Quality of service assessment of opportunistic spectrum access: a medium access control approach

INTRODUCTION

Static spectrum assignment, applied to radio fre-quencies for almost a century, results in a quasi-scarcity of spectrum. Finding a new unassignedfrequency slot pushes system designers to explorehigher and higher frequencies (e.g., 60 GHz).However, most of the already allocated frequen-cies are not used or are used sporadically. There-fore, it is logical to allow nonlicensed users to usethese frequencies when they are free at a specificplace and time. Theoretically, such an approachwill increase overall frequency reuse without anylicensing costs and boost the throughput forapplications that opportunistically use the emptyfrequencies. This communication technique iscalled opportunistic spectrum access (OSA).

The concept of OSA, although very promis-ing and attracting lots of attention, introducesnew challenges to the notion of dependablequality of service (QoS)-driven communication.It raises questions about the amount of interfer-ence a licensed user is willing to tolerate. Manyresearchers follow a strict definition and assumea frequency band should only be used whenthere is no licensed user present. As a result,most research effort focuses on detecting thepresence of a licensed user. In the example mea-surement we describe in the next section, 15.7percent of measured frequency bins were free oflicensed use over our whole measurement time,which means that OSA users could use this muchspectrum. When incorporating the licenseduser’s duty cycle in the computation, we obtaineda occupancy of only 6.8 percent, which meansthat theoretically an opportunistic user could use93.2 percent of the spectrum. This, however,means that the opportunistic user shares thespectrum in the time domain with the licenseduser. Clearly, in this case it is not possible toavoid all collisions between licensed and oppor-tunistic transmissions, leading to a potential per-formance degradation for the licensed user.

In this article we look at the trade-offbetween QoS for the licensed user vs. that forthe opportunistic user. There are many exampleswhere it makes sense to decrease licensed userQoS when this reduction results in a much largerQoS improvement for the opportunistic user.Consider, for instance, a licensed user interestedin broadcasting traffic information periodically.First, such broadcast is not that demanding interms of data throughput, so an opportunisticuser can take advantage of the idle periodsbetween broadcasts. Second, for a car interestedin the traffic information, missing a single broad-cast is not that harmful, and a considerableamount of collisions could hence be tolerated.With OSA, both licensed and opportunistic usersenjoy sufficient QoS, while the licensed user hasreduced spectrum cost.

The goal of this article is to quantify the QoSfor opportunistic users as a function of the

IEEE Wireless Communications • October 200820 1536-1284/08/$25.00 © 2008 IEEE

PU

(a)

PU Datatransfer 1

CTS1

RTS1PU

RTS3 PU

RTS CTS3PU Data

transfer 2

Data

Opportunistic Spectrum Access is apromising new spectrum managementapproach that willallow co-existence ofboth licensed andopportunistic users ineach spectrum band,potentially decreasingthe spectrum licensing costs for both classes of users.

DE P E N D A B I L I T Y I N T H E UB I Q U I T O U S WIRELESS ACCESS

PRZEMYSLAW PAWELCZAK, SOFIE POLLIN, HOI-SHEUNG WILSON SO, AHMAD BAHAI, R. VENKATESHA PRASAD, AND RAMIN HEKMAT

ABSTRACTOpportunistic spectrum access (OSA) is a

promising new spectrum management approachthat will allow coexistence of both licensed andopportunistic users in each spectrum band,potentially decreasing the spectrum licensingcosts for both classes of users. However, this hassignificant implications on the QoS experiencedby the licensed and opportunistic spectrumusers. In this article we investigate how tolerantto secondary user activity a licensed user shouldbe so as to provide dependable communicationwith sufficient QoS to an opportunistic user. Wealso look at key multichannel MAC features forsuch OSA networks proposed in the literature,and discuss how the design of control channelmanagement affects the QoS of opportunisticusers as a function of the tolerance of licensedusers. We quantify the trade-off betweendependability of the OSA network and thedependability of licensed users. The main con-clusion is that opportunistic users can indeedachieve good QoS, as long as the licensed usersare not highly active. For example, in one of thescenarios we studied, opportunistic users canachieve a delay below 100 ms if licensed useractivity stays below 30 percent.

QUALITY OF SERVICE ASSESSMENT OFOPPORTUNISTIC SPECTRUM ACCESS:

A MEDIUM ACCESS CONTROL APPROACH

PAWELCZAK LAYOUT 10/2/08 2:30 PM Page 20

Authorized licensed use limited to: National Chung Cheng University. Downloaded on January 12, 2009 at 04:25 from IEEE Xplore. Restrictions apply.

IEEE Wireless Communications • October 2008 21

licensed user activity level and collision toler-ance. The opportunistic user’s QoS is measuredin terms of delay and throughput. To quantifythis QoS, we need to make assumptions on theOSA medium access control (MAC) protocoldesign. Many different approaches are found inthe literature, varying in terms of the number ofradio front-ends used, the number of channels,the way the control information is exchanged,and the detection method for licensed users.Although we lack the space to give an exhaustiveoverview of all design options, we list the mostimportant ones in the literature. We model someof those options in more detail to relate to theQoS parameters. We compare differentapproaches to distributing the opportunisticuser’s transmissions among many different chan-nels using a multichannel MAC protocol.

The remainder of the article is organized asfollows. First, we introduce the concept ofdynamic spectrum sharing more systematically.Through measurements, we show that spectrumcan be used more effectively if opportunisticusers can time-share with licensed users. Thispossibility results in the fundamental question ofthe article: how much QoS degradation should alicensed user tolerate so that OSA becomesworthwhile for an OSA user? Quantifying thisrelation requires looking into details of thedesign of an OSA MAC protocol. In the follow-ing we discuss major OSA MAC design choices,and quantify their effects on the achieved QoSwhere possible. Finally, we list the main conclu-sions from this study in the final section.

