NeXt generation/dynamic spectrum access/cognitive radio ...cs752/papers/cognitive-001.pdf · The key enabling technology of xG networks is the cognitive radio. Cognitive radio techniques

Computer Networks 50 (2006) 2127–2159

www.elsevier.com/locate/comnet

NeXt generation/dynamic spectrum access/cognitiveradio wireless networks: A survey

Ian F. Akyildiz, Won-Yeol Lee, Mehmet C. Vuran *, Shantidev Mohanty

Broadband and Wireless Networking Laboratory, School of Electrical and Computer Engineering,

Georgia Institute of Technology, Atlanta, GA 30332, United States

Received 2 January 2006; accepted 2 May 2006Available online 17 May 2006

Responsible Editor: I.F. Akyildiz

Abstract

Today’s wireless networks are characterized by a fixed spectrum assignment policy. However, a large portion of theassigned spectrum is used sporadically and geographical variations in the utilization of assigned spectrum ranges from15% to 85% with a high variance in time. The limited available spectrum and the inefficiency in the spectrum usage neces-sitate a new communication paradigm to exploit the existing wireless spectrum opportunistically. This new networkingparadigm is referred to as NeXt Generation (xG) Networks as well as Dynamic Spectrum Access (DSA) and cognitiveradio networks. The term xG networks is used throughout the paper. The novel functionalities and current research chal-lenges of the xG networks are explained in detail. More specifically, a brief overview of the cognitive radio technology isprovided and the xG network architecture is introduced. Moreover, the xG network functions such as spectrum manage-ment, spectrum mobility and spectrum sharing are explained in detail. The influence of these functions on the performanceof the upper layer protocols such as routing and transport are investigated and open research issues in these areas are alsooutlined. Finally, the cross-layer design challenges in xG networks are discussed.� 2006 Elsevier B.V. All rights reserved.

Keywords: Next generation networks; Dynamic spectrum access networks; Cognitive radio networks; Spectrum sensing; Spectrummanagement; Spectrum mobility; Spectrum sharing

1. Introduction

Today’s wireless networks are regulated by afixed spectrum assignment policy, i.e. the spectrum

1389-1286/$ - see front matter � 2006 Elsevier B.V. All rights reserved

doi:10.1016/j.comnet.2006.05.001

* Corresponding author. Tel.: +1 404 894 5141; fax: +1 404 8947883.

E-mail addresses: [email protected] (I.F. Akyildiz), [email protected] (W.-Y. Lee), [email protected] (M.C.Vuran), [email protected] (S. Mohanty).

is regulated by governmental agencies and isassigned to license holders or services on a long termbasis for large geographical regions. In addition, alarge portion of the assigned spectrum is used spo-radically as illustrated in Fig. 1, where the signalstrength distribution over a large portion of thewireless spectrum is shown. The spectrum usage isconcentrated on certain portions of the spectrumwhile a significant amount of the spectrum remainsunutilized. According to Federal Communications

.

mailto:[email protected]

mailto:wylee@ ece.gatech.edu

mailto:wylee@ ece.gatech.edu



Fig. 1. Spectrum utilization.

2128 I.F. Akyildiz et al. / Computer Networks 50 (2006) 2127–2159

Commission (FCC) [20], temporal and geographicalvariations in the utilization of the assigned spectrumrange from 15% to 85%. Although the fixed spec-trum assignment policy generally served well in thepast, there is a dramatic increase in the access tothe limited spectrum for mobile services in therecent years. This increase is straining the effective-ness of the traditional spectrum policies.

The limited available spectrum and the inefficiencyin the spectrum usage necessitate a new communi-cation paradigm to exploit the existing wireless spec-trum opportunistically [3]. Dynamic spectrum accessis proposed to solve these current spectrum ineffi-ciency problems. DARPAs approach on DynamicSpectrum Access network, the so-called NeXt Gener-ation (xG) program aims to implement the policybased intelligent radios known as cognitive radios[67,68].

NeXt Generation (xG) communication net-works, also known as Dynamic Spectrum AccessNetworks (DSANs) as well as cognitive radio net-works, will provide high bandwidth to mobile usersvia heterogeneous wireless architectures anddynamic spectrum access techniques. The inefficientusage of the existing spectrum can be improvedthrough opportunistic access to the licensed bandswithout interfering with the existing users. xG net-works, however, impose several research challengesdue to the broad range of available spectrum as wellas diverse Quality-of-Service (QoS) requirements ofapplications. These heterogeneities must be cap-tured and handled dynamically as mobile terminalsroam between wireless architectures and along theavailable spectrum pool.

The key enabling technology of xG networks isthe cognitive radio. Cognitive radio techniques pro-vide the capability to use or share the spectrum in

an opportunistic manner. Dynamic spectrum accesstechniques allow the cognitive radio to operate inthe best available channel. More specifically, the cog-nitive radio technology will enable the users to (1)determine which portions of the spectrum is avail-able and detect the presence of licensed users whena user operates in a licensed band (spectrum sens-

ing), (2) select the best available channel (spectrum

management), (3) coordinate access to this channelwith other users (spectrum sharing), and (4) vacatethe channel when a licensed user is detected (spec-

trum mobility).Once a cognitive radio supports the capability to

select the best available channel, the next challengeis to make the network protocols adaptive to theavailable spectrum. Hence, new functionalities arerequired in an xG network to support this adaptivity.In summary, the main functions for cognitive radiosin xG networks can be summarized as follows:

• Spectrum sensing: Detecting unused spectrumand sharing the spectrum without harmful inter-ference with other users.

• Spectrum management: Capturing the best avail-able spectrum to meet user communicationrequirements.

• Spectrum mobility: Maintaining seamless com-munication requirements during the transitionto better spectrum.

• Spectrum sharing: Providing the fair spectrumscheduling method among coexisting xG users.

These functionalities of xG networks enable spec-trum-aware communication protocols. However,the dynamic use of the spectrum causes adverseeffects on the performance of conventional commu-nication protocols, which were developed consider-ing a fixed frequency band for communication. Sofar, networking in xG networks is an unexploredtopic. In this paper, we also capture the intrinsicchallenges for networking in xG networks and layout guidelines for further research in this area. Morespecifically, we overview the recent proposals forspectrum sharing and routing in xG networks aswell as the challenges for transport protocols. More-over, the effect of cross-layer design is addressed forcommunication in xG networks.

The xG network communication componentsand their interactions are illustrated in Fig. 2. It isevident from the significant number of interactionsthat the xG network functionalities necessitate across-layer design approach. More specifically, spec-

Fig. 2. xG network communication functionalities.

I.F. Akyildiz et al. / Computer Networks 50 (2006) 2127–2159 2129

trum sensing and spectrum sharing cooperate witheach other to enhance spectrum efficiency. In spec-trum management and spectrum mobility functions,application, transport, routing, medium access andphysical layer functionalities are carried out in acooperative way, considering the dynamic natureof the underlying spectrum.

This paper presents a definition, functions andcurrent research challenges of the xG networks. InSection 2, we provide a brief overview of the cognitiveradio technology. The xG network architectures onlicensed band and on unlicensed band are presentedin Section 3. In Section 4, we explain the existingwork and challenges in spectrum sensing. Then, wedescribe the xG network functionalities: spectrummanagement, spectrum mobility and spectrum shar-ing in Sections 5, 6, and 7, respectively. In Section 8,we investigate how xG features influence the perfor-mance of the upper layer protocols, i.e., routingand transport. Finally, we explain how xG functionscan be implemented in a cross-layer approach in Sec-tion 9 and conclude the paper in Section 10.

2. Cognitive radio

Cognitive radio technology is the key technologythat enables an xG network to use spectrum in adynamic manner. The term, cognitive radio, canformally be defined as follows [20]:

A ‘‘Cognitive Radio’’ is a radio that can change its

transmitter parameters based on interaction with

the environment in which it operates.

From this definition, two main characteristics ofthe cognitive radio can be defined [27,58]:

• Cognitive capability: Cognitive capability refersto the ability of the radio technology to captureor sense the information from its radio environ-ment. This capability cannot simply be realizedby monitoring the power in some frequency bandof interest but more sophisticated techniques arerequired in order to capture the temporal andspatial variations in the radio environment andavoid interference to other users. Through thiscapability, the portions of the spectrum thatare unused at a specific time or location can beidentified. Consequently, the best spectrumand appropriate operating parameters can beselected.

• Reconfigurability: The cognitive capability pro-vides spectrum awareness whereas reconfigu-rability enables the radio to be dynamicallyprogrammed according to the radio environ-ment. More specifically, the cognitive radio canbe programmed to transmit and receive on avariety of frequencies and to use different trans-mission access technologies supported by itshardware design [34].


The cognitive radio concept was first introducedin [45,46], where the main focus was on the radioknowledge representation language (RKRL) andhow the cognitive radio can enhance the flexibilityof personal wireless services. The cognitive radio isregarded as a small part of the physical world touse and provide information from environment.

The ultimate objective of the cognitive radio is toobtain the best available spectrum through cogni-tive capability and reconfigurability as described

Fig. 3. Spectrum hole concept.

Fig. 4. Physical architecture of the cognitive radio [12,34]: (a) Cognarchitecture.

before. Since most of the spectrum is alreadyassigned, the most important challenge is to sharethe licensed spectrum without interfering with thetransmission of other licensed users as illustratedin Fig. 3. The cognitive radio enables the usage oftemporally unused spectrum, which is referred toas spectrum hole or white space [27]. If this band isfurther used by a licensed user, the cognitive radiomoves to another spectrum hole or stays in the sameband, altering its transmission power level or mod-ulation scheme to avoid interference as shown inFig. 3.

In the following subsections, we describe thephysical architecture, cognitive functions and recon-figurability capabilities of the cognitive radiotechnology.

2.1. Physical architecture of the cognitive radio

A generic architecture of a cognitive radio trans-ceiver is shown in Fig. 4(a) [34]. The main compo-nents of a cognitive radio transceiver are the radiofront-end and the baseband processing unit. Eachcomponent can be reconfigured via a control bus

itive radio transceiver and (b) wideband RF/analog front-end


to adapt to the time-varying RF environment. Inthe RF front-end, the received signal is amplified,mixed and A/D converted. In the baseband process-ing unit, the signal is modulated/demodulated andencoded/decoded. The baseband processing unit ofa cognitive radio is essentially similar to existingtransceivers. However, the novelty of the cognitiveradio is the RF front-end. Hence, next, we focuson the RF front-end of the cognitive radios.

The novel characteristic of cognitive radio trans-ceiver is a wideband sensing capability of the RFfront-end. This function is mainly related to RFhardware technologies such as wideband antenna,power amplifier, and adaptive filter. RF hardwarefor the cognitive radio should be capable of tuningto any part of a large range of frequency spectrum.Also such spectrum sensing enables real-timemeasurements of spectrum information from radioenvironment. Generally, a wideband front-end archi-tecture for the cognitive radio has the following struc-ture as shown in Fig. 4(b) [12]. The components of acognitive radio RF front-end are as follows:

• RF filter: The RF filter selects the desired bandby bandpass filtering the received RF signal.

• Low noise amplifier (LNA): The LNA amplifiesthe desired signal while simultaneously minimiz-ing noise component.

• Mixer: In the mixer, the received signal is mixedwith locally generated RF frequency and con-verted to the baseband or the intermediate fre-quency (IF).

• Voltage-controlled oscillator (VCO): The VCOgenerates a signal at a specific frequency for agiven voltage to mix with the incoming signal.This procedure converts the incoming signal tobaseband or an intermediate frequency.

• Phase locked loop (PLL): The PLL ensures thata signal is locked on a specific frequency and canalso be used to generate precise frequencies withfine resolution.

• Channel selection filter: The channel selection fil-ter is used to select the desired channel and toreject the adjacent channels. There are two typesof channel selection filters [52]. The direct conver-

sion receiver uses a low-pass filter for the channelselection. On the other hand, the superheterodyne

receiver adopts a bandpass filter.• Automatic gain control (AGC): The AGC main-

tains the gain or output power level of an ampli-fier constant over a wide range of input signallevels.

In this architecture, a wideband signal is receivedthrough the RF front-end, sampled by the highspeed analog-to-digital (A/D) converter, and mea-surements are performed for the detection of thelicensed user signal. However, there exist some limi-tations on developing the cognitive radio front-end.The wideband RF antenna receives signals from var-ious transmitters operating at different power levels,bandwidths, and locations. As a result, the RF front-end should have the capability to detect a weak sig-nal in a large dynamic range. However, this capabil-ity requires a multi-GHz speed A/D converter withhigh resolution, which might be infeasible [12,13].

The requirement of a multi-GHz speed A/D con-verter necessitates the dynamic range of the signal tobe reduced before A/D conversion. This reductioncan be achieved by filtering strong signals. Sincestrong signals can be located anywhere in the widespectrum range, tunable notch filters are requiredfor the reduction [12]. Another approach is to usemultiple antennas such that signal filtering is per-formed in the spatial domain rather than in the fre-quency domain. Multiple antennas can receivesignals selectively using beamforming techniques[13].

As explained previously, the key challenge of thephysical architecture of the cognitive radio is anaccurate detection of weak signals of licensed usersover a wide spectrum range. Hence, the implementa-tion of RF wideband front-end and A/D converterare critical issues in xG networks.

2.2. Cognitive capability

The cognitive capability of a cognitive radioenables real time interaction with its environmentto determine appropriate communication parame-ters and adapt to the dynamic radio environment.The tasks required for adaptive operation in openspectrum are shown in Fig. 5 [27,46,58], which isreferred to as the cognitive cycle. In this section,we provide an overview of the three main steps ofthe cognitive cycle: spectrum sensing, spectrum anal-

ysis, and spectrum decision. The details and therelated work of these functions are described in Sec-tions 4 and 5.

The steps of the cognitive cycle as shown in Fig. 5are as follows:

1. Spectrum sensing: A cognitive radio monitors theavailable spectrum bands, captures their infor-mation, and then detects the spectrum holes.

Fig. 5. Cognitive cycle.


2. Spectrum analysis: The characteristics of thespectrum holes that are detected through spec-trum sensing are estimated.

3. Spectrum decision: A cognitive radio determinesthe data rate, the transmission mode, and thebandwidth of the transmission. Then, the appro-priate spectrum band is chosen according to thespectrum characteristics and user requirements.

Once the operating spectrum band is determined,the communication can be performed over this spec-trum band. However, since the radio environmentchanges over time and space, the cognitive radioshould keep track of the changes of the radio envi-ronment. If the current spectrum band in usebecomes unavailable, the spectrum mobility functionthat will be explained in Section 6, is performed toprovide a seamless transmission. Any environmen-tal change during the transmission such as primaryuser appearance, user movement, or traffic variationcan trigger this adjustment.

