A Decentralized MAC for OpportunisticSpectrum Access in Cognitive Wireless Networks

Junaid Ansari, Xi Zhang and Petri MähönenInstitute for Networked Systems, RWTH Aachen University,

Kackertstrasse 9, D-52072, Aachen, Germany{jan, xzh, pma}@inets.rwth-aachen.de

ABSTRACTCognitive MAC protocols are designed to efficiently uti-lize the spectral resources without affecting the performancecharacteristics of the primary users. The use of spectrum op-portunities, so called white spaces, can often require stochas-tic approaches due to difficulty in predicting their appear-ance. Infrastructure based coordinated access techniquesare not a viable option for all the applications and spectrumbands. In this paper, we describe a decentralized cogni-tive MAC protocol based on multi-channel preamble reserva-tion scheme. The protocol dynamically selects an availablecommunication channel using a distributed channel selectionscheme and allows nodes to be completely asynchronous toeach other. Our MAC does not need a common control chan-nel and cooperative infrastructure. We also address otherpractical issues such as mobility and traffic awareness in ourMAC design. We have carried out performance evaluationof our protocol on an SDR testbed consisting of WirelessOpen Access Research Platform (WARP). Experimental re-sults show that our protocol is able to achieve reliable datacommunication even in harshly interfering environments.

Categories and Subject DescriptorsC.2.1 [Network Architecture and Design]: Wireless com-munication

General TermsAlgorithms, Design, Experimentation, Performance

KeywordsMAC, Cognitive Radios, Implementation, Decentralized

1. INTRODUCTIONSpectrum sharing and symbiotic coexistence are becom-

ing inevitable due to the rapidly growing popularity of wire-less networks and technologies. Cognitive Radio (CR) tech-

Permission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.CoRoNet’10, September 20, 2010, Chicago, Illinois, USA.Copyright 2010 ACM 978-1-4503-0141-1/10/09 ...$10.00.

nologies aim at enabling higher spectrum utilization by op-portunistically using the spectral white spaces through Dy-namic Spectrum Access (DSA) schemes and cross-layer de-sign methodologies. Media access techniques in CR net-works allow nodes to efficiently communicate with each otherwithout affecting the performance characteristics of the Pri-mary User (PU).

In CR networks, channel sensing and spectrum accessare tightly coupled to the medium access—no matter if itis through random access, time slotted principle or hybridscheme. In recent years, a number of spectrum agile andcognitive MAC protocols have been proposed [4][12]. Thesesolutions have different requirements for network infrastruc-ture and hardware capabilities of Secondary Users (SUs).Most of the MAC designs rely on high degree of coopera-tion among the SUs and the centralized entity coordinatingthe DSA. If the spectrum dynamics are high, much highergranularity and accurate sensing is required for MACs [16].In situations, where accurate spectrum occupancies are hardto predict and cooperation from the PU is less likely, decen-tralized MAC protocols are desirable. We have designed andimplemented a solution, which neither makes any assump-tion on the infrastructure and the deployment environmentnor requires explicit synchronization among the nodes.

The main goal of our MAC solution is to provide a sim-ple learning based distributed MAC protocol that is able tochoose a new channel in a weighted manner if they have tovacate the existing channel for PU. The protocol is particu-larly useful for local area ad hoc type of applications and lowpower embedded networks requiring DSA capabilities. Ourdecentralized spectrum agile MAC protocol requires only ahalf-duplex radio interface without needing a common con-trol channel. Our MAC protocol is suitable for both licensedand ISM bands, where the protocol adapts its channel selec-tion to the stochastic spectrum occupancy behaviour of thePUs and mitigates the effects of the random interferences,respectively. Studies on the impact of frequency-agility ondynamic spectrum sharing show that radios which are ca-pable of using non-contiguous frequencies gives better per-formance over radios using single frequency for transmis-sion [13]. Our protocol uses a distributed channel selectionscheme, which adaptively expands and contracts the numberof frequency channels to be used depending upon the inter-ference conditions. The protocol is implemented on WARP[11] boards and is evaluated in terms of throughput andpacket delivery ratio with respect to different carrier sens-ing durations and number of used channels. A vector signalgenerator is used to model the behaviour of a PU. We believe

that our implementation and evaluation provides significantinsights into the practical aspects of cognitive radio MACs.This has a particular value as majority of the state-of-the-artdesigns remain at a theoretical and simulation stage.