OPPORTUNISTIC SPECTRUM ACCESS:A BRIEF OVERVIEW

Before going into the details of OSA depend-ability challenges, let us look at OSA from abroader perspective. Static frequency planningleads to scarcity of the spectrum since it allo-cates spectrum based on worst case use, notactual use. Since the spectrum is getting full,new dynamic spectrum management techniques

have emerged [1, 2]. Promising dynamic spec-trum management solutions are exclusive spec-trum management (ESM), the spectrumcommons (SC) sharing model, and hierarchicalspectrum management (HSM). In Fig. 1a impor-tant spectrum management techniques and theirhierarchy are introduced. The ESM model stillgives exclusive channel use to each user orprovider, but differs from a static assignment inthe sense that the channels are allocated dynam-ically among possible licensees. In the SC modeldifferent users compete for the assigned fre-quencies on equal terms. The HSM model givesprimary (licensed) users (PUs) more rights touse the spectrum than other secondary (nonli-censed) users (SUs). We can distinguish twoHSM approaches. In overlay HSM, only oneuser/system can use a frequency band at a partic-ular space and time, and the SU has to back offwhen a PU is present. However, when no PU ispresent, the SU can opportunistically use thefrequency band; hence, this technique is alsoreferred to as OSA. In underlay HSM, an SUcan transmit in an already occupied band if thistransmission does not increase the interferenceto the PU above a given threshold. A furtherclassification of overlay HSM (not shown in Fig1a) involves symmetric coexistence (when bothSU and PU networks adapt) and asymmetriccoexistence (when only the SU network adapts,obeying the PU requirements). In this article weconsider the case where the PU does not adaptto the operation of SU.

Since the SU can only access the spectrumwhen it is free of PU activity, this approachinherently puts limitations on the level of QoS itcan achieve. There is also some ambiguity aboutwhen a channel is considered to be free of PUactivity. Some definitions state that a channel isoccupied by a PU simply if the PU is present,irrespective of its duty cycle. Other definitionsassume that the SU can also time-share thechannel with a PU. In this article we use the sec-ond approach since clearly it can result in muchhigher spectrum availability for the SU, as moti-vated by the measurement discussed below.

n Figure 1. Modern spectrum management: a) classification with the application examples (see also [2, Fig. 1]); b) PU/SU QoS trade-off for different OSA MAC protocol designs.

Exclusive HierarchicalSpectrumcommons

(a) (b)

Opportunisticspectrum

access

Spectrummanagement

Static Dynamic

Overlay

Ultra-wideband

Low

Hig

hSU

qua

lity

of s

ervi

ce

Low

OptimalProtocol B

Protocol A

HighPU quality of service

ISM, UNIIaccess

-Dynamic frequency selection-Dynamic frequency sharing-Spectrum auctioning-Spectrum leasing-Negotiated spectrum access

Underlay



IEEE Wireless Communications • October 200822

WHAT TO EXPECT FROM PU SPECTRUM USE

Although many researchers claim that the spec-trum is used sparsely, it is in general very diffi-cult to obtain good information about realisticspectrum use. To obtain an example of PU spec-trum use, we measured the spectrum use in thefrequency range F = [446.04; 467.82] MHz on 13March 2007 at different times between 11 a.m.and 8 p.m. in the Electrical Engineering Depart-ment at the University of Twente in the Nether-lands. Following the Dutch radio spectrum map,these bands are assigned to public mobile com-munication channels, with the exception of thosechannels that are assigned to the Dutch Ministryof Defense for aviation communication. Wehave extracted periods of PU signal activity (ONperiods) and PU inactivity (OFF periods) foreach frequency bin of 100 kHz, which made itpossible to compute the PU activity metrics aslisted in Table 1.

Only 1.7 percent of the frequency bins werebusy the whole time, and could hence not beused by SUs at all. Also, 15.7 percent of all theobserved frequency bins were free during thewhole observation time. Therefore, the remain-ing 82.6 percent of all frequency bins showedON and OFF patterns (with mean ON and OFFtimes of 4.3 s and 58.9 s, respectively). As aresult, when an SU cannot time-share a frequen-cy bin with a PU, it can only achieve a spectrumutilization of 15.7 percent. A striking fact is thatthe total channel utilization of the measured fre-quency range F was only 6.8 percent. So, whentime-sharing is possible, the SU can achieve autilization of 93.2 percent, which is a significantimprovement compared to 15.7 percent.

Next we study the channel availability varia-tions. For the chosen frequency range F, theaverage difference in available free frequenciesbetween two consecutive time slots of 140 mswas 47 percent, which shows that the spectrum

available to the SU can vary significantly. Theminimum difference was 16.6 percent, whichmeans there was always a variation. The maxi-mum difference was 85 percent.

OSA QOS TRADE-OFFSTranslating the spectrum availability into anacceptable level of QoS for the SU is the keytask of the OSA MAC protocol. A network isonly dependable when the required level of QoS(i.e., delay or throughput) can be achieved con-sistently or with a high enough probability.Clearly there is a fundamental trade-off betweenthe PU spectrum requirements and the depend-ability that can be achieved, since no protocoldesign can deliver an acceptable QoS when noresources are available. We depict this trade-offin Fig. 1b.

The goal of this article is to quantify OSAdependability in terms of classical QoS parame-ters like throughput and delay as a function ofPU parameters such as load and tolerance tointerference or collisions from SUs. Intuitively,the OSA QoS will be improved when the PU ismore tolerant to interference or has a lowerload. Quantifying this, however, requires makingassumptions about the OSA MAC protocol,since the optimal MAC design will result in thebest joint SU-PU performance, as shown con-ceptually in Fig. 1b. Another question that henceneeds to be addressed is “How should the SUexploit the available spectrum to achieve a reli-able communication?” As we are focusing onMAC design here, we answer this question byfirst listing all features that are important forOSA networks and showing how these havebeen addressed in the literature. Where possible,we quantitatively assess which solution is optimaland hence results in the best SU QoS for a givenPU set of requirements.