2.3. Reconfigurability

Reconfigurability is the capability of adjustingoperating parameters for the transmission on thefly without any modifications on the hardware com-ponents. This capability enables the cognitive radioto adapt easily to the dynamic radio environment.There are several reconfigurable parameters thatcan be incorporated into the cognitive radio [20]as explained below:

• Operating frequency: A cognitive radio is capableof changing the operating frequency. Based onthe information about the radio environment,

the most suitable operating frequency can bedetermined and the communication can bedynamically performed on this appropriate oper-ating frequency.

• Modulation: A cognitive radio should reconfigurethe modulation scheme adaptive to the userrequirements and channel conditions. For exam-ple, in the case of delay sensitive applications, thedata rate is more important than the error rate.Thus, the modulation scheme that enables thehigher spectral efficiency should be selected. Con-versely, the loss-sensitive applications focus onthe error rate, which necessitate modulationschemes with low bit error rate.

• Transmission power: Transmission power can bereconfigured within the power constraints. Powercontrol enables dynamic transmission power con-figuration within the permissible power limit. Ifhigher power operation is not necessary, the cog-nitive radio reduces the transmitter power to alower level to allow more users to share the spec-trum and to decrease the interference.

• Communication technology: A cognitive radio canalso be used to provide interoperability amongdifferent communication systems.

The transmission parameters of a cognitive radiocan be reconfigured not only at the beginning of atransmission but also during the transmission.According to the spectrum characteristics, theseparameters can be reconfigured such that thecognitive radio is switched to a different spectrumband, the transmitter and receiver parameters arereconfigured and the appropriate communicationprotocol parameters and modulation schemes areused.

3. The xG network architecture

Existing wireless network architectures employheterogeneity in terms of both spectrum policiesand communication technologies [3]. Moreover,some portion of the wireless spectrum is alreadylicensed to different purposes while some bandsremain unlicensed. For the development of commu-nication protocols, a clear description of the xG net-work architecture is essential. In this section, the xGnetwork architecture is presented such that all pos-sible scenarios are considered.

The components of the xG network architecture,as shown in Fig. 6, can be classified in two groups asthe primary network and the xG network. The basic

Fig. 6. xG network architecture.


elements of the primary and the xG network aredefined as follows:

• Primary network: An existing network infrastruc-ture is generally referred to as the primary net-work, which has an exclusive right to a certainspectrum band. Examples include the commoncellular and TV broadcast networks. The compo-nents of the primary network are as follows:– Primary user: Primary user (or licensed user)

has a license to operate in a certain spectrumband. This access can only be controlled bythe primary base-station and should not beaffected by the operations of any other unli-censed users. Primary users do not need anymodification or additional functions for coex-istence with xG base-stations and xG users.

– Primary base-station: Primary base-station (orlicensed base-station) is a fixed infrastructurenetwork component which has a spectrumlicense such as base-station transceiver system(BTS) in a cellular system. In principle, theprimary base-station does not have any xGcapability for sharing spectrum with xG users.However, the primary base-station may berequested to have both legacy and xG proto-

cols for the primary network access of xGusers, which is explained below.

• xG network: xG network (or cognitive radio net-work, Dynamic Spectrum Access network, sec-ondary network, unlicensed network) does nothave license to operate in a desired band. Hence,the spectrum access is allowed only in an oppor-tunistic manner. xG networks can be deployedboth as an infrastructure network and an adhoc network as shown in Fig. 6. The componentsof an xG network are as follows:– xG user: xG user (or unlicensed user, cognitive

radio user, secondary user) has no spectrumlicense. Hence, additional functionalities arerequired to share the licensed spectrum band.

– xG base-station: xG base-station (or unli-censed base-station, secondary base-station)is a fixed infrastructure component with xGcapabilities. xG base-station provides singlehop connection to xG users without spectrumaccess license. Through this connection, an xGuser can access other networks.

– Spectrum broker: Spectrum broker (or sched-uling server) is a central network entity thatplays a role in sharing the spectrum resourcesamong different xG networks. Spectrum


broker can be connected to each network andcan serve as a spectrum information managerto enable coexistence of multiple xG networks[10,32,70].

Fig. 7. xG network on licensed band.

The reference xG network architecture is shownin Fig. 6, which consists of different types of net-works: a primary network, an infrastructure basedxG network, and an ad-hoc xG network. xG net-works are operated under the mixed spectrum envi-ronment that consists of both licensed andunlicensed bands. Also, xG users can either commu-nicate with each other in a multihop manner oraccess the base-station. Thus, in xG networks, thereare three different access types as explained next:

• xG network access: xG users can access their ownxG base-station both on licensed and unlicensedspectrum bands.

• xG ad hoc access: xG users can communicatewith other xG users through ad hoc connectionon both licensed and unlicensed spectrum bands.

• Primary network access: The xG users can alsoaccess the primary base-station through thelicensed band.

According to the reference architecture shown inFig. 6, various functionalities are required to sup-port the heterogeneity in xG networks. In Section3.1, we describe the xG network functions to sup-port the heterogeneity of the network environment.Moreover, in Sections 3.2 and 3.3, we overview xGnetwork applications and existing architectures.

3.1. xG network functions

As explained before, xG network can operate inboth licensed and unlicensed bands. Hence, the func-tionalities required for xG networks vary accordingto whether the spectrum is licensed or unlicensed.Accordingly, in this section, we classify the xG net-work operations as xG network on licensed bandand xG network on unlicensed band. The xG networkfunctions are explained in the following sectionsaccording to this classification.

3.1.1. xG network on licensed band

As shown in Fig. 1, there exist temporally unusedspectrum holes in the licensed spectrum band.Hence, xG networks can be deployed to exploitthese spectrum holes through cognitive communi-cation techniques. This architecture is depicted in

Fig. 7, where the xG network coexists with the pri-mary network at the same location and on the samespectrum band.

There are various challenges for xG networks onlicensed band due to the existence of the primaryusers. Although the main purpose of the xG net-work is to determine the best available spectrum,xG functions in the licensed band are mainly aimedat the detection of the presence of primary users.The channel capacity of the spectrum holes dependson the interference at the nearby primary users.Thus, the interference avoidance with primary usersis the most important issue in this architecture.Furthermore, if primary users appear in the spec-trum band occupied by xG users, xG users shouldvacate the current spectrum band and move to thenew available spectrum immediately, called spec-trum handoff.

3.1.2. xG network on unlicensed band

Open spectrum policy that began in the industrialscientific and medical (ISM) band has caused animpressive variety of important technologies andinnovative uses. However, due to the interferenceamong multiple heterogeneous networks, the spec-trum efficiency of ISM band is decreasing. Ulti-mately, the capacity of open spectrum access, andthe quality of service they can offer, depend on thedegree to which a radio can be designed to allocatespectrum efficiently.

xG networks can be designed for operation onunlicensed bands such that the efficiency isimproved in this portion of the spectrum. The xGnetwork on unlicensed band architecture is illustratedin Fig. 8. Since there are no license holders, allnetwork entities have the same right to access thespectrum bands. Multiple xG networks coexist inthe same area and communicate using the same por-tion of the spectrum. Intelligent spectrum sharing

Fig. 8. xG network on unlicensed band.


algorithms can improve the efficiency of spectrumusage and support high QoS.

In this architecture, xG users focus on detectingthe transmissions of other xG users. Unlike thelicensed band operations, the spectrum handoff isnot triggered by the appearance of other primaryusers. However, since all xG users have the same rightto access the spectrum, xG users should compete witheach other for the same unlicensed band. Thus,sophisticated spectrum sharing methods among xGusers are required in this architecture. If multiplexG network operators reside in the same unlicensedband, fair spectrum sharing among these networksis also required.

3.2. xG network applications

xG networks can be applied to the followingcases:

Leased network: The primary network can pro-vide a leased network by allowing opportunisticaccess to its licensed spectrum with the agreementwith a third party without sacrificing the servicequality of the primary user [56]. For example, theprimary network can lease its spectrum access rightto a mobile virtual network operator (MVNO).Also the primary network can provide its spectrumaccess rights to a regional community for the pur-pose of broadband access.

Cognitive mesh network: Wireless mesh networksare emerging as a cost-effective technology for pro-viding broadband connectivity [4]. However, as thenetwork density increases and the applicationsrequire higher throughput, mesh networks requirehigher capacity to meet the requirements of theapplications. Since the cognitive radio technologyenables the access to larger amount of spectrum,xG networks can be used for mesh networks that

will be deployed in dense urban areas with the pos-sibility of significant contention [38]. For example,the coverage area of xG networks can be increasedwhen a meshed wireless backbone network of infra-structure links is established based on cognitiveaccess points (CAPs) and fixed cognitive relay nodes(CRNs) [6]. The capacity of a CAP, connected via awired broadband access to the Internet, is distrib-uted into a large area with the help of a fixedCRN. xG networks have the ability to add tempo-rary or permanent spectrum to the infrastructurelinks used for relaying in case of high traffic load.

Emergency network: Public safety and emergencynetworks are another area in which xG networks canbe implemented [41]. In the case of natural disasters,which may temporarily disable or destroy existingcommunication infrastructure, emergency personnelworking in the disaster areas need to establish emer-

gency networks. Since emergency networks deal withthe critical information, reliable communicationshould be guaranteed with minimum latency. Inaddition, emergency communication requires a sig-nificant amount of radio spectrum for handling hugevolume of traffic including voice, video and data. xGnetworks can enable the usage of the existing spec-trum without the need for an infrastructure and bymaintaining communication priority and responsetime.

Military network: One of the most interestingpotential applications of an xG network is in a mil-itary radio environment [47]. xG networks canenable the military radios choose arbitrary, interme-diate frequency (IF) bandwidth, modulationschemes, and coding schemes, adapting to the vari-able radio environment of battlefield. Also militarynetworks have a strong need for security and pro-tection of the communication in hostile environ-ment. xG networks could allow military personnelto perform spectrum handoff to find secure spec-trum band for themselves and their allies.

3.3. Existing architectures

The main representative examples of the xG net-work architectures are described in this section.

Spectrum pooling: In [61,62], a centralizedspectrum pooling architecture is proposed basedon orthogonal frequency division multiplexing(OFDM). This architecture consists of an xGbase-station and mobile xG users. OFDM has theadvantage of feeding certain sub-carriers with zerosresulting in no emission of radio power on the

Table 1IEEE 802.22 WRAN system capacity and coverage [65]

RF channel bandwidth 6 MHzAverage spectrum efficiency 3 bit/s/HzChannel capacity 18 Mbit/sSystem capacity per subscriber

(forward)1.5 Mbit/s

System capacity per subscriber(return)

384 kbit/s

Forward/return ratio 3.9Over-subscription ratio 50Number of subscribers per

forward channel600

(FDD operation is assumed tomaximize system capacity for largecoverage distances, TDD would reducecapacity to 75% per TV channel)

Minimum number of subscribers 90 subs.Assumed early take-up rate 3 bit/s/HzPotential number of subscribers 1800 subs.Assumed number of persons

per household2.5 persons

Total number of personsper coverage area

4500 persons

WRAN base station EIRP 98.3 WRadius of coverage for WRAN system 30.7 kmMinimum population density covered 1.5 person/km2

Assuming 1.5 Mbit/s forward and 384 kbit/s return in a 6 MHzchannel with 3 bit/s/Hz and 50:1 over-subscription, each TVchannel can provide service to up to 600 subscribers. The areathat the Wireless Regional Area network (WRAN) operatorneeds to cover should include enough subscribers to make iteconomically viable early in the process when the take-up rate islow. This will define the potential subscriber base and the pop-ulation density for a sustainable business case.


carriers that are occupied by the licensed users. Thelicensed user detection is performed through detec-tion frames that are periodically broadcast by thebase-station. During the detection frames, themobile users perform spectrum sensing. The sensinginformation is then gathered at the base-station.Mobile terminals modulate a complex symbol atmaximum power in the sub-carriers where a licenseduser appears. Through this operation, the base-station receives an amplified signal on all sub-carri-ers with new licensed users. Physical and MAC layerissues, such as detection of spectral access, schedul-ing, and handoff are ongoing investigations in thisarchitecture.

CORVUS: In [8,14], a cognitive radio approachfor usage of virtual unlicensed spectrum (CORVUS)system is presented to exploit unoccupied licensedbands. In CORVUS, based on the local spectrumsensing, the primary user detection and the spec-trum allocation are performed in a coordinatedmanner. This cooperative effort greatly increasesthe system’s ability in identifying and avoiding pri-mary users. In CORVUS, a group of users form asecondary user group (SUG) to coordinate theircommunication. Each member of this group sensesthe spectrum pool, which is divided into sub-chan-nels. A universal control channel is used by allgroups for coordination and separate group controlchannels are used by the members of a group toexchange sensing information and establish second-ary user links. The performance of the physical andlink layers are evaluated through the CORVUS test-bed [44]. Moreover, recently, a reliable link mainte-nance protocol is proposed within CORVUS tomaintain the quality of secondary user communica-tion [64].

IEEE 802.22: IEEE 802.22 is the first worldwidestandard based on the cognitive radio technology[16,31] and is now in the process of standardization.This project, formally called the standard for wire-less regional area networks (WRAN), focuses onconstructing fixed point-to-multipoint WRAN thatwill utilize UHF/VHF TV bands between 54 and862 MHz. Specific TV channels as well as guardbands will be used for communication in IEEE802.22. The IEEE 802.22 system specifies a fixedpoint-to-multipoint wireless air interface wherebya base-station manages its own cell and all associ-ated users, which are denoted as consumer premiseequipments (CPEs). IEEE 802.22 base-station man-ages a unique feature of distributed sensing. This isneeded to ensure proper incumbent protection and

is managed by the base-station, which instructs thevarious CPEs to perform distributed measurementactivities. The IEEE 802.22 system specifies spectralefficiencies in the range of 0.5–5 bit/s/Hz. A distinc-tive feature of IEEE 802.22 WRAN as compared tothe existing IEEE 802 standards is the base-stationcoverage range, which can go up to 100 km if thepower is not an issue. Current specified coveragerange is 33 km at 4 W CPE effective isotropic radi-ated power (EIRP) [66]. Table 1 depicts the capacityand coverage of IEEE 802.22 WRAN system. IEEE802.22 working group was formed in 2004 and hasfinalized the specification of technical requirements.The first draft of IEEE 802.22 standard will beready around mid 2006.

