invited paper practicalissuesforspectrum managementwith ... · of the available spectrum. cognitive...

INV ITEDP A P E R

Practical Issues for SpectrumManagement WithCognitive RadiosThis paper discusses some of the practical spectrum management issues

that impact the adoption of cognitive radio technology.

By Stephen M. Dudley, Member IEEE, William Christopher Headley,

Marc Lichtman, Student Member IEEE, Eyosias Yoseph Imana,

Xiaofu Ma, Student Member IEEE, Mahi Abdelbar, Aditya Padaki, Student Member IEEE,

Abid Ullah, Munawwar M. Sohul, Taeyoung Yang, Member IEEE, and

Jeffrey H. Reed, Fellow IEEE

ABSTRACT | The policy of permanently assigning a frequency

band to a single application has led to extremely low utilization

of the available spectrum. Cognitive radio, with its ability to be

both intelligent and frequency agile, is thought to be one of the

prime contenders to provide the necessary capabilities needed

for dynamic spectrum access systems. With this in mind, this

paper discusses the practical issues inherent to the deployment

of spectrum management systems utilizing cognitive radios.

KEYWORDS | Cognitive radio; communication system security;

intelligent systems; radio spectrum management; wireless

communication; wireless networks

Nomenclature

3G Third generation (cellular).

3GPP Third-Generation Partnership Project (cellular).

4G Fourth generation (cellular).

AMT Aeronautical Mobile Telemetry.ADC Analog-to-digital convertor.

ASA Authorized shared access.

AWGN Additive white Gaussian noise.

BRAN Broadband radio access network.CDBS Consolidated database system (FCC).

CEPT Conference of Postal and Telecommunications

Administrations (European).

CR Cognitive radio.

DAC Digital-to-analog convertor.

DSA Dynamic spectrum access.

EC European Commission.

ETSI European Telecommunications StandardsInstitute.

EU European Union.

FBI Federal Bureau of Investigation (the United

States).

FCC Federal Communications Commission.

FSMS Federal Spectrum Management System.

GAA General authorized access.

GHz Gigahertz.GPS Global Positioning System.

HF High frequency.

IA Incumbent access.

IETF Internet Engineering Task Force.

ISM Industrial, scientific, and medical.

LDA Linear discriminant analysis (machine learning).

LSA Licensed shared access.

LTE Long-term evolution (cellular).MAC Media-access control.

MBAN Medical body area network.

MEMS Microelectromechanical systems.

Manuscript received September 30, 2013; accepted December 23, 2013. Date of

publication February 4, 2014; date of current version February 14, 2014. This work

was supported in part by the National Science Foundation (NSF) Enhancing Access

to the Radio Spectrum (EARS) Program under Projects 1343222 and 1247928, and by

the TWC under Project 1228902.

The authors are with the Electrical and Computer Engineering Department,

Virginia Polytechnic Institute and State University, Blacksburg, VA 24061-1019 USA

(e-mail: [email protected]).

Digital Object Identifier: 10.1109/JPROC.2014.2298437

0018-9219 � 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

242 Proceedings of the IEEE | Vol. 102, No. 3, March 2014

MHz Megahertz.Mil STD Military standard.

MIMO Multiple-input–multiple-output (antenna).

NIST National Institute of Standards (the United

States).

NPRM Notice of Proposed Rule Making (FCC).

NTIA National Telecommunications and Information

Administration.

OFDM Orthogonal frequency-division multiplexing.PA Priority access.

PAWS Protocol to Access White Space (IETF).

PCA Principle components analysis (machine

learning).

PCAST President’s Council of Advisors on Science and

Technology (the United States).

PHY Physical layer.

PUE Primary user emulation.REM Radio environment map.

RF Radio frequency.

RFIC Radio-frequency integrated circuit.

SAW Surface-acoustic wave.

SDR Software-defined radio.

SNR Signal-to-noise ratio.

SON Self-organizing networks (cellular).

TDM Time-division multiplexing.TV Television.

VoIP Voice-over-Internet Protocol.

Wi-Fi Wireless fidelity (IEEE 802.11).

WiMAX Worldwide interoperability for microwave

access.

WRAN Wireless regional area network.

I . INTRODUCTION

Some predict that demand for cellular spectrum could see

a 1000� rise over 3G demand [1]. Notwithstanding the

move to smaller cells and off-loading to Wi-Fi, not enough

spectrum appears to be available to meet this demand. Part

of the ‘‘spectrum scarcity’’ problem is that the regulatory

approach to spectrum allocation is extremely inefficient,

resulting in spectrum being allocated but unused, or usedonly occasionally, in many areas [2], [3].

A major factor driving us toward new spectrum-sharing

models is that the electromagnetic spectrum has been

shown to be a catalyst for economic development [4].

Eager to further spur their economies, nations are

competing for leadership in technologies that allow for

spectrum sharing [5]–[8]. CRs, which are capable of DSA,

have been claimed by proponents in the European Unionand the United States as the most effective solution to the

spectral congestion problem. CRs can be a critical

technology to spur positive change in the current

regulatory policies (see [2], [9]–[11], and article 4.1 of

[12]). Recently, there have been several efforts to

incorporate CR and DSA technologies into the develop-

ment of standards for a wide variety of applications (see

Table 1). While spectrum sharing is not new (e.g.,Automatic Link establishment), the use of CR technologies

may make it easier to develop and deploy systems that

share spectrum.

In some ways, DSA is a radical change from fixed

spectrum assignment. A change like this will not happen

overnight, nor will it happen through technology advance-

ment alone. Given these facts, this paper looks at some of

the practical issues in the deployment of DSA using CRs.

A. Paper OutlineThis paper is divided into six sections. Section II

provides some background information about new licens-

ing models currently being debated within regulatory

agencies, and an overview of a recently completed DSA

standardVIEEE 802.22. Section III considers spectrum

Table 1 Applications of DSA Technologies

Dudley et al. : Practical Issues for Spectrum Management With Cognitive Radios

Vol. 102, No. 3, March 2014 | Proceedings of the IEEE 243

management systems, and the various architectures that

have been suggested for providing legacy user protection.Section IV looks at how individual radio nodes achieve

frequency agility, and the issues that are faced in designing

systems with these capabilities. Section V provides a brief

excursion into machine learning techniques. Section VI

considers oversight and regulatory issues. Section VII

concludes with a discussion of the implications on future

standards of the topics considered in the paper.

II . BACKGROUND

One of the implications of dynamic spectrum sharing is the

need for new licensing models. Both the European Union

and the United States are taking steps to update their

regulatory processes to include new forms of licensing,

with commercial interests providing input [24]. In the areathat has been explored the longest, TV white space, the

IEEE 802.22 standard has been created.

A. Licensing ModelsThe U.S. PCAST issued a report on July 20, 2012,

relating to spectrum allocation [2]. Some of the high level

points made in that report are shown in Fig. 1 and are

addressed in this section. The report recommended athree-tiered Federal Spectrum Access System that involves

classifying users into one of the following categories: 1) IA;

2) secondary access; and 3) GAA.

The incumbent users (also know as primary users)

would receive protection from interference from all other

types of access. Secondary access users would be issued

short-term priority operating rights in a specific geograph-

ical area, but must vacate the band when an incumbent is

present.The report noted that receiver performance can limit

efficient use of the spectrum and recommended the

management of receivers as well as transmitters. Declar-

ing a receiver’s tolerance for interfering signals makes it

easier to manage spectrum decisions when new entrants

arrive.

It was also recommended that the model for spectrum

allocation be changed from the current ‘‘semipermanentassignment’’ system to a more flexible system that adds

short-term and medium-term licenses. Medium-term

licenses are envisioned to be three or fewer years and

have no automatic rights of renewal. This scheme would

allow firms to experiment with networks and business

models. It would also provide assurance to government

agencies that may need additional spectrum in the near

future. A short-term license, on the other hand, involves afixed set of standards, terms, and conditions, so as to

accelerate the agreement process. The idea of a short-term

license is new, and the meaning of ‘‘short term’’ will likely

be shaped by the market over time. An example short-term

license might be renting spectrum over the course of a few

days when the Super Bowl is in a city in order to provide

adequate cellular service. The addition of medium-term

and short-term licensing will likely spur new and innovatewireless applications.

ASA is an alternative spectrum-sharing structure that

involves two tiers. ASA started as a joint project by seven

companies (Digital Europe, Ericsson, Huawei, Intel,

Nokia, NSN, and Qualcomm) in order to promote the

use of unused spectrum in Europe through DSA and CR

Fig. 1. Licensing models [2], [24].



technology [12]. The ASA concept has been named LSA bythe Radio Spectrum Policy Group within the European

Union. LSA seeks to grant shared access licenses to operate

on bands that have been licensed to other users [12]. One

operator can pay another operator for a certain amount of

bandwidth, at a certain quality of service. This type of

system uses the idea of a licensed secondary user, in which

the secondary user has rights to the spectrum in locations

(be it time, frequency, and/or geography) where theprimary user is not present. A licensed primary and

secondary user can be thought of as two licensed primary

users, where one has priority over the other. Having

licenses for shared access allows operators to predict the

quality of service.

The U.S. FCC is evaluating schemes for DSA [25]. The

NPRM for the 3.5-GHz band echoes the PCAST three-

tiered structure [8]. The NPRM describes IA in the 3.5-GHzband as belonging to authorized federal users and grand-

fathered fixed satellite services. The second tier is the PA,

which includes users with a definite need for guaranteed

access to broadband spectrum at specific facilities such as

hospitals, utilities, government installation, and public safety

entities. Tier three is theGAA, for opportunistic use on

noninterfering bases in specified geographical areas.

The European Commission is also taking spectrum-sharing initiatives. The Radio Spectrum Policy Group, an

advisory group that assists the European Commission,

released a report that explores dynamic spectrum-sharing

approaches and the identification of higher frequency

bands that could be candidates for shared spectrum [12].

Similar to PCAST, this report proposes multiple modes of

spectrum sharing in order to remain flexible. The

European Commission’s proposal differs from PCAST’s inthat Europe is targeting all spectrum with its sharing

initiative, while the United States is so far primarily

interested in sharing of TV white space and spectrum

currently allocated for government use [12].

B. IEEE 802.22 Wide-Area Regional NetworkIEEE 802.22 is a standard for WRANs using the TV

white space [26]. The primary application of thistechnology is providing wireless broadband access to rural

homes. Using DSA techniques, IEEE 802.22 allows the

sharing of spectrum allocated to TV broadcasters, as well

as other incumbents such as wireless microphone systems.

IEEE 802.22 uses a point-to-multipoint wireless access

system (similar to cellular), formed by a base station and

connected users. Spectrum sensing data are fused with

occupancy data retrieved from a database service and usedby the base station to determine which channels are

available for operation. As such, each base station must

include a GPS receiver for geolocation. To deal with low-

power incumbents (e.g., wireless microphones), the IEEE

802.22.1 amendment was created to describe a beacon

signal that can be transmitted in order to indicate the

incumbent’s presence. In IEEE 802.22, the spectrum

manager acts as the cognitive engine inside each base

station, and is responsible for tasks such as maintaining

spectrum availability information, channel selection and

management, scheduling spectrum sensing, enforcing

regulatory policies, and allowing self-coexistence. Fig. 2

shows the logical interfaces of the IEEE 802.22 spectrum

manager.The standard’s PHY implementation, which is OFDM

based, can support a cell radius up to 30 km. To support

multiple IEEE 802.22 cells in close proximity, self-

coexistence procedures are defined. The IEEE working

group for IEEE 802.22 was formed in 2004, and the first

standard was published in July 2011. As of the writing of

this paper, FCC rules require each spectrum manager to

check the FCC database daily, but no rules have yet to beestablished to cover sensing applications (e.g., detecting

wireless microphones operating in the same band, for

example).

C. IEEE 802.15.4j Medical Body Area NetworkCurrent MBAN solutions are mostly deployed in the

2.4-GHz ISM band. Since the ISM band is becoming

more and more crowded, interference could limit large-

scale deployment of MBAN applications. In the United

States, the FCC recently allocated the 2360–2400-MHz

spectrum for MBAN use on a secondary basis, sharing thespectrum of aeronautical mobile telemetry. The IEEE

802.15 Task Group 4j (TG 4j) has been developing an

amendment to the IEEE 802.15.4 standard to extend its

PHY and MAC layer solutions to the MBAN spectrum.