2. RELATED WORKThere are both centralized and distributed MAC designs

for CRs [4]. MAC protocols in infrastructure based networksrequire a central controller, e.g., for management of the net-work activities, gathering and distribution of data, synchro-nization among nodes. IEEE 802.22[9] standard for WirelessRegional Area Network is a good example in this category.As compared to the distributed protocols, centralized MACapproaches typically demand simpler hardware and softwarecapabilities for SUs (CR nodes). MAC protocols relying ona centralized infrastructure, at the same time, impose muchstringent requirements in terms of sensing and coordinationon the infrastructure. These include various mechanismsfor spatio-temporal spectrum sensing and active coordina-tion among the nodes. Numerous centralized protocols, suchas [2][10] have been designed. The protocol in [2] uses theCSMA principle for underlay systems [6], where a SU ac-tivity is permitted with simultaneous PU activity as longas the interference caused to the PU stays within a radiotechnology dependent threshold. The protocol enables co-existence by adjusting the transmit power and data rates ofthe CR nodes. The intelligence for adapting the protocolparameters lies at the controlling infrastructure.

The other category of MAC protocols for CR networksis designed for decentralized operation. The cognitive usersgather spectrum information either locally or through coop-eration from the neighboring nodes. These protocols havedifferent assumptions on the hardware platform and the net-work capabilities. Some of these schemes assume that cogni-tive users are equipped with multiple transceivers and nodeshave the capability to access multiple channels simultane-ously and select the best channel as described in [1], [7]and [8]. Some of the decentralized MACs assume that allthe entities in the network have an always available com-mon control channel for exchanging control information andestablishing an agreement on the selection of data transmis-sion channel between the transmitter and receiver pairs. Adetailed taxonomy of these protocols is given in [5]. Theuse of a common control channel has its limitations. Anin-band common control channel is exposed to PU activitiesand thus its reliability and availability are not guaranteedwhile an out-of-band common control channel requires ei-ther an extra radio interface or active switching betweenthe frequency bands. Furthermore, a common control chan-nel needs to be allocated by a regulator and agreed throughstandardization.

Simulation based studies have been conducted to showthe theoretical effectiveness of the above mentioned pro-tocols. However, these lack performance measurements inreal environments and network conditions. To the best ofour knowledge, C-MAC [3] is the first cognitive MAC whichcomes with a full prototype implementation. Instead of fix-ing a common control channel, C-MAC uses a rendezvouschannel, which is selected based on the reliability of theavailable channels. It uses the TDMA principle, where eachCR node has a distinct slot for periodic beaconing dura-tion. This work is definitely a way forward and shows thefeasibility of implementing a cognitive MAC. However, the

Figure 1: Multichannel sensing and packet transmis-sion. As the two (unsynchronized) receivers detectthe transmission activity, they stay in the channeluntil a complete packet is received.

evaluation of the prototype is not provided. The implemen-tations, SoftMAC [18], MultiMAC [17], etc., mainly targetthe modularity and flexibility aspects instead of focusing onthe primary performance metrics such as throughput andlatency or specific details of supporting DSA operations.