DIMSUMnet: The dynamic intelligent manage-ment of spectrum for ubiquitous mobile network(DIMSUMnet) [10] implements statistically multi-


plexed access (SMA) to spectrum in the coordinatedaccess band (CAB). While the CAB improves thespectrum access efficiency and fairness, the SMAis focused on improving the spectrum utilization.CAB is a contiguous chunk of spectrum reservedby regulating authorities. A spectrum brokerpermanently owns the CAB and leases it accordingto requests. DIMSUMnet uses a centralized, regio-nal network level brokering mechanism that aimsto significantly improve spectrum utilization whilereducing the complexity and the agility require-ments of the deployed system. The base-station reg-isters with its designated radio access networkmanager (RANMAN), which negotiates a leasewith a spectrum information and management(SPIM) broker for an appropriate portion of thespectrum. If the lease is successfully obtained, theRANMAN configures the leased spectrum inthe base-station. The base-station sends the spec-trum information received from the RANMAN toits users for the configuration of client. The spec-trum utilization of DIMSUMnet is currently mea-sured in existing Code Division Multiple Access(CDMA) and Global System for Mobile communi-cation (GSM) cellular networks, aimed at character-izing feasibility of CAB and SMA [35]. Recent workfocuses on the spectrum pricing and allocation func-tions for spectrum brokers [11].

DRiVE/OverDRiVE project: The EuropeanDynamic Radio for IP Services in Vehicular Envi-ronments (DRiVE) project focuses on dynamicspectrum allocation in heterogeneous networks byassuming a common coordinated channel [75]. Thefollow-up project, Spectrum Efficient Uni- and Mul-ticast Over Dynamic Radio Networks in VehicularEnvironments (OverDRiVE) aims at UMTSenhancements and coordination of existing radionetworks into a hybrid network to ensure spectrumefficient provision of mobile multimedia services[26]. Two aspects of dynamic spectrum allocationwere investigated in DRiVE/OverDRiVE, i.e.,temporal dynamic spectrum allocation (DSA), andspatial DSA [39]. In the case of the temporalDSA, a radio access network (RAN) can use thespectrum that is not currently being used by otherRANs, at that time. On the other hand, spatialDSA allows spectrum allocations to adapt to regio-nal fluctuations in traffic demands. The efficiency ofthese DSA schemes depends on the ability to predictthe traffic load. Although these projects have shownsignificant potential for increasing spectral effi-ciency, the reconfigurable system implementation

for temporal and spatial DSA is still a majorchallenge.

Nautilus: Nautilus project is designed to empha-size distributed coordination enabled spectrum shar-ing, without relying on centralized control [48]. Inthe Nautilus project, a distributed, scalable and effi-cient coordination framework for open spectrum adhoc networks is proposed, which accounts for spec-trum heterogeneity while not relying on the existenceof a pre-defined common channel for control traffic[73,74]. Based on this framework, three different col-laborative spectrum access schemes are presented. In[73], a graph coloring based collaborative spectrumaccess scheme is proposed, where a topology-opti-mized allocation algorithm is used for the fixedtopology. In mobile networks, however, the networktopology changes due to the node mobility. Usingthis global optimization approach, the networkneeds to completely recompute spectrum assign-ments for all users after each change, resulting inhigh computational and communication overhead.Thus, distributed spectrum allocation based on localbargaining is proposed in [15], where mobile usersnegotiate spectrum assignment within local self-organized groups. For the resource constrained net-works such as sensor and ad hoc network, rule-baseddevice centric spectrum management is proposed,where unlicensed users access the spectrum indepen-dently according to both local observation and pre-determined rules. Currently, this project focuses onselecting the best channel for data transmissionusing proposed distributed coordination framework.

OCRA network: In [5], an OFDM-based cogni-tive radio (OCRA) network is proposed. OCRAnetwork considers all possible deployment scenariosover the heterogeneous xG network environmentand develops cross-layer operations for the OFDMbased dynamic spectrum access. OCRA networkarchitecture and its components are shown inFig. 6. For the spectrum decision and the spectrumhandoff, OCRA network provides a novel conceptof an OFDM-based spectrum management overthe heterogeneous spectrum environment. Basedon this physical layer (PHY) structure, a dual-modespectrum sharing framework is proposed, whichenables access to existing networks as well as coor-dination between xG users. Furthermore, a newrouting paradigm that considers joint re-routingand spectrum handoff is proposed. Moreover,OCRA network introduces multi-spectrum trans-port techniques to exploit the available butnon-contiguous wireless spectrum for high quality


communication. In [5], the testbed design for evalu-ation and integration of the OCRA network is pro-posed. The OCRA testbed is based on IEEE802.11a/g technology, which exploits the OFDMtechnology. Moreover, a standalone cognitive sens-ing unit is developed to emulate the spectrum sens-ing capabilities of the cognitive radio.

4. Spectrum sensing

An important requirement of the xG network isto sense the spectrum holes. As explained in Section2, a cognitive radio is designed to be aware of andsensitive to the changes in its surrounding. The spec-trum sensing function enables the cognitive radio toadapt to its environment by detecting spectrumholes.

The most efficient way to detect spectrum holes isto detect the primary users that are receiving datawithin the communication range of an xG user. Inreality, however, it is difficult for a cognitive radioto have a direct measurement of a channel betweena primary receiver and a transmitter. Thus, the mostrecent work focuses on primary transmitter detec-tion based on local observations of xG users.

Generally, the spectrum sensing techniques canbe classified as transmitter detection, coopera-tive detection, and interference-based detection, asshown in Fig. 9. In the following sections, wedescribe these spectrum sensing methods for xG net-works and discuss the open research topics in thisarea.

4.1. Transmitter detection (non-cooperative

detection)

The cognitive radio should distinguish betweenused and unused spectrum bands. Thus, the cognitiveradio should have capability to determine if a signal

Fig. 9. Classification of spectrum sensing techniques.

from primary transmitter is locally present in a cer-tain spectrum. Transmitter detection approach isbased on the detection of the weak signal from a pri-mary transmitter through the local observations ofxG users. Basic hypothesis model for transmitterdetection can be defined as follows [25]:

xðtÞ ¼nðtÞ H 0;

hsðtÞ þ nðtÞ H 1;

(ð1Þ

where x(t) is the signal received by the xG user, s(t)is the transmitted signal of the primary user, n(t) isthe AWGN and h is the amplitude gain of the chan-nel. H0 is a null hypothesis, which states that there isno licensed user signal in a certain spectrum band.On the other hand, H1 is an alternative hypothesis,which indicates that there exist some licensed usersignal.

Three schemes are generally used for the trans-mitter detection according to the hypothesis model[53]. In the following subsections, we investigatematched filter detection, energy detection and cyclo-

stationary feature detection techniques proposedfor transmitter detection in xG networks.

4.1.1. Matched filter detection

When the information of the primary user signalis known to the xG user, the optimal detector in sta-tionary Gaussian noise is the matched filter since itmaximizes the received signal-to-noise ratio (SNR)[53]. While the main advantage of the matched filteris that it requires less time to achieve high process-ing gain due to coherency, it requires a priori know-ledge of the primary user signal such as themodulation type and order, the pulse shape, andthe packet format. Hence, if this information isnot accurate, then the matched filter performspoorly. However, since most wireless network sys-tems have pilot, preambles, synchronization wordor spreading codes, these can be used for the coher-ent detection.

4.1.2. Energy detection

If the receiver cannot gather sufficient informa-tion about the primary user signal, for example, ifthe power of the random Gaussian noise is onlyknown to the receiver, the optimal detector is anenergy detector [53]. In order to measure the energyof the received signal, the output signal of bandpassfilter with bandwidth W is squared and integratedover the observation interval T. Finally, the outputof the integrator, Y, is compared with a threshold, k,


to decide whether a licensed user is present or not[17].

If the energy detection can be applied in a non-fading environment where h is the amplitude gainof the channel as shown in (1), the probability ofdetection Pd and false alarm Pf are given as follows[17]:

P d ¼ PfY > kjH 1g ¼ Qm

ffiffiffiffiffi2c

p;ffiffiffikp� �

; ð2Þ

P f ¼ PfY > kjH og ¼Cðm; k=2Þ

CðmÞ ; ð3Þ

where c is the SNR, u = TW is the time bandwidthproduct, C(Æ) and C(Æ, Æ) are complete and incompletegamma functions and Qm( ) is the generalized Mar-cum Q-function. From the above functions, whilea low Pd would result in missing the presence ofthe primary user with high probability which in turnincreases the interference to the primary user, a highPf would result in low spectrum utilization sincefalse alarms increase the number of missed opportu-nities. Since it is easy to implement, the recent workon detection of the primary user has generallyadopted the energy detector [23,53].

In [25], the shadowing and the multi-path fadingfactors are considered for the energy detector. Inthis case, while Pf is independent of C, when theamplitude gain of the channel, h, varies due tothe shadowing/fading, Pd gives the probability ofthe detection conditioned on instantaneous SNRas follows:

P d ¼Z

xQm

ffiffiffiffiffi2c

p;ffiffiffikp� �

fcðxÞdx; ð4Þ

where fc(x) is the probability distribution functionof SNR under fading.

However, the performance of energy detector issusceptible to uncertainty in noise power. In orderto solve this problem, a pilot tone from the pri-mary transmitter is used to help improve theaccuracy of the energy detector in [53]. Anothershortcoming is that the energy detector cannot dif-ferentiate signal types but can only determine thepresence of the signal. Thus, the energy detector isprone to the false detection triggered by the unin-tended signals.

4.1.3. Cyclostationary feature detectionAn alternative detection method is the cyclosta-

tionary feature detection [12,22,57]. Modulated sig-nals are in general coupled with sine wave carriers,pulse trains, repeating spreading, hopping sequen-

ces, or cyclic prefixes, which result in built-in period-icity. These modulated signals are characterized ascyclostationarity since their mean and autocorrela-tion exhibit periodicity. These features are detectedby analyzing a spectral correlation function. Themain advantage of the spectral correlation func-tion is that it differentiates the noise energy frommodulated signal energy, which is a result of the factthat the noise is a wide-sense stationary signal withno correlation, while modulated signals are cyclosta-tionary with spectral correlation due to the embed-ded redundancy of signal periodicity. Therefore, acyclostationary feature detector can perform betterthan the energy detector in discriminating againstnoise due to its robustness to the uncertainty in noisepower [57]. However, it is computationally complexand requires significantly long observation time.

For more efficient and reliable performance, theenhanced feature detection scheme combining cyclicspectral analysis with pattern recognition based onneural networks is proposed in [22]. Distinct fea-tures of the received signal are extracted using cyclicspectral analysis and represented by both spectralcoherent function and spectral correlation densityfunction. The neural network, then, classifies signalsinto different modulation types.

4.2. Cooperative detection

The assumption of the primary transmitter detec-tion is that the locations of the primary receivers areunknown due to the absence of signalling betweenprimary users and the xG users. Therefore, the cog-nitive radio should rely on only weak primary trans-mitter signals based on the local observation of thexG user [72,73]. However, in most cases, an xGnetwork is physically separated from the primarynetwork so there is no interaction between them.Thus, with the transmitter detection, the xG usercannot avoid the interference due to the lack ofthe primary receiver’s information as depicted inFig. 10(a). Moreover, the transmitter detectionmodel cannot prevent the hidden terminal problem.An xG transmitter can have a good line-of-sight to areceiver, but may not be able to detect the transmit-ter due to the shadowing as shown in Fig. 10(b).Consequently, the sensing information from otherusers is required for more accurate detection.

In the case of non-cooperative detection ex-plained in Section 4.1, the xG users detect the pri-mary transmitter signal independently through theirlocal observations. Cooperative detection refers to

Fig. 10. Transmitter detection problem: (a) Receiver uncertainty and (b) shadowing uncertainty.


spectrum sensing methods where information frommultiple xG users are incorporated for primary userdetection. Cooperative detection can be imple-mented either in a centralized or in a distributedmanner [23,71]. In the centralized method, the xGbase-station plays a role to gather all sensing infor-mation from the xG users and detect the spectrumholes. On the other hand, distributed solutionsrequire exchange of observations among xG users.

Cooperative detection among unlicensed users istheoretically more accurate since the uncertainty ina single user’s detection can be minimized [25].Moreover, the multi-path fading and shadowingeffect are the main factors that degrade the perfor-mance of primary user detection methods [44].However, cooperative detection schemes allow tomitigate the multi-path fading and shadowingeffects, which improves the detection probability ina heavily shadowed environment [25].

In [55], the limitation of non-cooperative spec-trum sensing approaches is investigated. Generally,the data transmission and sensing function are co-located in a single xG user device. However, thisarchitecture can result in suboptimal spectrum deci-sion due to possible conflicts between data transmis-sion and sensing. In order to solve this problem, in[55], two distinct networks are deployed separately,i.e., the sensor network for cooperative spectrumsensing and the operational network for data trans-mission. The sensor network is deployed in thedesired target area and senses the spectrum. A cen-tral controller processes the spectrum informationcollected from sensors and makes the spectrumoccupancy map for the operational network. Theoperational network uses this information to deter-mine the available spectrum.

While cooperative approaches provide moreaccurate sensing performance, they cause adverseeffects on resource-constrained networks due tothe additional operations and overhead traffic.Furthermore, the primary receiver uncertaintyproblem caused by the lack of the primary receiverlocation knowledge is still unsolved in the coopera-tive sensing. In the following section, we explaininterference-based detection methods, which aimto address these problems.

4.3. Interference-based detection

Interference is typically regulated in a transmit-ter-centric way, which means interference can becontrolled at the transmitter through the radiatedpower, the out-of-band emissions and location ofindividual transmitters. However, interference actu-ally takes place at the receivers, as shown inFig. 10(a) and (b).

Therefore recently, a new model for measuringinterference, referred to as interference temperature

shown in Fig. 11 has been introduced by the FCC[21]. The model shows the signal of a radio stationdesigned to operate in a range at which the receivedpower approaches the level of the noise floor. Asadditional interfering signals appear, the noise floorincreases at various points within the service area, asindicated by the peaks above the original noisefloor. Unlike the traditional transmitter-centricapproach, the interference temperature modelmanages interference at the receiver through theinterference temperature limit, which is representedby the amount of new interference that the receivercould tolerate. In other words, the interference tem-perature model accounts for the cumulative RF

Power atReceiver

Original Noise Floor

Interference Temperature Limit

Licensed Signal

New Opportunitiesfor

Minimum ServiceRange with

Interference Cap

Service Range atOriginal Noise Floorriginal Noise FloorPower at

Receiver

Original Noise Floor

Interference Temperature Limit

Licensed Signal

New Opportunitiesfor Spectrum Access

Minimum ServiceRange with

Interference Cap

Service Range atO

Fig. 11. Interference temperature model [21].


energy from multiple transmissions and sets a max-imum cap on their aggregate level. As long as xGusers do not exceed this limit by their transmissions,they can use this spectrum band.

However, there exist some limitations in measur-ing the interference temperature. In [9], the interfer-ence is defined as the expected fraction of primaryusers with service disrupted by the xG operations.This method considers factors such as the type ofunlicensed signal modulation, antennas, ability todetect active licensed channels, power control, andactivity levels of the licensed and unlicensed users.However, this model describes the interference dis-rupted by a single xG user and does not considerthe effect of multiple xG users. In addition, if xGusers are unaware of the location of the nearby pri-mary users, the actual interference cannot bemeasured using this method.