DSA protocols play an important role in allowing

MBAN operations since aeronautical mobile telemetry

devices are mostly receive-only. To avoid requiring explicit

coordination between primary and secondary users on acase-by-case basis, a beacon signal, called a hub, is placed

near the patient, and low-power transmitters on the body

can transmit only when they receive the beacon. The hub

gets its authority to transmit from a key that it receives as

part of the general coordination process that grants

spectrum access to the medical facility. For its part, the

FCC decided not to require all users to apply for

Fig. 2. IEEE 802.22 system functions [26].



authorization. MBAN users operate on a secondary user

basis, meaning they must not cause harmful interference

and they must accept all interference from incumbents.

III . SYSTEMS FOR SPECTRUMMANAGEMENT

Fig. 3 provides a pictorial overview of the spectrum

management system assumed in this paper.Introduction of CRs into spectrum sharing and

spectrum management opens up many avenues for

efficient system design. It also introduces new challenges,

some of which will be discussed in this section. As shown

in the figure, this paper assumes that CRs have frequency

agility and spectrum sensing capabilities. The location of

cognitive engines and knowledge bases typically depends

on the nature of the spectrum management system.This paper assumes a classification of spectrum

management systems into the categories of supervised

systems and unsupervised systems. For details, see Table 2.

(Note that other taxonomies exist [27].)

A. Legacy Protection and Commercial ViabilityWithout protection for legacy users, spectrum sharing

will never be allowed by regulatory authorities. This makeslegacy assurance the most important practical issue for

DSA. The nature of the incumbent user and the potential

secondary user significantly affects willingness of the

incumbent user to enter into a sharing relationship.

Research has not yet provided a reliable mechanism to

gauge the level of protection that each type of DSA system

can provide. This is still a very new field, and it is

impossible, simply by doing thought experiments or

running simulations, to know what real problems will

come up when systems are actually deployed.

Given that DSA is not just a single system, but rather a

general philosophy for many different kinds of systems,there are many cases for which metrics need to be

identified, as well as decisions made about mechanisms to

gauge effectiveness of legacy protection. What we have

seen so far is willingness of some users (e.g., AMT) to

allow sharing with some organizations (e.g., the medical

community) to allow a specific application (e.g., MBAN)

Fig. 3. Example spectrum management system.

Table 2 Spectrum Management System Types



where the primary user has a relatively straightforwardmechanism to identify and hold the organization account-

able (e.g., allow only one user, or conform to a

coordination process). It is unclear how long it will take

to reach the point where all spectrum can be accessed by

any user simply by petitioning the spectrum management

system and complying with its rules. It should be expected

that there will be many trials of different types of spectrum

management systems, with different kinds of primary andsecondary users, before confidence in such systems will

reach that tipping point.

B. Choosing the Right ArchitectureChoosing the right architecture is both a function of

the technical characteristics of the system, and the

political realities of the parties involved. The question of

tangible benefits to primary users has directed some of thediscussion about regulatory policy and the types of

cooperation that can be achieved [28]. If the primary

user has no mechanism to benefit from sharing spectrum,

then cooperation is unlikely. For example, a system that

incorporates spectrum trading may not be viable if the

primary user is a federal organization that is forbidden by

law from receiving compensation for spectrum.

The ‘‘right’’ architecture is still being debated in bothresearch and commercial communities. For many incum-

bents, even if they are using spectrum inefficiently, they

see huge risks in admitting secondary users. Although early

research favored unsupervised systems using a pure

sensing-based approach, the failure to provide solid

guarantees of protection led to a shift toward researching

centrally supervised systems and databases. Research into

decentralized approaches is just beginning [29]. Researchis ongoing with all three types (unsupervised, centrally

supervised, and distributed supervision). Whether there

will be many flavors of systems, each tailored to a specific

circumstance, or whether there will be adrift toward a

general strategy for all DSA systems is not clear at this

point.

An unsupervised approach does not suffer issues with

latency in decision making, and is relatively low cost.However, selfish optimization without the knowledge of

the network can result in low overall spectrum efficiency.

In addition, there are types of primary users for which this

type of system would be unsuitable. Given these facts, the

design of algorithms for these systems also includes the

study of game theory, in which games are devised that

attempt to provide minimum performance guarantees

[30], [31].Decentralized supervision uses network information

from neighbors to make better decisions than could be made

if restricted to its own sensors and data (see Section III-C). Of

the three general approaches, it is the one most suited for

implementation with CRs since it requires collaboration to

achieve its goals. However, as noted in [32], a decentralized

approach may not achieve the global optimum solution if it is

not omniscient. A research direction for decentralizedapproaches is the utilization of distributed databases to

create more reliable REMs [29].

Centrally supervised architectures are commonly

touted as most easily accommodating legacy assurance.

This is due to the fact that they can be made to comply with

arbitrary policies or levels of security and do not require

radios to have a high degree of cognitive capabilities (yet

can accommodate those that do). However, there is asignificant tradeoff between complexity, cost, and latency.

Design approaches for these kinds of systems are still in

their infancy. IEEE 802.22 is an example of a centrally

supervised system, as would be systems that involve the

sharing of spectrum with cellular service providers (e.g., a

proposal for allowing cellular providers to use naval radar

spectrum at inland locations).

Whether systems are unsupervised, have distributedsupervision, or are centrally supervised, there may be a

need for sensing. Likewise, there may be a role for the use

of databases to maintain information about the environ-

ment. That information could be static (e.g., TV station

locations) or dynamic (e.g., sensing of wireless micro-

phone usage). Issues with sensing and databases will be

discussed in subsequent sections.

C. Collaborative Sensing IssuesCollaborative sensing is one of the solutions proposed

for several sensing problems: sensing accuracy, spectrum

coverage, processing complexity, and energy efficiency

[33]–[38]. The intelligent combining of the spectrum

sensing data and/or decisions from multiple CRs can

improve the sensing accuracy (with respect to the

probability of detection and false alarm) while potentiallylowering the sensing requirements of all of the CRs in the

group (with respect to sensing time, sensing approach, and

energy usage) [38], [39].

Issues faced by collaborative sensing systems are the

overhead that collaboration creates, the management of

sensors (sensor clustering), fusion of data from sensors,

and distributed spectrum search. Each will be considered

in subsequent sections.

1) Overhead: There are two main approaches to

collaborative spectrum sensing: centralized approaches

[40] and distributed approaches [41]. In centralized

approaches, the cooperating CRs send their spectrum

sensing information to a data fusion center. The data

fusion center then makes global spectrum decisions for all

of the cooperating CRs. In distributed approaches, each CRmakes its own sensing decisions based on sensing

information from cooperating radios. In each of these

approaches, there is a natural tradeoff between the per-

formance benefits gained from collaboration and the

amount of overhead required [42]. This overhead takes the

form of information that must be passed among the CRs

(in distributed approaches) or to a data fusion center (in



centralized approaches), thus increasing network trafficand energy needs.

In the literature [38], [43], it is typically assumed that

the collaboration channels (channels by which the

cooperating CRs communicate with the fusion center

and/or each other) are ideal. An issue not typically

considered in the literature that needs additional research

is the impact of nonideal channels [44], [45].

2) Appropriate Clustering of Sensors: It is known that

there are diminishing returns on the performance gains

that can be achieved through collaboration as the number

of collaborating CRs increases. Therefore, when develop-

ing collaborative approaches to spectrum sensing, it is

important to consider how best to cluster the CRs to

achieve adequate performance gains while keeping energy

needs reasonable. In other words, determining what is theappropriate way to group CRs together for the best

cooperation results [46]–[48].

3) Fusion of Potentially Diverse Sensing Data: There is a

natural tradeoff between collaborative sensing perfor-

mance and overhead between using hard and soft

decisions. Hard decisions, in which each CR sends a

distinct, or ‘‘hard,’’ decision to the fusion center, will tendto lower overhead with respect to data transmissions in the

cooperation channels. However, there is a fundamental

performance loss that will occur as a result of losing some

of the detailed sensing information [43].

On the other hand, soft decisions provide more of the

detailed sensing information to the fusion center, leading

to more intelligent cooperative spectrum sensing decisions

to be made. Naturally, each CR sending more detailedspectrum sensing information will lead to an increase in

data transmission overhead. Additionally, soft decisions

are more prone to errors caused by nonideal collaboration

channels [49].

In an attempt to balance these tradeoffs between hard

and soft decisions, so-called softened hard, or hybrid,

decisions have been proposed for data fusion; see, for

example, [50]. These approaches can provide lower overheadthan purely soft decisions, while improving upon the sensing

performance with respect to purely hard decisions.

Another critically important practical issue with

collaborative data fusion deals with the consideration of

nonidentical, or heterogeneous, CR sensing capabilities

when performing collaboration. In the literature [49], the

majority of works on collaborative spectrum sensing

assume that the sensors have identical sensing capabilitiesand/or sensing channels. For example, they typically

assume each sensor uses the same sensing approach

(cyclostationarity, likelihood-based, cumulants, etc.) or is

impacted by the same type of channel.

4) Distributed Spectrum Search: Sharing between local

databases has been proposed as an enabling approach for

spectrum search in the wide spectrum bands [51]. Theissue is that understanding the entire radio environment is

a challenging task for a single node. Distributing the task of

spectrum search among multiple nodes reduces the impact

to each individual node.

Conducting a spectrum search can be a multilayer

process [52]. A receiver can be dedicated to scanning the

spectrum, and the results maintained in a database. In

addition, some MAC protocols have a learning feature thatcan characterize availability of channels based on past

decisions [52], [53]. Optimizing the placement of learning

processes, and the use of databases and cognitive engines

to exploit this information, is an ongoing research area.

D. Database ReliabilityInformation reliability is critical for database systems.

As an example, geolocation databases are envisioned tomaintain data about primary user locations and their

operation parameters (e.g., transmit power, antenna

parameters, protection contours, and operational periods)

[54]–[57]. Spectrum availability is calculated based on

these parameters [58]. Unreliable data would lead to

allocation of spectrum bands that are already in use by

primary users.

Given that the spectrum management system isdatabase driven, primary user data are acquired by

database administrators from regulatory databases (e.g.,

the FCC CDBS [59]) and potentially through web

registration methods. The database administrators [56],

[57] and regulators (e.g., FCC, NTIA, and CEPT) must

ensure that primary user registration and update mechan-

isms provide complete and accurate data. The reliability of

database information is a subject of research [60]. Metricsare being devised to measure that reliability. These include

utility, objectivity, and integrity of data. Utility is the

usefulness of information to its intended user. Objectivity

is the presentation of information in an accurate, clear,

complete, and unbiased manner. Integrity is the protection

of information from unauthorized access or revision in

order to ensure information is not compromised through

corruption or falsification.The IETF PAWS outlines secondary user device

registration to a database, as well as spectrum requests

and database responses [61]. The protocol also provides

certain security measures for protection against PUE

attacks, spoofing the database, modifying or jamming a

query request/response, and tracking the white space

device location and identity. Significant effort is being

applied to improve the reliability of primary user’sspectrum usage data in the regulatory spectrum licensing

databases [59], [62].

Conventional data reliability methods cannot address

the problem of primary user parameter inference through

normal secondary user database queries [63], [64]. For

example, is it necessary to prevent a secondary user from

inferring the presence or capabilities of another user (say



radar, or FBI communications)? These techniques createprimary user privacy concerns in spectrum-sharing data-

bases, some of which will be addressed in Section VI-D.

Research is still needed in the area of primary user

operational privacy concerns within spectrum management

systems.

The database enabled spectrum sharing will be divided

into categories of federal and nonfederal spectrum users.

The NTIA FSMS will maintain operational data aboutfederal primary users like radar systems, government

agencies, etc., and the FCC CDBS will maintain data about

nonfederal spectrum users like TV stations, satellite

systems, cellular systems, etc. NTIA will make available

bits and pieces of spectrum that is not used by federal

systems in frequency, space, and time to the geolocation

database manager for secondary use. This categorization of

spectrum users will address the concerns of the federalagencies whose operational privacy is one of the major

hurdles holding back the federal government from opening

up some of its spectrum for sharing with commercial

systems.

E. System/Network LatencyLatency in system interactions is a practical issue facing

developers of spectrum management systems. As with anycontrol system, the latency of the decision cycle can affect

system stability, as well as limit the maximum number of

clients supported. The system can have two kinds of

latencies. The first relates to how long it takes to establish

an initial security and trust association. The second relates

to how long it takes to query the database/server and

receive a response. If the system is expected to provide

guarantees of maximum latency in order to respond toevents when all the nodes need to (re)register with a

database at once, designing the servers can become an

economic issue involving both sizing of the server

infrastructure and the transport network. Although there

is no strict need for research into better strategies to

implement the network pieces of the system (since such

network design techniques are well known), choosing the

best strategy requires careful engineering, and may benefitfrom future research in the area of forming security and

trust relations.