3. PROTOCOL DESIGNOur MAC design is targeted for infrastructureless envi-

ronments. The MAC protocol uses a distributed channelselection strategy, which allows it to handle network mobil-ity in an effective manner. Our MAC design tries to solvetwo issues at once. First, how to sense possible channelswithout introducing undue latency and poll busy channelswith active PUs. Second, avoiding overloading the samechannel with SUs. The protocol uses multichannel carriersensing principle where a node scans all the channels in thepool sequentially. The transmitting node ensures that thetransmission in the selected channel lasts for long enough du-ration that the asynchronous receiving nodes detect it whenscanning that particular channel. Upon detecting a packettransmission activity, the receiving nodes do not scan sub-sequent channels and keep on listening to the channel un-til a data packet is received. In order to engage the chan-nel, the transmitter repeats the data packet back to backas shown in Figure 1. The total number of packet repe-titions, Npkt required to be sent governs an upper bound,Npkt ≥ (TCS+Tswitch)Nch/Tpkt, where TCS is the carrier sens-ing duration, Tswitch is the channel switching duration, Nch

is the number of channels and Tpkt is the time required tosend a packet.

A transmitter first scans all the channels in the pool toensure that there is no other on going packet transmissionbefore attempting to send a packet. This also justifies the re-ceiver(s) for not sensing subsequent channels upon detectinga packet activity. Scanning the channels prior to transmis-sion helps in avoiding cases of multiple simultaneous trans-missions to the same receiver(s) though there exists possi-bilities to utilize an additional available bandwidth. Sincean activity in the medium can also be because of an inter-ferer or a PU, the protocol uses a timeout scheme in orderto characterize an interferer. Please note that in this paper,in terms of terminology, we do not distinguish between aPU and an interferer1. If channel activity is detected andno valid packet is received within an interval of two maxi-mum sized packet transmissions, the channel is identified as

1In this work, we do not take signal classification and feature detec-

tion into account. However, without losing generality, such a capabil-ity can be trivially included into our MAC.

interfering. A timeout duration of two packet transmissioninterval allows receiving a packet transmitted back to backwith any packet boundary offset.

The protocol uses a heuristics based method for channelselection similar to [19]. In each sensing cycle, all the nodesscan the available channels sequentially for potential spec-trum activity. Weights are associated with the channels,which are updated in each sensing cycle based on the typeof the activity. If a particular channel is found free and datacommunication is established, the weight associated with thechannel is increased. On the contrary, if a channel is foundinterfering, its weight is decreased. Channel weight history isalso maintained, which helps identifying interfering channelsand blacklist them. Before a packet transmission, the sendercarries out dynamic channel selection in order to opportunis-tically utilize an available spectrum hole. Nodes are ableto communicate with each other even without having anyprior knowledge about the sensed spectrum characteristics ofother nodes in the network. However, nodes do exchange thesensed channel characteristics (compressed in a binary for-mat, called as“Channel maps”) inside the preamble, which istransmitted before data. A channel map is essentially a bitencoded information indicating whether a particular chan-nel has a weight higher than a predefined threshold, T2 (c.f.Algorithm 3.1). Channel maps of neighbors allow a node toavoid spatially local interferers by benefiting from their radioenvironment. Scanning a larger number of channels adds la-tency to the communication and reduces the throughput asconfirmed by our experiments in Section 5. Therefore, ourchannel selection scheme tries to maintain a smaller numberof channels with higher weights in the pool and delete theblacklisted interfering channels. Also if the quality of chan-nels (expressed through the median channel weight) deteri-orates, the channel pool is replenished by replacing the low-weighted channels from the pool. New channels with theirweights initialized to the threshold T2, are included in thepool. Keeping a channel history prevents adding previouslydeleted channels into the pool. Precedence is given to thechannels which have not been blacklisted before and to chan-nels with the oldest blacklisting timestamp. The dynamicexpansion and contraction of channel pool allows keepingdiversity in the channel pool and simultaneously lowers thelatency in data communication (c.f. Figure 10). Owing tothe distributed nature of the channel selection algorithm,there is a danger that neighboring nodes may converge tonon-overlapping channel pools. However, our experimentalresults indicate that this is unlikely to happen in practice dueto the exchange of channel maps. Algorithm 3.1 describesthe distributed channel weighting scheme. The values of thehysteresis thresholds, T1 and T2 are empirically chosen tobe 15 and 40, when the ceiling channel weight value is setto be 100. Since the channels are sorted in the descendingorder of their weights, this scheme inherently allows to scanand use the least interfering channels first.