In [63], a direct receiver detection method is pre-sented, where the local oscillator (LO) leakage poweremitted by the RF front-end of the primary receiveris exploited for the detection of primary receivers. Inorder to detect the LO leakage power, low-costsensor nodes can be mounted close to the primaryreceivers. The sensor nodes detect the leakage LOpower to determine the channel used by the primaryreceiver and this information is used by the unli-censed users to determine the operation spectrum.

4.4. Spectrum sensing challenges

There exist several open research challenges thatneed to be investigated for the development of thespectrum sensing function.

• Interference temperature measurement: The diffi-culty of this receiver detection model lies in effec-

tively measuring the interference temperature.An xG user is naturally aware of its transmitpower level and its precise location with the helpof a positioning system. With this ability, however,its transmission could cause significant interfer-ence at a neighboring receiver on the same fre-quency. However, currently, there exists nopractical way for a cognitive radio to measure orestimate the interference temperature at nearbyprimary receivers. Since primary receivers are usu-ally passive devices, an xG user cannot be aware ofthe precise locations of primary receivers. Further-more, if xG users cannot measure the effect of theirtransmission on all possible receivers, a usefulinterference temperature measurement may notbe feasible.

• Spectrum sensing in multi-user networks: Usually,xG networks reside in a multi-user environmentwhich consists of multiple xG users and primaryusers. Furthermore, the xG networks can alsobe co-located with other xG networks competingfor the same spectrum band. However, currentinterference models [9,21] do not consider theeffect of multiple xG users. Multi-user environ-ment makes it more difficult to sense the primaryusers and to estimate the actual interference.Hence, spectrum sensing functions should bedeveloped considering the possibility of multi-user/network environment. In order to solve themulti-user problem, the cooperative detectionschemes can be considered, which exploit the spa-tial diversity inherent in a multi-user network.

• Detection capability: One of the main require-ments of xG networks is the detection of theprimary users in a very short time [24,53].OFDM-based xG networks are known to beexcellent fit for the physical architecture of xG


networks [5,57,62]. Since multi-carrier sensingcan be exploited in OFDM-based xG networks,the overall sensing time can be reduced. Once aprimary user is detected in a single carrier, sens-ing in other carriers is not necessary. In [57], apower-based sensing algorithm in OFDM net-works is proposed for detecting the presence ofa primary user. It is shown that the overall detec-tion time is reduced by collecting informationfrom each carrier. However, this necessitatesthe use of a large number of carriers, whichincreases the design complexity. Hence, novelspectrum sensing algorithms need to be devel-oped such that the number of samples neededto detect the primary user is minimized within agiven detection error probability.

5. Spectrum management

In xG networks, the unused spectrum bands willbe spread over wide frequency range including bothunlicensed and licensed bands. These unused spec-trum bands detected through spectrum sensingshow different characteristics according to not onlythe time varying radio environment but also thespectrum band information such as the operatingfrequency and the bandwidth.

Since xG networks should decide on the bestspectrum band to meet the QoS requirements overall available spectrum bands, new spectrum man-agement functions are required for xG networks,considering the dynamic spectrum characteristics.We classify these functions as spectrum sensing,spectrum analysis, and spectrum decision. Whilespectrum sensing, which is discussed in Section 4,is primarily a PHY layer issue, spectrum analysis

and spectrum decision are closely related to theupper layers. In this section, spectrum analysis andspectrum decision are investigated.

5.1. Spectrum analysis

In xG networks, the available spectrum holesshow different characteristics which vary over time.Since the xG users are equipped with the cognitiveradio based physical layer, it is important to under-stand the characteristics of different spectrumbands. Spectrum analysis enables the characteriza-tion of different spectrum bands, which can beexploited to get the spectrum band appropriate tothe user requirements.

In order to describe the dynamic nature of xGnetworks, each spectrum hole should be character-ized considering not only the time-varying radioenvironment and but also the primary user activityand the spectrum band information such as operat-ing frequency and bandwidth. Hence, it is essentialto define parameters such as interference level,channel error rate, path-loss, link layer delay, andholding time that can represent the quality of a par-ticular spectrum band as follows:

• Interference: Some spectrum bands are morecrowded compared to others. Hence, the spectrumband in use determines the interference character-istics of the channel. From the amount of theinterference at the primary receiver, the permissi-ble power of an xG user can be derived, which isused for the estimation of the channel capacity.

• Path loss: The path loss increases as the operatingfrequency increases. Therefore, if the transmis-sion power of an xG user remains the same, thenits transmission range decreases at higher fre-quencies. Similarly, if transmission power isincreased to compensate for the increased pathloss, then this results in higher interference forother users.

• Wireless link errors: Depending on the modula-tion scheme and the interference level of the spec-trum band, the error rate of the channel changes.

• Link layer delay: To address different path loss,wireless link error, and interference, differenttypes of link layer protocols are required at dif-ferent spectrum bands. This results in differentlink layer packet transmission delay.

• Holding time: The activities of primary users canaffect the channel quality in xG networks. Hold-ing time refers to the expected time duration thatthe xG user can occupy a licensed band beforegetting interrupted. Obviously, the longer theholding time, the better the quality would be.Since frequent spectrum handoff can decreasethe holding time, previous statistical patterns ofhandoff should be considered while designingxG networks with large expected holding time.

Channel capacity, which can be derived from theparameters explained above, is the most importantfactor for spectrum characterization. Usually SNRat the receiver has been used for the capacity estima-tion. However, since SNR considers only localobservations of xG users, it is not enough to avoidinterference at the primary users. Thus, spectrum


characterization is focused on the capacity estima-tion based on the interference at the licensed receiv-ers. The interference temperature model [21] givenin Section 4.3 can be exploited for this approach.The interference temperature limit indicates anupper bound or cap on the potential RF energy thatcould be introduced into the band. Consequently,using the amount of permissible interference, themaximum permissible transmission power of anxG user can be determined.

In [63], a spectrum capacity estimation methodhas been proposed that considers the bandwidthand the permissible transmission power. Accord-ingly, the spectrum capacity, C, can be estimatedas follows:

C ¼ B log 1þ SN þ I

� �; ð5Þ

where B is the bandwidth, S is the received signalpower from the xG user, N is the xG receiver noisepower, and I is the interference power received atthe xG receiver due to the primary transmitter.

Estimating spectrum capacity has also beeninvestigated in the context of OFDM-based cogni-tive radio systems in [57]. Accordingly, the spectrumcapacity of the OFDM-based xG networks isdefined as follows [57]:

C ¼Z

X

1

2log2 1þ Gðf ÞS0

N 0

� �df ; ð6Þ

where X is the collection of unused spectrum seg-ments, G(f) is the channel power gain at frequencyf, S0 and N0 are the signal and noise power per unitfrequency, respectively.

The recent work on spectrum analysis, as dis-cussed above, only focuses on spectrum capacityestimation. However, besides the capacity, otherfactors such as delay, link error rate, and holdingtime also have significant influence on the qualityof services. Moreover, the capacity is closely relatedto both interference level and path loss. However, acomplete analysis and modeling of spectrum in xGnetworks is yet to be developed. In order to decideon the appropriate spectrum for different types ofapplications, it is desirable and an open researchissue to identify the spectrum bands combining allcharacterization parameters described above.

5.2. Spectrum decision

Once all available spectrum bands are character-ized, appropriate operating spectrum band should

be selected for the current transmission consideringthe QoS requirements and the spectrum characteris-tics. Thus, the spectrum management function mustbe aware of user QoS requirements.

Based on the user requirements, the data rate,acceptable error rate, delay bound, the transmissionmode, and the bandwidth of the transmission can bedetermined. Then, according to the decision rule, theset of appropriate spectrum bands can be chosen. In[73], five spectrum decision rules are presented,which are focused on fairness and communicationcost. However, this method assumes that all chan-nels have similar throughput capacity. In [36], anopportunistic frequency channel skipping protocolis proposed for the search of better quality channel,where this channel decision is based on SNR. Inorder to consider the primary user activity, the num-ber of spectrum handoff, which happens in a certainspectrum band, is used for spectrum decision [37].Spectrum decision constitutes rather important butyet unexplored issues in xG networks, which are pre-sented in the following subsection.

5.3. Spectrum management challenges

There exist several open research issues that needto be investigated for the development of spectrumdecision function.

• Decision model: Signal to noise ratio (SNR) is notsufficient to characterize the spectrum band in xGnetworks. Besides the SNR, many spectrum char-acterization parameters would affect the quality,as investigated in Section 5.1. Thus, how tocombine these spectrum characterization para-meters for the spectrum decision model is stillan open issue. Moreover, in OFDM based xGnetworks, multiple spectrum bands can be simul-taneously used for the transmission. For thesereasons, a decision framework for the multiplespectrum bands should be provided.

• Multiple spectrum band decision: In xG networks,multiple spectrum bands can be simultaneouslyused for the transmission. Moreover, the xG net-works do not require the selected multiple bandsto be contiguous. Thus, an xG user can sendpackets over non-contiguous spectrum bands.This multi-spectrum transmission shows lessquality degradation during the spectrum handoffcompared to the conventional transmission onsingle spectrum band [5]. For example, if a pri-mary user appears in a particular spectrum band,


the xG user has to vacate this band. However,since the rest of spectrum bands will maintainthe communication, abrupt service quality degra-dation can be mitigated. In addition, transmis-sion in multiple spectrum bands allows lowerpower to be used in each spectrum band. As aresult, less interference with primary users isachieved, compared to the transmission on singlespectrum band [5]. For these reasons, a spectrummanagement framework should support multiplespectrum decision capabilities. For example, howto determine the number of spectrum bands andhow to select the set of appropriate bands are stillopen research issues in xG networks.

• Cooperation with reconfiguration: The cognitiveradio technology enables the transmissionparameters of a radio to be reconfigured for opti-mal operation in a certain spectrum band. Forexample, if SNR is fixed, the bit error rate(BER) can be adjusted to maintain the channelcapacity by exploiting adaptive modulation tech-niques, e.g., cdma2000 1x EVDO [1,18]. Hence, acooperative framework that considers both spec-trum decision and reconfiguration is required.

• Spectrum decision over heterogenous spectrum

bands: Currently, certain spectrum bands arealready assigned to different purposes while somebands remain unlicensed. Thus, the spectrumused by xG networks will most likely be a combi-nation of exclusively accessed spectrum and unli-censed spectrum. In case of licensed bands, the xGusers need to consider the activities of primaryusers in spectrum analysis and decision in ordernot to influence the primary user transmission.Conversely, in unlicensed bands, since all the xGusers have the same spectrum access rights,sophisticated spectrum sharing techniques arenecessary. In order to decide the best spectrumband over this heterogeneous environment, xGnetwork should support spectrum decisionoperations on both the licensed and the unli-censed bands considering these different charac-teristics.

6. Spectrum mobility

xG networks target to use the spectrum in adynamic manner by allowing the radio terminals,known as the cognitive radio, to operate in the bestavailable frequency band. This enables ‘‘Get the

Best Available Channel’’ concept for communication

purposes. To realize the ‘‘Get the Best Available

Channel’’ concept, an xG radio has to capture thebest available spectrum. Spectrum mobility isdefined as the process when an xG user changesits frequency of operation. In the following sections,we describe the spectrum handoff concept in xG net-works and discuss open research issues in this newarea.

6.1. Spectrum handoff

In xG networks, spectrum mobility arises whencurrent channel conditions become worse or a pri-mary user appears. Spectrum mobility gives rise toa new type of handoff in xG networks that we referto as spectrum handoff. The protocols for differentlayers of the network stack must adapt to the chan-nel parameters of the operating frequency. More-over, they should be transparent to the spectrum

handoff and the associated latency.As pointed out in earlier sections, a cognitive

radio can adapt to the frequency of operation.Therefore, each time an xG user changes its fre-quency of operation, the network protocols aregoing to shift from one mode of operation toanother. The purpose of spectrum mobility manage-ment in xG networks is to make sure that such tran-sitions are made smoothly and as soon as possiblesuch that the applications running on an xG userperceive minimum performance degradation duringa spectrum handoff. It is essential for the mobilitymanagement protocols to learn in advance aboutthe duration of a spectrum handoff. This informationshould be provided by the sensing algorithm. Oncethe mobility management protocols learn about thislatency, their job is to make sure that the ongoingcommunications of an xG user undergo only mini-mum performance degradation.

Consequently, multi-layer mobility managementprotocols are required to accomplish the spectrummobility functionalities. These protocols supportmobility management adaptive to different typesof applications. For example, a TCP connectioncan be put to a wait state until the spectrum handoffis over. Moreover, since the TCP parameters willchange after a spectrum handoff, it is essential tolearn the new parameters and ensure that the transi-tion from the old parameters to new parameters arecarried out rapidly. For a data communication e.g.,FTP, the mobility management protocols shouldimplement mechanisms to store the packets thatare transmitted during a spectrum handoff, whereas


for a real-time application there is no need to storethe packets as the stored packets, if delivered later,will be stale packets and can not be used by the cor-responding application.

6.2. Spectrum mobility challenges in xG networks

The followings are the open research issues forefficient spectrum mobility in xG networks.

• At a particular time, several frequency bandsmay be available for an xG user. Algorithmsare required to decide the best available spectrumbased on the channel characteristics of the avail-able spectrum and the requirements of the appli-cations that are being used by an xG user.

• Once, the best available spectrum is selected, thenext challenge is to design new mobility and con-nection management approaches to reduce delayand loss during spectrum handoff.

• When the current operational frequency becomesbusy (this may happen if a licensed user starts touse this frequency) in the middle of a communi-cation by an xG user, then applications runningon this node have to be transferred to anotheravailable frequency band. However, the selectionof new operational frequency may take time.Novel algorithms are required to ensure thatapplications do not suffer from severe perfor-mance degradation during such transitions.

• Spectrum handoff may occur due to reasonsother than the detection of the primary user.Thus, there exist various other spectrum handoffschemes in xG networks. If an xG user movesfrom one place to another, spectrum handoffmay occur just because the available spectrumbands change. Thus the desired spectrum handoffscheme should integrate inter-cell handoff. Apartfrom this, spectrum handoff between differentnetworks, referred to as vertical handoff is alsolikely to occur in xG networks. Under such adiverse environment, it is essential that spectrumhandoff scheme takes all the above mentionedpossibilities into consideration.

• Spectrum mobility in time domain: xG networksadapt to the wireless spectrum based on availablebands on the spectrum. Since these availablechannels change over time, enabling QoS in thisenvironment is challenging. The physical radioshould ‘‘move’’ through the spectrum to meetthe QoS requirements.