IV. THE RADIO

A. Cognitive RadioFor the purposes of this paper, a CR is defined as a

radio that has sensing capabilities, frequency agility, and is

operating in a spectrum management system under

cognitive control. A practical CR measures the state of

the spectrum using a locally implemented spectrum sensor

and/or by probing a network level database. The assump-

tion in this section is that CRs have a sensing receiver. If

there are quiet periods in the operation, a CR can use its

main receiver as a sensing receiver so a secondary receivechain is not always required. It is also noted that since the

purpose of this receiver is detection of a signal, not

decoding of the message, that processing gains can

sometimes be achieved.

B. Radio ComplexityImplementing DSA techniques in the radio, as

discussed in this section, has an impact on radio cost.One aspect of this cost is software. DSA applications often

need the ability to be reconfigured in real time. Although

software techniques to do this are well known, real-time

configuration software is more difficult to write. This

comes partly from the challenges imposed by software

validation and general testing, partly from the need to

address new security issues, and partly from the large

number of use cases inherent to DSA.DSA strategies within the radio also have to compete

directly with simple packaging exercises that place

multiple RF chains in a single device such as is currently

done for smartphones that operate with 3G, 4G, LTE,

Bluetooth, and Wi-Fi. In the sections that follow will be a

brief discussion of hardware issues. This includes:

achieving frequency flexibility in the RF chain, searching

the spectrum for primary users, tradeoffs related tospectrum sensing, and the availability of propagation

models for higher frequencies.

C. Achieving Frequency FlexibilityCRs need to be operational over multiple frequency

bands. Using a sufficiently selective flexible RF removes

the need for high-performance ADCs and DACs in CRs.

Frequency flexibility can be achieved through the follow-ing: wideband tunable oscillators, wideband antennas, and

wideband tunable filters in transmitters and receivers (see

Fig. 4 and discussion below).

1) Tunable Oscillators: RFICs with wide tuning range

have already been reported and are available in the market

[65]. However, oscillator noise is still a problem that needs

more research.

2) Wideband Antennas and RF Amplifiers: There is a

fundamental limit theory addressing the well-known

tradeoff between antenna size and achievable instanta-

neous gain bandwidth for transmit antennas [66]–[69].

For receive antennas different limits apply. The funda-

mental limit of receiver antenna size is related to SNR

bandwidth and the noise figure of the antenna [70]. Noiseand selectivity of RF front–ends are dominant factors

causing practical spectrum sensing issues in the receiver.

Mitigation of coupling effects between small form-

factor MIMO systems remains a challenge. However,

ultrawideband electrically small antenna designs covering

3.1–10.6-GHz band have been extensively investigated in

the literature (see [71] and [72] among others), and



commercial off-the-shelf products are available in the

market. Practical issues for designing receive antennas

come from performing tradeoffs between selectivity and

sensitivity of the RF front–end as opposed to being unable

to design wideband antennas.

Superwideband low-noise amplifiers and mixers with

good linearity performance, as well as wideband power

amplifiers with good linearity and efficiency performance,have been proposed [73], [74]. However, a similar level of

progress has not been achieved in RF filtering. The

problem of RF filtering is a major challenge for making CR

a reality [75]. This is mainly because of the technical

tradeoffs between high selectivity and broad tuning range

in RF filters.

3) RF Bandpass Filters in Receivers: Pre-selector(bandpass) filters are often placed at the input of the

main receiver and the sensing receiver. The pre-selector

filters attempt to maximally reject all undesired adjacent

channel signals with minimal effect on the desired signal.

Without pre-selector filters, an excess amount of undesired

signal enters the receiver and distorts the desired signal.

This degrades both the quality of the signal for demodu-

lation purposes and for spectrum sensing purposes.The selectivity specification of the pre-selector filter

can be very stringent when adjacent channels are very

close to the desired frequency band. The selectivity

specification can be even more challenging in full duplex

systems where the full transmit power may appear on the

receive antenna. Noise generated by a wideband transmit

power amplifier, as well as leakage from the transmitted

signal, can easily degrade the sensitivity of the receivers.Therefore, selectivity of wideband tunable RF filters in

DSA capable receivers must be comparable to fixed

frequency RF filters found in devices such as cell phones.

There are two traditional approaches to construct

frequency flexible RF filters in receivers, multiband filters

and tunable filters. A nontraditional approach is relaxing

the selectivity specifications of the filters.

• Multiband filters (using multiple frequency-

dedicated filters): A major challenge with this

strategy is the insertion loss of the switchingnetwork (which affects the SNR), and the physical

size of each filter. Miniaturization of surface

acoustic wave and bulk acoustic wave filter tech-

nologies is addressing the latter problem. For

example, some commercial smartphones support

a dozen frequency bands on a single device using

very small fixed filters. However, as the number of

required frequency bands increases, managingthem becomes more costly, and at some point

becomes impractically large.

• Tunable filter: RF filter implementation technol-

ogies such as SAW-based devices and ceramics are

not suitable for making tunable filters. For this

reason, different technologies such as semicon-

ductor varactors, ferroelectric materials, and

MEMS are used to make tunable filters. However,flexible RF filter technologies provide significantly

inferior selectivity performance compared to fixed

RF filters [76].

• Relaxing selectivity specification: When RF filters

cannot be sufficiently selective, approaches that

relax the selectivity specifications can be consid-

ered. For example, designing the network to be

half-duplex significantly relaxes the selectivityspecifications of the RF filters. In [77], Marshall

proposed a network level technique in which

Fig. 4. Achieving frequency selectivity over a wideband.



transmission frequencies are selected such that theexpected nonlinearity distortion in the receiver is

minimized. Marshall showed that this technique

has the potential of significantly relaxing the

selectivity and linearity specifications of the

receiver.

Incorporating cognitive control of receiver parameters,

such as sampling frequency and local oscillator frequency

of the receiver, relative to the state of the spectrum, hasbeen proposed. This technique potentially relaxes the

selectivity and linearity specifications of the receiver since

it avoids worst case situations that may be a lot worse than

the expected cases. Techniques to reconstruct the effects

of RF front–end nonlinearity in wideband spectrum

sensing have also been proposed that would allow

mitigation to be applied, thus further reducing the worst

case nonlinearity.1

4) RF Bandpass Filters in Transmitters: For efficient

spectrum sharing, each transmitter needs to emit the

desired signal with minimal interference on other

frequency bands. Undesired transmissions can originate

from harmonics, spectral sidelobes due to discontinuities

in the pulsing of the signal, and spectral regrowth due to

power amplifier nonlinearity.The same mitigation strategies for receivers (multiband

filters or tunable filters) can be used to eliminate

transmissions outside the desired band. Because harmo-

nics are relatively far off in frequency, the specification of

the transmitter’s RF filter is not as strict as that of the

receiver. Tunable filters are not mature enough for RF

filtering in the transmitter.

In terms of relaxing the need for tight filter specifica-tions, the transmitter can also implement one or more of

the following: digital pulse-shaping filters to reduce

spectral sidelobes, backoff in the power amplifier to

reduce spectral regrowth, and implementing half-duplex

(e.g., time-division duplex) versus full duplex systems.

D. Spectrum SearchFast and reliable spectrum search is an important

aspect of white space characterization. One way to detect

available white space is to test for primary users operating

within the communication range of a secondary user.Searching for primary users over a wideband spectrum

sensing allows nodes to search for primary user signals in a

wide spectrum band [78].

Initial approaches to wideband spectrum sensing used

multiresolution spectrum radios based on wavelet trans-

form [79] and fast Fourier transform processors [80]. The

idea is to estimate the power spectral density of the

wideband signal using an RF front–end with a widebandarchitecture. Most of the existing literature on wideband

sensing focuses on improving the detection reliability [81]of the primary user signal.

A significant practical issue is the time it takes for

reliable detection. A common research goal is to speed up

the channel search process by designing a sensing policy

that identifies the desired signals quickly. In [82], random

and serial search mechanisms are proposed for indepen-

dent and correlated channel occupancies. The perfor-

mance of these search schemes is compared in terms of themean time to detect the primary user signal. In [83] and

[84], a two-stage channel search is proposed to reduce the

average channel search time by focusing on frequency

channels which are more likely to be vacant. These

approaches perform better than conventional methods

when the spectrum is crowded. In [85], a wideband

channel clustering scheme is proposed to cluster users

with smaller bandwidth requirements in order to avoidunnecessary blockage to users with larger bandwidth

requirements.

E. Sensing Constraints and TradeoffsAt the expense of implementation complexity, multi-

layer and clustering-based channel search can be used to

reduce the average channel search time by giving unequal

focus on different channels based on a priori information[83], [84]. The secondary user applies single-channel

sensing on each individual channel to determine if it is

vacant. Practical issues involved in single-channel sensing

are discussed below.

1) Time Constraints: There are two types of CR

architectures typically discussed that are relevant to

spectrum sensing, single-receiver designs, and dual-receiver designs [86]. In single-receiver designs, spectrum

sensing must be accomplished during the so-called ‘‘quiet

times’’ when data transmission is idle. This architecture

inherently requires a time slot for sensing and a time slot

for data transmission. In dual-receiver designs, these quiet

times are only necessary when sensing must be performed

on the band being used for data transmission. The

necessity of quiet times to perform spectrum sensingleads to a tradeoff between spectrum sensing accuracy and

data throughput [87].

Detrimental channel effects, such as shadowing and

multipath, determine how long spectrum sensing decisions

will remain relevant (due to Doppler spread and channel

coherence effects, for example), thus dictating the needed

update rate. For example, spectrum sensing in mobile

applications, where the wireless channel is changingrapidly, requires much faster sensing capabilities than

static or slowly varying channels.

In many cases, the specific spectrum management

application will provide constraints upon the availability

and length of quiet times. For example, in IEEE 802.22, it

is specified that the maximum time it should take to detect

licensed primary users within a given channel, during

1This technique is developed at Wireless@Virginia Tech. Papersdescribing the technique were recently submitted for publication.



normal operation, should be much less than 2 s, under theconstraints of over 90% probability of incumbent detection

and a false alarm rate of less than 10% [26]. In more

unpredictable and critical spectrum-sharing applications,

where a variety of wireless devices and wireless commu-

nication standards are simultaneously in operation (such

as in hospital environments), the sensing time and update

rate need to be much faster [88].

2) Performance Tradeoffs: There are a large number of

spectrum sensing algorithms to choose from. In general,

low complexity spectrum sensing algorithms (e.g., energy

detection [89] and eigenvalue decomposition [90]) per-

form poorly in bad channel conditions, but may be the only

choice given receivers with low processing power. The

setting of detection thresholds for these cases is still a

highly researched topic [91], [92]. On the other hand,more accurate and highly complex algorithms (e.g.,

cyclostationarity detection [93], [94], wavelet approaches

[95], and cumulants [96]) provide more information about

the signal than mere detection. These algorithms typically

require some sort of pattern recognition approach in order

to perform decision making, leading to additional

complexity and overhead. For more in-depth details of

different spectrum sensing algorithms and their associatedtradeoffs, the reader is referred to [31], [33], [34], [86],

and [97]–[99].

When performing spectrum sensing, the performance

of the sensing algorithm improves as the observation time

of the signal of interest increases. This tradeoff between

performance and observation interval is typically referred

to as processing gain. In DSA applications, the optimum

point is dictated by the tolerance of the primary spectrumuser to interference. In other words, the time required to

sense the primary user and vacate its spectrum is a

deciding factor for system viability.

3) Environmental Constraints: An important practical

issue to consider when choosing a spectrum sensing

algorithm is the sensing channel. Although AWGN and

flat-fading assumptions are sufficient under some condi-tions (e.g., in sparsely cluttered line-of-sight environ-

ments), they do not suit more general wireless operating

environments, where multipath and frequency-selective

fading effects may be prominent. As a means of dealing

with these issues, MIMO systems have been discussed as a

possible solution [100], [101]. Exploiting resolvable

multipath with modified spectrum sensing algorithms

such as modified energy detectors and equal gain detectioncan actually improve the detection performance. However,

these approaches introduce their own practical issues (i.e.,

interantenna spacing to guarantee independency of signal

streams).

Another issue that has a severe effect on sensing

reliability is the hidden node problem [86]. If the DSA

transceiver is unable to detect the primary user, either

because of physical location, severe multipath fading, orshadowing, this will cause unwanted interference to the

primary user. Cooperative sensing is proposed in the

literature for handling this kind of practical issue [102].

F. Need for New Propagation ModelsPropagation models are crucial for spectrum manage-

ment systems to ascertain the spectral reuse factor and to

obtain interference maps. Ample propagation models areavailable at conventional frequency bands, typically up to

3 GHz. However, propagation models and vector channel

models for frequencies greater than 3 GHz is an area that

is still developing. Several attempts have been made by

researchers to obtain the propagation model with the

popularization of the WiMAX standard. However, there is

an ample scope for developing better propagation models.