KEY FEATURES OF OSA MACSQuantifying dependability for the scenario whereSUs and PUs share a set of channels in time andfrequency requires making assumptions aboutOSA network operation. We have listed manyimportant OSA MAC proposals found in the lit-erature and identified a set of key featuresrequired to enable OSA operation. Our focus ison decentralized MAC protocols only; that is, inwhich each OSA node locally decides when andhow to access the channel. In addition, manycentralized solutions have been proposed wherea coordinator organizes the channel access. Forinstance, the current proposal for IEEE 802.22wireless radio access network (WRAN) [3] is anexample of such an OSA protocol. We are alsoaware of proprietary OSA MACs found in theOSA devices of Shared Spectrum Company,Philips, and Microsoft, but since their specifica-tions are not public, we were not able to includethem in the survey.

We briefly introduce the identified features,as listed in Table 2, for the protocols found inthe literature. Before an SU network can startoperating, it should decide on the set of chan-nels to use. This bootstrapping is hence a firstSU MAC feature that deserves attention. Next,after the set of possible channels is identified,

n Table 1. Results of the PU channel observa-tions.

Feature Value

Mean channel utilization 6.8%

Total number of busy bins 1.7%

Total number of free bins 15.7%

Total number of bins with PU dutycycles 82.6%

Slot-to-slot difference in available bins 47%

Number of free bins within frequencypool (Max) 85%

Number of free bins within frequencypool (Min) 16.6%

Average ON time 4.3 s

Average OFF time 58.9 s

Since the SU canonly access the spectrum when it isfree from PU activity,this approach inherently puts limitations on thelevel of QoS it canachieve. There isalso some ambiguityabout when a channel is consideredto be free from PU activity.




the network should decide on how to organizeSU communication over those channels. Themore channels of a given bandwidth are used,the more throughput the SU network canachieve. Also, since each channel can potentiallybe claimed by a PU, the probability that an SUloses all its channels decreases when using morechannels. We hence assume a multichannel OSAMAC, and selecting a MAC type is consideredthe next important feature. Next, OSA operationrequires information about the presence of PUs,and how this is implemented is a third importantdesign choice. Depending on the multichannelMAC type and the organization of the scanning,more or fewer front-ends are required to workin parallel, which is a fourth design choice.Finally, a policy is required to establish the coex-istence rules with the PU. The stricter the policy,the more difficult it becomes for the SU. Below,we discuss each feature in more detail and quan-tify the effect on QoS where possible.

BOOTSTRAPPINGBootstrapping is the process during which an SUnode decides which PU channels are suited toopportunistic spectrum communication. In onescenario third parties provide information aboutsuch channels so that the SU node only has toconsult such a third party when it wants to startor join a network. Other scenarios assume thateach node finds those channels locally, whichcan involve a significant amount of spectrumscanning. Next to finding the channels, eachnode should distribute its set of channels toother users in the network. Interestingly, only ahandful of proposed OSA MACs consider boot-strapping: C-MAC [6], AS-MAC [10], DOSS[11] (only for a control channel), and HD-MAC[15]. Usually MAC designers assume that eachOSA node has a preprogrammed list of PUchannels for use. In the rest of this article weassume that each node has decided the set ofopportune channels, and that this set of channelsis available to each node in the SU network.Hence, ach SU node operates on the same set ofchannels. Motivated by the measurement exam-ple discussed earlier, we assume that a channelis opportune when a PU is not using it constant-ly (i.e., channels with no PU present and chan-nels with some PU activity). In most cases thisgives us a set with more than one channel. In thenext section we discuss how to organize the SUcommunication across those channels.

CONTROL CHANNEL DESIGNAfter the bootstrapping procedure, the SU net-work has decided on a set of possible channels.Now, for each data packet transmission, the SUtransmitter and receiver have to coordinatewhich channel and time slot they will use forthat transmission. This coordination is typicallyimplemented with a (common) control channel(CC). From a reliability viewpoint, this CC is avery crucial element of the MAC design, sinceno SU data communication is possible when it isobstructed.

Using the approach defined in [16, Sec. II]for general multichannel MACs, we can identifyfour types of CC implementation, as listed inFig. 2.

1. Dedicated (common) CC (DCC), whereone SU channel is dedicated solely to the trans-port of control messages. All nodes should over-hear the control data exchange, even during thedata exchange. As a result, one radio front-end(RFE) needs to be dedicated to the exchange ofcontrol data. When only one RFE is used, trans-mission of control and data packets is time divid-ed, but then the operation of the protocol getsmore complex. The drawback of the DCCapproach in the context of OSA is that when aPU is active on the CC, all communication isobstructed. It is hence often assumed that theCC should always be available or free from PU.We discuss this issue in more detail below.

2. Hopping CC (HCC), where all nodes hopbetween all channels following a predefined pat-tern. When both sender and receiver successfullyexchange control messages on the current chan-nel, they stop hopping and start transmittingdata. After that, they come back to the originalhopping pattern. HCC has the advantage that ituses all channels for transmission and control,whereas in DCC the CC can be used to transfercontrol packets only. Also, HCC does not requirea single channel to be free from PU activity.

3. Split phase CC (SPCC), where time isdivided into control and data phases. During thecontrol phases, all nodes switch their RFEs tothe dedicated CC and decide on the channels touse for the upcoming data transfers. After eachcontrol phase, a data phase allows for data trans-missions on the agreed channels. The advantageis that the CC can be used during the data phas-es. Also, compared to DCC, no extra RFE forthe CC is needed. On the other hand, SPCCneeds stronger synchronization to identify con-trol and data phases.