A transmitting node sends a repeated sequence of pack-ets in the selected channel. The header is encoded withthe base rate modulation and contains control informationsuch as the destination address, source address, the mod-ulation scheme for data payload and the channel map. Anon-addressed node is also able to gather the spatial spectralcharacteristics of the transmitting node and other relevantmeta-data by overhearing the header. In order to efficientlysupport higher data traffic loads at a node, we transmit the

Figure 2: Multi-packet transmission scheme to sup-port higher traffic volumes at a node.

queued data packets back to back without repetition aftera repetitive transmission of only the first packet. The firstpacket repetition serves to implicitly reserve the medium forfurther packet transmissions by a node. Since our protocol isdesigned for platforms with a single radio transceiver, whilethe transmission is underway at a node, it cannot detect ifthe channel is re-engaged by a PU. In order to prevent caus-ing interference to the PU in the case when higher data vol-umes are to be supported, the transmitter quickly sniffs thechannel between consecutive packet transmissions to ensurethat the channel is still available. This scheme is illustratedin Figure 2.

Algorithm 3.1: Channel Selection Algorithm()

ch.wt← T2for i← init to max channel

do

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

CarrierSensing(ch[i])if ch[i] = Freethen ch[i].wt← ch[i].wt + 1

if Interferencethen ch[i].wt← ch[i].wt− 7

if Tx or Rxthen ch[i].wt← ch[i].wt + 5

if Rx

then

⎧⎨⎩

for j ← init to max channeldo if channel map = 1then ch[j].wt← ch[j].wt + 3

if ch[i].wt < T1 and ch contraction enabledthen Delete(ch[i])

cond← Truefor each chdo if ch.wt < T2then cond← False

if cond = True and ChannelWtMedian(ch) < 1.5T1

then

{Refresh(ch)TimeStampChannelActivity(ch)

Sort(ch)comment: Sort all channels in descending order of weights.

4. IMPLEMENTATION DETAILSWe have implemented the MAC protocol on Rice Univer-

sity’s WARP boards [11]. Our MAC implementation usesWARP OFDM Reference Design 14 and is based on thecomponent based MAC development framework [14]. Ta-ble 1 lists the parameter values used for our prototype im-plementation. We expose different MAC parameters to theapplication, like the initial channel pool, contracted channelpool size, the weighting metric thresholds, CCA durations,

Table 1: PHY/MAC parameter values.Parameter ValueMax. packet size used (Lmax pkt) 1000 bytesMax. packet transmission time (Tmax pkt) 1.48 msChannel switching interval (Tchannel switch) 35 usInterferer timeout interval (Tint timeout) 2.96 ms

Figure 3: Snapshot of the WARP testbed.

backoff window sizes, persistency values, etc. which can betuned to the application requirements. These parametershave their implications on the MAC performance character-istics as we showed in a demonstration [15].

5. PERFORMANCE EVALUATIONWe have carried out the evaluation of the MAC proto-

col on an 6-node WARP testbed as shown in Figure 3.The WARP boards were connected to Gigabit Ethernet andRS232 interfaces to a PC in order to control the parame-ter settings and gather results during our experiments. Weconfigured Agilent’s E4438C Vector Signal Generator to gen-erate different types of interference patterns in order to em-ulate the primary user(s). We monitored the spectrum oc-cupancy characteristics using Agilent’s E4440A SpectrumAnalyzer and WiSpy DBx spectrum scanner. In order tomeasure and set precise timings of different radio parame-ters, we used Agilent’s Infiniium DSO8104A Oscilloscope.

