensedprimary radio (PR) users.

1128 H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139

A CRN has unique characteristics that distinguish itfrom conventional multi-channel wireless networks. Un-like these networks, which typically operate over contig-uous bands [2,3], a CRN is expected to operate overwidely-separated, non-contiguous frequency bands. Com-munications on such bands exhibit different RF attenua-tion and interference behaviors. CRN users must operateusing a regulated transmission power so as to avoiddegrading the performance of PR users. They should fre-quently sense their operating channels for active PR sig-nals, and should vacate these channels if a PR signal isdetected. Although many MAC protocols have been pro-posed for traditional multi-channel wireless networks,these protocols are not well suited to the peculiar charac-teristics of CRNs. Specifically, the absence of PR users inmulti-channel wireless networks makes their protocolsfundamentally different from CRN MAC protocols [2,4].Hence, new CRN MAC protocols are needed for efficientspectrum utilization.

1.1. Previous research

Recently, several attempts were made to develop MACprotocols for CRNs (e.g., [3,5–10]). DDMAC [3] is a spec-trum-sharing protocol for CRNs that attempts to maxi-mize the CRN throughput through a novel probabilisticchannel assignment algorithm that exploits the depen-dence between the signal’s attenuation model and thetransmission distance while considering the prevailingtraffic and interference conditions. In [11], the conceptof a time-spectrum block is introduced to model spec-trum reservation in a CRN. Based on this concept, theauthors presented centralized and distributed CRN proto-cols with a common control channel for spectrum alloca-tion. In [9], the authors proposed a decentralized channel-sharing mechanism for CRNs based on a game-theoreticapproach for both cooperative and non-cooperativescenarios.

It is worth mentioning that most of these protocols as-sume that each CR is equipped with multiple transceivers,which may not often be the case. This assumption comes atthe cost of extra hardware, although it greatly simplifiesthe task of MAC design. In addition, these protocols are of-ten based on a greedy channel assignment strategy, whichselects the ‘‘best’’ available channel (or channels) for a gi-ven transmission [12,13]. The best channel is often definedas the one that supports the highest rate. Hereafter, we re-fer to this strategy as the best multi-channel (BMC) ap-proach. As shown later, when the BMC approach isemployed in a CRN, the blocking probability for CR trans-missions increases, leading to a reduction in networkthroughput. In contrast, in our work, we investigate the de-sign of new MAC protocols for single-hop and multi-hopCRNs, assuming a single half-duplex transceiver per CRuser. Our primary objective is to maximize the number offeasible concurrent CR transmissions with respect to bothchannel assignment and transmission power subject totarget rate demand and interference constraints. Energyconservation is also treated, but as a secondary objective.Our optimization follows a ‘‘fall back’’ approach, wherebythe secondary objective is optimized over the set of feasi-

ble channel assignments that are found to be optimal withrespect to the primary objective.

1.2. Contributions

The contributions of this paper are as follows. We firstformulate the optimal channel assignment and power allo-cation problem under both binary-level and multi-levelspectrum opportunity frameworks. Then, we present anoptimal centralized algorithm for this problem based onbipartite matching that applies to any power-rate relation-ship. For a single-hop CRN, we develop a CSMA-based MACprotocol, called AW-MAC, which realizes the centralizedalgorithm in a distributed manner. The centralized algo-rithm requires global information, which is hard to obtainin a multi-hop environment. Accordingly, for a multi-hopCRN, we present an efficient distributed channel assign-ment that relies only on information provided by the twocommunicating users. Our distributed scheme improvesthe CRN throughput performance through cooperativeassignment among neighboring CR users. Specifically, aCR user that intends to transmit has to account for potentialfuture transmissions in its neighborhood. Based on thisdistributed scheme, we then develop a novel CSMA-basedMAC protocol, called WFC-MAC, for multi-hop ad hoc CRNswith a single half-duplex radio per node. WFC-MAC ex-ploits the ‘‘dual-receive single-transmit’’ capability ofradios (i.e., each radio is capable of receiving over twochannels simultaneously, but can transmit over one chan-nel at a time), thus alleviating various channel access prob-lems that are common to multi-channel designs. Ourprotocols do not require interaction with PR networks(PRNs), and can be adapted to existing multi-channel sys-tems (using currently available hardware) with little extraprocessing overhead.

To evaluate the performance of our protocols, we con-duct simulations for a single-hop and a multi-hop CRNwith mobile users. Simulation results show that our proto-cols significantly improve the network throughput overtwo previously proposed schemes (i.e., BMC-MAC [13,12]and DDMAC [3]). The results also indicate that our proto-cols preserve (even slightly improve) throughput fairness.For single-hop scenarios, we show that AW-MAC achievesbetter throughput (up to 50% improvement over BMC-MACscheme) at no additional cost in energy consumption. Inmulti-hop scenarios, WFC-MAC achieves better through-put at the cost of energy consumption.

1.3. Organization

The rest of the paper is organized as follows. In Section2, we introduce our system model, state our assumptions,and formulate the optimal channel assignment/power con-trol problem. Section 3 introduces the centralized channelassignment algorithm. Section 3.2 describes the proposedAW-MAC protocol. In Section 4, we introduce the distrib-uted channel assignment algorithm and the proposedWFC-MAC protocol. In Section 5, we analysis the through-put of our proposed protocols. Section 6 presents our sim-ulation results. Our concluding remarks are presented inSection 7.

H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139 1129

2. Problem formulation and design constraints

2.1. Network model

We consider a distributed opportunistic CRN that geo-graphically coexists with M different PRNs. The PRNs are li-censed to operate on non-overlapping frequency bands,each of Fourier bandwidth W. For k = 1, . . . ,M, the carrierfrequency associated with the kth PRN is fk (in Hz). LetM denote the set of all non-overlapping channels in allPRNs (i.e., M ¼ jMj).

CR users continuously identify potential spectrumholes and exploit them for their transmissions. They em-ploy power control to avoid harmful interference with PRreceptions. Specifically, CR users adopt a regulated trans-mission power strategy, whereby for band i, i = 1, . . . ,M,the maximum CR transmission power is 0 if any PR useroperates on band i, or limited to PðiÞmax if no PR signal isdetected. PðiÞmax is the smaller of the FCC regulatory maxi-mum transmission power over band i and the maximumpower supported by the CR’s battery (PCR). Note thatidentifying the list of idle channels that is potentiallyavailable for CR transmissions at a given time and in a gi-ven geographical location is a challenging problem. Todeal with this challenge, the FCC recently adopted threeprincipal methods that can be used to determine the listof idle channels that is potentially available for CR trans-missions at a given time and in a given geographical loca-tion [14]. The first method requires determining thelocation of a CR user and then accessing a database of li-censed services (internal or external database) to identifybusy/idle PR channels. The second method is to integratespectrum sensing capabilities in the CR device. The thirdmethod is to periodically (or on-demand) transmit con-trol information from a professionally installed fixedbroadcast CR station. This control information containsthe list of idle channels. Under this method, a CR trans-mitter can only transmit when it receives a control infor-mation that positively identifies idle PR channels.According to the FCC, this control information can alsobe transmitted by external entities, such as PR base sta-tions (e.g., broadcast TV and radio stations). For our pur-poses, we assume that the control signal method is inplace for determining the list of idle channels.