4. Multiple rendezvous CC (MRCC), where

n Table 2. Survey of representative OSA MACs.

Protocol name Bootstrap Type Scan. No. RFEs Policies

BB-OSA [4] No DCC No 1 —

ESCAPE [5] No DCC Yes 1 P1, P2

C-MAC [6] Yes DCC Yes 1 —

MMAC-CR [7] No DCC Yes 1 P1

Choi et al. [8] No DCC No 2 —

Shu et al. [9] No DCC No 2 P1

AS-MAC [10] Yes DCC Yes 1 —

DOSS [11] Yes DCC Yes 3 —

HC-MAC [12] No DCC Yes 1 —

Su et al. [13] No DCC No 2 —

SRAC [14] No SPCC No 1 P1

HD-MAC [15] Yes SPCC Yes 1 P1




multiple nodes can exchange control informationat the same time, using all available channels.Each node knows the hopping pattern of theothers (such a hopping pattern is based on theseed of a pseudo-random generator), whichmakes control exchanges possible by followingthe intended receiver on its hopping sequence.MRCC maximally spreads both control and dataexchanges across the channels in a very randomway. As a result, MRCC seems to be the mostrobust to PU activities on any of the channels.We illustrate this quantitatively below. However,MRCC also requires more stringent synchro-nization between hopping users since users haveto keep track of meeting times.

We further study the different MAC types.First, we assess the impact of PU activity on the

control messages exchange. Next, delay andthroughput of the SU network are determinedfor each of the four MAC types.

1. PU activity on the CC: For HCC andMRCC, the exchange of control messages on PUchannels is inevitable since the exchange of con-trol data is spread among all channels in the SUnetwork. This certainly affects network availabil-ity and communication reliability. However, forDCC and SPCC, the CC does not necessarilyneed to be implemented on a channel with PUactivity. More specifically, a single DCC thatdoes not suffer from PU activity can be builtusing a proprietary non-PU channel (e.g., ISMor UNII channels), or a wideband transmissiontechnique such as code-division multiple access(CDMA) or ultra-wideband (UWB). The first

n Figure 2. Illustration of the operation of different multichannel MAC types, with PU activity on eachchannel: a) DCC; b) HCC; c) SPCC; d) MRCC.

Respective default hoppingsequences of node 1 and 2

PU PUDatatransfer 2

Datatransfer 2

Channel 2(data)

(a)

PU PU PUDatatransfer 1

Datatransfer 1

Channel 1(data)

RTS1

CTS1

RTS1

RTS2

CTS2

CTS2PU PUPUChannel 0

(CC)

PU RTS3

CTS3PU PUData

transfer 2Channel 2

(data)

RTS3

CTS3PU PUData

transfer 2Data

transfer 2Channel 1

(data)

RTS1

CTS1 PU PU PUData

transfer 1Data

transfer 1Data

transfer 1Channel 0

(data)

(b)


Channel 2(data)


Datatransfer 1

Channel 1(data)

RTS1

CTS1

RTS2

CTS3

CTS2PU PU

Control phase Data phase

PUPU RTS3

Channel 0(data)

PU PU PU PUChannel 2(data)

(c)

(d)

PU PU PU PUChannel 1(data)

Respective actual hoppingsequences of node 1 and 2

PU PU

RTS/CTS

Hoppingresumes

Data exchange

PUChannel 0(data)

For HCC and MRCCthe exchange of control messages onPU channels isinevitable since theexchange of controldata is spreadamong all channelsin the SU network.This certainly affectsnetwork availabilityand communicationreliability.




approach is the most used in the literature [6,15]. Spreading the CC across a wide bandwidthis very robust against PU activity, but limits theoperating range of the network since UWBthroughput decreases strongly with distance.When it is not possible to use a proprietary orwideband channel, the concept of a backup CChas been proposed [6]. Indeed, the probabilitythat both CCs are occupied by a PU simultane-ously is smaller. However, this solution isresource inefficient.

Since DCC is a very popular choice for OSAMAC protocols (Table 2), we quantify theimpact of PU activity in the CC on the through-put that can be achieved in an SU networkusing DCC. This will allow us to assess howimportant it is for an SU network to get a pro-prietary channel for its operation. For this, wehave extended the analytical model for multi-

channel MACs proposed in [16] with a moredetailed physical layer model to capture theimpact of PU and SU interference. Also wehave implemented the PU presence and the PUscanning process (detailed later in this article).In Fig. 3a we plot the impact of PU presence inthe 2 Mb/s dedicated control channel on theSU throughput as function of SU data packetsize. In the simulation, PU presence was mod-eled as a Bernouilli process with average pres-ence rate qcc. This presence is detected with aprobability of 0.99 and with a probability of0.03 the SU falsely assumes the PU to be pre-sent on the channel and pauses control messageexchanges. A total of three SU channels areconsidered, and one of those is the CC. ThePU was assumed to be present in the CC only.The interesting conclusion is that the SU cancontrol how dependent it is on the control

n Figure 3. QoS assessment for 3 PU channels and 20 SU users: a) analytical throughput of an OSA network as a function of data packetlength for different levels of PU activity qcc on CC for DCC MAC; b) simulated impact of PU channel occupancy rates for four differentclasses of OSA MACs in terms of delay; c) throughput; dashed and solid lines represent different PU packet sizes (see text for more expla-nation).