A propagation model specifically designed for the3.5-GHz range is the Stanford University Interim model

formulated for the IEEE 802.16 working group [103]. The

ECC-33 model was a new propagation model developed by

fitting the Okumura data, and works similar to the popular

Cost Hata 231 model. The WINNER project has developed

channel models, which are classified as 3GPP spatial

channel models for MIMO [104]. However, they are

geometry-based channel models and not specific to 3.5 GHzor small cells. Phase 2 of the project included several

system level simulation studies and presented an in-depth

study of outdoor-to-indoor communication channels [105].

Not all of these models are specific to high-frequency

channels being considered for commercial use on a sharing

basis. At present, the Stanford University Interim or

ECC-33 models provide reasonable fits, but not good

enough to ensure an adequate quality of service. Moreover,exhaustive building penetration studies and vector channel

models for MIMO systems need to be carried out for

successful system implementations at these frequencies.

One intriguing idea suggested by Ronan Farrell is the

use of CRs to improve models by comparing what the

model predicts with what the radio is actually seeing. In

essence, this is a call for a longitudinal study of radio

models, looking for inferences about when or where themodels do not match actual conditions.

G. Decision Reliability in a DSA EnvironmentTraditionally, propagation models (sensing papers)

have only concerned themselves with the channel between

the primary user transmitter and the sensor. From this

channel model, they determine a probability of detection

and probability of false alarm of the primary user. MichaelMarcus has proposed that this view be amended to include

all three channels between the primary user transmitter,

secondary user, and primary user receivers [106]. The

reasoning behind this is that there may be correlation

between two or more of the channels. This makes it more

difficult to answer the question of how likely it is that the

failure to detect a primary user signal down to some



arbitrary level (say �120 dBm) means that the secondaryusers’ transmission will be below some threshold (say

�80 dBm) by the time it reaches any primary user.

V. COGNITIVE ENGINES

Cognitive engines are implied for spectrum management

systems that employ CRs. Cognitive engines may exist

either in the radio (for unsupervised systems or distributed

supervision systems) or in designated servers somewhere

in the network. Cognitive engines are based on either

machine learning or optimization techniques, and themost common techniques are listed in Table 3. The tasks

that cognitive engines are assigned include achieving

situation awareness, parameter adaption, and pattern

recognition. Each of these tasks, and issues that they

engender, will be considered in this section.

A. Achieving Situation AwarenessAs shown in Fig. 5, situation awareness is the first step

in the cognitive cycle. A given situation can be described

by the environment, user’s needs, or any other observable

problem. The goal of situation awareness is to use previous

knowledge that was gathered in the past under the same,or similar, situations for decision making purposes.

Therefore, a major component in situation awareness is

the ability to classify current situations in order to see if itis similar to one or more previous situations.

The use of situational awareness leads to a cognitive

engine that has a case-based reasoning component to it.

This type of cognitive engine uses optimization augmented

by learning. For example, a cognitive engine may use a

genetic algorithm for optimization and a neural network to

perform situation classification, but a case-based reasoner

to store what it has learned. The result is a radio that cansave time and/or reach better solutions by leveraging

previously gathered knowledge. A practical issue

concerning case-based reasoning is the amount of memory

required to hold previous cases, actions, and results.

Solutions include forgetting the oldest case (treating it like

a first-in–first-out buffer) and eliminating cases that had

poor results when compared to similar cases [108].

B. Choosing Utility Functions forParameter Adaption

In CR applications, prior knowledge of a given situationis typically in the form of optimization results; in other

words, identifying the best set of knobs (adjustable

parameters) for a given situation. The best set of knobs

can then be immediately used by the radio, or can act as a

seed for additional optimization [108]. Table 4 lists

common knobs and meters in CR applications.

When performing these multiobjective optimizations, a

utility function must be chosen. When multiple factors areinvolved, the typical strategy is to turn the problem into a

single objective optimization by assigning weights to each

of the factors. Although there are numerous optimization

algorithms available, genetic algorithms have gained wide

use in CR applications [108], [111].

Commonly used utility functions include weighted

sum, exponential, and constant elasticity of substitution

[108]. The weights in the chosen utility function aretypically chosen manually, based on the importance of

each objective for the given application. Unfortunately, the

importance of each objective is not always clear.

Additional complexity may be added when the weights of

each objective change based on the situation. For example,

when sending VoIP data, the system will have strict latency

requirements but may not require high throughput.

Conversely, for streaming video or large file transfer,throughput may be an issue, and not latency. For a survey

on intelligence in CR parameter adaption, we refer the

reader to [111].

C. Choosing Features for Pattern RecognitionPatterns in the spectrum access of incumbents,

opportunistic users, or malicious users can be used to

improve situation awareness, leading to improved com-

munications and threat avoidance. For example, an

opportunistic user in another channel could use incum-

bent patterns to predict where the next channel is likely to

Table 3 Cognition Toolbox



be available, in the event that it has to immediately

evacuate the current channel.

Pattern recognition can often be formulated as amachine learning classification problem, using algorithms

such as k-nearest neighbors, support vector machines, or

neural networks [112]. When developing a pattern

recognition system, one practical issue is choosing the

appropriate set of features. Another is having enough

training data to represent real-world conditions.

In any pattern recognition algorithm, there is the risk

of using too many features. For example, in signalclassification, one might take into account a signal’s

modulation scheme, bandwidth, received power, center

frequency, etc. If the goal is to classify incumbent signals,

then all of these features may not be required as inputs to

the classifier. In machine learning and statistics, the

process of dimensionality reduction is used to reduce the

number of random variables, or features, under consider-

ation. The goal of dimensionality reduction is to find asubset of the original features that provides equal or

similar performance. A linear combination of two or more

features may also be used (this can be thought of as

merging features prior to classification).

Dimensionality reduction is rarely performed in real

time. Instead, it is used during the design of the classifier,

in order to reduce the number of features the classifiermust process. By reducing the number of features, the

computational complexity of training and using the

classifier is reduced, and less system memory is required.

In order to perform dimensionality reduction during the

design process, a large and realistic set of training data

must be obtained. If the training data do not reflect real-

world conditions, then an important feature may be

unknowingly removed from the set and the accuracy ofdecision making degraded.

Dimensionality reduction is a well-researched area of

statistics and machine learning. Common procedures

include PCA [113] and LDA [114]. When the training data

set is small (which often occurs in CR), PCA can outperform

LDA [115]. For a tutorial on PCA, we refer the reader to [113].

D. Self-Organizing NetworksOne of the strategies for dealing with spectrum

congestion for LTE deployments is to move toward a

heterogeneous network (Het-Net) of small cells and

macrocells. This results in a large number of network

elements with complex interactions among them. Manag-

ing the complexity of manual approaches to managing the

radio access network leads to SONs [116], [117]. Three

focus areas exist within this umbrella: self-configuration,self-optimization, and self-healing. Self-configuration is

needed to reduce the costs of installing new nodes and

managing larger network structures such as domains. Self-

optimization is needed given that channel conditions and

connectivity situations are different over geography and

time. Self-healing is needed to control network manage-

ment costs [117]–[121].

Table 4 Common Knobs and Meters in CR

Fig. 5. Cognitive cycle [108].



The concepts associated with CR are suited for the taskof implementing SONs. SON typically applies to the

automatic configuration and optimization of LTE base

stations. Therefore, the SON algorithms could be im-

plemented on the base stations themselves or on a

centralized server that communicates with the base

stations. For example, when a new base station is added,

surrounding base stations could automatically reconfigure

their emission power and antenna tilt using optimizationtechniques.

Implementing SON has its own practical issues. One of

the practical issues in SON implementations is reliability

in a multivendor scenario [122]. Another issue is resolving

goals. Nine use cases have been identified for SON. The

goals of energy savings, load balancing, coverage, and

interference mitigation are conflicting. Research is begin-

ning to identify algorithms that allow an operator tochoose the desired behavior (see, for example, [123] for a

discussion of load balancing and handover tradeoffs and

[124] for node optimization interactions).

VI. OVERSIGHT

A. SecuritySecurity has typically not been an issue on the critical

path leading to deployment. More often, it is considered as

a postdeployment mitigation process. However, most of

the bands available for spectrum sharing are controlled by

Federal or civilian agencies (defense, aeronautical, mari-

time, public safety, etc.) whose mission cannot afford to be

compromised. It should be expected that all reasonable

security threats and mitigation approaches be identifiedbefore sharing will be allowed. What follows are some

threats that must be considered for DSA (individual

approaches may also incur other threats).

1) Primary User Emulation: PUE involves an adversary or

malicious node emulating the signal of the incumbent in a

given band, so that the opportunistic users will evacuate

the channel(s) [125]. A greedy node may perform PUE inorder to improve its own throughput, or an adversary may

perform the attack with the goal of causing denial of

service. Thus, PUE threatens the objectives of availability

and access control.

PUE detection strategies can be split up into location

based and nonlocation based [126]. Location-based strat-

egies involve knowing the location of all primary users, and

using some metric to estimate location at each radio.Nonlocation-based strategies assume either that the

adversary is moving, or that the adversary is much closer

than the primary user, or that the primary user has a digital

signature incorporated into its signal [127]. In many

dynamic spectrum applications, these assumptions will not

be valid, thus the PUE attack is a practical issue that needs

additional research. Fig. 6 illustrates the PUE attack and

the spectrum sensing data falsification attack, which is

described in the following section.

2) Spectrum Sensing Data Falsification: In any DSA

application that involves sharing sensing data between

nodes, an attack known as spectrum sensing data

falsification is possible. The decision engine in CRs relies

heavily on valid spectrum occupancy information. Whenthis information is fused between nodes, the spectrum

sensing data falsification attack involves injecting false

sensing information [126]. This could be done by a greedy

node who wants to improve its own throughput, a

malicious node causing denial of service, or even a

malfunctioning node that is unintentionally sending wrong

information (e.g., broken hardware).

Most proposed detection strategies involve building atrust or reputation value for each node based on past

reports, in which outliers are used to indicate a possible

lack of trust [126]. The ability for a radio to identify

malicious or malfunctioning entities is a practical issue in

dynamic spectrum sharing, and these detection strategies

should be considered during the design process.

3) Cognitive Control Channel Attack: DSA protocols oftenhave a common control channel used to share information,

which could be corrupted by jamming or by packet

flooding. Avoiding such points of failure is a practical issue

in the design of DSA systems. One research direction is the

use of a rendezvous strategy instead of a statically allocated

common control channel [129].

4) SDR Hacking: Policies used in DSA will likely evolveover time, and CRs need a way to adjust to policies that are

created after the design process. If nodes are designed to

be updated remotely through the Internet, then there are

clear hacking vulnerabilities. Furthermore, if users have

Fig. 6. Security threats associated with dynamic spectrum

sharing [128].



control over software on their own radios, then they mayeither deliberately or inadvertently disable policies

intended to protect incumbent users. It may even be

possible for an adversary to hack radios and turn them into

hostile jammers. Research into secure downloading

systems for firmware or policy updates is, therefore,

needed.

B. Industry and Government RegulationThe issue of standardization occurs at multiple levels,

and several of these levels will be addressed here. There is

a clear need to standardize the expression of capabilities

and needs. This shows up both in the form of device

classifications and in dealing with policies. There is also a

need to minimize the amount of information that needs to

be transmitted in order to convey meaning. Additionally,

there is a need for standardization of metrics to facilitateinteroperability and regulation. Each of these will be

considered.

1) Device Classification and Certification: There are, as of

yet, a relatively small number of devices capable of

supporting DSA. From a search of the FCC database [130],

only 336 approvals for SDR capable devices were given

between 2005 and 2012 (compared to a total of �16 000approvals per year for all radio devices).

It is likely that new classifications will be required, but

the fear that the approval process may be a bottleneck is

likely more perceived than real. Organizations like the

FCC maintain a knowledge base that contains guidance for

cases that come up between rule making events (for

example, see [131]). The European Union maintains a set

of what are called harmonized standards (see [132] and[133]). When new cases come up, there is a process to

provide a notified body of opinion from a certification lab.

Each has mechanisms for dealing with the kinds of changes

envisioned.

Receiver performance characteristics play a crucial role

in spectral efficiency and quality of service. Receiver

characterization has traditionally not been considered in

spectrum management processes in favor of a system basedon characterizing only the transmitters. However, DSA

creates new situations, such as spectrum placement of new

users adjacent to incumbent users.

A literature survey reveals that most studies related to

this have been carried out by regulatory and government

bodies (FCC, Ofcom, etc.). One of the initial studies into

receiver performance standards was carried out in [134]. A

detailed analysis and methodology of receiver perfor-mance, and the relevant influencing factors, has been

presented in [135].