2.2. Feasibility constraints

For a CR transmission j, transmitter and receiverneed to cooperatively select an appropriate channeland transmission power while meeting the followingconstraints:

1. Exclusive channel occupancy policy: The selected channelcannot be assigned to more than one CR transmission inthe same neighborhood (inline with the CSMA/CAmechanism).

2. One transceiver per CR user: Each CR user can transmit orreceive on one channel only. The operation is half-duplex, i.e., a CR user cannot transmit and receive atthe same time.

3. Maximum transmission power: For a CR transmission jand idle channel i, the transmission power PðiÞj is limitedto PðiÞmax. If channel i is occupied by a PR user, PðiÞj ¼ 0.

4. Rate demand: Each CR transmission j requires a givendata rate Rj. If none of the idle channels can supportRj, the CR transmission j will be blocked.

2.3. Problem formulation

At a given time t, let NðtÞ andMIdleðtÞ#M respectivelydenote the set of all CR transmission requests and the set ofall jMIdleðtÞj ¼ MIdleðtÞ idle channels in a given neighbor-hood. Let NðtÞ ¼ jN ðtÞj. It has been shown that neighboringCR users in a given locality typically share a similar view ofspectrum opportunities (i.e., the set of common idle chan-nels) [15,2].

Given the rate demands ðRj; 8j 2 NðtÞÞ and the set ofidle channels MIdleðtÞ, our goal is to compute a feasiblechannel assignment that assigns channels and transmis-sion powers to CR requests such that the number of simul-taneous CR transmissions is maximized subject to thepreviously mentioned constraints. If multiple solutions ex-ist for this optimization problem, we seek the one that re-quires the least amount of energy. Because we focus oncomputing a feasible channel assignment at a given timet, in what follows, we drop the time subscript (t) for nota-tional convenience. Let aðiÞj be a binary variable that is de-fined as follows:

aðiÞj ¼1; if channel i is assigned to transmission j

0; otherwise:

�

ð1Þ

The resource assignment problem is stated as follows:

maximizeaðiÞj;PðiÞ

j

Xi2MIdle

Xj2N

aðiÞj 1½rðiÞj P Rj� �1

Ptot

Xi2MIdle

Xj2N

aðiÞj PðiÞj

SubjecttoXj2N

aðiÞj 6 1;8i 2 MIdle

Xi2MIdle

aðiÞj 6 1;8j 2 N

0 6 PðiÞj 6 PðiÞmax;8i 2 MIdle and 8j 2 N ð2Þ

where 1[�] is the indicator function, rðiÞj ¼ f ðPðiÞj Þ is the datarate for link j on channel i, f(�) is monotonically non-decreasing rate-power function (it can be Shannons capac-ity or any other power-rate function), and Ptot ¼

Pi2MPðiÞmax.

The second term in the objective function ensures that ifmultiple solutions exist for the optimization problem, theone with the least amount of total transmission power willbe selected. Note that the first two constraints in (2) ensurethat at most one channel can be assigned per transmissionand a channel cannot be assigned to more than one trans-

mission. The third constraint ensures thatPðiÞ

j

PðiÞmax6 1. Given

the above three constraints and noting that MIdle #M,the second term of the objective function is always <1(i.e., 1

Ptot

Pi2MIdle

Pj2NaðiÞj PðiÞj < 1). So, for any two feasible

assignment X1 with N1 of admitted CR transmissions andX2 with N2 < N1 of admitted CR transmissions, the above

1130 H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139

formulation will also selects X1 over X2, irrespective of thetotal transmission power.

The optimization problem in (2) is a mixed integer non-linear program (MINLP). Due to integrality constraints, oneexpects such a problem to be NP-hard. However, we showthat this MINLP is not NP-hard and may be solved opti-mally in polynomial time. Specifically, we show that thisproblem is the same as assigning channels to independent(distinct) links such that the number of CR transmissions ismaximized while using the minimum total transmissionpower. In Section 3, we propose an algorithm that trans-forms this optimization problem into the well-knownmaximum weighted perfect bipartite matching problem,which has a polynomial-time solution [16].

Remark. For multi-transceiver case, the joint channel/power assignment problem is known to be NP-hard [3,17].

1 2 3

w1

(1)

w1

(2)

w1

(3)

w3

(3)

w3

(1)

w3

(2)

w2

(1)

w2

(2)

w3

(3)

Links

3. Optimal channel assignment

In this section, we first present a centralized algorithmfor the channel assignment problem based on bipartitematching. The objective of this algorithm is to maximizethe total number of simultaneous CR transmissions bymeans of power management. Note that centralized algo-rithms are easy to implement in single-hop networkswhere all users are within radio range of each other. Basedon this centralized algorithm, we develop a CSMA-basedMAC protocol that can be executed in a distributedmanner.

3.1. Proposed algorithm

In our context, a centralized algorithm implies that theinstantaneous SINR values, location information, and ratedemand are known to the decision-making entity that as-signs channels and transmission powers. For a finite num-ber of available channels and given rate demands, a CRuser can compute the minimum required power over eachchannel. Using this fact and noting that the graph connect-ing the set of CR transmission requests and the set of avail-able channels is a bipartite graph,1 our optimizationproblem can be transformed into a bipartite perfect match-ing problem. The maximum matching of this bipartite graphproblem is the set containing the maximum number of CRtransmissions that can proceed simultaneously. If there aremultiple feasible channel assignments with maximummatching, the one requiring the smallest total transmissionpower will be selected. In the following, we develop an algo-rithm that transforms our optimization problem into abipartite perfect matching problem. Formally, the algorithmproceeds as follows:

Step 1. Compute the minimum required powers: Forevery CR transmission request j 2 N and every idlechannel i 2 MIdle, the algorithm computes the

1 A bipartite graph is a graph whose vertex set can be decomposed intotwo disjoint sets such that no two vertices in the same set are connected.

minimum required transmission power PðiÞj;req thatcan support the rate demand Rj, i.e.,

Cha

PðiÞj;req ¼ f�1ðRjÞ: ð3Þ

where f�1(�) is the inverse of the rate-power function f(�).Then, the algorithm identifies prohibited (infeasible) chan-nel/transmission combination (i, j) whose PðiÞj;req does notsatisfy the maximum transmission power constraint (i.e.,PðiÞj;req > PðiÞmax).Step 2. Formulate and solve the perfect bipartite

matching problem: The algorithm creates MIdle

nodes, each corresponding to one of the idle chan-nels. Let these nodes constitute the channel set C.The algorithm also creates N nodes to representthe CR transmission requests. Let these nodes con-stitute the request set R. If N > MIdle, the algorithmcreates N �MIdle additional nodes CD ¼ fMIdleþ1; . . . MIdle þ Ng to represent dummy channels andupdates C as C ¼ C

SCD. On the other hand, if

N < MIdle, the algorithm creates MIdle � N additionalnodes RD ¼ fN þ 1; . . . MIdle þ Ng to representdummy requests and updates R as R ¼ R

SRD.