• Spectrum mobility in space: The available bandsalso change as a user moves from one place toanother. Hence, continuous allocation of spec-trum is a major challenge. in xG networks.

7. Spectrum sharing

In xG networks, one of the main challenges inopen spectrum usage is the spectrum sharing. Spec-trum sharing can be regarded to be similar to gen-eric medium access control (MAC) problems inexisting systems. However, as we will investigate inthis section, substantially different challenges existfor spectrum sharing in xG networks. The coexis-tence with licensed users and the wide range ofavailable spectrum are two of the main reasons forthese unique challenges. In this section, we delveinto the specific challenges for spectrum sharing inxG networks, overview the existing solutions anddiscuss open research areas.

In order to provide a directory for different chal-lenges during spectrum sharing, we first enumeratethe steps in spectrum sharing in xG networks. Thechallenges and the solutions proposed for thesesteps will then be explained in detail. The spectrumsharing process consists of five major steps.

1. Spectrum sensing: An xG user can only allocate aportion of the spectrum if that portion is not usedby an unlicensed user. In Section 4, the solutionsand the challenges for this problem, i.e., spec-trum sensing, are described. Accordingly, whenan xG node aims to transmit packets, it firstneeds to be aware of the spectrum usage aroundits vicinity.

2. Spectrum allocation: Based on the spectrumavailability, the node can then allocate a channel.This allocation not only depends on spectrumavailability, but it is also determined based oninternal (and possibly external) policies. Hence,the design of a spectrum allocation policy toimprove the performance of a node is an impor-tant research topic.

3. Spectrum access: In this step, another majorproblem of spectrum sharing comes into picture.Since there may be multiple xG nodes trying toaccess the spectrum, this access should also becoordinated in order to prevent multiple userscolliding in overlapping portions of the spectrum.

4. Transmitter-receiver handshake: Once a portion ofthe spectrum is determined for communication,


the receiver of this communication should also beindicated about the selected spectrum. Hence, atransmitter-receiver handshake protocol is essen-tial for efficient communication in xG networks.Note that the term handshake by no meansrestricts this protocol between the transmitterand the receiver. A third party such as a central-ized station can also be involved.

5. Spectrum mobility: xG nodes are regarded as‘‘visitors’’ to the spectrum they allocate. Hence,if the specific portion of the spectrum in use isrequired by a licensed user, the communica-tion needs to be continued in another vacantportion. As a result, spectrum mobility is alsoimportant for successful communication betweenxG nodes.

The existing work in spectrum sharing in xG net-works aims to provide solutions for each stepexplained above. The existing solutions constitutea rich literature for spectrum sharing in xG net-works. In Section 7.1, we classify the spectrum shar-ing techniques and describe the fundamental resultsabout these techniques in xG networks. These workprovide insight about how a spectrum sharing pro-tocol can be designed. Accordingly, in Sections 7.2and 7.3, we overview the solutions for spectrumsharing among multiple coexisting xG networks(inter-network spectrum sharing), and inside an xGnetwork (intra-network spectrum sharing), respec-tively. Finally, in Section 7.4, the open researchissues for spectrum sharing in xG networks arediscussed.

7.1. Overview of spectrum sharing techniques

The existing solutions for spectrum sharing in xGnetworks can be mainly classified in three aspects:i.e., according to their architecture assumption, spec-

trum allocation behavior, and spectrum access tech-

nique as shown in Fig. 12. In this section, wedescribe these three classifications and present thefundamental results that analyze these classifica-

Fig. 12. Classification of spectrum sharing in xG networks based ontechnique.

tions. The analysis of xG spectrum sharing tech-niques has been investigated through two majortheoretical approaches. While some work uses opti-mization techniques to find the optimal strategiesfor spectrum sharing, game theoretical analysishas also been used in this area.

The first classification for spectrum sharing tech-niques in xG networks is based on the architecture,which can be described as follows:

• Centralized spectrum sharing: In these solutions, acentralized entity controls the spectrum alloca-tion and access procedures [7,51,70]. With aidto these procedures, generally, a distributed sens-ing procedure is proposed such that each entity inthe xG network forward their measurementsabout the spectrum allocation to the centralentity and this entity constructs a spectrum allo-cation map.

• Distributed spectrum sharing: Distributed solu-tions are mainly proposed for cases where theconstruction of an infrastructure is not preferable[15,29,40,54,71–73]. Accordingly, each node isresponsible for the spectrum allocation andaccess is based on local (or possibly global)policies.

The second classification for spectrum sharingtechniques in xG networks is based on the accessbehavior. More specifically, the spectrum accesscan be cooperative or non-cooperative as explainedbelow:

• Cooperative spectrum sharing: Cooperative (orcollaborative) solutions consider the effect ofthe node’s communication on other nodes[7,15,29,40,71]. In other words, the interferencemeasurements of each node are shared amongother nodes. Furthermore, the spectrum alloca-tion algorithms also consider this information.While all the centralized solutions can be regardedas cooperative, there also exist distributed cooper-ative solutions.

architecture, spectrum allocation behavior, and spectrum access


• Non-cooperative spectrum sharing: Contrary tothe cooperative solutions, non-cooperative (ornon-collaborative, selfish) solutions consideronly the node at hand [54,72,73]. These solutionsare also referred to as selfish. While non-cooper-ative solutions may result in reduced spectrumutilization, the minimal communication require-ments among other nodes introduce a tradeofffor practical solutions.

These two solutions have generally been com-pared through their spectrum utilization, fairness,and throughput. The utilization and fairness in spec-trum access has been investigated in [50], where thespectrum allocation problem is modeled as a graphcoloring problem and both centralized and distrib-uted approaches are investigated. Using this model,an optimization framework is developed. In thisframework, secondary users allocate channelsaccording to the interference that will be caused bythe transmission. Both cooperative and non-cooper-ative approaches are considered such that coopera-tive approaches also consider the effect of thechannel allocation on the potential neighbors. Thesimulation results show that cooperative approachesoutperform non-cooperative approaches as well asclosely approximating the global optimum. More-over, the comparison of centralized and distributedsolutions reveals that distributed solution closely fol-lows the centralized solution. A similar analysis hasalso been provided in [74], where the effects of col-laboration in spectrum access is investigated. Animportant assumption in these work is that second-ary users know the location and transmit power ofprimary users so that the interference calculationscan be performed easily. However, such an assump-tion may not always be valid in xG networks.

Game theory has also been exploited for perfor-mance evaluation of xG spectrum access schemes.Especially, the comparison between cooperativeand non-cooperative approaches has been presentedin [49] through game theoretical analysis. In [49],game theory is exploited to analyze the behaviorof the cognitive radio for distributed adaptive chan-nel allocation. It is assumed that users deployCDMA and determine the operating channel andthe coding rate by keeping transmission power con-stant. It is shown that the cooperative case can bemodeled as an exact potential game, which con-verges to a pure strategy Nash equilibrium solution.However, this framework has been shown not to beapplicable for non-cooperative spectrum sharing

and a learning algorithm has been proposed. Theevaluations reveal that Nash equilibrium point forcooperative users is reached quickly and results ina certain degree of fairness as well as improvedthroughput. On the other hand, the learningalgorithm for non-cooperative users converge to amixed strategy allocation. Moreover, the fairnessis degraded when non-cooperative approach is used.While this approach results in slightly worse perfor-mance, the information exchange required by selfishusers is significantly low.

Finally, the third classification for spectrum shar-ing in xG networks is based on the access technol-ogy as explained below:

• Overlay spectrum sharing: Overlay spectrum shar-ing refers to the spectrum access technique used.More specifically, a node accesses the networkusing a portion of the spectrum that has not beenused by licensed users [7,15,40,54,71–73]. As aresult, interference to the primary system isminimized.

• Underlay spectrum sharing: Underlay spectrumsharing exploits the spread spectrum techniquesdeveloped for cellular networks [29]. Once a spec-trum allocation map has been acquired, an xGnode begins transmission such that its transmitpower at a certain portion of the spectrum isregarded as noise by the licensed users. This tech-nique requires sophisticated spread spectrumtechniques and can utilize increased bandwidthcompared to overlay techniques.

The effects of underlay and overlay approaches ina cooperative setting are investigated in [19], wherenon-cooperative users are analyzed using a gametheoretical framework. Using this framework, it isshown that frequency division multiplexing is opti-mal when interference among users is high. As aresult, the overlay approach becomes more efficientthan underlay when interference among users ishigh. The lack of cooperation among users,however, necessitates an overlay approach. Thecomparative evaluations show that the performanceloss due to the lack of cooperation is small, and van-ishes with increasing SNR. However, in this frame-work, the cost and inaccuracies of informationexchange between users are not considered.

Another comparison of underlay and overlayapproaches is provided in [42]. The comparison isbased on the influence of the secondary system onthe primary system in terms of outage probability


and three spectrum sharing techniques have beenconsidered. The first technique (spreading basedunderlay) requires secondary users to spread theirtransmit power over the full spectrum such asCDMA or Ultra Wide Band (UWB). The secondtechnique (interference avoidance overlay) requiresnodes to choose a frequency band to transmit suchthat the interference at a primary user is minimized.Also an hybrid technique (spreading based underlaywith interference avoidance) is investigated where anode spreads its transmission over the entire spec-trum and also null or notch frequencies where a pri-mary user is transmitting. Consequently, first, theinterference statistics for each technique are deter-mined for outage probability analysis. Then, theoutage probability for each technique is derivedassuming no system knowledge, perfect systemknowledge, and limited system knowledge. Similarto other existing work, when perfect system knowl-edge is assumed, the overlay scheme outperformsthe underlay scheme in terms of outage probability.However, when interference avoidance is incorpo-rated into spectrum sharing, the underlay schemewith interference avoidance guarantees smaller out-age probability than the pure interference avoid-ance. In a more realistic case, when limited systemknowledge is considered, the importance of thehybrid technique is exacerbated. The overlayschemes result in poor performance due imperfec-tions at spectrum sensing. More specifically, a nodecan transmit at a channel where a primary user istransmitting. However, when underlay with interfer-ence avoidance is used, the interference caused tothe primary user is minimized. Another importantresult is that a higher number of secondary users

Fig. 13. Inter-network and intra-network

can be accommodated by the hybrid scheme thanthe pure interference avoidance scheme.

The theoretical work on spectrum access in xGnetworks reveals important tradeoffs for the designof spectrum access protocols. As expected, it hasbeen shown that cooperative settings result in higherutilization of the spectrum as well as fairness. How-ever, this advantage may not be so high consideringthe cost of cooperation due to frequent informationexchange among users. On the other hand, the spec-trum access technique, i.e., whether it is overlay orunderlay, affects the performance in each setting.While an overlay technique focuses on the holes inthe spectrum, dynamic spreading techniques arerequired for underlay techniques for interference-free operation between primary and secondary sys-tems. Considering the tradeoff between system com-plexity and performance, hybrid techniques may beconsidered for the spectrum technique. In the follow-ing two sections, we explain the existing spectrumsharing techniques that are combinations of thethree classifications we have discussed in this section.

7.2. Inter-network spectrum sharing

xG networks are envisioned to provide opportu-nistic access to the licensed spectrum using unli-censed users. This setting enables multiple systemsbeing deployed in overlapping locations and spec-trum as shown in Fig. 13. Hence, spectrum sharingamong these systems is an important research topicin xG networks. Up to date, inter-network spectrumsharing has been regulated via static frequencyassignment among different systems or centralizedallocations between different access points of a sys-

spectrum sharing in xG networks.


tem in cellular networks. In ad-hoc networks, onlythe interference issues in the ISM band has beeninvestigated focusing mostly on the coexistence ofWLAN and Bluetooth networks. Consequently,intra-network spectrum sharing in xG networksposes unique challenges that have not been consid-ered before in wireless communication systems. Inthis section, we overview the recent work in thisresearch area.

7.2.1. Centralized inter-network spectrum sharing

As a first step for the coexistence of open spec-trum systems, in [33], the common spectrum coordi-nation channel (CSCC) etiquette protocol isproposed for coexistence of IEEE 802.11b and802.16a networks. The reason we do not considerthis work as a complete solution for xG networksis that it necessitates modifications in users usingboth of the networks. More specifically, each nodeis assumed to be equipped with a cognitive radioand a low bit-rate, narrow-band control radio.The coexistence is maintained through the coordi-nation of these nodes with each other by broadcast-ing CSCC messages. Each user determines thechannel it can use for data transmission such thatinterference is avoided. In case channel selection isnot sufficient to avoid interference, power adapta-tion is also deployed. The evaluations reveal thatwhen there is vacant spectrum to use frequencyadaptation, CSCC etiquette protocol improvesthroughput by 35–160% via both frequency andpower adaptation. Another interesting result is thatwhen nodes are clustered around IEEE 802.11baccess points, which is a realistic assumption,the throughput improvement of CSCC protocolincreases.

In addition to the competition for the spectrum,competition for the users is also considered in [32].In this work, a central spectrum policy server(SPS) is proposed to coordinate spectrum demandsof multiple xG operators. In this scheme, each oper-ator bids for the spectrum indicating the cost it willpay for the duration of the usage. The SPS thenallocates the spectrum by maximizing its profit fromthese bids. The operators also determine an offer forthe users and users select which operator to use for agiven type of traffic. When compared to a casewhere each operator is assigned an equal share ofthe spectrum, the operator bidding scheme achieveshigher throughput leading to higher revenue for theSPS, as well as a lower price for the users accordingto their requirements. This work opens a new per-

spective by incorporating competition for users aswell as the spectrum in xG networks.

7.2.2. Distributed inter-network spectrum sharingA distributed spectrum sharing scheme for wire-

less Internet service provides (WISPs) that sharethe same spectrum is proposed in [43], where a dis-tributed QoS based dynamic channel reservation(D-QDCR) scheme is used. The basic conceptbehind D-QDCR is that a base station (BSs) of aWISP competes with its interferer BSs accordingto the QoS requirements of its users to allocate aportion of the spectrum. Similar to the CSCC proto-col [33], the control and data channels are sepa-rated. The basic unit for channel allocation inD-QDCR is called Q-frames. When a BS allocatesa Q-frame, it uses the control and data channelsallocated to it for coordination and data communi-cation between the users. The competition betweenBSs are performed according to the priority of eachBS depending on a BSs data volume and QoSrequirement. Moreover, various competition poli-cies are proposed based on the type of traffic a userdemands. Although thorough evaluations are notprovided in [43], the D-QDCR scheme serves animportant contribution for inter-network spectrumsharing.

The inter-network spectrum sharing solutions sofar provide a broader view of the spectrum sharingsolution including certain operator policies for thedetermination of the spectrum allocation. A majorproblem for the existing solutions in the xG networkarchitecture is the requirement for a common con-trol channel. We detail the potential problems andopen research issues in this aspect in Section 7.4.