Data packet length (bits)

(a)

0.200.4

0.6

Ave

rage

thr

ough

put

(b/s

)

0.8

1

1.2

1.4

1.6

1.8x106

0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2x104 Offered PU load qp (% total channel capacity)

(b)

00

20

Ave

rage

per

pac

ket

dela

y (m

s)

40

60

80

100

120

140

160

180

200

5 4010 15 20 25 30 35

Offered PU load qp (% total channel capacity)

(c)

00

1

Ave

rage

thr

ough

put

(Mb/

s)

2

3

4

5

6

70605040302010

qcc=0qcc=30%qcc=50%qcc=70%qcc=90%

SPCCDCCHCCMRCC

SPCCDCCHCCMRCC




channel by tuning its data size. Indeed, for larg-er data packets, less control messages need tobe exchanged, so the impact of the controlchannel is smaller. When the data size needs tobe smaller, the impact of PU activity is larger,and in this case the concept of a backup CCcould be helpful.

2. PU activity vs. SU throughput and delay:Because of their opportunistic nature, it is gen-erally assumed that SU networks should be high-ly tolerant to delays. Indeed, it can happen thatall channels are used by the PU, causing thecommunication to be suspended. Let us nowquantify how large the SU delay becomes as afunction of PU activity on the channels. This willgive us important information on the type ofapplications that can be supported on OSA net-works, or alternatively how the PU activityshould be limited to be able to support a target-ed SU application.

From Table 2, it is clear that the majority ofOSA MAC proposals use DCC and only twouse SPCC. Surprisingly, we were not able toidentify HCC and MRCC proposals in the OSAliterature. To be complete, however, we haveconsidered these four MAC classes. We haveimplemented all the protocols in a coarse time-slotted simulator. It allows us to capture all theintrinsic features of the considered MACs,especially the way the control data exchange isorganized. For a more detailed description ofthe simulator readers are referred to [16, Sec.V]. We have extended the simulator with PUactivity patterns. In Fig. 3b we plot the simulat-ed delay of the SU applications as a function ofPU activity for each of the MACs. In the simu-lation the OSA network consists of 20 devices;every user was generating traffic following aPoisson distribution with average transmissionrate of 150 kb/s (the total SU load is 50 percentof the total bandwidth). There are three chan-nels. Every channel has a fixed bandwidth of 2Mb/s. PU activity was modeled through a geo-metrically distributed on-off process. The aver-age PU packet was 0.8 ms (solid line in Fig. 3b)or 8 ms (dashed line in Fig. 3b), while the aver-age off time was varied from 0.8 ms to 160 ms,resulting in a range of average PU activity lev-els. We have assumed perfect detection of PUactivity on each individual channel. The strikingfact is that MRCC is the best MAC of all, what-ever the PU activity level. Its immunity to tem-poral nonavailabilities of the channel andefficient use of the whole channel capacity pre-sents this type of MAC as a candidate for real-life implementation. This is because MRCCrandomizes both control and data exchange sig-nificantly. Another observation is that the delayof all MAC classes becomes higher with increas-ing PU packet size. Because of its randomizingproperties, MRCC suffers less. As noted previ-ously in our measurements, PU traffic can haveON times or packet durations on the order ofseconds. For the given scenario, even when thePU activity reaches 30 percent, the delay expe-rienced by the SU is still lower than 100 ms.This delay could even fit within the bounds forpacket voice communication, where the round-trip delay for a voice conversation should notexceed 400 ms according to International

Telecommunication Union — Telecommunica-tion Standardization Sector (ITU-T) Recom-mendation G.114.

Next, in Fig. 3c we assess OSA networkthroughput as a function of PU activity for asimilar scenario (the solid line is now 8 ms anddashed 80 ms PU packet size). As expected, theaverage SU throughput decreases linearly withPU activity. SPCC performs the worst in thiscase, since it wastes a lot of bandwidth on thedata channels during the control phase. MRCCis still the best MAC design. The averagethroughput does not vary a lot with PU packetsize.

SCANNING PROCESSSince an SU cannot use the channel when a PUis present, it should obtain information aboutPU activities on each channel. Typically, this isimplemented using PU detectors [17]. Alterna-tively, PU activity information can be assumed tobe broadcast by a central device. We can thusclassify OSA MAC protocols into sensing andnon-sensing OSAs. From Table 2 we can con-clude that the majority of the considered proto-cols assume having the scanning under theircontrol.

Unfortunately, scanning increases the over-head since nodes cannot transmit when they arescanning. Since it is often difficult to distinguishSU and PU signals, the whole SU network hasto be quiet during sensing, which requires quietperiod management [6]. Scanning, or quieting thenetwork, can be done periodically or before eachtransmission attempt. The distance between twoconsecutive sensing intervals varies, and is oftena function of the policy. The more tolerant thePU to interference, the less often sensing shouldbe done. Noise, fading, multipath shadowing,and low PU signal levels make a reliable detec-tion process difficult. Suboptimal detectors affectnot only the PU QoS levels, but the SU QoS aswell.

Scanning performance is measured in termsof the probability of detecting a PU when pre-sent, and the probability of falsely detecting aPU. In the former case, both SU and PU QoSis degraded, since an SU will transmit and col-lide with the PU, resulting in packet loss forboth PU and SU. In the latter case SU QoS isdegraded, since an SU will not transmit whenthe channel was actually free. It is well knownthat scanning performance improves withincreasing scanning length [18], and in Fig. 4awe investigate the optimal sensing time interms of SU QoS for a scenario with DCCMAC. Scanning is performed using energydetection before each transmission attempt,and Rayleigh fading is assumed. SU through-put indeed improves with increasing detectionreliability. When detection performance isacceptable, the throughput starts to decreasesince the scanning overhead dominates. Thiseffect is less visible with high PU activity, sincean OSA network will not have enough oppor-tunities to communicate; therefore, it will notlose much of the already small PU channelcapacity. The impact on PU QoS is discussedlater since PU QoS can be considered to be apolicy constraint.