5.1 Sensing Reliability on WARP BoardsIn order to empirically measure the reliable sensitivity of

the hardware without channel impairments, we fed a sig-nal directly from the vector signal generator to the antennainput of a WARP board over a coaxial cable. We alsocalibrated the cable attenuation losses in this experiment.We varied the strength of the signal power and the WARP

−96 −94 −92 −90 −88 −86 −84 −820

10

20

30

40

50

60

70

80

90

100

Energy Detection Threshold [dBm]

Fals

e D

etec

tions

(%)

−80dBm−85dBm−90dBm−95dBm−100dBm

Figure 4: False negative detections on a WARPboard for 10000 samples.

board’s energy detection threshold level. Figure 4 shows theaverage false detections percentages for 10000 samples. Theresult show that the WARP board shows a stable behaviourover a range of detection threshold values and a natural in-crease in the percentage of false detections as the thresholdgoes up. This result indicates the stable operating point forour measurements.

5.2 Interference AvoidanceWe configured a signal generator to transmit random in-

terfering signal in different channels for a certain durationusing different transmit power levels. WARP boards run-ning our MAC protocol with 4 channels in the pool wereplaced in the interference range. As can be seen from thespectrogram in Figure 5 that our MAC protocol is alwaysable to detect the interferer and quickly hops away to anavailable free channel for data communication.

Cognitive MAC Primary User

Figure 5: Spectrogram showing that the MAC pro-tocol is able to dynamically select interference freechannels when subjected to random interferencepatterns emulating a primary user.

5.3 Effects of Channel Pool Size and CarrierSensing Duration

In order to study the effects of the channel pool size on thegoodput and packet success ratios, we use a transmitter andreceiver pair without the presence on any interferer. Thetransmitting node tries to send packets with payload size of1000 bytes as fast as possible without using the smart chan-nel reservation scheme. Figure 6 shows the achieved nor-malized goodput with a normalizing factor of 2.393Mbps.The results show a tradeoff between transmission and sens-ing. It is evident from the figure that the goodput goesdown exponentially as the number of channels increases inthe pool and as the CCA duration increases. This is becauseof the larger sensing duration; spending longer time in scan-ning more number of channels and the longer per channelscanning interval. The results indicate that if a node canmaintain a smaller pool of channels, the total capacity is ofcourse larger. Figure 7 shows that the corresponding packetdelivery ratio does not suffer any significant change with theincrease in the number of channels in the pool and with theCCA duration. These two graphs also serve as the bench-mark performance characteristics while studying the good-put and packet delivery ratio in the presence of differentinterference patterns.

5.4 Goodput and Packet Success RatiosIn order to study the empirical goodput characteristics

with respect to the number of channels in the pool and the

0 5 10 15 20 2510

−3

10−2

10−1

100

Carrier Sensing Duration [ms]

Nor

mal

ized

Goo

dput

No. of Channels = 1No. of Channels = 2No. of Channels = 4No. of Channels = 8

Figure 6: Normalized goodput behaviour with re-spect to the number of channels in the pool and thechannel sensing duration.

0 5 10 15 20 250.75

0.8

0.85

0.9

0.95

1

Carrier Sensing Duration [ms]

Pack

et D

eliv

ery

Ratio

No. of Channels = 1No. of Channels = 2No. of Channels = 4No. of Channels = 8

Figure 7: Effects of the channel pool size and thechannel sensing duration on the successful packetdelivery ratio.

channel assessment duration, we used a WARP transmitter-receiver pair placed in the vicinity of an interferer. The sig-nal generator is configured to generate a frequency sweepingsignal with 100ms in the channels used by the WARP board.Figure 8 shows the normalized goodput achieved when thetransmitter tries to send as many packets as it can with apayload size of 1000 bytes. The normalization factor usedis 2.393Mbps. It can be seen that the goodput has gonedown as compared to the interference free case as shown inFigure 6. The slight degradation in the goodput is becauseof the timeouts for interference characterization. Comparedto the case with no interference in the medium, the goodputvalues for the channel pool size of one suffers the most. Thechannel pool size of two performed better than the rest,which shows that channel diversity is good but keeping ahigh channel diversity is not necessary in all the cases sinceit adds to the control overhead. This result corresponds tothe earlier theoretical findings in [13].