Then, the algorithm connects the nodes in C tothe nodes in R. Any (i, j) assignment that containsa dummy node is also a prohibited assignment. LetwðiÞj denote the arc weight of link (i, j) on the bipar-tite graph. For all prohibited assignments, thealgorithm sets wðiÞj to a very large number C� PCR.Formally,

wðiÞj ¼ PðiÞj;req; if PðiÞj;req 6 PðiÞmax; j 2 R and i 2 C

wðiÞj ¼ C; if PðiÞj;req > PðiÞmax; j 2 R and i 2 C:

8<: ð4Þ

Fig. 1 shows an example of the bipartite graph withMIdle = N = 3.

The above bipartite graph construction transforms theassignment problem into a weighted perfect bipartitematching (because the number of CR transmissions is equalto the number of channels, and every node in the requestset is connected to every node in the channel set). It isworth mentioning that the global optimal solution of suchmatching problem can be found using the Hungarian algo-rithm, which has a polynomial-time complexity (i.e.,Oð

ffiffiffiffiKp

KÞ, where K = max{N,MIdle} [16]) and codes are read-ily available for its implementation [18]. Note that the ob-tained optimal solution is a one-to-one assignment for the

1 2 3nnels

Fig. 1. Bipartite graph with MIdle = N = 3.

AW

Ctrl

Data

Tdata

Data+Ack

MIdleTctrl

Tctrl

……..

tto AS

Fig. 2. Basic operation of AW-MAC.

H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139 1131

constructed weighted max{MIdle,N} �max{MIdle,N} bipar-tite graph. To find the optimal feasible assignment thatmaximizes the number of possible concurrent CR trans-missions while selecting the minimum transmission pow-ers, all prohibited assignments in the obtained one-to-oneassignment should be removed.

3.2. Channel access protocol for single-hop CRNs

Based on the channel assignment algorithm presentedin Section 3.1, we now propose a distributed multi-channelMAC protocol for single-hop ad hoc CRNs with a singlehalf-duplex radio per node. Before describing our protocolin detail, we first state our main assumptions.

3.2.1. AssumptionsFor each frequency channel, we assume that its gain is

stationary for the duration of a few control packets andone data packet. This assumption holds for typical mobilitypatterns and transmission rates [19]. We also assume sym-metric gains between two users, which is a commonassumption in RTS/CTS-based protocols, including the IEEE802.11 scheme. Our protocols assume the availability of aprespecified common control channel. Such a channel isnot necessarily dedicated to the CRN. It may, for example,be one of the unlicensed ISM bands. Note that the exis-tence of a common control channel is a characteristic ofmany MAC protocols proposed for CRNs (e.g., [3,20,12,21]).

3.2.2. Operational detailsTo execute the centralized algorithm presented in the

previous section in a distributed manner, we require theinstantaneous SINR information and rate demands of allcontending CR users in a given locality to be available toCR users in that locality before assigning channels andtransmission powers. In a single-hop network, this issuecan be handled during the ‘‘admission phase’’ by introduc-ing a contention period known as the access window (AW).The AW consists of MIdle fixed-duration access slots (AS). Aseries of control packet exchanges take place during theseslots, after which several data transmissions can com-mence concurrently. We note here that the use of an AWfor contention was originally proposed in the MACA-P pro-tocol [22] and was later integrated in the design of POW-MAC [19]. However, in both protocols the objective wasnot to address spectrum sharing (channel assignment),but rather to prevent collisions between control and datapackets (in MACA-P) and to address single-channel trans-mission power control (in POWMAC). During the AW, com-municating CR users announce their instantaneous SINRinformation. A CR user that has packets to transmit andthat is not aware of any already established AW in itsneighborhood can asynchronously initiate an AW. EachAS consists of the sum of an RTS duration, a CTS duration,and a maximum backoff interval (explained below), andtwo fixed short interframe spacing (SIFS) periods.2 Controlpackets are sent at the maximum (known) power Pctrl. This

2 As defined in the IEEE 802.11b standard [2], a SIFS period consists of theprocessing delay for a received packet plus the turnaround time.

Pctrl is constrained by the maximum permissible transmis-sion power imposed on the control channel. Upon receiv-ing an RTS packet from a CR user, say A, that is initiatingan AW, other CR users in the network synchronize theirtime reference with A’s AW.

Suppose that a CR user C overhears A’s RTS, and has adata packet to send. C contends for the control channel inthe next access slot of A’s AW as follows. It first backs offfor a random duration of time (T) that is uniformly distrib-uted in the interval [0,Tmax]; Tmax is a system-wide backoffcounter. After this waiting time and if no carrier is sensed,user C sends its RTS packet in the current AS. Note that Tmax

is in the order of a few microseconds whereas a time slot isin milliseconds, so the backoff mainly serves to preventsynchronized RTS attempts. For illustration purposes,Fig. 2 shows a time diagram of the channel access process,assuming fixed data-packet sizes and equal rate demands.Tctrl and Tdata in the figure denote the durations (in seconds)of one RTS/CTS packet exchange and one data plus ACKpackets transmissions, respectively.

After all the control packets have been exchanged, thechannel assignment and power management algorithm ofSection 3 is executed at every communicating node.

3.3. Remarks and design variants

3.3.1. Granularity of channel assignmentDepending on channel availability due to PR dynamics,

the proposed channel assignment can be performed at thegranularity of a packet or a link. In the latter case, theassignment applies to all packets of the current connectionbetween the two end points of a link.

3.3.2. Fairness properties of AW-MACAccording to AW-MAC, CR users contend over the con-

trol channel using a variant of the CSMA/CA mechanism.This gives all CR users the same probability of accessingchannels, irrespective of their rate demands. Thus, ourAW-MAC protocol preserves fairness among CR users. Inour simulations (Section 6), we compare the fairness prop-erties of AW-MAC to that of a typical multi-channel CSMA-based protocols. The results show that AW-MAC preserves(slightly improves) the network fairness.

3.3.3. RTS/CTS handshake in AW-MACIt should be noted that the RTS/CTS handshake is essen-

tial in multi-channel systems (e.g., CRNs). Besides mitigat-

AW

Ctrl

Data

Tdata

Data+AckMoTctrl

Tctrl

……..

tto AS

Tctrl

……..

AWM1Tctrl

t1

……..

……..

Fig. 3. Basic operation of 2-radio AW-MAC (Note that MIdle(to) = Mo andMIdle(t1) = M1).

1132 H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139

ing the hidden-terminal problems, there are two othermain objectives for the use of RTS/CTS: (1) conductingand announcing the channel assignment, and (2) prompt-ing both the transmitter and the receiver to tune to theagreed on channels before transmission commences. Sim-ulation studies have shown that using RTS/CTS packetsfor data packets larger than 250 bytes is beneficial [23].3

It is also worth mentioning that our AW-MAC protocol relieson passive learning. When CR users are not transmitting/receiving (i.e., during the admission control phase), they al-ways listen to the control channel and overhear any control-packet exchanges, including those not destined to them.Contending CR users use the control information to extractthe required rate demand and instantaneous SINR informa-tion. Thus, AW-MAC does not introduce any additional con-trol message overhead beyond the classic two-way (RTS/CTS) handshake, which is indeed needed for any multi-chan-nel CSMA/CA MAC protocol.