7.3. Intra-network spectrum sharing

A significant amount of work on spectrum shar-ing focuses on intra-network spectrum sharing,where the users of an xG network try to access theavailable spectrum without causing interference tothe primary users. In this section, we overview theexisting work and the proposed solutions in thisarea while providing a classification of existingprotocols in terms of the classification provided inSection 7.1.

7.3.1. Cooperative intra-network spectrum sharing

A cooperative local bargaining (LB) scheme isproposed in [15] to provide both spectrum utiliza-tion and fairness. The local bargaining framework


is formulated based on the framework in [50,74].Local bargaining is performed by constructing localgroups according to a poverty line that ensures aminimum spectrum allocation to each user andhence focuses on fairness of users. The evaluationsreveal that local bargaining can closely approximatecentralized graph coloring approach at a reducedcomplexity. Moreover, localized operation viagrouping provides an efficient operation between afully distributed and a centralized scheme.

Another approach that considers local groups forspectrum sharing is provided in [71], where a heter-ogeneous distributed MAC (HD-MAC) protocol isproposed. A potential problem in the solution pro-vided in LB [15] is that a common control channelmay not exist in xG networks or can be occupiedby a primary user. In [71], it is shown that for agiven topology, very limited number of commonchannels exist for each of the users in a network.However, when local neighbors are considered, anode shares many channels with its neighbors.Based on this observation, a clustering algorithmis proposed such that each group selects a commonchannel for communication, and distributed sensingand spectrum sharing is provided through thischannel. Moreover, if this channel is occupied bya primary user at a specific time, the nodes reorga-nize themselves to use another control channel.The performance evaluations show that the distri-buted grouping approach outperforms commoncontrol channel approaches especially when thetraffic load is high.

The notion of busy tones, which are mainly usedin some ad-hoc network protocols, is extended tothe xG networks in [40] with the dynamic open spec-trum sharing MAC (DOSS-MAC) protocol. As aresult, when a node is using a specific data channelfor communication, both the transmitter and thereceiver send a busy tone signal through the associ-ated busy tone channel. In order to further eliminatecontrol channel communication, FFT-based radioand the noncoherent modulation/demodulation-based radio designs are proposed which theoreti-cally enable receivers to detect the carrier frequencyand the bandwidth of a signal without any controlinformation.

In addition to spectrum allocation, transmitpower determination is also included in the spec-trum sharing protocol in [29]. In this work both sin-gle channel and multi-channel asynchronousdistributed pricing (SC/MC-ADP) schemes are pro-posed, where each node announces its interference

price to other nodes. Using this information fromits neighbors, a node can first allocate a channeland in case there exist users in that channel, then,determine its transmit power. As a result, thisscheme can be classified as a hybrid of underlayand overlay techniques. While there exist usersusing distinct channels, multiple users can sharethe same channel by adjusting their transmit power.Furthermore, the SC-ADP algorithm provideshigher rates to users when compared to selfish algo-rithms where users select the best channel withoutany knowledge about their neighbors’ interferencelevels. Finally, it is shown that under high interfer-ence, the proposed algorithm outperforms underlaytechniques.

So far, we have presented distributed solutionswhere a fixed infrastructure is not assumed. In [7],dynamic spectrum access protocol (DSAP), whichis a centralized solution for spectrum sharing inxG networks, is presented. This solution is similarto the SPS approach [32] described in Section 7.2with a focus on intra-network spectrum sharing.The dynamic spectrum access protocol (DSAP) pro-posed in this work enables a central entity to leasespectrum to users in a limited geographical region.DSAP consists of clients, DSAP server, and relaysthat relay information between server and clientsthat are not in the direct range of the server. More-over, clients inform the server their channel condi-tions so that a global view of the network can beconstructed at the server. By exploiting cooperativeand distributed sensing, DSAP servers construct aRadioMap. This map is used for channel assign-ments which are leased to clients for a limitedamount of time.

7.3.2. Non-cooperative intra-network spectrum

sharing

An opportunistic spectrum management scheme isproposed in [73], where users allocate channels basedon their observations of interference patterns andneighbors. In the device centric spectrum manage-ment scheme (DCSM), the communication overheadis minimized by providing five different system rulesfor spectrum allocation. As a result, users allocatechannels according to these rules based on theirobservations instead of collaborating with otherusers. In case more than one node chooses the samechannel in close proximity, random access techniquesare used to resolve the contention. The comparativeanalysis of this scheme with the cooperative schemesshow that rule-based spectrum access results in


slightly worse performance. However, the communi-cation overhead is reduced significantly.

A spectrum sharing protocol for ad-hoc xG net-works, (AS-MAC), is proposed in [54]. AS-MACexploits the RTS-CTS exchange and Network Allo-cation Vector (NAV) concepts of the IEEE 802.11MAC protocol [30] in an open spectrum setting.Moreover, a common control channel is used suchthat transmitter receiver handshake is initiatedthrough this channel. In this work, the xG networkis assumed to coexist with a GSM network. Eachnode first listens to the broadcast channel of theGSM network as well as the control channel ofthe xG network, and each node then constructs itsNAV and selects channels accordingly.

In addition to the spectrum allocation methods, atransmitter receiver handshake method is proposedin [72] as a part of a cross-layer decentralized cogni-tive MAC (DC-MAC) protocol. The details of thiswork is explained in Section 9. In the transmitter-receiver handshake method, each user is assigned aset of channels is continuously monitored by theuser. A transmitter selects one of those channelsand initiates communication. The actual data chan-nel selection is then performed through this initialhandshake channel.

7.4. Spectrum sharing challenges

In the previous sections, the theoretical findingsand solutions for spectrum sharing in xG networksare investigated. Although there already exists avast amount of research in spectrum sharing, thereare still many open research issues for the realiza-tion of efficient and seamless open spectrum opera-tion. In the following, we detail the challenges forspectrum sharing in xG networks along with somepossible solutions.

7.4.1. Common control channel (CCC)

Many spectrum sharing solutions, either central-ized or distributed, assume a CCC for spectrumsharing [7,40,54]. It is clear that a CCC facilitatesmany spectrum sharing functionalities such as trans-mitter receiver handshake [40], communication witha central entity [7], or sensing information exchange.However, due to the fact that xG network users areregarded as visitors to the spectrum they allocate,when a primary user chooses a channel, this channelhas to be vacated without interfering. This is alsotrue for the CCC. As a result, implementation of afixed CCC is infeasible in xG networks. Moreover,

in a network with primary users, a channel commonfor all users is shown to be highly dependent on thetopology, hence, varies over time [71]. Consequently,for protocols requiring a CCC, either a CCC mitiga-tion technique needs to be devised or local CCCsneed to be exploited for clusters of nodes [71]. Onthe other hand when CCC is not used, the transmit-ter receiver handshake becomes a challenge. For thischallenge, receiver driven techniques proposed in[72] may be exploited.

7.4.2. Dynamic radio rangeRadio range changes with operating frequency

due to attenuation variation. In many solutions, afixed range is assumed to be independent of theoperating spectrum [15,71]. However, in xG net-works, where a large portion of the wireless spec-trum is considered, the neighbors of a node maychange as the operating frequency changes. Thiseffects the interference profile as well as routingdecisions. Moreover, due to this property, thechoice of a control channel needs to be carefullydecided. It would be much efficient to select controlchannels in the lower portions of the spectrumwhere the transmission range will be higher and toselect data channels in the higher portions of thespectrum where a localized operation can be utilizedwith minimized interference. So far, there exists nowork addressing this important challenge in xG net-works and we advocate operation frequency awarespectrum sharing techniques due the direct interde-pendency between interference and radio range.

7.4.3. Spectrum unit

Almost all spectrum sharing techniques discussedin the previous sections consider a channel as thebasic spectrum unit for operation. Many algorithmsand methods have been proposed to select the suit-able channel for efficient operation in xG networks.However, in some work, the channel is vaguelydefined as ‘‘orthogonal non-interfering’’ [73],‘‘TDMA, FDMA, CDMA, or a combination ofthem’’ [50], or ‘‘a physical channel as in IEEE802.11, or a logical channel associated with a spec-trum region or a radio technology’’ [71]. In otherwork, the channel is simply defined in the frequencydimension as frequency bands [33,40,43,49,54]. It isclear that the definition of a channel as a spectrumunit for spectrum sharing is crucial in further devel-oping algorithms. Since a huge portion of the spec-trum is of interest, it is clear that properties of achannel may not be constant due to the effects of


operating frequency. On the contrary, a channel isusually assumed to provide the same bandwidth asother channels [15,29,42,50,73].

Furthermore, the existence of primary users andthe heterogeneity of the networks that are availableintroduce additional challenges to the choice of aspectrum unit/channel. Hence, different resourceallocation units such as CSMA random access,TDMA time slots, CDMA codes, as well as hybridtypes can be allocated to the primary users. In orderto provide seamless operation, these properties needto be considered in the choice of a spectrum unit. In[28], the necessity of a spectrum space for a spectrumunit is also advocated. The possible dimensions ofthe spectrum space are classified as power, fre-quency, time, space, and signal. Although notorthogonal, these dimensions can be used to distin-guish signals [28].

For this purpose, we describe a three dimensionalspace model for modeling network resources thathas been proposed in [59]. Although this workfocuses on heterogeneity in next generation/fourthgeneration (NG/4G) networks, as discussed in[59], it can be easily incorporated into xG networks.Based on this three dimensional resource-space, anovel Virtual Cube concept has been proposed inorder to evaluate the performance of each network.The Virtual Cube concept defines a unit structurebased on the resource allocation techniques usedin existing networks.

The resource is modeled in a three dimensionalresource-space with time, rate, and power/codedimensions as shown in Fig. 14. The time dimension

models the time required to transfer information.

Fig. 14. Virtual cube model.

The rate dimension models the data rate of the net-work. Thus, the capacity of different networks withthe same connection durations but different datarates are captured in the rate dimension. Further-more, in the case of CDMA networks, the band-width increase due to the multi-code transmissionsis also captured in this dimension. The power/code

dimension models the energy consumed for trans-mitting information through the network. Notethat, the resource in terms of available bandwidthcan be modeled using the time and rate dimensions.However, the cost of accessing different networksvary in terms of the power consumed by the wirelessterminal. Hence, a third dimension is required. Eachnetwork type requires different power levels fortransmission of the MAC frames because of variousmodulation schemes, error coding and channelcoding techniques. As a result, the resource differ-ences in these aspects are captured in the powerdimension.

Using this model, heterogeneous access types inexisting networks as well as xG network spectrumcan be modeled based on a generic spectrum unit.We advocate that determining a common spectrumunit is crucial for efficient utilization of the wirelessspectrum and seamless operability with existing pri-mary networks.

8. Upper layer issues

In addition to the unique challenges of xG net-works in terms of spectrum sensing, spectrum man-agement, spectrum mobility, and spectrum sharing,upper layer issues such as routing, flow control andcongestion control are also important for the reali-zation of dynamic spectrum networks. In thissection, we overview the challenges related to theseareas.

8.1. Routing challenges

Routing constitutes a rather important but yetunexplored problem in xG networks. Especially inxG networks with multi-hop communicationrequirements, the unique characteristics of the openspectrum phenomenon necessitates novel routingalgorithms to be developed. So far, the researchon xG networks is primarily on spectrum sensingtechniques and spectrum sharing solutions. How-ever, we emphasize that the need for routing algo-rithms in open spectrum environment constitutesan important topic in xG network research. In this


section, we overview two existing solutions for rout-ing and discuss open research topics in this area.

A major design choice for routing in xG net-works is the collaboration between routing andspectrum management. The dynamic spectrum thatis intermittent in terms of both time and spacenecessitates such an approach [60,69]. Simulation-based comparisons are performed in [60,69] forcross-layer and decoupled approaches for routingand spectrum management. The results in both ofthese work reveal that a cross-layer solution thatconstructs routes and determines the operatingspectrum jointly for each hop outperforms asequential approach where routes are selected inde-pendent of the spectrum allocation [60,69].

In [60], the inter-dependence between route selec-tion and spectrum management is investigated.First, a decoupled route selection and spectrummanagement methodology is proposed. In thisscheme, the route selection is performed independentof the spectrum management using the shortest-pathalgorithm. The spectrum sharing is performed usingthe scheme in [71]. In this scheme, routing layerinvokes path discovery to select routes. The spec-trum management is then performed on each hop.A cross-layer solution that considers joint routeselection and spectrum management is also pro-posed. In this approach, each source node usesDSR to find candidate paths and schedules a timeand channel for each hop. This source-based routingtechnique is performed centrally using a global viewof the network to show the upper bound in achiev-able performance. A similar comparison of layeredand cross-layer approach is presented in [69] usinga novel graph modeling technique, which will be dis-cussed in the following. The simulations in both[60,69] reveal that cross-layer approach is advanta-geous for routing in xG networks since the availabil-ity of spectrum directly affects the end-to-endperformance. These two solutions for routingin xG networks clearly show that cross-layerapproaches that jointly consider route and spectrumselection is necessary for xG networks.

Another unique challenge for routing in xG net-works is the development tools for analytical evalu-ation of routing protocols. Traditionally, routingprotocols for ad hoc networks are analyzed usinggraph models. However, in these networks, thecommunication spectrum is fixed and continuouscontrary to the dynamic nature of xG networks.Hence, a node can use the same set of static chan-nel(s) for communication with all neighbors [69].

On the contrary, modeling network topology andconnectivity of an xG network is challenging. In[69], a layered graph model is proposed for this chal-lenge. In this model, each layer corresponds to achannel in the network. In each layer, each node isrepresented by subnodes forming the vertex in eachlayer. This model is exploited in [69] to constructroutes. Using different cost functions for each edge,required constraints for routes such as interferenceavoidance can be achieved. This model providesan interesting solution for xG network modelingfor relatively static link properties. However, as willbe discussed in the following, time-varying nature ofavailable links necessitates time dependent modelsfor a complete analysis of xG networks.

Although these solutions provide interestingresults for routing in xG networks, there are stillmajor challenges and open research topics that hasnot been addressed before. Below, we summarizethe open research issues for routing in xG networks:

• Common control channel: It has also been empha-sized in Section 7.4 that the lack of a commoncontrol channel (CCC) in xG networks consti-tutes a major problem. Traditional routing pro-tocols require either local or global broadcastmessages for specific functionalities such asneighbor discovery, route discovery and routeestablishment. However, even broadcasting inxG networks is a major problem due to the lackof a CCC. Hence, solutions considering this factis required in xG networks.

• Intermittent connectivity: In xG networks, thereachable neighbors of a node may change rap-idly. This is due to two reasons. First, the avail-able spectrum may change or vanish as licensedusers exploit the network. Moreover, once a nodeselects a channel for communication it is nolonger reachable through other channels. As aresult, the connectivity concept used for wirelessnetworks is different in xG networks and dependson the spectrum. For this purpose channel-basedmodels such as the one in [69] is required as wellas time-based solutions.