Because of theiropportunistic nature,it is generallyassumed that SUnetworks should behighly tolerant todelays. Indeed, itcan happen that allchannels are used bythe PU, causing thecommunication to besuspended.




RADIO FREQUENCY FRONT-ENDS

The exact multichannel MAC operation andscanning implementation degrees of freedomdepend on the number of front-ends that areavailable in each SU node. Indeed, when multi-ple RFEs are available, it is possible to use mul-tiple channels simultaneously for transmission.Alternatively, spare RFEs can be used for scan-ning only, decreasing the impact of scanning onthe network throughput. When only one RFE isavailable, sensing and communication should besplit in time. Of course, increasing the numberof RFEs increases the reliability of the systemand decreases delay, but simultaneously increas-es the total cost. Typically this number variesfrom 1 to 3 (Table 2). In this article, we assumetwo RFEs for the DCC and a single RFE for theother MAC types.

INTERFERENCE MANAGEMENT POLICIESSince it is impossible to detect PU presence withcertainty, harmful interference to the PU cannotbe avoided. The maximum level of interferenceis typically specified through interference poli-cies (IPs), which define how SUs can behave incertain PU bands while maintaining the QoSrequirements of the PU. The more relaxed theIPs, the better the SU can take advantage ofspectrum opportunities. In other words, policiesare rules that determine the trade-off betweenSU and PU QoS. Defining such IPs, however, isa very difficult task. In this article we want to seehow much an SU could benefit from morerelaxed PU policies.

From our literature search (Table 2), we canenlist three major policy classes for OSA net-works:P1 Time-based: These policies define time met-

rics that regulate SU transmissions. An exam-ple metric is the evacuation time that defines

how fast an SU should vacate a channel aftera PU is detected.

P2 Power-based: These policies define the powerlimits each SU needs to take into accountwhen using PU channels. Example metrics aremaximum (peak) power, power mask, andaverage transmit power.

P3 Collision-based: These policies are defined atthe MAC layer, usually assuming packet-basedtransmissions. They define collision probabili-ty limits, bounding the probabilities that anSU packet will harm a PU packet.Depending on the PU system, one of these

policies is most appropriate (e.g., policy P3 canonly be applied to packet-based networks). Also,a given policy can often be described, or imple-mented, differently. The exact description oftensignificantly impacts the usability and cost of theSU network. For example, policy P2 can also bedefined as a maximum distance between PU andSU, which requires the OSA network to embedexpensive localization capabilities. We note thatthe definition of policies for OSA networks is avery hard problem and an ongoing topic ofresearch.

Next to the policy format, its level of PU pro-tection can be too restraining. For a given policy(we use the P3 policy since we assume both SUand PU networks are packet-based), we investi-gate what QoS the SU can achieve (Fig. 4b).The probability of collision with a PU packetand the SU throughput have both been comput-ed as a function of the scanning duration (andhence scanning quality). A stricter collision con-straint is only achieved with improved detectionperformance, requiring the SU to scan over avery long time. When the PU constraint isrelaxed, the SU can scan for a shorter time,resulting in throughput improvement of the SU.

The P3 policy, avoiding collisions with thePU, can be implemented using listen-before-

n Figure 4. QoS assessment for 3 PU channels and 20 SU users: a) OSA network analytical throughput as a function of scanning lengthfor DCC MAC; b) analytical relation between level of interference to PU and SU network throughput for DCC MAC. Throughput andinterference have been computed as a function of scanning length varying from 1 to 45 ms (resulting in decrease of probability of falsealarm from 0.23 to 0.024 and increase in probability of detection from 0.82 to 0.92) and three different levels of PU activity qp on allchannels.

Scanning length (s)

(a)

1

1.5

Ave

ratg

e th

roug

hput

(b/

s)

1

2

2.5

3

3.5

1.5

x10-5

x106

2 2.5 3 3.5 4 4.5Interference probability (P3 policy)

(b)

Ave

ratg

e th

roug

hput

(b/

s)

1.2

1.4

1.6

1.8

2

2.2

2.4x106

0.0105 0.01250.011 0.0115 0.0120.01

qp = 0%qp = 30%qp = 50%qp = 70%

qp = 0%qp = 40%qp = 50%




send scanning. Since the PU does not scan forSU presence, it is possible that the PU will starta new packet during SU transmission. This canonly be avoided by assuming small SU packets,since it is not realistic to assume any synchro-nization between the PU and SU networks.However, assuming such synchronization is veryconvenient for analysis [15], and we have alsoassumed such synchronization in our models.The only OSA MAC that can actually assumesuch synchronization is AS-MAC [10], which wasspecifically designed for operation on GSMchannels, where slot boundaries can be capturedeasily. In general, since it is very hard to pre-serve SU/PU synchronization, certain policieslike P1 and P3 have to be defined very carefully.

DISCUSSION AND CONCLUSION

In the previous section many features that areimportant for OSA MAC design are listed anddiscussed, first by means of reviewing the pro-posals found in the literature and also by quanti-tatively assessing the impact of some features onPU or SU QoS. A first conclusion is that most ofthe proposed solutions do not cover many of thecrucial elements of a proper OSA MAC protocoldesign. Indeed, since the operating conditions ofOSA networks are typically unknown during thedesign time phase, the bootstrapping procedureto set up the network before communication isvery important. However, it is omitted in manyof the protocol designs. In the case of OSA net-working, this bootstrapping cannot be consid-ered to be a one-time effort at the start of thecommunication network, so it is crucial to makeit as efficient as possible and embed it in theMAC protocol design. Also, the required scan-ning for the presence of the PU is sometimesomitted in the protocol design or performanceanalysis. More important, the specification ofpolicies to regulate the coexistence with PUs isoften described very vaguely or even fully omit-ted. It can be concluded that although manyindividual contributions can be found, it isimportant to assess how these subtasks can beintegrated together into a complete solution tobe able to fully assess the expected QoS of OSAnetworks.