3.3.4. Channel assignment with a multi-level frequency-dependent power constraint

The problem of identifying spectrum holes and select-ing appropriate channels/powers is overcomplicated bythe presumingly non-cooperative nature of PRNs, whichusually do not provide feedback (e.g., interference mar-gins) to CR users. To address this problem, a multi-leveltime-varying frequency-dependent power mask Pmask ¼ð

Pð1Þmask; Pð2Þmask; . . . ; PðMÞmask

n oÞ on the CR transmissions is often

adopted (e.g., [3,25]). Enforcing such a power mask allowsfor spectrum sharing between neighboring CR and PRusers. According to this approach, CR users can exploitboth idle as well as partially-utilized bands, potentiallyleading to better spectrum utilization. However, the deter-mination of an appropriate multi-level power mask is stillan open issue, which has been recently investigated undercertain simplifying assumptions (e.g., [12,2]). Although ourproposed algorithm assumes a binary-level power con-straint on CR transmissions, the algorithm is still valid forthe case of a multi-level frequency-dependent power maskby setting the maximum CR transmission power overchannel i to PðiÞmax ¼min PðiÞmask; PCR

n o;8i 2 M.

3.3.5. AW-MAC with two transceiversAnother design possibility that can achieve improve-

ment in the CRN throughput is to use two half-duplextransceivers per CR user: a control transceiver and a datatransceiver. The control transceiver only operates on thecontrol channel to exchange control packets with otherCR users and to obtain the right to access data channels.The data transceiver dynamically switches to one of thedata channels in MIdle to transmit data packets. In such acase, each CR user can transmit or receive only on one datachannel at a time. Because there is no interference betweendata and control transmissions (the two are separated infrequency), CR users always listen to the control channel,and accordingly reservations of the subsequent AW canbe conducted while current data transmissions are taking

3 The RTS threshold depends on the number of users in the network[23,24]. It should be reduced for a large number of users.

place (i.e., mimicing a full-duplex operation). This reducesthe control overhead and improves the overall throughputat the cost of an additional transceiver. We refer to thechannel access mechanism that uses AW assignment withone transceiver as AW-MAC, and the one that uses AWassignment with two transceivers as 2-radio AW-MAC.Fig. 3 shows the basic operation of 2-radio AW-MAC. InSection 5, we study the potential throughput improvementof 2-radio AW-MAC due to its reduced control overheadover AW-MAC.

4. Distributed channel assignment for multi-hop CRNs

In this section, we present a distributed channel assign-ment scheme for a multi-hop CRN. It attempts to improvespectrum utilization in a purely distributed manner whilerelying only on information provided by the two commu-nicating nodes. We first identify the key challenges in-volved in realizing the centralized algorithm in adistributed manner. Then, we describe our distributedscheme in detail.

4.1. Challenges

To execute our centralized algorithm in a multi-hopenvironment, the algorithm must run in a distributed man-ner at each CR device in a given locality (i.e., contention re-gion). This implies that each CR user that belongs to acontention region must exchange instantaneous SINRinformation and rate demands with other neighboring CRusers in that region before selecting channels and powers.This incurs high control overhead and delay. Moreover, in amulti-hop environment, CR users may belong to multiplecontention regions that differ in their views of the spec-trum opportunity. To overcome such challenges, we devel-op a heuristic channel assignment scheme that provides asuboptimal solution with low complexity and thatachieves good spectrum utilization.

4.2. Channel assignment

The main consideration in our distributed scheme is toenable cooperation among neighboring CR users. A CR userthat intends to transmit has to account for potential futuretransmissions in its neighborhood. It does that in a purelydistributed manner while relying only on information

H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139 1133

provided by the two communicating nodes4 by assigningto its transmission the worst feasible channel, i.e., the least-capacity available channel that can support the required ratedemand.5 We refer to this approach as the worst feasiblechannel (WFC) scheme. Note that a user determines theworst feasible channel for its transmission using only localinformation. WFC scheme preserves better channels for po-tential future CR transmissions. Even though WFC requires apair of CR users to communicate on a channel that may notbe optimal from one user’s perspective, such a channel givesmore room to other CR transmissions to take place simulta-neously, especially under moderate to high traffic loads.Compared with previously proposed channel assignmentschemes (evaluated in Section 6), our approach avoidsunnecessary blocking of CR transmissions and has a greatpotential to improve network throughput by means of coop-erative channel assignment.

4.3. Channel access protocol

4.3.1. Protocol overviewBased on the WFC algorithm, we propose a distributed

multi-channel MAC protocol for multi-hop ad hoc CRNswith a single half-duplex radio per node. The proposed pro-tocol is an extension of the single channel RTS-CTS-DATA-ACK handshaking scheme used in the 802.11 standard. Itdiffers from previous designs in that it exploits the ‘‘dual-re-ceive single-transmit’’ capability of radios (i.e., each radio iscapable of receiving over two channels simultaneously, butcan transmit over one channel at a time). The operation ishalf-duplex, i.e., while transmitting, the radio cannot re-ceive/listen, even over other channels. It can be imple-mented using one transceiver with slight upgrade in thereceive chains of the circuitry. This capability is readilyavailable in some recent radios. For example, QUALCOMM’sRFR6500 radio [26] supports ‘‘simultaneous hybrid dual-re-ceive operation, which allows for 1X paging signal monitor-ing during a 1xEV-DO connection, while monitoring otherfrequency bands for hand-off’’. Another example is Ken-wood’s TH-D7A Dual-Band Handheld Transceiver [27],which supports simultaneous reception over both dataand voice channels using a single antenna. Though a simpleenhancement of the transceiver circuitry, the dual-receivecapability makes the MAC design much easier. In particular,if we assume a common (or coordinated) control channel, aCR user that is not transmitting any data can tune one of itstwo receive branches to the control channel while receivingdata over the other receive branch. This way, the multi-channel hidden-terminal problem can be alleviated.

4.3.2. Operational detailsTo facilitate multi-channel contention and reduce the

likelihood of CR collisions, each CR user, say A, maintains

4 Queueing channel access requests and using intelligent scheduling willimprove spectrum utilization. However, executing such queueing andscheduling in a multi-hop environment incurs high control overhead anddelay.