The U.K. Office of Communications has reported the

various receiver characteristics specific to several receiver

architectures in [136]. The PCAST report [2] also gives a

brief description about receiver regulations and the

specific receiver characteristics needed to aid in better

spectrum management. In a recent report by the U.S.Government Accountability Office [137], a survey of the

receiver performance, and their impact on spectrum, has

been described. The Receivers and Spectrum Working

Group of the FCC has reported elaborate means to describe

the interference limits, and the means to incorporate them

as a policy, in [138].

A final issue with regard to certification is the question

of how and when certification is required. The currentsystem is based either on onetime certification testing

(usually with a very small number of units), or on

providing manufacturing guarantees of low power. With a

radio able to operate in bands with sensitive primary

users, the question may become how to ensure over time

that the radio is still fit for duty across all bands; passive

filters will no longer be a guarantee of noninterference.

The definition of ‘‘over time’’ includes both whether theyare still being manufactured to meet the original specs as

well as whether fielded units are still meeting those

specs.

2) Policy Drift Over Geography and Time: As spectrum

management systems are developed, the question of how

to deal with differences in policy becomes more signifi-

cant. Barring the existence of a universal set of policiesapplicable worldwide, a CR that crosses a national

boundary will be required to adhere to different policies.

It may also need to communicate with a different type of

spectrum management system, as well as with common

control channels operating at different frequencies. The

question of whether this will be a balkanizing issue has yet

to be answered and, while it is not a technical issue per se,

it will affect device design and the willingness ofmanufacturers to invest.

Another practical issue in dynamic spectrum sharing is

how to deal with evolving capabilities of new radios and

with policy changes that occur after a radio is designed and

manufactured. Making the radio flexible enough to deal

with future policy changes has been proposed as a way of

dealing with this.

At the heart of the question of how to deal with policyover time is the tradeoff between expressiveness and

complexity. In general, the more expressive the mecha-

nism, the more flexible it is. Given this fact, a significant

amount of research has been devoted to the concept of

policy reasoners that can interpret whatever policies they

are given (see [139]–[143]). However, at the other

extreme, there is caution about trying to create more

complexity than is really needed. Systems that come withexpensive processing requirements will not be acceptable

to the community. As expressed in [143], ‘‘In this effort,

two fundamental dangers must be avoided: development of

a perfect language that nobody wants to use (similar to Ada

or Esperanto) and development of a large number of non-

compatible languages that each solve one piece of the

problem but not an efficient composite solution.’’



The ability to change configuration of the radios, and/or their policies, introduces the issue of maintaining

synchronization between a radio’s configuration and the

configuration needed to operate correctly with the

spectrum management system. More research in this

area is needed.

3) Metrics for Regulatory Agencies: In the past, regulatory

agencies made band plans and system allocations withknowledge of component limitations of specific electronic

implementations. With DSA, there are no specific

passband filters or RF characteristics that all hardware

will contain. Therefore, it is not possible to achieve

spectrum management goals (like transmit spectrum

bounds) simply by controlling hardware characteristics.

The metrics used by regulatory agencies for determining

suitability become more complex with DSA.Governments are mandating spectrum efficiency when

considering purchase of spectrum-dependent systems [6],

[12]. However, one of the findings of the PCAST report is

that the traditional metric used for determining spectrum

efficiency, bits per second per hertz, has not guaranteed

that all available spectrum is being used. New metrics need

to be defined, and not just for efficiency. Other parameters

are important, such as receiver performance, time taken tovacate a band, interference tolerance metrics, probability

that spectrum is available, etc. New models also have to be

developed to deal with issues such as propagation

characteristics.

C. Enforcement MethodologiesThe traditional method of enforcing spectrum usage is

based on equipping engineers with spectrum analyzers andsending them to an area where violations are continuously

occurring. With a static allocation process, it is easy to

determine that any user transmitting in an unauthorized

band is a violator. With DSA, the difficulty greatly

increases. Not only is every secondary user device a

potential violator, but also they may only be causing a

problem episodically, so having engineers sitting and

waiting for an event could be very costly.Research into mechanisms that could assist enforce-

ment agencies takes several forms. There is research into

locating users by receive signal strength, time-of-arrival

differences, watermarking transmitters (inserting some-

thing into the signal to provide identity), and spectral

fingerprinting (identifying radio characteristics unique to a

transmitter).

Regulatory requirements specify a framework forsharing spectrum bands that can be used for proactive

enforcement using regulatory databases [2], [144], [145].

The use of a database in this manner is foreseen by the FCC

CDBS [59]. It contains a list of stations that require

protection.

Additional research is needed to decide the full details

of what should be included in a regulatory database, and

how it should be used. One potential use case is that aspectrum manager could access the database and use a

regulatory-approved interference prediction model for

authorizing access, or for optimizing the spectrum

allocation process. In this mode, it might get items such

as the following from the database: interference power

spectral density limits for incumbents, channel models,

spectrum masks for secondary user transmitters, underlay

masks for receivers, cochannel and adjacent channelinterference limits for receivers, etc. It could then run

calculations locally to coordinate spectrum assignments

between secondary users, or even to detect noncompliance

and enforce interference limits.

D. Information PrivacyA database service that holds spectral occupancy

information could be tampered with or probed in orderto gain information about the location of important

entities (e.g., Navy ships). An adversary could be a valid

CR that joins a network under normal conditions, but

requests spectrum occupancy information beyond what is

appropriate, such as information in a geographical area

that the radio has no intention to be in. Although

database security is a well-researched area, the question

of how to provide information from the database withoutunintentionally disclosing sensitive information is an area

in need of further research. For example, military users

may not want their location or capability information in a

database. This could be a fundamental problem with

applying a database approach to spectrum shared with

military users.

Another issue deals with who is authorized to collect

and analyze data coming from another radio. As anexample, collecting and analyzing spectrum data from a

military system is considered a felony in some countries.

Does the presence of sensors on a radio in the same band as

a military system, and the ability to either store data or

process data via cognitive engines on a radio, constitute a

violation of those laws? If so, the impact could be that

systems that share spectrum with military radios may not

be able to exploit learning (because of data storage),spectrum fingerprinting, or perhaps even be equipped with

a cognitive engine that could be used to process signals. It

would also impact the ability to exploit collaboration

between nodes and/or spectrum managers.

The 2013 Presidential Memorandum on Expanding

America’s Leadership in Wireless Innovation states that

‘‘All policies, practices, and templates shall be subject to

safeguards reasonably necessary to protect classified,sensitive, and proprietary data’’ [6]. A similar injunction

exists in the EU Commission Implementing Decision of

April 23, 2013, to ‘‘ensure the protection of information

which is considered confidential or classified by a Member

state’’ [146]. Research into algorithms and approaches that

deal with the necessity of sharing position and capability

information are needed to meet these requirements.



E. Creating Confidence: Test BedsA serious challenge facing implementation of DSA

systems is proving that the system will work as advertised.

Incumbent users and regulatory agencies need confidence

that dynamic sharing of spectrum will not cause harmful

interference to existing equipment before they will allow

systems to be deployed. Commercial entities need to feel

that opportunistic spectrum access is reliable, both from a

technology standpoint and a regulatory standpoint, beforethey will pursue significant investment. This dependency

was recognized in the PCAST spectrum report. It

recommended that the NTIA and the NIST provide test

services to support the development of policies and

underlying technologies [2]. In terms of scale, a ‘‘test city’’

and a ‘‘mobile test service’’ were suggested as components.

VII. IMPLICATIONS FORFUTURE STANDARDS

It is clear that regulators and commercial interests alike

believe that spectrum allocations in the future will stress-

sharing mechanisms. IEEE 802.22 is one of the first DSA-

based standards, and is paving the road for next-generation

DSA technology. The two market places that have proven to

be lucrative in the past, Wi-Fi and cellular, are the next tar-gets in standards efforts. The IEEE 802.11 standard is getting

a DSA update in the form of 802.11af, and LTE is looking to

exploit its carrier aggregation feature for DSA. A brief

discussion of IEEE 802.11af, LTE, and cognitive radar follows.

A. IEEE 802.11af White-FiIEEE 802.11af is a technical standard currently under

development with the goal of providing Wi-Fi in the TVwhite space. The standard will define modifications to

both the 802.11 physical layers and the 802.11 MAC layers,

to meet the legal requirements for channel access and

coexistence in the unused spectrum between TV stations.

The TV white spaces occur below 1 GHz, which means

802.11af has the potential for much greater distances than

the previous versions of 802.11. Future devices that

include 802.11af technologies will most likely also includeother Wi-Fi technologies such as 802.11ac and 802.11n.

The physical layer of 802.11af is based on IEEE 802.11ac,

except scaled down to work with 6-, 7-, and 8-MHz

channels [147]. Multiple channels can be in an aggregated

contiguous or noncontiguous manner. One of the design

requirements of 802.11af is an adjacent channel leakage

ratio of �55 dB, which is much higher than the current

Wi-Fi requirements [12]. The MAC layer protocol will alsobe based on previous Wi-Fi technology, with the addition

of spectrum sensing and a method of communicating with

a central database containing occupancy information.

B. LTE and LTE-AdvancedEven though the 3GPP LTE and LTE-Advanced

specifications do not include CR mechanisms such as

spectrum sensing, it is envisioned that the evolution ofLTE will lead toward use in shared spectrum. There is an

abundance of literature proposing CR and DSA schemes

for LTE, in both frequency-division duplex and time-

division duplex modes [148]. While most of the proposed

schemes involve spectrum sharing, it is possible to

implement a simple version of dynamic spectrum sharing

by requiring all LTE base stations to query a spectrum

occupancy database before transmitting. This databasemethod will likely represent the first wave of DSA-enabled

LTE, simply because it does not require as many additions

to the specifications.

Release 10 introduced a feature known as carrier

aggregation, which allows the user and base station to

communicate on multiple blocks of spectrum, known as

component carriers. These component carriers can be

adjacent, nonadjacent and in the same band, or nonadja-cent and in different bands. Each component carrier can

have a bandwidth of 1.4, 3, 5, 10, 15, or 20 MHz and a

maximum of five component carriers can be aggregated (for a

max of 100 MHz). Initially, carrier aggregation will only be

specified for a few specific operating bands. This means that

the technology cannot be directly applied to any given piece

of spectrum that allows opportunistic sharing. However, it is

one step closer to realizing DSA-enabled LTE.An example scenario that combines carrier aggregation

and dynamic spectrum sharing involves an LTE network

which operates in spectrum owned by the network

provider, but includes aggregate carriers in shared bands

which turn on and off based on the spectrum availability. It

is possible to turn off an aggregate carrier while the system

is running, given that vital control information is carried

on the primary carrier operating in the network operator’sspectrum. Turning off ‘‘extra carriers’’ simply reduces the

capacity of the link. Although LTE is not specifically a DSA

technology, it may evolve into a technology with DSA

capabilities, and features such as carrier aggregation and

SONs will surely aid in this evolution.

C. Cognitive RadarThe idea of using cognition in radar systems is an area

of research seeing increasing interest in recent years

[149]–[151]. The basic idea is that performance improve-

ments can be achieved over traditional radar systems

through the added capabilities of observation, memory,

and learning of the radar’s geographical, temporal, and

spectral environments. More specifically, it is envisioned

that a cognitive radar system could, in real time, makeintelligent adjustments to its system parameters in order to

deal with changes in its environment, to more closely track

its targets of interest, and to adjust for potential

interference sources [152]. It is due to this last point that

cognitive radar is thought of as a key technology for the

successful development of spectrum-sharing applications

in the government controlled radar bands.



D. Other AreasThere is an ongoing debate within the public safety

community about how to deal with the spectrum allocated

to FirstNet [17] for building a United States-wide public

safety network. Building a complete network is a veryexpensive proposition, and given that it is most likely to be

based on LTE, there have been a number of proposals for

the spectrum to be shared between public safety and

commercial carriers. If this comes to pass, LTE standards

would be affected.

Although hospitals have long been the place where you

were most likely to be asked to turn off your cell phone,

the hospital room of the future is destined to be one of thedensest RF environments anywhere. A large number of

medical devices are moving toward wireless interfaces, and

RFID tagging is beginning to be used to track equipment

within the hospital (e.g., to reduce costs of finding

equipment when it must be calibrated or updated as part

of a safety recall). Using strictly the ISM bands (e.g., Wi-Fi)

will not be sufficient for the volumes of data produced in

these environments when doctors begin to place wearablemedical monitoring devices on their patients. MBANs are

the first, but not likely the only, foray into spectrum sharing

for hospitals.