• Re-routing: In xG networks, due to the intermit-tent connectivity, a route established for a flowcan change due to the available spectrum in addi-tion to mobility. Hence, the re-routing algo-rithms considering the dynamic spectrum isnecessary for routing in xG networks. A spec-trum-aware routing adapts route selection tospectrum fluctuations [73].


• Queue management: The queue management inxG networks is another challenge which has notbeen addressed to date. An xG terminal mayhave multiple interfaces for communication withdifferent nodes. Since the available spectrum var-ies over time, these interfaces may becomeunavailable requiring the packets served throughthat interface moved to other interfaces. In addi-tion, the quality of service requirements maydeploy various priorities on different traffic types.Hence, the implementation of a single queue ormultiple queues for each traffic type of each inter-face needs to be investigated.

8.2. Transport layer challenges

Transport protocols constitute an unexploredarea for xG networks since there exist no work onthis area yet. Several solutions have been proposedto improve the performance of TCP and UDP inconventional wireless networks in recent years [2].These studies focus on mechanisms to limit the per-formance degradation of TCP and UDP that arisebecause of wireless link errors and access delays.However, the xG networks impose unique chal-lenges for transport protocols as explained below:

The performance of TCP depends on the packetloss probability and the round trip time (RTT).Wireless link errors and, hence, the packet loss prob-ability not only depends on the access technology,but also on the frequency in use, interference level,and the available bandwidth. Therefore, the wirelessTCP and UDP protocols that are designed for exist-ing wireless access technologies cannot be used indynamic spectrum assignment based xG networks.

On the other hand, RTT of a TCP connectiondepends indirectly on the frequency of operation.For example, if the packet error rate (or equiva-lently, the frame error rate) is higher at a particularfrequency band, a higher number of link layerretransmissions are required to successfully trans-port a packet across the wireless channel. Moreover,the wireless channel access delay in xG networksdepends on the operation frequency, the interfer-ence level, and the medium access control protocol.These factors influence the RTT of a TCP connec-tion. Therefore, based on the frequency of opera-tion, RTT and packet loss probability observed bya TCP protocol will vary. Hence, transport proto-cols need to be designed to adapt to thesevariations.

The operation frequency of a cognitive radio inxG networks may vary from time to time due tospectrum handoff as explained in Section 6. Whenan xG terminal changes its operating frequency, thisresults in a finite amount of delay before the newfrequency can be operational. This is referred toas the spectrum handoff latency. The spectrum hand-off latency can increase the RTT, which leads toretransmission timeout (RTO). Conventional trans-port protocols can perceive this RTO as packet lossand invoke its congestion avoidance mechanismresulting in reduced throughput. To eliminate theadverse effects of spectrum mobility, transport pro-tocols need to be designed such that they are trans-parent to spectrum handoff.

9. Cross-layer design

The communication challenges outlined abovenecessitate new communication protocols to bedesigned for spectrum-aware communication inxG networks. As explained before, the performanceof xG networking functionalities directly depend onthe properties of the spectrum band in use. Thisdirect relationship necessitates a cross-layer designin the entire xG networking protocol stack. Inparticular, the effects of the selected spectrum bandand the changes due to spectrum mobility need to becarefully considered in the design of communicationprotocols. Moreover, the spectrum managementfunctionalities such as spectrum sensing and spec-trum handoff should work in collaboration withthe communication protocols. Based on thismotivation, in the following sections, we overviewthe challenges for the cross-layer design in xGnetworks.

9.1. Cross-layer challenges in spectrum

management

The dynamic nature of the underlying spectrumin xG networks necessitates communication proto-cols to adapt to the wireless channel parameters.Moreover, the behavior of each protocol affectsthe performance of other protocols. For example,different medium access techniques used in xG net-works, directly affect the round trip time (RTT)for the transport protocols. Similarly, when re-rout-ing is done because of link failures arising fromspectrum mobility, the RTT and error probabilityin the communication path change accordingly.The change in error probability also affects the


performance of the medium access protocols. Con-sequently, all these changes affect the overall qualityof the user applications.

The spectrum management function cooperateswith the communication layers, as shown inFig. 2. In order to decide the appropriate spectrumband, the spectrum management requires the infor-mation regarding QoS requirement, transport, rout-ing, scheduling, and sensing. Hence, theseinterdependencies among functionalities of the com-munication stack, and their close coupling with thephysical layer necessitate a cross layer spectrummanagement function which considers mediumaccess, routing, and transport, and applicationrequirements as well as available spectrum in theselection of the operating spectrum.

9.2. Cross-layer challenges in spectrum handoff

Spectrum handoff results in latency, which affectsthe performance of the communication protocols.Thus, the main challenge in spectrum handoff is toreduce the latency for spectrum sensing. Thespectrum handoff latency has adverse effects onthe performance of transport protocols. Moreover,during spectrum handoff, the channel parameterssuch as path loss, interference, wireless link error,and link layer delay are influenced by the dynamicuse of the spectrum. On the other hand, the changesin the PHY and MAC channel parameters can initi-ate spectrum handoff. Moreover, the user applica-tion may request spectrum handoff to find a betterquality spectrum band.

As shown in Fig. 2, the spectrum mobility func-tion cooperates with spectrum management func-tion and spectrum sensing to decide on anavailable spectrum band. In order to estimate theeffect of the spectrum handoff latency, informationabout the link layer and sensing delays are required.Transport and application layer should also beaware of the latency to reduce the abrupt qualitydegradation. In addition, the routing informationis also important for the route recovery using spec-trum handoff. For these reasons, the spectrumhandoff is closely related to the operations in allcommunication layers.

9.3. Cross-layer challenges in spectrum sharing

The performance of spectrum sharing directlydepends on the spectrum sensing capabilities ofthe xG users. Spectrum sensing is primarily a

PHY layer function. However, in the case of coop-erative detection, xG users should use ad hoc con-nection for exchanging sensing information, whichnecessitates a cross-layer design between spectrumsharing and spectrum sensing. It is clear that theperformance of communication protocols dependson spectrum sensing, i.e., getting information aboutthe spectrum utilization. Two major challenges existin this aspect.

The first challenge is the interference mitigation.While interference occurs at a receiver, spectrumscanning alone only provides information abouttransmitters [33]. Hence, cooperative techniquesthat necessitate transmitters to consider both theirinterference to other users and the interference attheir receivers are required. The superiority ofcooperative techniques in terms of system perfor-mance has already been demonstrated in manystudies [15,29,40,71]. On the other hand, such acollaboration increases the communication over-head and may lead to overall system performancedegradation when channel capacity or energyconsumption is considered. Consequently, effectivespectrum sharing techniques that enable efficientcollaboration between different xG nodes in termsof spectrum sensing information sharing arerequired.

The second challenge about spectrum sensing isthat the whole spectrum cannot be sensed all thetime. More specifically, due to the huge range ofspectrum foreseen for xG networks, a significantamount of time is required to sense the whole spec-trum. Since sensing consumes energy this processhas to be carefully scheduled. As a result, it maynot be practical to assume that a node has an accu-rate knowledge about the spectrum at all times.Moreover, current radio technologies prevent con-tinuous spectrum sensing if only a single radio isdeployed on a device. During communication, spec-trum sensing has to be stopped to switch to therequired channel and perform communication.Hence, this operation requires cross-layer interac-tion between the physical layer and the upper layers.More specifically, communication attempts need tobe coordinated with spectrum sensing events. Asan alternative, the effect of using multiple radioshas been investigated in [54], where a two trans-ceiver operation is considered such that a trans-ceiver always listens to the control channel forsensing. This operation improves the system perfor-mance, however, the complexity and device costsare high.


The spectrum sharing techniques discussed inSection 7 decouples spectrum allocation and spec-trum sensing. In decentralized cognitive MAC(DC-MAC) scheme proposed in [72], a cross-layerapproach for spectrum allocation and spectrumsensing is considered. Contrary to many work, in[72], limited spectrum sensing is addressed, whereonly a portion of the whole spectrum can be sensed.Moreover, spectrum sensing is jointly performedwith spectrum allocation and application layer suchthat the node senses a portion of the spectrum onlyif it will allocate a channel due to a request from theapplication layer. First, an optimal approach forchannel allocation is provided followed by a greedysuboptimal approach. A node is assumed to be ableto sense one channel among N possible channels.Through performance evaluation, it is shown thatthe greedy approach closely approximates the opti-mal solution. Moreover, this solution is also robustto inaccuracies in spectrum sensing and limitedknowledge.

9.4. Cross-layer challenges in upper layers

Since available spectrum bands in xG networkswith multi-hop communication are different foreach hop, spectrum sensing information is requiredfor topology configuration in xG networks. More-over, a major design choice for routing in xG net-works is the collaboration between routing andspectrum decision. If the optimal route from anxG user to another results in interference to the pri-mary users, end-to-end latency or packet losses canbe affected along the route due to this interaction.To mitigate the degradation, multiple spectruminterfaces at intermediate nodes may be selected.Therefore, end-to-end route may consist of multiplehops traversing different spectrum bands.

Finally, re-routing needs to be performed in across-layer fashion. If a link failure occurs due tospectrum mobility, the routing algorithm needs todifferentiate this failure from a node failure. Fur-thermore, intermediate xG users can perform re-routing by exploiting the spectrum informationavailable from spectrum sensing functionality selectbetter routes method.

Due to the dynamic frequency of operation, theRTT and packet loss rates vary in xG networks,which leads to variations in packet transmissiondelay. Moreover, medium access schemes introduceaccess delay. All these factors influence the roundtrip time of a connection, which in turn affect the

performance of transport protocols. Moreover, thelatency associated with a spectrum handoff increasesthe instantaneous RTT of the packet transmission.Consequently, transport protocols for xG networksshould be designed in a spectrum aware approachthat introduces cooperative operation with the othercommunication layers.

10. Conclusions

xG networks are being developed to solve currentwireless network problems resulting from the lim-ited available spectrum and the inefficiency in thespectrum usage by exploiting the existing wirelessspectrum opportunistically. xG networks, equippedwith the intrinsic capabilities of the cognitive radio,will provide an ultimate spectrum-aware communi-cation paradigm in wireless communications. In thissurvey, intrinsic properties and current researchchallenges of the xG networks are presented. Weinvestigate the unique challenges in xG networksby a bottom-up approach, starting from the capa-bilities of cognitive radio techniques to the commu-nication protocols that need to be developed forefficient communication. Moreover, novel spectrummanagement functionalities such as spectrum sens-ing, spectrum analysis, and spectrum decision aswell as spectrum mobility are introduced.

The discussions provided in this survey stronglyadvocate spectrum-aware communication protocolsthat consider the spectrum management functional-ities. This cross-layer design requirement necessitatesa rethinking of the existing solutions developed forwireless networks. Many researchers are currentlyengaged in developing the communication technolo-gies and protocols required for xG networks.However, to ensure efficient spectrum-aware commu-nication, more research is needed along the linesintroduced in this survey.

References

[1] 3GPP2, cdma2000 High rate packet data air interfacespecification, TS C.S20024 V2.0, October 2000.

[2] O.B. Akan, I.F. Akyildiz, ATL: an adaptive transport layerfor next generation wireless internet, IEEE Journal onSelected Areas in Communications (JSAC) 22 (5) (2004)802–817.

[3] I.F. Akyildiz, Y. Altunbasak, F. Fekri, R. Sivakumar,AdaptNet: adaptive protocol suite for next generationwireless internet, IEEE Communications Magazine 42 (3)(2004) 128–138.

[4] I.F. Akyildiz, X. Wang, W. Wang, Wireless mesh networks:a survey, Computer Networks Journal 47 (4) (2005) 445–487.


[5] I.F. Akyildiz, Y. Li, OCRA: OFDM-based cognitive radionetworks, Broadband and Wireless Networking LaboratoryTechnical Report, March 2006.

[6] L. Berlemann, S. Mangold, B.H.Walke, Policy-basedreasoning for spectrum sharing in cognitive radio net-works, in: Proc. IEEE DySPAN 2005, November 2005,pp. 1–10.

[7] V. Brik, E. Rozner, S. Banarjee, P. Bahl, DSAP: a protocolfor coordinated spectrum access, in: Proc. IEEE DySPAN2005, November 2005, pp. 611–614.

[8] R.W. Brodersen, A. Wolisz, D. Cabric, S.M. Mishra, D.Willkomm, Corvus: a cognitive radio approach for usage ofvirtual unlicensed spectrum, Berkeley Wireless ResearchCenter (BWRC) White paper, 2004.

[9] T.X. Brown, An analysis of unlicensed device operation inlicensed broadcast service bands, in: Proc. IEEE DySPAN2005, November 2005, pp. 11–29.

[10] M.M. Buddhikot, P. Kolody, S. Miller, K. Ryan, J. Evans,DIMSUMNet: new directions in wireless networking usingcoordinated dynamic spectrum access, in: Proc. IEEEWoWMoM 2005, June 2005, pp. 78–85.

[11] M.M. Buddhikot, K. Ryan, Spectrum management incoordinated dynamic spectrum access based cellular net-works, in: Proc. IEEE DySPAN 2005, November 2005, pp.299–307.

[12] D. Cabric, S.M. Mishra, R.W. Brodersen, Implementationissues in spectrum sensing for cognitive radios, in: Proc. 38thAsilomar Conference on Signals, Systems and Computers2004, November 2004, pp. 772–776.

[13] D. Cabric, R.W Brodersen, Physical layer design issuesunique to cognitive radio systems, in: Proc. IEEE PersonalIndoor and Mobile Radio Communications (PIMRC) 2005,September 2005.

[14] D. Cabric, S.M. Mishra, D. Willkomm, R. Brodersen, A.Wolisz, A Cognitive radio approach for usage of virtualunlicensed spectrum, in: Proc. 14th IST Mobile and WirelessCommunications Summit, June 2005.

[15] L. Cao, H. Zheng, Distributed spectrum allocation via localbargaining, in: Proc. IEEE Sensor and Ad Hoc Communi-cations and Networks (SECON) 2005, September 2005, pp.475–486.

[16] C. Cordeiro, K. Challapali, D. Birru, S. Shankar, IEEE802.22: the first worldwide wireless standard based oncognitive radios, in: Proc. IEEE DySPAN 2005, November2005, pp. 328–337.

[17] F. Digham, M. Alouini, M. Simon, On the energy detectionof unknown signals over fading channels, in: Proc. IEEEICC 2005, vol. 5, May 2003, pp. 3575–3579.

[18] E. Esteves, The high data rate evolution of the cdma2000cellular systemMobility and Teletraffic for Wireless Com-munications, vol. 5, Kluwer Academic Publishers, 2000, pp.61–72.