5 Recall that, in this paper, we consider a CRN with a target rate demandper CR user (i.e., each CR transmission j requires a given data rate Rj; if noneof the idle channels can support Rj, the transmission j will be blocked).

a free-channel list (FCL) and a busy-node list (BNL). TheFCL (A) represents idle PR channels that are not occupiedby other CR users within the A’s one-hop communicationrange. BNL(A) consists of the IDs of CR users that are cur-rently busy transmitting/receiving data packets in A’sneighborhood. The FCL(A) and BNL(A) are continuously up-dated according to the channel access dynamics and over-heard control packets. The proposed protocol followssimilar interframe spacings and collision avoidance strate-gies of the 802.11 scheme (implemented here over thecontrol channel) by using physical carrier sensing andbackoff before initiating control-packet exchanges. Uponaccessing the control channel, communicating CR usersperform a three-way handshake, during which theyexchange control information, conduct the channel assign-ment, and announce the outcome of this channel assign-ment to their neighbors.

The details of the channel access mechanism are nowdescribed. Suppose that CR user A has data to transmit toCR user B at a rate demand RA. If A does not sense a carrierover the control channel for a randomly selected backoffperiod, it proceeds as follows:

� If FCL (A) is empty or B is busy (based on BNL(A)), Abacks off and attempts to access the control channellater. Otherwise, A sends an RTS message at power Pctrl.The RTS packet includes FCL (A) and RA.� A’s neighbors other than B, that can correctly decode the

RTS will stay silent until either they receive anothercontrol packet from A, denoted by FCTS (explainedbelow), or until the expected time for the FCTS packetexpires.� Upon receiving the RTS packet, B determines the com-

mon channel list that is available for A ? B transmission,denoted by CCL(A,B). Then, B proceeds with the channelassignment process, whose purpose is to determinewhether or not there exists a feasible channel assign-ment that can support RA.� Depending on the outcome of the channel assignment

process, B decides whether or not A can transmit. Ifnot (i.e., none of the channels in CCL(A,B) can supportRA), then B does not respond to A, prompting A to backoff, with an increased backoff range that is similar to802.11, and retransmit later. Otherwise, B sends a CTSmessage to A that contains the assigned channel, thetransmit power, and the duration (Tpkt(A)) needed toreserve the assigned channel. The CTS implicitlyinstructs B’s CR neighbors to refrain from transmittingover the assigned channel for the duration Tpkt(A).� Once A receives the CTS, it replies back with a ‘‘Feasible-

Channel-to-Send’’ (FCTS) message, informing its neigh-bors of the assigned channel and Tpkt(A). Such a three-way handshake is typically needed in multi-channelCSMA/CA protocols designed for multi-hop networks(e.g., [13,3,12]). For single-hop networks, where allusers can hear each other, there is no need for the FCTSpacket. Likewise, in single-channel multi-hop networks,the FCTS packet is also not needed.� After completing the RTS/CTS/FCTS exchange, the trans-

mission A ? B proceeds. Once completed, B sends backan ACK packet to A over the assigned data channel.

……..

t

Ctrl

CH 1

Tctrl

CH 2

CH MIdle

Tdata

MIdle Tctrl

CH MIdle-1

Tctrl

.

.

.

.

.

………...

………...

………...

………...

………...

(MIdle -1)T ctrl

(MIdle-2)T ctrl

……

Tctrl

.

.

.

.

.

Fig. 4. Basic operation of the distributed spectrum access scheme.

1134 H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139

When used with the WFC assignment, the above proto-col is referred to as WFC-MAC. Note that, while receiving adata packet over a given data channel, a CR user still listensto other control packet exchanges taking place over thecontrol channel, and can update its FCL and BNL accord-ingly. However, a CR user that is transmitting a data packetwill not be able to listen to the control channel, so its FCLand BNL may become outdated. We refer to this problemas transmitter deafness, which is primarily caused by thehalf-duplex nature of the radios. To remedy this problem,when the receiver sends its ACK, it includes in this ACKany changes in the FCL and BNL that may have occurredduring the transmission of the data packet. The transmitteruses this information to update its own tables.

Because there is no interference between data and con-trol packets, a CR user that hears the RTS (CTS) packet de-fers its transmission only until the end of the controlpacket handshaking. This allows for more parallel trans-missions to take place in the same vicinity.

5. Throughput analysis

In this section, we use simplified analysis to evaluatethe maximum achievable throughput of various channelaccess schemes in single-hop topologies. We assume thata CR user transmits data in the form of fixed-size packetsat a fixed transmission rate. Recall that Tctrl denotes thetransmission duration of one RTS plus one CTS packets,and Tdata denotes the duration of one data plus one ACKpackets. Assume that Tctrl can be expressed in terms of Tdata

as Tctrl = dTdata. It is worth mentioning that according to theIEEE 802.11 specifications, Tdata is at least an order of mag-nitude larger than Tctrl (i.e., 0 < d� 1). As an example, con-sider data- and control-packet sizes of 4-KB and 120 bits,respectively [28]. Also consider a transmission rate of5 Mbps. Then, d 0.0073. We now provide expressionsfor the maximum achievable throughput under the variousschemes assuming the availability of MIdle channels and aper-packet channel assignment. The maximum achievablethroughput is defined as the maximum number of simulta-neous CR transmissions that can be supported in aTdata þMIdleTctrl ¼ ð1þMIdledÞTdata duration.

For the single-transceiver AW-MAC, according to Fig. 2,the maximum number of data packets that can be poten-tially transmitted in a Tdata + MIdleTctrl duration is MIdle. Un-der 2-radio AW-MAC, at steady state, the maximumnumber of data packets that can be potentially transmittedin the same duration is MIdle þ

PMIdlei¼1 MIdle

TctrlTdata¼

MIdle þM2Idled ¼ MIdleð1þMIdledÞ (see Fig. 3). Under both

WFC-MAC and BMC-MAC (similar to WFC-MAC but usesthe BMC channel assignment), for a given channel i, Fig. 4shows that an RTS/CTS exchange can immediately followthe transmission of the previous data packet over thatchannel. Thus, the maximum achievable throughput inthe Tdata + MIdleTctrl duration is MIdle þ

PMIdle�1i¼1 ðMIdle � i�

1Þ TctrlTdata¼ MIdle þ dð

PMIdle�1i¼1 ðMIdle � 1Þ �

PMIdle�1i¼1 iÞ ¼ MIdleþ

dððMIdle � 1ÞðMIdle � 1Þ �PMIdle�1

i¼1 iÞ. By summing

the series using the identityPm

i¼1i ¼ mðmþ1Þ2 , this quantity

can be written as MIdle þ dððMIdle � 1ÞðMIdle � 1Þ�

ðMIdle�1ÞMIdle2 Þ ¼ MIdle þ dðMIdle � 1ÞðMIdle

2 � 1Þ ¼ MIdle þ dðM2Idle2 �

32 MIdle þ 1Þ.

Computing the maximum achievable throughput in thisway is rather optimistic since we are assuming that for 2-radio AW-MAC/AW-MAC, all AW slots result in successfulRTS/CTS exchanges, and that for the BMC-MAC/WFC-MACand a given data channel, an RTS/CTS exchange followsimmediately the transmission of the previous data packetover that channel.