Figure 9 shows the packet delivery ratios of a randomchannel selection scheme and our cognitive MAC protocolin the presence of an interferer that cyclically sweeps fre-

0 5 10 15 20 2510

−3

10−2

10−1

100

Carrier Sensing Duration [ms]

Nor

mal

ized

Goo

dput

No. of Channels = 1No. of Channels = 2No. of Channels = 4No. of Channels = 8

Figure 8: Effects of the channel pool size and thechannel sensing duration on the goodput in the pres-ence of a sweeping frequency interferer.

0 5 10 15 20 250.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Carrier Sensing Duration [ms]

Pack

et D

eliv

ery

Ratio

No. of Channels = 2; Interference FreeNo. of Channels = 2; CogMAC; Cyclic InterfererNo. of Channels = 2; Random MAC; Cyclic InterfererNo. of Channels = 4; Interference FreeNo. of Channels = 4; CogMAC; Cyclic InterfererNo. of Channels = 4; Random MAC; Cyclic Interferer

Figure 9: Packet delivery ratio comparison of arandom channel selection scheme and our cognitiveMAC in the presence of a sweeping interferer.

quency in the channels used by the WARP boards with achannel dwell time of 100ms. It can be observed from thefigure that our protocol is able to achieve remarkably highpacket delivery ratio as compared to the random channelselection based scheme in the presence of the interferer. Atlarger CCA durations, the random channel selection schemeimproves the delivery ratio since longer channel assessmentdurations makes sensing more reliable.

5.5 Multihop LatencyThe multihop latency of the MAC protocol was measured

using a linear topology with no interferer in the networkand a CCA duration of 0.961ms. Figure 10 shows that themultihop latencies show a linearly increasing behaviour withthe number of hops and roughly a linear increasing trendwith the number of channels in the pool.

5.6 Coexistence of Multiple CogNetsIn order to study the coexistence behaviour of two cog-

nitive networks using our MAC protocol in the same inter-fering environment, we used a transmitter-receiver pair ofWARP boards with BPSK encoded preamble and data mod-

2 3 4 5 6 7 8 9 100

20

40

60

80

100

120

140

160

180

Number of hops

End−

to−

end

Late

ncy

[ms]

No. of Channels = 1No. of Channels = 2No. of Channels = 4No. of Channels = 8

Figure 10: Multihop latency.

ulation while the other pair with QPSK encoded preambleand data modulation. In this way, the two pairs interferedeach other without able to communicate. We forced the fournodes to use the same set of channels with a pool size of two.We placed the four nodes at a square grid and interfererin the center, jumping randomly in the two channels. Weobserved that the two pairs of WARP boards dynamicallyhopped onto the available free channel. Figure 11 shows thepacket delivery ratios of the two networks. It can be ob-served that our MAC allows symbiotic coexistence of bothnetworks and is able to provide reliable communication.

0 5 10 15 20 250.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

Carrier Sensing Duration [ms]

Pack

et D

eliv

ery

Ratio

Modulation: BPSK; Interference freeModulation: BPSK; Cyclic InterfererModulation: QPSK; Interference freeModulation: QPSK; Cyclic Interferer

Figure 11: Coexistence of different CogNets.

6. CONCLUSIONS AND FUTURE WORKWhite spaces in the spectrum can be hard to predict

and tight cooperation among different networks sharing thecrowded spectrum is often infeasible. In order to addressthese issues, we have designed and implemented a decentral-ized cognitive MAC protocol, which allows nodes to commu-nicate reliably even in highly interfering environments. Inthis paper, we have described the design rationale and imple-mentation details of our MAC protocol on WARP boards.Our protocol dynamically selects an interference minimalchannel using a distributed channel selection strategy. Anavailable wireless channel can further be utilized in an op-portunistic fashion with only a little overhead for potentiallysubsequent transmissions by keeping the nodes listening for

a short duration to the available channel. Performance eval-uation experiments conducted on a WARP testbed showthat our protocol is able to deliver packets with high relia-bility and throughput in interfering environments. On thecontrary, a random channel selection scheme shows very lowpacket delivery ratios. We have also conducted empiricalstudies on multihop latencies, effects of channel pool sizesand carrier sensing durations on the MAC performance char-acteristics. Our current work in progress includes the use ofmore advanced machine learning techniques to improve thechannel selection algorithm. We are also planning to have apossible public release of the code for interested parties.