Given the convenience and cost reduction often

afforded by going wireless, and the increased traffic flow

occasioned by machine-to-machine communications in the

Internet of Things, it should be expected that new types of

spectrum-sharing arrangements will be created in thefuture. So far, we have seen the development of standards

for a centrally supervised database system (IEEE 802.22)

and an unsupervised beacon-based system (IEEE 802.14.4j).

We expect to see centrally supervised database systems for

LTE and for IEEE 801.11af. The former would manage a

potentially large number of bands from 1 to 3.5 GHz and the

latter would manage TV white space. As other bands are

considered, the legacy protection requirements of the

incumbents, and the business models of the secondary

users, could see development of standards for some of the

other architectures (distributed supervision, or sense-only,

systems).

VIII . CONCLUSION

There is currently a paradigm shift underway in the

management of RF spectrum. Regulatory bodies are

coming face to face with the realization that the currentpolicies of fixed spectrum allocations are spectrally

inefficient and undesirable given the current explosion

of demand for wireless technologies. Regulatory policy

makers, as well as the research community at large,

envision DSA technologies as one of the key enablers to

this move away from rigid spectrum assignments.

The development of spectrum management systems is

still in its infancy and introduces unique issues typicallyunconsidered in traditional systems. These issues will need

to be addressed by equipment designers, network devel-

opers, policy makers, and even end users to guarantee

success. Given these facts, the purpose of this paper has

been to review some of the immediate practical issues

inherent to the development of spectrum management

systems utilizing CRs for DSA. A summary of the issues

discussed can be found in Fig. 7.While this paper could not possibly cover all of the

possible practical issues, the authors are hopeful that it will

aid in the up-and-coming spectral evolution/revolution. h

Acknowledgment

The authors would like to thank the many reviewers who

provided them with feedback about on how to approach some

of these topics, including M. Marcus, R. Farrell, P. Stanforth,

S. Hasan, and R. Pavlak.

Fig. 7. Summary of practical issues for spectrum management.



REF ERENCE S

[1] Qualcomm, ‘‘The 1000� data challenge,’’2013. [Online]. Available: http://www.qualcomm.com/solutions/wireless-networks/technologies/1000x-data.

[2] President’s Council of Advisors onScience and Technology, ‘‘Report to thePresident realizing the full potential ofgovernment-held spectrum to spur economicgrowth,’’ Jul. 2012.

[3] J. A. Stine, ‘‘Model based spectrummanagementVPart 1: Modeling andcomputational manual,’’ MITRECorporation, McLean, VA, USA, 2011.

[4] P. Ryan, ‘‘Treating the wireless spectrum as anatural resource,’’ Environ. Law Reporter,vol. 35, p. 10620, 2005.

[5] White House Office of the Press Secretary,‘‘Presidential MemorandumVUnleashingthe Brtoadband Revolution,’’ 2010.

[6] White House Office of the Press Secretary,‘‘Presidential MemorandumVExpandingAmerica’s Leadership in WirelessInnovation,’’ 2013.

[7] European Commission, ‘‘Digital agenda forEuropeVRadio spectrum policy documentarchive,’’ Jul. 3, 2013. [Online]. Available:http://ec.europa.eu/digital-agenda/radio-spectrum-policy-document-archive.

[8] U.S. Federal Communications Commission,‘‘Enabling innovative small cell use in3.5 GHz band NPRM & order,’’ 2012.

[9] J. A. Stine, ‘‘Spectrum consumptionmodels,’’ in Cognitive Radio Technology,B. A. Fette, Ed. Burlington, MA, USA:Academy Publishers (Imprint of Elsevier),2009, pp. 645–666.

[10] FCC Spectrum Policy Task Force, ‘‘Report ofthe spectrum efficiency working group,’’2002.

[11] J. H. Reed, J. T. Bernhard, and P. Jung-Min,‘‘Spectrum access technologies: The past,the present, and the future,’’ Proc. IEEE,vol. 100, no. Special Centennial Issue,pp. 1676–1684, May 13, 2012.

[12] European Parliament, DecisionNo 243/1012/EU of the European Parliamentand the Council of 14 March 2012establishing a multiannual radio spectrumpolicy programme,’’ Strasbourg, France,243/2012/EU, 2012.

[13] C. Stevenson, G. Chouinard, L. Zhongding,W. Hu, S. J. Shellhammer, and W. Caldwell,‘‘IEEE 802.22: The first cognitive radiowireless regional area network standard,’’IEEE Commun. Mag., vol. 47, no. 1,pp. 130–138, Jan. 2009.

[14] K. G. Shin, J. Hyoil, A. W. Min, andA. Kumar, ‘‘Cognitive radios for dynamicspectrum access: From concept to reality,’’IEEE Wireless Commun., vol. 17, no. 6,pp. 64–74, Dec. 2010.

[15] IEEE Standard for InformationTechnologyVLocal and Metropolitan AreaNetworksVSpecific RequirementsVPart 11:Wireless LAN Medium Access Control (MAC)and Physical Layer (PHY) SpecificationsAmendment 3: 3650–3700 MHz Operationin USA, IEEE STD 802-11y-2008, 2008,pp. 1-90.

[16] D. Wang, D. Evans, and R. Krasinski, ‘‘IEEE802.15.4J: Extend IEEE 802.15.4 radiointo the MBAN spectrum,’’ IEEE WirelessCommun., vol. 19, no. 5, pp. 4–5, Oct. 2012.

[17] National Telecommunications andInformation Administration, ‘‘FirstResponder Network Authority,’’ 2012.[Online]. Available: http://www.ntia.doc.gov/category/firstnet.

[18] T. L. Doumi, ‘‘Spectrum considerations forpublic safety in the United States,’’ IEEECommun. Mag, vol. 44, no. 1, pp. 30–37,Jan. 2006.

[19] I. F. Akyildiz, D. M. Gutierrez-Estevez, andE. C. Reyes, ‘‘The evolution to 4G cellularsystems: LTE-Advanced,’’ Phys. Commun.,vol. 3, pp. 217–244, 2010.

[20] S. Haykin, ‘‘Cognitive radar: A way of thefuture,’’ IEEE Signal Process. Mag., vol. 23,no. 1, pp. 30–40, Jan. 2006.

[21] U.S. Department of Energy, ‘‘Communicationsrequirements of smart grid technologies,’’Oct. 5, 2010.

[22] U.S. Department of Defense,‘‘Mil STD-188-141B: Interoperability andperformance standards for medium and highfrequency radio systems,’’ 1999.

[23] European Telecommunications StandardsInstitute, ‘‘Broadband wireless access,’’[Online]. Available: http://www.etsi.org/technologies-clusters/technologies/broadband-wireless-access.

[24] Qualcomm, ‘‘Authorized shared access: Anevolutionary spectrum authorization schemefor sustainable economic growth andconsumer benefit,’’ Jul. 3, 2013. [Online].Available: www.cept.org/Documents/cg-crs/363/CGRS(11)07_Presentation-of-ASA-concept-(FM(11)116-Attachment).

[25] M. M. Buddhikot, ‘‘Understanding dynamicspectrum access: Models, taxonomy andchallenges,’’ in Proc. 2nd IEEE Int. Symp. NewFrontiers Dyn. Spectrum Access Netw., 2007,pp. 649–663.

[26] IEEE Standard for InformationTechnologyVLocal and Metropolitan AreaNetworksVSpecific RequirementsVPart 22:Cognitive Wireless RAN Medium AccessControl (MAC) and Physical Layer (PHY)Specifications: Policies and Proceduresfor Operation in the TV Bands, IEEEStd. 802.22-2011, 2011, pp. 1-680.

[27] B. Gao, J. Park, and Y. Yang, ‘‘A taxonomy ofcoexistence mechanisms for heterogeneouscognitive radio networks operating in TVwhite spaces,’’ IEEE Wireless Commun.,vol. 19, no. 4, pp. 41–48, Aug. 2012.

[28] M. J. Marcus, ‘‘CR: Cooperative radio orconfrontational radio,’’ in Proc. 2nd IEEE Int.Symp. New Frontiers Dyn. Spectrum AccessNetw., 2007, pp. 208–211.

[29] Y. Zhao, ‘‘Enabling cognitive radiosthrough radio environment maps,’’ Ph.D.dissertation, Bradley Dept. Electr. Comput.Eng., Virginia Tech., Blacksburg, VA, USA,2007.

[30] B. Wang, ‘‘Dynamic spectrum allocation andsharing in cognitive cooperative networks,’’Ph.D. dissertation, Electr. Eng. Dept., Univ.Maryland, College Park, MD, USA, 2009.

[31] B. Wang and K. J. R. Liu, ‘‘Advances incognitive radio networks: A survey,’’ IEEEJ. Sel. Topics Signal Process., vol. 5, no. 1,pp. 5–23, Feb. 2011.

[32] S. Haykin, ‘‘Cognitive radio: Brain-empoweredwireless communications,’’ IEEE J. Sel. AreasCommun., vol. 23, no. 2, pp. 201–220,Feb. 2005.

[33] B. Ahsant and R. Viswanathan, ‘‘A review ofcooperative spectrum sensing in cognitiveradios,’’ in Advancement in SensingTechnology. New York, NY, USA:Springer-Verlag, 2013, pp. 69–80.

[34] R. Viswanathan and B. Ahsant, ‘‘A review ofsensing and distributed detection algorithmsfor cognitive radio systems,’’ Int. J. SmartSens. Intell. Syst., vol. 5, pp. 177–190, 2012.

[35] N. Pratas, N. R. Prasad, A. Rodrigues, andR. Prasad, ‘‘Cooperative spectrum sensing:State of the art review,’’ in Proc. 2nd Int. Conf.Wireless Commun. Veh. Technol. Inf. TheoryAerosp. Electron. Syst. Technol., 2011, DOI:10.1109/WIRELESSVITAE.2011.5940902.

[36] G. Ganesan and L. Ye, ‘‘Cooperativespectrum sensing in cognitive radio, Part II:Multiuser networks,’’ IEEE Trans. WirelessCommun., vol. 6, no. 6, pp. 2214–2222,Jun. 2007.

[37] G. Ganesan and L. Ye, ‘‘Cooperativespectrum sensing in cognitive radio, Part I:Two user networks,’’ IEEE Trans. WirelessCommun., vol. 6, no. 6, pp. 2204–2213,Jun. 2007.

[38] I. F. Akyildiz, B. F. Lo, and R. Balakrishnan,‘‘Cooperative spectrum sensing in cognitiveradio networks: A survey,’’ Phys. Commun.,vol. 4, pp. 40–62, 2011.

[39] Z. Quan, S. Cui, and A. H. Sayed, ‘‘Optimallinear cooperation for spectrum sensing incognitive radio networks,’’ IEEE J. Sel. TopicsSignal Process., vol. 2, no. 1, pp. 28–40,Feb. 2008.

[40] W. Zhang, R. K. Mallik, and K. Letaief,‘‘Optimization of cooperative spectrumsensing with energy detection in cognitiveradio networks,’’ IEEE Trans. WirelessCommun., vol. 8, no. 12, pp. 5761–5766,Dec. 2009.

[41] Z. Li, F. R. Yu, and M. Huang,‘‘A distributed consensus-based cooperativespectrum-sensing scheme in cognitiveradios,’’ IEEE Trans. Veh. Technol., vol. 59,no. 1, pp. 383–393, Jan. 2010.

[42] S. Atapattu, C. Tellambura, and J. Hai,‘‘Energy detection based cooperativespectrum sensing in cognitive radionetworks,’’ IEEE Trans. Wireless Commun.,vol. 10, no. 4, pp. 1232–1241, Apr. 2011.

[43] J. Unnikrishnan and V. V. Veeravalli,‘‘Cooperative sensing for primary detectionin cognitive radio,’’ IEEE J. Sel. TopicsSignal Process., vol. 2, no. 1, pp. 18–27,Feb. 2008.

[44] S. Chaudhari, J. Lunden, V. Koivunen, andH. V. Poor, ‘‘Cooperative sensing withimperfect reporting channels: Harddecisions or soft decisions?’’ IEEE Trans.Signal Process., vol. 60, no. 1, pp. 18–28,Jan. 2012.

[45] T. C. Aysal, S. Kandeepan, and R. Piesiewicz,‘‘Cooperative spectrum sensing with noisyhard decision transmissions,’’ in Proc. IEEEInt. Conf. Commun., 2009, DOI: 10.1109/ICC.2009.5198863.

[46] A. S. Cacciapuoti, I. F. Akyildiz, and L. Paura,‘‘Correlation-aware user selection forcooperative spectrum sensing in cognitiveradio ad hoc networks,’’ IEEE J. Sel. AreasCommun., vol. 30, no. 2, pp. 297–306,Feb. 2012.