Often, the solutions proposed for the sub-tasks are suboptimal. In this article focus hasbeen on the organization of the control channelsince this is a very important aspect of multi-channel OSA networking. Although the solu-tions proposed in the literature always assumethe availability of a fixed channel for controlinformation exchange (for DCC this channel isonly for control, in SFCC the channel is alsoused for data), we show that this is not necessar-ily optimal. Indeed, especially when there are alot of possible channels to use, a fixed controlchannel easily becomes the bottleneck. Also,when no channel can be assumed to be free ofPU activity, it is best to spread the controlexchanges over different channels as much aspossible. As a result, we show that the MRCCactually outperforms DCC, HCC, and SFCCover a broad range of PU traffic conditions.

Very often, researchers stick to a given designfor OSA MAC since it is the most practical, and

it does not make sense to make the design morecomplicated in the absence of measurements ordetailed performance analysis. In this articl wehave included the results of a simple measure-ment campaign since we want to emphasize thatit is important to build conclusions using realisticassumptions. Those measurements have shownthat it is not impossible to find a channel that isnot used for a very long time, facilitating the useof a dedicated control channel. However, it wasalso shown that the instantaneous capacity variesa lot, advocating the use of a protocol that caneasily take advantage of these variations.

Finally, we want to emphasize that no solu-tions found so far in the literature assess the QoSgiven to the secondary network in detail. This is,however, a very crucial area of study since theintroduction of OSA networks only makes senseif a sufficient level of QoS can be expected. Inthis article we attempt to study the delay andthroughput performance of a broad range ofOSA designs as a function of PU activity. Also,we assess the fundamental trade-off between PUQoS and SU QoS. The more freedom is given tothe SU to access the channel, the more capacityit can use and the better its performance. Howev-er, more freedom to the SU means less guaran-tees for the PU, and the success of OSAnetworking will depend on how well we can opti-mize this trade-off with a given policy.

ACKNOWLEDGMENTSThis work has been supported by the FreebandAAF project sponsored by the Dutch Ministry ofForeign Affairs. Sofie Pollin is supported by theMarie Curie OIF fellowship of the EU.

REFERENCES[1] R. V. Prasad et al., “Cognitive Functionality in Next Genera-

tion Wireless Networks: Standardization Efforts,” IEEE Com-mun. Mag., vol. 46, no. 5, Apr. 2007, pp. 72–78.

[2] Q. Zhao and B. M. Sadler, “A Survey of Dynamic Spec-trum Access: Signal Processing, Networking, and Regu-latory Policy,” IEEE Signal Processing Mag., vol. 24, no.3, May 2007, pp. 79–89.

[3] S. Sengupta et al., “Enhancements to Cognitive RadioBased IEEE 802.22 Air-Interface,” Proc. IEEE ICC ’07,Glasgow, UK, June 24–28, 2007.

[4] G. Auer, H. Haas, and P. Omiyi, “Interference AwareMedium Access for Dynamic Spectrum Sharing,” Proc.IEEE DySPAN ’07, Dublin, Ireland, Apr. 17–20, 2007.

[5] X. Liu and Z. Ding, “ESCAPE: A Channel Evacuation Pro-tocol for Spectrum-Agile Networks,” Proc. IEEE DySPAN’07, Dublin, Ireland, Apr. 17–20, 2007.

[6] C. Cordeiro and K. Challapali, “C-MAC: A Cognitive MACProtocol for Multi-Channel Wireless Networks,” Proc. IEEEDySPAN ’07, Dublin, Ireland, Apr. 17–20, 2007.

[7] M. Timmers et al., “A Distributed Multichannel MACProtocol for Cognitive Radio Networks with PrimaryUser Recognition,” Proc. IEEE CrownCom ’07, FL, Aug.1–3, 2007.

[8] N. Choi, M. Patel, and S. Venkatesan, “A Full DuplexMulti-Channel MAC Protocol for Multi-Hop CognitiveRadio Networks,” Proc. IEEE CrownCom ’06, MykonosIsland, Greece, June 8–10, 2006.

[9] T. Shu, S. Cui, and M. Krunz, “Medium Access Controlfor Multi-Channel Parallel Transmission in CognitiveRadio Networks,” Proc. IEEE GLOBECOM ’06, San Fran-cisco, CA, Nov. 27–Dec. 1, 2006.

[10] S. Sankaranarayanan, P. Papadimitratos, and A.Mishra, “A Bandwidth Sharing Approach to ImproveLicensed Spectrum Utilization,” IEEE Commun. Mag.,vol. 43, no. 12, Dec. 2005, pp. S10–S14.

[11] L. Ma, X. Han, and C.-C. Shen, “Dynamic Open Spec-trum Sharing MAC Protocol for Wireless Ad Hoc Net-works,” Proc. IEEE DySPAN ’05, Baltimore, MA, Nov.8–11, 2005.

It can be concludedthat although manyindividual contributions can befound, it is importantto assess how thesesubtasks can be integrated togetherinto a complete solution to be ableto fully assess theexpected QoS ofOSA networks.




[12] J. Jia, Q. Zhang, and X. Shen, “HC-MAC: A Hardware-Con-strained Cognitive MAC for Efficient Spectrum Manage-ment,” IEEE JSAC, vol. 26, no. 1, Jan. 2008, pp. 106–17.

[13] H. Su and X. Zhang, “Cross-Layer Based OpportunisticMAC Protocols for QoS Provisioning over CognitiveRadio Wireless Networks,” IEEE JSAC, vol. 26, no. 1,Jan. 2008, pp. 118–29.