Fig. 5 shows the maximum achievable throughput as afunction of MIdle for two data-packet sizes and variouschannel access schemes. For practical data- and control-packet sizes [28], where d� 1, the figures reveal that var-ious channel access schemes achieve comparable through-put performance. More importantly, the use of two half-duplex transceivers per CR user (with one transceivertuned to the common control channel) provides a minorimprovement in the system throughput over a single-transceiver design. The figures also demonstrate that thethroughput gain due to two transceivers is larger at smal-ler data-packet sizes (i.e., larger d) and larger MIdle. This isbecause a larger d (or MIdle) means larger AW duration,which results in more overhead for the single-transceiversolution.

6. Performance evaluation

We now evaluate the performance of the proposed pro-tocols via simulations. Our proposed protocols (AW-MACand WFC-MAC) are compared with two multi-channelMAC protocols: BMC-MAC [12,13] and DDMAC [3]. Asmentioned before, BMC-MAC selects the best availablechannel for data transmission. DDMAC is a CSMA-basedspectrum-sharing protocol for CRNs. It attempts to maxi-mize the CRN throughput through a probabilistic channelassignment algorithm that exploits the dependence be-tween the signal’s attenuation model and the transmissiondistance while considering current traffic and interferenceconditions. For a fair comparison, in BMC-MAC, WFC-MAC,and DDMAC, CR users employ the same channel access

5 10 15 200

5

10

15

20

25

MIdle

Max

imum

Ach

ieva

ble

Thro

ughp

ut2−radio AW−MAC

AW−MAC

WFC/BMC−MAC

5 10 15 200

5

10

15

20

25

MIdle

Max

imum

Ach

ieva

ble

Thro

ughp

ut

2−radio AW−MAC

AW−MAC

WFC/BMC−MAC

Fig. 5. Maximum achievable throughput (in packet/(Tdata + MIdleTctrl)) vs. total number of idle channels (control-packet size = 120 bits).

4 8 12 16 20 24 28 32 3620

25

30

35

40

45

λ (Packet/Sec)

CR

Blo

ckin

g ra

te (%

)

WFC−MACBMC−MACDDMACAW−MAC2−radio AW−MAC

0 4 8 12 16 20 24 28 32 360

6

18

30

42

λ (Packet/sec)

Thro

ughp

ut (M

bps)

WFC−MACBMC−MACDDMACAW−MAC2−radio AW−MAC

50%

18%1%

4 8 12 16 20 24 28 32 360

0.02

0.04

0.06

0.08

λ (Packet/sec)

Ep

(mJ)

WFC−MACBMC−MACDDMACAW−MAC2−radio AW−MAC

Fig. 6. CRN performance in single-hop scenarios.

H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139 1135

mechanism described in Section 4.3. They differ in thechannel assignment approach. The maximum achievablethroughput under DDMAC channel access is the same asthe one obtained in Section 5 for WFC-MAC/BMC-MAC

and is comparable to the one for AW-MAC (see Fig. 5). Notethat, in all protocols, if there is no feasible channel assign-ment that can support the rate demand, no channel will beassigned, prompting the transmitter to back off. It is worth

1136 H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139

mentioning that DDMAC involves more processing over-head, as it requires distance and traffic estimation. In ourevaluation, we first study the network performance in asingle-hop CRN, where all users can hear each other. Then,we study it in a multi-hop mobile CRN. Our results arebased on simulation experiments conducted using CSIM,a C-based, process-oriented, discrete-event simulationpackage [29].

0 4 8 12 16 20 24 28 32 360.75

0.8

0.85

0.9

0.95

1

λ (Packet/sec)

Fairn

ess

Inde

x

AW−MACWFC−MACBMC−MACDDMAC

Fig. 7. Fairness index in single-hop scenarios (2-radio AW-MAC depictedsimilar behavior as AW-MAC).

1 2 3 4 5 60

0.05

0.1

0.15

High

Ch

1 2 3 4 5 60

0.05

0.1

0.15Low

1 2 3 4 5 60

0.05

0.1

0.15

Moder

Cha

nnel

Usa

ge (%

)

BMC−MAC WFC−MAC AW−

Fig. 8. CR channel usage in single-hop scenarios (2-radio

6.1. Simulation setup

We consider four PRNs and one CRN that coexist in a100 m � 100 m field. Users in each PRN are uniformly dis-tributed. The PRNs operate in the 600 MHz, 900 MHz,2.4 GHz, and 5.7 GHz bands, respectively. Each PRN con-sists of three 2.5-MHz-wide channels, resulting in a maxi-mum of 12 channels for opportunistic transmissions. Wedivide the time into slots, each of length 6.6 ms. A time slotcorresponds to the transmission of one data packet of size4-KB at a transmission rate of 5 Mbps. Each user in the kthPRN acts as an ON/OFF source, where it is ON while trans-mitting and OFF otherwise. The source is further character-ized by the distribution of its ON and OFF periods, whichare both taken to be exponential. We set the average ONand OFF periods for the four PRNs to be the duration of10 and 190 time slots, respectively. The number of PR linksin each PRN is 20. Each active link in the kth PRN transmitsover one of the three channels in its own band. Thus, theavailable spectrum opportunity in each PR band is 66.7%.For the CRN, we consider 200 mobile users. The randomwaypoint model is used for mobility, with the speed of aCR user uniformly distributed between 0 and 2 m/s. Foreach generated packet, the destination node is selectedrandomly. Each CR user generates fixed-size (4-KB) datapackets according to a Poisson process of rate k (in pack-et/time slot). Each user requires a transmission rate of5 Mbps. We set the CRN SINR threshold to 5 dB and thethermal noise power density to PðiÞth ¼ 10�21 Watt/Hz for

7 8 9 10 11 12

Load

. No.

7 8 9 10 11 12

Load

7 8 9 10 11 12

ate Load

MAC DDMAC

AW-MAC depicted similar behavior as AW-MAC).

0.05 0.1 0.15 0.2 0.2535

40

45

50

55

λ (Packet/time slot)

CR

Blo

ckin

g R

ate

(%)

WFC−MACBMC−MACDDMAC

0 0.05 0.1 0.15 0.2 0.250

5

10

15

20

25

λ (Packet/time slot)

Thro

ughp

ut (M

bps)

WFC−MACBMC−MACDDMAC

20%

9%

0.05 0.1 0.15 0.2 0.25

0.8

1

1.2

1.4

1.6

1.8

λ (Packet/time slot)

Ep

(mJ)

WFC−MACBMC−MACDDMAC

Fig. 9. CRN performance in multi-hop scenarios.

H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139 1137

all channels. We set the maximum transmission power toPð1Þmax ¼ Pð2Þmax ¼ . . . ¼ Pð12Þ

max ¼ 50 mW and the control-packetsize to 120 bits. The data rate of a CR transmission over agiven channel is calculated according to Shannon’s for-mula.6 The reported results are averaged over 100 runs.Our performance metrics include: (1) the networkthroughput, (2) the CR blocking rate, (3) the average en-ergy consumption for successfully transmitting one datapacket (Ep), and (4) the fairness index. The CR blocking rateis defined as the percentage of CR requests that are blockeddue to the unavailability of a feasible channel. We useJain’s fairness index [30] to quantify the fairness of ascheme according to the throughput of all the CR users inthe network.