[19] R. Etkin, A. Parekh, D. Tse, Spectrum sharing for unli-censed bands, in: Proc. IEEE DySPAN 2005, November2005, pp. 251–258.

[20] FCC, ET Docket No 03-222 Notice of proposed rule makingand order, December 2003.

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

[22] A. Fehske, J.D. Gaeddert, J.H. Reed, A new approach tosignal classification using spectral correlation and neural

networks, in: Proc. IEEE DySPAN 2005, November 2005,pp. 144–150.

[23] G. Ganesan, Y.G. Li, Cooperative spectrum sensing incognitive radio networks, in: Proc. IEEE DySPAN 2005,November 2005, pp. 137–143.

[24] G. Ganesan, Y.G. Li, Agility improvement through coop-erative diversity in cognitive radio networks, in: Proc.GLOBECOM, November 2005, pp. 2505–2509.

[25] A. Ghasemi, E.S. Sousa, Collaborative spectrum sensing foropportunistic access in fading environment, in: Proc. IEEEDySPAN 2005, November 2005, pp. 131–136.

[26] D. Grandblaise, D. Bourse, K. Moessner, P. Leaves,Dynamic spectrum allocation (DSA) and reconfigurability,in: Proc. Software-Defined Radio (SDR) Forum, November2002.

[27] S. Haykin, Cognitive radio: brain-empowered wireless com-munications, IEEE Journal on Selected Areas in Communi-cations 23 (2) (2005) 201–220.

[28] W.D. Horne, Adaptive spectrum access: using the fullspectrum space, in: Proc. Telecommunications PolicyResearch Conference (TPRC), September 2003.

[29] J. Huang, R.A. Berry, M.L. Honig, Spectrum sharing withdistributed interference compensation, in: Proc. IEEEDySPAN 2005, November 2005, pp. 88–93.

[30] IEEE 802.11, Wireless LAN medium access control (MAC)and physical layer (PHY) specifications, 1999.

[31] IEEE 802.22 Working group on wireless regional areanetworks, Available from: <http://www.ieee802.org/22/>.

[32] O. Ileri, D. Samardzija, N.B. Mandayam, Demand respon-sive pricing and competitive spectrum allocation via spec-trum server, in: Proc. IEEE DySPAN 2005, November 2005,pp. 194–202.

[33] X. Jing, D. Raychaudhuri, Spectrum co-existence of IEEE802.11b and 802.16a networks using CSCC etiquette proto-col, in: Proc. IEEE DySPAN 2005, November 2005, pp. 243–250.

[34] F.K. Jondral, Software-defined radio-basic and evolution tocognitive radio, EURASIP Journal on Wireless Communi-cation and Networking 2005.

[35] T. Kamakaris, M.M. Buddhikot, R. Iyer, A case forcoordinated dynamic spectrum access in cellular networks,in: Proc. IEEE DySPAN 2005, November 2005, pp. 289–298.

[36] V. Kanodia, A. Sabharwal, E. Knightly, MOAR: a multi-channel opportunistic auto-rate media access protocol for adhoc networks, in: Proc. IEEE BROADNETS 2004, October2004, pp. 600–610.

[37] S. Krishnamurthy, M. Thoppian, S. Venkatesan, R. Prak-ash, Control channel based MAC-layer configuration, rout-ing and situation awareness for cognitive radio networks,in: Proc. IEEE MILCOM 2005, October 2005.

[38] P. Kyasanur, X. Yang, N.H. Vaidya, Mesh networkingprotocols to exploit physical layer capabilities, in: Proc. FirstIEEE Workshop on Wireless Mesh Networks (WiMesh),September 2005.

[39] P. Leaves, K. Moessner, R. Tafazoli, D. Grandblaise, D.Bourse, R. Tonjes, M. Breveglieri, Dynamic spectrumallocation in composite reconfigurable wireless networks,IEEE Comm. Magazine, vol. 42, May 2004, pp. 72–81.

[40] L. Ma, X. Han, C.-C. Shen, Dynamic open spectrum sharingMAC protocol for wireless ad hoc network, in: Proc. IEEEDySPAN 2005, November 2005, pp. 203–213.

http://www.ieee802.org/22/


[41] D. Maldonado, B. Lie, A. Hugine, T.W. Rondeau, C.W.Bostian, Cognitive radio applications to dynamic spectrumallocation, in: Proc. IEEE DySPAN 2005, November 2005,pp. 597–600.

[42] R. Menon, R.M. Buehrer, J.H. Reed, Outage probabilitybased comparison of underlay and overlay spectrum sharingtechniques, in: Proc. IEEE DySPAN 2005, November 2005,pp. 101–109.

[43] G. Marias, Spectrum scheduling and brokering based onQoS demands of competing WISPs, in: Proc. IEEE DySPAN2005, November 2005, pp. 684–687.

[44] S.M. Nishra, D. Cabric, C. Chang, D. Willkomm, B.Schewick, A. Wolisz, R.W. Brodersen, A real time cognitiveradio testbed for physical and link layer experiments, in:Proc. IEEE DySPAN 2005, November 2005, pp. 562–567.

[45] J. Mitola III, Cognitive radio for flexible mobile multimediacommunication, in: Proc. IEEE International Workshop onMobile Multimedia Communications (MoMuC) 1999,November 1999, pp. 3–10.

[46] J. Mitola III, Cognitive radio: an integrated agent architec-ture for software defined radio, Ph.D Thesis, KTH RoyalInstitute of Technology, 2000.

[47] R. Murty, Software-defined reconfigurability radios: smart,agile, cognitive, and interoperable, Technology@Intel Mag-azine, July 2003.

[48] Nautilus Project Website. Available from: <http://www.cs.ucsb.edu/htzheng/cognitive/nautilus.html>.

[49] N. Nie, C. Comaniciu, Adaptive channel allocation spectrumetiquette for cognitive radio networks, in: Proc. IEEEDySPAN 2005, November 2005, pp. 269–278.

[50] C. Peng, H. Zheng, B.Y. Zhao, Utilization and fairness inspectrum assignment for opportunistic spectrum access, in:ACM Mobile Networks and Applications (MONET), 2006.

[51] C. Raman, R.D. Yates, N.B. Mandayam, Scheduling vari-able rate links via a spectrum server, in: Proc. IEEEDySPAN 2005, November 2005, pp. 110–118.

[52] B. Razavi, RF Microelectronics, Prentice Hall, 1997.[53] A. Sahai, N. Hoven, R. Tandra, Some fundamental limits in

cognitive radio, Allerton Conf. on Commun., Control andComputing 2004, October 2004.

[54] S. Sankaranarayanan, P. Papadimitratos, A. Mishra, S.Hershey, A bandwidth sharing approach to improve licensedspectrum utilization, in: Proc. IEEE DySPAN 2005, Novem-ber 2005, pp. 279–288.

[55] S. Shankar, Spectrum agile radios: utilization and sensingarchitecture, in: Proc. IEEE DySPAN 2005, November 2005,pp. 160–169.

[56] J.A. Stine, Spectrum management: the killer application ofad hoc and mesh networking, in: Proc. IEEE DySPAN 2005,November 2005, pp. 184–193.

[57] H. Tang, Some physical layer issues of wide-band cognitiveradio system, in: Proc. IEEE DySPAN 2005, November2005, pp. 151–159.

[58] R.W. Thomas, L.A. DaSilva, A.B. MacKenzie, Cognitivenetworks, in: Proc. IEEE DySPAN 2005, November 2005,pp. 352–360.

[59] M.C. Vuran, I.F. Akyildiz, AMAC: adaptive medium accesscontrol for next generation wireless terminals, IEEE/ACMTransactions on Networking, in press.

[60] Q. Wang, H. Zheng, Route and spectrum selection indynamic spectrum networks, in: IEEE Consumer Commu-

nications and Networking Conference (CNCC), January2006.

[61] T.A. Weiss, J. Hillenbrand, A. Krohn, F.K. Jondral,Efficient signaling of spectral resources in spectrum poolingsystems, in: Proc. 10th Symposium on Communications andVehicular Technology (SCVT), November 2003.

[62] T.A. Weiss, F.K. Jondral, Spectrum pooling: an innovativestrategy for the enhancement of spectrum efficiency, IEEERadio Communication Magazine 42 (March) (2004) 8–14.

[63] B. Wild, K. Ramchandran, Detecting primary receivers forcognitive radio applications, in: Proc. IEEE DySPAN 2005,November 2005, pp. 124–130.

[64] D. Willkomm, J. Gross, A. Wolisz, Reliable link mainte-nance in cognitive radio systems, in: Proc. IEEE DySPAN2005, November 2005, pp. 371–378.

[65] IEEE 802.22 Working group, WRAN Reference Model, DocNum. 22-04-0002-12-0000.

[66] RAN Requirements, Doc Num. 22-05-0007-46-0000.[67] DARPA XG WG, The XG Architectural Framework V1.0,

2003.[68] DARPA XG WG, The XG Vision RFC V1.0, 2003.[69] C. Xin, A novel layered graph model for topology formation

and routing in dynamic spectrum access networks, in: Proc.IEEE DySPAN 2005, November 2005, pp. 308–317.

[70] S.A. Zekavat, X. Li, User-central wireless system: ultimatedynamic channel allocation, in: Proc. IEEE DySPAN 2005,November 2005, pp. 82–87.

[71] J. Zhao, H. Zheng, G.-H. Yang, Distributed coordination indynamic spectrum allocation networks, in: Proc. IEEEDySPAN 2005, November 2005, pp. 259–268.

[72] Q. Zhao, L. Tong, A. Swami, Decentralized cognitive MACfor dynamic spectrum access, in: Proc. IEEE DySPAN 2005,November 2005, pp. 224–232.

[73] H. Zheng, L. Cao, Device-centric spectrum management, in:Proc. IEEE DySPAN 2005, November 2005, pp. 56–65.

[74] H. Zheng, C. Peng, Collaboration and fairness in opportu-nistic spectrum access, in: Proc. IEEE ICC 2005, vol. 5, May2005, pp. 3132–3136.

[75] L. Xu, R. Tonjes, T. Paila, W. Hansmann, M. Frank, M.Albrecht, DRiVE-ing to the internet: dynamic radio for ipservices in vehicular environments, in: Proc. of 25th AnnualIEEE Conference on Local Computer Networks, November2000, pp. 281–289.

Ian F. Akyildiz received the B.S., M.S.,and Ph.D. degrees in Computer Engi-neering from the University of Erlangen-Nuernberg, Germany, in 1978, 1981 and1984, respectively.

Currently, he is the Ken Byers Dis-tinguished Chair Professor with theSchool of Electrical and ComputerEngineering, Georgia Institute of Tech-nology, Atlanta, and Director ofBroadband and Wireless Networking

Laboratory. He is an Editor-in-Chief of Computer Networks

Journal (Elsevier) as well as the founding Editor-in-Chief of the
AdHoc Network Journal (Elsevier). His current research interestsare in next generation wireless networks, sensor networks andwireless mesh networks.
http://www.cs.ucsb.edu/htzheng/cognitive/nautilus.html

http://www.cs.ucsb.edu/htzheng/cognitive/nautilus.html


He received the ‘‘Don Federico Santa Maria Medal’’ for hisservices to the Universidad of Federico Santa Maria, in 1986.From 1989 to 1998, he served as a National Lecturer for ACMand received the ACM Outstanding Distinguished LecturerAward in 1994. He received the 1997 IEEE Leonard G. AbrahamPrize Award (IEEE Communications Society) for his paperentitled ‘‘Multimedia Group Synchronization Protocols forIntegrated Services Architectures’’ published in the IEEE Journalof Selected Areas in Communications (JSAC) in January 1996.He received the 2002 IEEE Harry M. Goode Memorial Award(IEEE Computer Society) with the citation ‘‘for significant andpioneering contributions to advanced architectures and protocolsfor wireless and satellite networking’’. He received the 2003 IEEEBest Tutorial Award (IEEE Communication Society) for hispaper entitled ‘‘A Survey on Sensor Networks,’’ published inIEEE Communications Magazine, in August 2002. He alsoreceived the 2003 ACM Sigmobile Outstanding ContributionAward with the citation ‘‘for pioneering contributions in the areaof mobility and resource management for wireless communica-tion networks’’. He received the 2004 Georgia Tech FacultyResearch Author Award for his ‘‘outstanding record of publi-cations of papers between 1999 and 2003’’. He also received the2005 Distinguished Faculty Achievement Award from School ofECE, Georgia Tech. He has been a Fellow of the Association forComputing Machinery (ACM) since 1996.

Won-Yeol Lee received his B.S. and M.S.degrees from Department of ElectronicEngineering, Yonsei University, Seoul,Korea in 1997 and 1999, respectively.From 1999 to 2004, he was a researchengineer of Network R&D Center andWireless Multimedia Service Division atLG Telecom, Seoul, Korea. Currently heis a Graduate Research Assistant in theBroadband and Wireless NetworkingLaboratory, Georgia Institute of Tech-

nology, pursuing his Ph.D. degree under the supervision of Prof.

Ian F. Akyildiz. His current research interests include cognitiveradio networks, next generation wireless networks, and wirelesssensor networks.

Mehmet C. Vuran received his B.Sc.degree in Electrical and ElectronicsEngineering from Bilkent University,Ankara, Turkey, in 2002. He is currentlya Research Assistant in the Broadbandand Wireless Networking Laboratoryand pursuing his Ph.D. degree at theSchool of Electrical and ComputerEngineering, Georgia Institute of Tech-nology, Atlanta, GA under the guidanceof Prof. Ian F. Akyildiz. His current

research interests include adaptive and cross-layer communica-tion protocols for heterogeneous wireless architectures, next
generation wireless networks, and wireless sensor networks.
Shantidev Mohanty received his B. Tech.(Hons.) degree from the Indian Instituteof Technology, Kharagpur, India in2000. He received his M.S. and Ph.D.degrees from the Georgia Institute ofTechnology, Atlanta, Georgia, in 2003and 2005, respectively, both in electricalengineering. He is currently workingwith Intel Corporation, Portland, Ore-gon. His current research interestsinclude wireless networks, mobile com-

munications, mobility management, ad-hoc and sensor networks,and cross-layer protocol design. From 2000 to 2001, he worked as
a mixed signal design engineer for Texas Instruments, Bangalore,India. He worked as a summer intern for Bell Labs, LucentTechnologies, Holmdel, New Jersey, during the summers of 2002and 2003 and for Applied Research, Telcordia Technologies,Piscataway, New Jersey, during the summer of 2004.

Documents

NeXt generation/dynamic spectrum access/cognitive radio ...cs752/papers/cognitive-001.pdf · The key enabling technology of xG networks is the cognitive radio. Cognitive radio techniques