[14] L. Ma, C.-C. Shen, and B. Ryu, “Single-Radio AdaptiveChannel Algorithm for Spectrum Agile Wireless Ad HocNetworks,” Proc. IEEE DySPAN’07, Dublin, Ireland, Apr.17–20, 2007.

[15] Q. Zhao et al., “Decentralized Cognitive MAC forOpportunistic Spectrum Access in Ad Hoc Networks: APOMDP Framework,” IEEE JSAC, vol. 25, no. 3, Apr.2007, pp. 589–600.

[16] J. Mo et al., “Comparison of Multi-Channel MAC Pro-tocols,” IEEE Trans. Mobile Comp., vol. 7, no. 1, Jan.2008, pp. 50–65.

[17] T. Yucek and H. Arslan, “A Survey of Spectrum SensingAlgorithms for Cognitive Radio Applications,” IEEECommun. Surveys and Tutorials, Apr. 2007.

[18] F. F. Digham, M.-S. Alouini, and M. K. Simon, “On theEnergy Detection of Unknown Signals over FadingChannels,” Proc. IEEE ICC ’03, Anchorage, AK, May11–15, 2003.

BIOGRAPHIESPRZEMYSLAW PAWELCZAK [M] ([email protected])graduated with honors in electronics and telecommuni-cations engineering from Wroclaw University of Tech-nology, Poland, in 2004. Between 2004 and 2005 hewas a staff member of Siemens Software DevelopmentCenter, Wroclaw, Poland. Since 2005 he has been pur-suing his Ph.D. studies at Delft University of Technologyin the field of dynamic spectrum access networks. Dur-ing fall 2007 he was a visiting scholar at the Connectiv-ity Laboratory of the Univers ity of Cal i fornia (UC),Berkeley. In 2008 he received an annual KIVI NIRIA Tele-com Prize for best Ph.D. student in telecommunicationsin the Netherlands. He is the co-originator and an orga-nizing committee member of the Cognitive Radio work-shops collocated with IEEE ICC 2007, 2008, and 2009.He is a member of the IEEE Technical Committee onCognitive Networks and IEEE SCC41 StandardizationCommittee.

SOFIE POLLIN ([email protected]) received a eegreein electrical engineering in 2002 and a Ph.D. degree in2006 (with honors) from Katholieke Universiteit Leuven,Belgium. Since October 2002 she has been a researcher atthe Wireless Research group of the Interuniversity Micro-electronics Center (IMEC). In the summer of 2004 she wasa visiting scholar at National Semiconductor, Santa Clara,California. In the summer of 2005 she was a visitor at UCBerkeley. Currently, she is a post-doctoral researcher at UCBerkeley working on coexistence issues in wireless commu-nication networks.

HOI-SHEUNG WILSON SO ([email protected]). [M] receiveda B.S. degree in computer science from Cornell Universityin 1997, and M.Sc. and Ph.D. degrees in computer sciencefrom UC Berkeley, in 2000 and 2006, respectively. He iscurrently with the Siemens Technology-to-Business Center,Berkeley. His current research interests include wireless pro-tocol design and multichannel media access control proto-cols for wireless networks.

AHMAD R. S. BAHAI ([email protected] ) received his M.Sc.degree from Imperial College, University of London in 1988and his Ph.D. degree from UC Berkeley in 1993, both inelectrical engineering. He is currently a professor-in-resi-dence at UC Berkeley, a consulting professor at StanfordUniversity, and executive advisor to National Semiconduc-tor. He has served as a fellow and CTO of National Semi-conductor for five years. He was technical manager ofAdvanced Wireless Technology Group at AT&T Bell Labora-tories until 1997. His research interest includesadaptive/mixed signal processing and communication sys-tem design. He co-invented multicarrier spread spectrumtheory, which is used in most modern wireless systems andstandards. He is the author of the first textbook on OFDM,Multicarrier Digital Communications, and served as an Asso-ciate Editor of IEEE Communication Letters for five years.

R. VENKATESHA PRASAD (vprasad @ewi.tudelft.nl) obtained aB.Sc. degree in electronics and communication engineeringfrom the University of Mysore, India, in 1991. In 1994 hereceived an M. Tech. degree in industrial electronics and in2003 a Ph.D. degree from the University of MysoreandIndian Institute of Science (IISc) Bangalore, respectively.During 1994 and 1996 he worked as a consultant and pro-ject associate for ERNET Laboratory of ECE at IISc. Whilepursuing his Ph.D degree, from 1999 to 2003, he was alsoworking as a consultant for CEDT, IISc, Bangalore for VoIPapplication developments, as part of a Nortel Networkssponsored project. From 2003 to 2005 he headed a teamof engineers at Esqube Communication Solutions Pvt. Ltd.for the development of various real-time networking appli-cations. Since 2005 he has been with the Wireless andMobile Communications group at Delft University of Tech-nology, working on the EU funded projects MAGNET/MAG-NET Beyond and PNP 2008, and guiding students. He isalso a consultant to Esqube on a part-time basis.

RAMIN HEKMAT ([email protected]) received an M.Sc.degree in electrical engineering from Delft University ofTechnology (TU Delft) in 1990. He has worked since thenfor several telecommunication companies in the Nether-lands and the United States in research and developmentas well as managerial positions. In September 2005 heobtained a Ph.D. degree for his work related to ad hocnetworks from TU Delft. Currently he is working as anassistant professor in the Faculty of Electrical Engineering,Mathematics and Computer Science of TU Delft. His primeresearch interests include multi-user communication sys-tems, wireless communications, and peer-to-peer networks.

We want to emphasize that nosolutions found so

far in the literatureassess the QoS given

to the secondary network in detail.This is however avery crucial study

since the introductionof OSA networks

only makes sense if a sufficient level

of QoS can beexpected.



Documents

Quality of service assessment of opportunistic spectrum access: a medium access control approach