6.2. Single-hop network

We first study the throughput performance. Fig. 6a andb show that 2-radio AW-MAC provides only minorimprovement in the network throughput over the single

6 Other rate-vs-power relationships, such as a staircase function, can beused for calculating the achievable data rates.

transceiver AW-MAC (this result is inline with the analysisin Section 5). Because both 2-radio AW-MAC and AW-MACuse the same channel assignment algorithm and providecomparable throughput performance, in the following,we focus on the performance of AW-MAC and compare itwith the performance of the other protocols. Specifically,Fig. 6a and b show that under moderate and high trafficloads, AW-MAC significantly outperforms the other proto-cols. At steady state, AW-MAC reduces the CR blocking rateand improves the overall one-hop throughput by up to 50%compared to BMC-MAC, 18% compared to DDMAC, and 12%compared to WFC-MAC. This improvement is mostlyattributed to the increase in the number of simultaneousCR transmissions. WFC-MAC outperforms both BMC-MACand DDMAC. This is because WFC-MAC attempts to servea given CR transmission first using the worst feasible chan-nel and preserves better channels for potential futuretransmissions. Under light loads, all protocols achievecomparable throughput performance.

In Fig. 6c, we study the impact of the channel assign-ment strategy on Ep. It is clear that WFC-MAC and DDMACperform the worst in terms of energy consumption. At thesame time, the figure reveals that 2-radio AW-MAC, AW-

1138 H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139

MAC, and BMC-MAC have comparable performance withrespect to Ep. Thus, the throughput advantage of AW-MAC does not come at the expense of additional energyconsumption.

Fig. 7 shows that all schemes achieve comparable fair-ness. This can be attributed to the fact that in all of theseschemes CR users contend over the control channel usinga variant of the CSMA/CA mechanism.

Finally, Fig. 8 depicts the channel usage, defined as thefraction of time in which a specific channel is used for CRtransmissions. For WFC-MAC and DDMAC, channel usageis roughly evenly distributed among all channels, irrespec-tive of the traffic load. For AW-MAC and BMC-MAC, underlow and moderate traffic loads, channels with lower carrierfrequencies are favored for CR transmissions (lower atten-uation). On the other hand, under high traffic load, thereare no significant differences in channel usage among allchannels.

6.3. Multi-hop network

In order to study the performance in a multi-hop envi-ronment, we use the same simulation setup described inSection 6.1, but with the following changes:

� A 500 m � 500 m field is considered for the 200 mobileCR users.� The maximum transmission power is set to Pð1Þmax ¼

Pð2Þmax ¼ � � � ¼ Pð12Þmax ¼ 100 mW.

� Each CR user generates 4-KB data packets according to aPoisson process of rate k. For each generated packet, thedestination node is randomly selected to be any node inthe network. We use a min-hop routing policy, but weignore the routing overhead. For all schemes (BMC-MAC, WFC-MAC, and DDMAC), the next-hop candidatesare nodes that are within the transmission range of thetransmitter.

The purpose behind these changes in the setup is to giverise to hidden terminals. Our simulations take into accountthe effect of the hidden-terminal problem due to imperfectcontrol and inaccurate ACL at both the receiver and trans-mitter by considering the interference from active neigh-boring CR transmissions that use common channels (ifany).

As shown in Fig. 9a and b, WFC-MAC achieves lower CRblocking rate and higher end-to-end network throughputthan the other two protocols under moderate and hightraffic loads. On the other hand, under low traffic load, allprotocols achieve comparable throughput performance.Fig. 9c shows that BMC-MAC outperforms WFC-MAC andDDMAC in terms of Ep under different traffic loads. Similarfairness and channel usage properties to the single-hopscenarios are also observed here.

Note that no single strategy is always best in all trafficregimes. Under light traffic, BMC-MAC provides the samethroughput performance as WFC-MAC and DDMAC, butoutperforms them in terms of Ep. However, under moder-ate and high traffic loads, WFC-MAC performs better interms of throughput at the cost of Ep.

7. Conclusion

In this paper, we investigated the design of cooperativedynamic channel assignment for single-transceiver CR de-vices that employ adaptive power management. Our solu-tions attempt to maximize the network throughput as aprimary objective, followed by minimizing energy con-sumption as a secondary objective. We first presented cen-tralized and distributed channel assignment algorithms.For single-hop CRNs, we developed a CSMA-based MACprotocol with access window (AW) for exchanging controlmessages. Our AW-MAC realizes the optimal centralizedchannel assignment in a distributed manner. Based onour heuristic distributed assignment, we also developed adistributed, asynchronous MAC protocol (WFC-MAC) formulti-hop CRNs. We studied the performance of our proto-cols and contrasted them with two previously proposedMAC protocols (i.e., BMC-MAC and DDMAC). We showedthat for single-hop CRNs, AW-MAC performs the best interms of throughput and energy consumption under vari-ous traffic conditions. Under moderate-to-high trafficloads, AW-MAC achieves about 50% increase in throughputover BMC-MAC at no additional cost in energy. It achievesabout 18% throughput improvement over DDMAC, witheven less energy consumption and processing overhead.For multi-hop scenarios, our results show that WFC-MACis the best strategy in terms of throughput at the cost ofenergy consumption under different traffic loads. In addi-tion, under low traffic load, we found that BMC-MAC is agood scheme in terms of energy consumption, as bothBMC-MAC and WFC-MAC have the same throughput per-formance in such traffic regime.

References

[1] FCC, spectrum policy task force report, ET Docket No. 02-155,November 2002.

[2] H.B. Salameh, M. Krunz, Channel access protocols for multihopopportunistic networks: challenges and recent developments, IEEENetwork 23 (4) (2009) 14–19.

[3] H.B. Salameh, M. Krunz, O. Younis, Distance- and traffic-awarechannel assignment in cognitive radio networks, in: Proceedings ofthe IEEE SECON Conference, 2008, pp. 10–18.

[4] I. Akyildiz, W.-Y. Lee, M. Vuran, S. Mohanty, Next generationdynamic spectrum access cognitive radio wireless networks: asurvey, Computer Networks 50 (13) (2006) 2127–2159.

[5] Y. Xing, R. Chandramouli, S. Mangold, S. Shankar, Dynamic spectrumaccess in open spectrum wireless networks, IEEE Journal on SelectedAreas in Communications 24 (3) (2006) 626–637.

[6] Y. Xing, C. Mathur, M. Haleem, R. Chandramouli, K. Subbalakshmi,Dynamic spectrum access with QoS and interference temperatureconstraints, IEEE Transactions on Mobile Computing 6 (4) (2007)423–433.

[7] R. Menon, R. Buehrer, J. Reed, Outage probability based comparisonof underlay and overlay spectrum sharing techniques, in:Proceedings of the IEEE DySPAN Conference, 2005, pp. 101–109.

[8] S. Sankaranarayanan, P. Papadimitratos, A. Mishra, S. Hershey, Abandwidth sharing approach to improve licensed spectrumutilization, in: Proceedings of the IEEE DySPAN Conference, 2005,pp. 279–288.