[47] Y. Chongjoon, K. Hongkyu, and H. Jun,‘‘Cooperative TV spectrum sensing incognitive radio for Wi-Fi networks,’’ IEEETrans. Consumer Electron., vol. 57, no. 1,pp. 62–67, Feb. 2011.

[48] A. C. Malady and C. R. C. M. da Silva,‘‘Clustering methods for distributedspectrum sensing in cognitive radiosystems,’’ in Proc. IEEE Military Commun.Conf., 2008, DOI: 10.1109/MILCOM.2008.4753432.

[49] W. Han, J. Li, Z. Li, J. Si, and Y. Zhang,‘‘Efficient soft decision fusion rule incooperative spectrum sensing,’’ IEEE Trans.Signal Process., vol. 61, no. 8, pp. 1931–1943,Apr. 15, 2013.



[50] J. Ma, G. Zhao, and Y. Li, ‘‘Soft combinationand detection for cooperative spectrumsensing in cognitive radio networks,’’ IEEETrans. Wireless Commun., vol. 7, no. 11, pt. 2,pp. 4502–4507, Nov. 2008.

[51] J. Li and Y. Zhao, ‘‘Radio environmentmap-based cognitive Doppler spreadcompensation algorithms for high-speed railbroadband mobile communications,’’EURASIP J. Wireless Commun. Netw.,vol. 2012, pp. 1–18, 2012.

[52] P. Ren, Y. Wang, Q. Du, and J. Xu, ‘‘A surveyon dynamic spectrum access protocols fordistributed cognitive wireless networks,’’EURASIP J. Wireless Commun. Netw.,vol. 2012, p. 60, 2012.

[53] Q. Zhao, L. Tong, A. Swami, andY. Chen, ‘‘Decentralized cognitive mac foropportunistic spectrum access in ad hocnetworks: A POMDP framework,’’ IEEE J.Sel. Areas Communications, vol. 25, no. 3,pp. 589–600, Apr. 2007.

[54] Y. Zhao, B. Le, and J. H. Reed, ‘‘NetworksupportVThe radio environment map,’’ inCognitive Radio Technology, B. A. Fette, Ed.Burlington, MA, USA: Newnes, 2006,pp. 325–366.

[55] FARAMIR, A European FP7 project for radioenvironment mapping. [Online]. Available:http://www.ict-faramir.eu/.

[56] Spectrum Bridge Inc., Database.[Online]. Available: http://www.spectrumbridge.com/Home.aspx.

[57] Google, Whitespace database.[Online]. Available: http://www.google.org/spectrum/whitespace/.

[58] Guidelines for Channel Calculations for WhiteSpaces, U.S. Federal CommunicationsCommission, 2013.

[59] U.S. Federal Communications Commission,Consolidated database system (CDBS).[Online]. Available: http://licensing.fcc.gov/prod/cdbs/pubacc/prod/cdbs_pa.htm.

[60] U.S. Government Accountability Office,‘‘Report on assessing the reliability ofcomputer-processed data,’’ Jul. 2009.

[61] E. A. Mancuso, S. Probasco, and B. Patil,RFC 6953VProtocol to Access White SpaceDatabase (PAWS): Use cases and requirements.[Online]. Available: http://datatracker.ietf.org/wg/paws/.

[62] National Telecommunications andInformation Administration (NTIA),‘‘Planning and processes need strengtheningto promote the efficient use of spectrum byFederal agencies,’’ 2011. [Online]. Available:http://www.gao.gov/products/GAO-11-352.

[63] Y. Chen and W. W. Chu, ‘‘Protection ofdatabase security via collaborative inferencedetection,’’ IEEE Trans. Knowl. Data Eng.,vol. 20, no. 8, pp. 1013–1027, Aug. 2008.

[64] B. Barak and J. Park, ‘‘Location privacy ofprimary users in geolocation database-drivenspectrum sharing,’’ 2014, to be published.

[65] J. Craninckx, M. Liu, D. Hauspie,V. Giannini, T. Kim, J. Lee, M. Libois,B. Debaillie, C. Soens, M. Ingels,A. Baschirotto, J. Van Driessche,L. Van der Perre, and P. Vanbekbergen, ‘‘Afully reconfigurable software-defined radiotransceiver in 0.13 �m CMOS,’’ in Dig. Tech.Papers IEEE Int. Solid-State Circuits Conf.,2007, pp. 346–607.

[66] L. J. Chu, ‘‘Physical limitations ofomni-directional antennas,’’ J. Appl. Phys.,vol. 19, pp. 1163–1175, 1948.

[67] R. Collin and S. Rothschild, ‘‘Evaluation ofantenna Q,’’ IEEE Trans. Antennas Propag.,vol. AP-12, no. 1, pp. 23–27, Jan. 1964.

[68] J. S. McLean, ‘‘A re-examination of thefundamental limits on the radiation Q ofelectrically small antennas,’’ IEEE Trans.Antennas Propag., vol. 44, no. 5, pp. 672–676,May 1996.

[69] W. Davis, T. Yang, E. Caswell, andW. Stutzman, ‘‘Fundamental limits onantenna size: A new limit,’’ IET Microw.Antennas Propag., vol. 5, pp. 1297–1302,2011.

[70] T. Yang, W. A. Davis, W. L. Stutzman,S. Shajedul Hasan, R. Nealy, C. B. Dietrich,and J. H. Reed, ‘‘Antenna design strategy anddemonstration for software-defined radio(SDR),’’ Analog Integr. Circuits Signal Process.,vol. 69, pp. 161–171, 2011.

[71] H. Schantz, The Art and Science ofUltrawideband Antennas. Boston, MA, USA:Artech House, 2005.

[72] W. L. Stutzman and G. A. Thiele, AntennaTheory and Design, 3rd ed. Hoboken, NJ,USA: Wiley, 2013.

[73] Z. Heng and E. Sanchez-Sinencio,‘‘Linearization techniques for CMOS lownoise amplifiers: A tutorial,’’ IEEE Trans.Circuits Syst. I, Reg. Papers, vol. 58, no. 1,pp. 22–36, Jan. 2011.

[74] H. Zhang, X. Fan, and E. S. Sinencio,‘‘A low-power, linearized, ultra-widebandLNA design technique,’’ IEEE J. Solid-StateCircuits, vol. 44, no. 2, pp. 320–330,Feb. 2009.

[75] P. Pawelczak, K. Nolan, L. Doyle, O. Ser Wah,and D. Cabric, ‘‘Cognitive radio: Ten years ofexperimentation and development,’’ IEEECommun. Mag., vol. 49, no. 3, pp. 90–100,Mar. 2011.

[76] E. Imana, T. Yang, and J. Reed, ‘‘Relaxingpre-selector filter selectivity requirementsusing cognitive RF front-end control,’’ AnalogIntegr. Circuits Signal Process., Aug. 24, 2013,DOI: 10.1007/s10470-013-0110-2.

[77] P. F. Marshall, ‘‘Cognitive radio as amechanism to manage front-end linearityand dynamic range,’’ IEEE Commun. Mag.,vol. 47, no. 3, pp. 81–87, Mar. 2009.

[78] Z. Quan, S. Cui, A. Sayed, and H. Poor,‘‘Wideband spectrum sensing in cognitiveradio networks,’’ in Proc. IEEE Int.Conf. Commun., Beijing, China, 2008,pp. 901–906.

[79] Y. Hur, J. Park, W. Woo, K. Lim, C. Lee,H. Kim, and J. Laskar, ‘‘A widebandanalog multi-resolution spectrum sensing(MRSS) technique for cognitive radio (CR)systems,’’ in Proc. IEEE Int. Symp. CircuitsSyst., Island of Kos, 2006, DOI: 10.1109/ISCAS.2006.1693528.

[80] N. Neihart, S. Roy, and D. Allstot, ‘‘A parallelmulti-resolution sensing technique formultiple antenna cognitive radios,’’ in Proc.IEEE Int. Symp. Circuits Syst., New Orleans,LA, USA, 2007, pp. 2530–2533.

[81] Z. Quan, S. Cui, A. Sayed, and H. Poor,‘‘Optimal multiband joint detection forspectrum sensing in cognitive radionetworks,’’ IEEE Trans. Signal Process.,vol. 57, no. 3, pp. 1128–1140, Mar. 2009.

[82] L. Luo and S. Roy, ‘‘Analysis of searchschemes in cognitive radio,’’ in Proc. IEEECommun. Soc. Conf. Sensor Mesh Ad HocCommun. Netw., San Diego, CA, USA, 2007,pp. 647–654.

[83] L. Luo, N. Neihart, S. Roy, and D. Allstot,‘‘A two-stage sensing technique for dynamicspectrum access,’’ in Proc. IEEE Int.Conf. Commun., Beijing, China, 2008,pp. 3028–3037.

[84] S. Fazeli-Dehkordy, K. Plataniotis, andS. Pasupathy, ‘‘Two-stage spectrum detectionin cognitive radio networks,’’ in Proc. IEEEInt. Conf. Acoust. Speech Signal Process., 2010,pp. 3118–3121.

[85] L. Luo and S. Roy, ‘‘Analysis of dynamicspectrum access with heterogeneousnetworks: Benefits of channel packingscheme,’’ in Proc. 28th IEEE Conf. GlobalTelecommun., 2009, pp. 3508–3514.

[86] T. Yucek and H. Arslan, ‘‘A survey ofspectrum sensing algorithms for cognitiveradio applications,’’ IEEE Commun. Surv. Tut.,vol. 11, no. 1, pp. 116–130, First Quarter2009.

[87] Y.-C. Liang, Y. Zeng, E. C. Y. Peh, andA. T. Hoang, ‘‘Sensing-throughputtradeoff for cognitive radio networks,’’IEEE Trans. Wireless Commun., vol. 7, no. 4,pp. 1326–1337, Apr. 2008.

[88] M. Patel and W. Jianfeng, ‘‘Applications,challenges, and prospective in emergingbody area networking technologies,’’ IEEEWireless Commun., vol. 17, no. 1, pp. 80–88,Feb. 2010.

[89] D. Cabric, S. M. Mishra, andR. W. Brodersen, ‘‘Implementation issues inspectrum sensing for cognitive radios,’’ inConf. Rec. 38th Asilomar Conf. Signals Syst.Comput., Nov. 2004, vol. 1, pp. 772–776.

[90] Z. Yonghong and L. Ying-Chang,‘‘Eigenvalue-based spectrum sensingalgorithms for cognitive radio,’’ IEEE Trans.Commun., vol. 57, no. 6, pp. 1784–1793,Jun. 2009.

[91] X. Ling, B. Wu, H. Wen, P.-H. Ho, Z. Bao,and L. Pan, ‘‘Adaptive threshold controlfor energy detection based spectrumsensing in cognitive radios,’’ IEEE WirelessCommun. Lett., vol. 1, no. 5, pp. 448–451,Oct. 2012.

[92] L. Yixian, Z. Chunyan, W. Hongjiang, andW. Gang, ‘‘Energy detection thresholdoptimization for cooperative spectrumsensing,’’ in Proc. 2nd Int. Conf. Adv. Comput.Control, Mar. 2010, pp. 566–570.

[93] S. Haykin, D. J. Thomson, and J. H. Reed,‘‘Spectrum sensing for cognitive radio,’’ Proc.IEEE, vol. 97, no. 5, pp. 849–877, May 2009.

[94] K. Kim, I. A. Akbar, K. K. Bae, J.-S. Um,C. M. Spooner, and J. H. Reed,‘‘Cyclostationary approaches to signaldetection and classification in cognitiveradio,’’ in Proc. 2nd IEEE Int. Symp. NewFrontiers Dyn. Spectrum Access Netw.,Apr. 2007, pp. 212–215.

[95] Z. Tian and G. B. Giannakis, ‘‘A waveletapproach to wideband spectrum sensingfor cognitive radios,’’ in Proc. 1st Int. Conf.Cogn. Radio Oriented Wireless Netw. Commun.,Jun. 2006, DOI: 10.1109/CROWNCOM.2006.363459.

[96] J. Wang, X. Jin, G. Bi, Z. Cao, and J. Huang,‘‘Multiple cumulants based spectrum sensingmethods for cognitive radios,’’ IEEE Trans.Commun., vol. 60, no. 12, pp. 3620–3631,Dec. 2012.

[97] Y. Zeng, Y.-C. Liang, A. T. Hoang, andR. Zhang, ‘‘A review on spectrum sensing forcognitive radio: Challenges and solutions,’’EURASIP J. Adv. Signal Process., 2010,DOI: 10.1155/2010/381465.

[98] L. Lu, X. Zhou, U. Onunkwo, and G. Y. Li,‘‘Ten years of research in spectrum sensingand sharing in cognitive radio,’’ EURASIPJ. Wireless Commun. Netw., 2012,DOI: 10.1186/1687-1499-2012-28.