AcknowledgmentWe thank partial financial support from Deutsche Forschungs-gemeinschaft (DFG) through UMIC Research Centre andEU through ARAGORN, FARAMIR and 2PARMA projects.

7. REFERENCES[1] A. Nasipuri, J. Zhuang and S. Das A multichannel CSMA MAC

protocol for multihop wireless networks. In Proc. of WCNC,New Orleans, USA, 1999.

[2] A.-Y. Lien, C.-C. Tseng and K.-C. Chen. Carrier sensing basedmultiple access protocols for cognitive radio networks. In Proc.of ICC, Beijing, China, 2008.

[3] C. Cordeiro and K. Challapali. C-MAC: A cognitive MACprotocol for multi-channel wireless networks. In Proc. ofDySPAN, pages 147–157, 2007.

[4] C. Cormio and K. Chowdhury. A survey on MAC protocols forcognitive radio networks. Ad Hoc Networks, 7(7):15–29, 2009.

[5] I.F. Akyildiz, W.-Y. Lee, and K.R. Chowdhury. CRAHNs:Cognitive radio ad hoc networks. In Ad Hoc Networks 7, pages810–836, 2009.

[6] J. Xiang, Y. Zhang and T. Skeie. Medium access controlprotocols in cognitive radio networks. In Wireless Commun.Mob. Comput., 10(1):31-49, 2010.

[7] N. Jain and S. Das. A multichannel CSMA MAC protocol withreceiver-based channel selection for multihop wireless networks.In Proc. of ICCCN, Scottsdale, USA, 2001.

[8] S.-L. Wu et al.. A new multi-channel MAC protocol withon-demand channel assignment for multi-hop mobile ad hocnetworks. In Proc. of I-SPAN, Texas, USA, 2000.

[9] IEEE 802.22 working group on wireless regional area networks.http://www.ieee802.org/22 [Visted on 20.05.10]

[10] C. Zhou and C. Chigan. A game theoretic DSA-driven MACframework for cognitive radio networks. In Proc. of ICC,Beijing, China, 2008.

[11] WARP: Wireless Open Access Research Platformhttp://warp.rice.edu/trac. [Visited on 01.06.10]

[12] I.F. Akyildiz et al. A survey on spectrum management incognitive radio networks. InComm. Mag., IEEE, April 2008,Vol. 46, No. 4 pp. 40–48.

[13] L.L. Cao, L. Yang and H. Zheng The Impact ofFrequency-Agility on Dynamic Spectrum Sharing InProc. ofDySPAN, 2010, Singapore.

[14] J. Ansari et al., Decomposable MAC Framework for HighlyFlexible and Adaptable MAC Realizations.Demonstrationabstract. InProc. of DySPAN, 2010, Singapore.

[15] J. Ansari, X. Zhang and P. Mahonen Demo Abstract: ADecentralized MAC Protocol for Opportunistic SpectrumAccess in Cognitive Wireless Networks. Demonstrated at IEEEINFOCOM, 2010, San Diego, CA, USA.

[16] M. Wellens, J. Riihijarvi and P. Mahonen Evaluation ofAdaptive MAC-Layer Sensing in Realistic Spectrum OccupancyScenarios. InProc. of DySPAN, 2010, Singapore.

[17] C. Doerr et al. MultiMAC - an adaptive MAC framework fordynamic radio networking. InProc. of DySPAN 2005, Baltimore

[18] G. Nychis et al. Enabling MAC protocol implementations onsoftware-defined radios. InProc. of NSDI, 2009.

[19] J. Ansari, T. Ang and P. Mahonen Spectrum Agile MediumAccess Control Protocol for Wireless Sensor Networks. InProc.of IEEE SECON, 2010.