[9] N. Nie, C. Comaniciu, Adaptive channel allocation spectrum etiquettefor cognitive radio networks, in: Proceedings of the IEEE DySPANConference, 2005, pp. 269–278.

[10] H.B. Salameh, Rate-maximization channel assignment scheme forcognitive radio networks, in: Proceedings of the IEEE GLOBECOMConference, 2010.

[11] Y. Yuan, P. Bahl, R. Chandra, T. Moscibroda, Y. Wu, Allocatingdynamic time-spectrum blocks in cognitive radio networks, in:

H.B. Salameh, M. Krunz / Ad Hoc Networks 9 (2011) 1127–1139 1139

Proceedings of the ACM International Symposium on Mobile and Ad-Hoc Networking and Computing (MobiHoc), 2007, pp. 130–139.

[12] H.B. Salameh, M. Krunz, O. Younis, Mac protocol for opportunisticcognitive radio networks with soft guarantees, IEEE Transactions onMobile Computing 8 (10) (2009) 1339–1352.

[13] N. Jain, S. Das, A. Nasipuri, A multichannel CSMA MAC protocol withreceiver-based channel selection for multihop wireless networks, in:Proceedings of the 9th International Conference on ComputerCommunications and Networks (IC3N), 2001, pp. 432–439.

[14] Second Report and Order and Memorandum Opinion and Order, ETDocket No. 04-186;FCC 08-260, 2008.

[15] J. Zhao, H. Zheng, G.-H. Yang, Distributed coordination in dynamicspectrum allocation networks, in: Proceedings of the IEEE DySPANConference, 2005, pp. 259–268.

[16] R. Sedgewick, Algorithms in C, Part 5: Graph Algorithms, third ed.,Addison-Welsy, 2002.

[17] A. Behzad, I. Rubin, Multiple access protocol for power-controlledwireless access nets, IEEE Transactions on Mobile Computing 3 (4)(2004) 307–316.

[18] V. Dendeit, H. Emmons, Max-Min matching problems with multipleassignments, Journal of Optimization Theory and Application 91 (2)(1996) 491–511.

[19] A. Muqattash, M. Krunz, POWMAC: a single-channel power controlprotocol for throughput enhancement in wireless ad hoc networks,IEEE Journal on Selected Areas in Communications 23 (5) (2005)1067–1084.

[20] Y. Yuan, P. Bahl, R. Chandra, P. Chou, J. Ferrell, T. Moscibroda, S.Narlanka, Y. Wu, Knows: Kognitive networking over white spaces,in: Proceedings of the IEEE DySPAN Conference, 2007, pp. 416–427.

[21] H.B. Salameh, M. Krunz, O. Younis, Dynamic spectrum accessprotocol without power mask constraints, in: INFOCOM, 2009, pp.2322–2330.

[22] A. Acharya, A. Misra, S. Bansal, MACA-P: a MAC for concurrenttransmissions in multi-hop wireless networks, in: Proceedings of theFirst IEEE PerCom 2003 Conference, 2003, pp. 505–508.

[23] B. Crow, I. Widjaja, J. Kim, P. Sakai, IEEE 802.11 wireless local areanetworks, IEEE Communications Magazine 42 (3) (1997) 116–126.

[24] G. Bianchi, Performance analysis of the IEEE 802.11 distributedcoordination function, IEEE Journal on Selected Areas inCommunications 18 (3) (2000) 535–547.

[25] F. Wang, M. Krunz, S. Cui, Price-based spectrum management incognitive radio networks, IEEE Journal of Selected Topics in SignalProcessing 2 (1) (2008) 74–87.

[26] Qualcomm Announces Sampling of the Industry’s First Single-ChipReceive Diversity Device for Increased CDMA2000 NetworkCapacity. <http://www.qualcomm.com/press/releases/2005/050504-rfr6500.html>.

[27] Kenwood TH-D7A dual-band handheld transceiver. <http://www.kenwoodusa.com/Communications/Amateur-Radio/Portables/TH-D7A(G)>.

[28] The Cisco Aironet 350 Series of wireless LAN. <http://www.cisco.com/warp/public/cc/pd/witc/ao350ap>.

[29] Mesquite Software Incorporation. <www.mesquite.com>.[30] R. Jain, The Art of Computer System Performance Analysis, John

Wiley & Sons, New York, 1991.

Haythem A. Bany Salameh received the Ph.D.degree in electrical and computer engineeringfrom the University of Arizona, Tucson, in2009. He is currently an Assistant Professor ofelectrical and computer engineering with theHijjawi Faculty for Engineering Technology,Yarmouk University (YU), Irbid, Jordan. Hejoined YU in August 2009, after a brief post-doctoral position with the University of Ari-zona. His current research interests are insystem architecture and communication pro-tocol designs for wireless networks with

emphasis on dynamic spectrum access, radio resource management, androuting/MAC protocol design. His research covers a wide variety of

wireless systems, including cognitive radio networks, wireless sensornetworks, mobile ad hoc networks, and cellular networks. In summer2008, he was a member of the R&D LTE (Long Term Evolution) Develop-ment Group, QUALCOMM, Inc., San Diego, USA. He serves as a reviewerfor many IEEE conferences and journals.

Marwan Krunz is a professor of electrical andcomputer engineering at the University ofArizona. He directs the wireless and net-working group and is also the UA site directorfor Connection One, a joint NSF/state/industryIUCRC cooperative center that focuses on RFand wireless communication systems andnetworks. Dr. Krunz received his Ph.D. degreein electrical engineering from Michigan StateUniversity in 1995. He joined the University ofArizona in January 1997, after a brief post-doctoral stint at the University of Maryland,

College Park. He previously held visiting research positions at INRIA, HPLabs, University of Paris VI, and US West (now Qwest) Advanced Tech-nologies. His research interests lie in the fields of computer networking

and wireless communications. His current research is focused on cogni-tive radios and SDRs; distributed radio resource management in wirelessnetworks; channel access and protocol design; MIMO and smart-antennasystems; UWB-based personal area networks; energy management andclustering in sensor networks; media streaming; QoS routing; and faultmonitoring/detection in optical networks. He has published more than150 journal articles and refereed conference papers, and is a co-inventoron three US patents. M. Krunz is a recipient of the National ScienceFoundation CAREER Award (1998). He currently serves on the editorialboards for the IEEE Transactions on Mobile Computing and the ComputerCommunications Journal. He previously served on the editorial board forthe IEEE/ACM Transactions on Networking (2001–2008). He was a guestco-editor for special issues in IEEE Micro and IEEE Communicationsmagazines. He served as a technical program chair for various interna-tional conferences, including the IEEE WoWMoM 2006, the IEEE SECON2005, the IEEE INFOCOM 2004, and the 9th Hot Interconnects Symposium(2001). He has served and continues to serve on the executive andtechnical program committees of many international conferences and onthe panels of several NSF directorates. He gave keynotes and tutorials, andparticipated in various panels at premier wireless networking confer-ences. He is a consultant for a number of companies in the telecommu-nications sector.