[99] D. D. Ariananda, M. K. Lakshmanan, andH. Nikookar, ‘‘A survey on spectrum sensingtechniques for cognitive radio,’’ in Proc. 2nd



Int. Workshop Cogn. Radio Adv. SpectrumManage., May 2009, pp. 74–79.

[100] Y. Huang and X. Huang, ‘‘Optimal spectrumsensing over multipath channels,’’ in Proc.IEEE Int. Conf. Commun., Jun. 2012,pp. 1522–1527.

[101] Y. Huang and X. Huang, ‘‘Spectrum sensingover frequency-selective fading channel withtap and spatial correlations,’’ in Proc. IEEE23rd Int. Symp. Pers. Indoor Mobile RadioCommun., Sep. 2012, pp. 2143–2148.

[102] S. A. Astaneh and S. Gazor, ‘‘Cooperativespectrum sensing over mixture-Nakagamichannels,’’ IEEE Wireless Commun. Lett.,vol. 2, no. 3, pp. 259–262, Jun. 2013.

[103] IEEE 802.16 Working Group, ‘‘Channelmodels for fixed wireless applications,’’Tech. Rep. IEEE 802.16.3c-01/29r4, 2001.

[104] WINNER Channel Model, ‘‘Spatial channelmodel for multiple input multiple output(MIMO) simulations,’’ 2003.

[105] P. Kyosti, J. Meinila, L. Hentila, X. Zhao,T. Jamsa, C. Schneider, M. Narandzic,M. Milojevic, A. Hong, J. Ylitalo,V.-M. Hollappa, M. Alatossava, R. Bultitude,Y. de Jong, and T. Rautiainen, ‘‘WINNER IIChannel Models, IST-WINNER D1.1.2,’’2007. [Online]. Available: http://www.ist-winner.org/deliverables.html

[106] M. J. Marcus, ‘‘Wireless research topics withspectrum policy significance,’’ IEEE WirelessCommun., vol. 17, no. 6, p. 4, Dec. 2010.

[107] G. Wiederhold and J. McCarthy, ‘‘ArthurSamuel: Pioneer in machine learning,’’ IBMJ. Res. Develop., vol. 36, no. 3, pp. 329–331,May 1992.

[108] T. W. Rondeau, ‘‘Application of artificialintelligence to wireless communications,’’Ph.D. dissertation, Electr. Comput. Eng.Dept., Virginia Polytechnic Inst. State Univ.,Blacksburg, VA, USA, 2007.

[109] T. W. Rondeau, C. J. Rieser, T. M. Gallagher,and C. W. Bostian, ‘‘Online modeling ofwireless channels with hidden Markovmodels and channel impulse responses forcognitive radios,’’ in IEEE MTT-S Int. Microw.Symp. Dig., Jun. 2004, vol. 2, pp. 739–742.

[110] C. Ghosh, C. Cordeiro, D. P. Agrawal, andM. B. Rao, ‘‘Markov chain existence andhidden Markov models in spectrum sensing,’’in Proc. IEEE Int. Conf. Perv. Comput.Commun., Mar. 2009, DOI: 10.1109/PERCOM.2009.4912868.

[111] A. He, B. Kyung Kyoon, T. R. Newman,J. Gaeddert, K. Kim, R. Menon,L. Morales-Tirado, J. J. Neel, Y. Zhao,J. H. Reed, and W. H. Tranter, ‘‘A survey ofartificial intelligence for cognitive radios,’’IEEE Trans. Veh. Technol., vol. 59, no. 4,pp. 1578–1592, May 2010.

[112] K. M. Thilina, C. Kae Won, N. Saquib, andE. Hossain, ‘‘Pattern classification techniquesfor cooperative spectrum sensing incognitive radio networks: SVM and W-KNNapproaches,’’ in Proc. IEEE Global Commun.Conf., Dec. 2012, pp. 1260–1265.

[113] J. Shlens, ‘‘A tutorial on principal componentanalysis,’’ Syst. Neurobiol. Lab., Univ.California San Diego, La Jolla, CA, USA,2005.

[114] S. Balakrishnama and A. Ganapathiraju,‘‘Linear discriminant analysisVA brieftutorial,’’ Inst. Signal Inf. Process., 1998.

[115] A. M. Martinez and A. C. Kak, ‘‘PCA versusLDA,’’ IEEE Trans. Pattern Anal. Mach. Intell.,vol. 23, no. 2, pp. 228–233, Feb. 2001.

[116] Z. Zhang, K. Long, and J. Wang,‘‘Self-organization paradigms andoptimization approaches for cognitive radio

technologies: A survey,’’ IEEE WirelessCommun., vol. 20, no. 2, pp. 36–42,Apr. 2013.

[117] S. Feng and E. Seidel, ‘‘Self-organizingnetworks (SON) in 3GPP long termevolution,’’ Nomor Research GmbH WhitePaper, 2008.

[118] H. Hu, Z. J, X. Zheng, and Y. Yang,‘‘Self-configuration and self-optimization forLTE networks,’’ IEEE Commun. Mag., vol. 48,no. 2, pp. 94–100, Feb. 2010.

[119] G. Americas, ‘‘The benefits of SON in LTE:Self-optimizing and self-organizingnetworks,’’ White Paper, 2009.

[120] D. Aziz, A. Ambrosy, L. Ho, L. Ewe,M. Gruber, and H. Bakker, ‘‘Autonomousneighbor relation detection and handoveroptimization in LTE,’’ Bell Labs Tech. J.,vol. 15, no. 3, pp. 63–83, Dec. 2010.

[121] L. Lindbom, R. Love, S. Krishnamurty,C. Yao, N. Miki, and V. Chanrashekar,Enhanced Inter-Cell Interference Coordinationfor Heterogeneous Networks in LTE-Advanced:A Survey. Ithaca, NY, USA: Cornell Univ.Library, Dec. 2011.

[122] M. M. S. Marwangi, N. Fisal, S. K. S. Yusof,R. A. Rashid, A. S. A. Ghafar, F. A. Saparudin,and N. Katiran, ‘‘Challenges and practicalimplementation of self-organizing networksin LTE/LTE-Advanced systems,’’ in Proc.Int. Conf. Inf. Technol. Multimedia, KualaLumpur, Nov. 2011, DOI: 10.1109/ICIMU.2011.6122748.

[123] L. Zhiqiang, P. Hong, K. Xue, and M. Peng,‘‘Conflict avoidance between mobilityrobustness optimization and mobility loadbalancing,’’ in Proc. IEEE Global Telecommun.Conf., Dec. 2010, DOI: 10.1109/GLOCOM.2010.5683861.

[124] A. Awada, B. Wegmann, D. Rose, I. Viering,and A. Klein, ‘‘Towards self-organizingmobility robustness optimization ininter-RAT scenario,’’ in Proc. IEEE 73rd Veh.Technol. Conf., Budapest, Hungary,May 2011, DOI: 10.1109/VETECS.2011.5956234.

[125] T. C. Clancy and N. Goergen, ‘‘Security incognitive radio networks: Threats andmitigation,’’ in Proc. 3rd Int. Conf. Cogn. RadioOriented Wireless Netw. Commun., May 2008,DOI: 10.1109/CROWNCOM.2008.4562534.

[126] A. G. Fragkiadakis, E. Z. Tragos, andI. G. Askoxylakis, ‘‘A survey on securitythreats and detection techniques in cognitiveradio networks,’’ IEEE Commun. Surv. Tut.,vol. 15, no. 1, pp. 428–445, First Quarter,2013.

[127] C. Ruiliang, P. Jung-Min, and J. H. Reed,‘‘Defense against primary user emulationattacks in cognitive radio networks,’’ IEEE J.Sel. Areas Commun., vol. 26, no. 1, pp. 25–37,Jan. 2008.

[128] C. Ruiliang, P. Jung-Min, Y. T. Hou, andJ. H. Reed, ‘‘Toward secure distributedspectrum sensing in cognitive radionetworks,’’ IEEE Commun. Mag., vol. 46,no. 4, pp. 50–55, Apr. 2008.

[129] L. A. DaSilva and I. Guerreiro,‘‘Sequence-based rendezvous for dynamicspectrum access,’’ in Proc. 3rd IEEE Symp.New Frontiers Dyn. Spectrum Access Netw.,Oct. 2008, DOI: 10.1109/DYSPAN.2008.52.

[130] U.S. Federal Communications Commission,‘‘Office of Engineering and TechnologyLaboratory Division Knowledge Database(KDB),’’ 2013. [Online]. Available: https://apps.fcc.gov/oetcf/kdb/index.cfm.

[131] U.S. Federal Communications Commission,‘‘FCC Office of Engineering and TechnologyEquipment Authorization Search,’’ 2013.

[Online]. Available: https://apps.fcc.gov/oetcf/eas/reports/GenericSearch.cfm.

[132] European Commission, ‘‘EuropeanstandardsVHarmonized standards,’’ 2013.[Online]. Available: http://ec.europa.eu/enterprise/policies/european-standards/documents/harmonised-standards-legislation/list-references/.

[133] European Commission, Enterprise andindustry, 2013. [Online]. Available: http://ec.europa.eu/enterprise/index_en.htm.

[134] B. Joiner, ‘‘Receiver spectrum standards:Phase 1 Summary of research into existingstandards,’’ NTIA Rep. 03-404, Nov. 2003.

[135] Communications Receiver PerformanceDegradation Handbook, DSO Joint SpectrumCenter, Oct. 2010, JSC-CR-10-004.

[136] L. Davies and P. Winter, ‘‘Study of currentand future receiver performance,’’ in FinalRep. OFCOM, Jan. 2010.

[137] U.S. Government Accountability Office,‘‘Spectrum management: Furtherconsideration of options to improvereceiver performance needed,’’ Rep. Congr.Committees GAO-13-265, Feb. 2013.

[138] FCC Technological Advisory Council,‘‘Interference limits policy: The use ofharm claim thresholds to improve theinterference tolerance of wireless systems,’’Receivers and Spectrum Working Group,Feb. 2013.

[139] D. Wilkins, G. Denker, M. O. Stehr,D. Elenius, R. Senanayake, and C. Talcott,‘‘Policy-based cognitive radios,’’ IEEEWireless Commun., vol. 14, no. 4, pp. 41–46,Aug. 2007.

[140] J. Riihijarvi, M. Petrova, V. Atanasovski, andL. Gavrilovska, ‘‘Extending policy languageswith utility and prioritization knowledge:The CAPRI approach,’’ in Proc. IEEE Symp.New Frontiers Dyn. Spectrum, Apr. 2010,DOI: 10.1109/DYSPAN.2010.5457838.

[141] B. Bahrak, A. Deshpande, M. Whitaker, andJ.-M. Park, ‘‘BRESAP: A policy reasonerfor processing spectrum access policiesrepresented by binary decision diagrams,’’ inProc. IEEE Symp. New Frontiers Dyn.Spectrum, Apr. 2010, DOI: 10.1109/DYSPAN.2010.5457867.

[142] B. Bahrak, J.-M. Park, and H. Wu,‘‘Ontology-based spectrum access policies forpolicy-based cognitive radios,’’ in Proc. IEEEInt. Symp. Dyn. Spectrum Access Netw.,Oct. 2012, pp. 489–500.

[143] M. M. Kokar, D. Hillman, L. Shujun,B. Fette, P. Marshall, M. Cummings,T. Martin, and J. Strassner, ‘‘Towards aunified policy language for futurecommunication networks: A process,’’ inProc. 3rd IEEE Symp. New Frontiers Dyn.Spectrum Access Netw., Oct. 2008,DOI: 10.1109/DYSPAN.2008.74.

[144] U.S. Department of Commerce, ‘‘Anassessment of the near-term viability ofaccommodating wireless broadband systemsin the 1675–1710 MHz, 1755–1780 MHz,3500–3650 MHz, and 4200–4220 MHz,4380–4400 MHz bands,’’ Oct. 2010.

[145] National Telecommunications andInformation Administration, ‘‘CommerceSpectrum Management AdvisoryCommittee Final Report,’’ Working Group1-1695-1710 MHz Meteorological-Satellite.[Online]. Available: www.ntia.doc.gov/files/ntia/publications/wg-1_report_v2.pdf.

[146] European Parliament, ‘‘CommissionImplementing Decision on 23 April 2013Defining the Practical Arrangements,Uniform Formats and a Methodology inRelation to the Radio Spectrum Inventory



Established by Decision No 243/2012/EUof the European Parliament and of theCouncil Establishing a MultiannualRadio Spectrum Policy Programme,’’ 2013.[Online]. Available: http://ec.europa.eu/digital-agenda/radio-spectrum-policy-document-archive#DEC.

[147] IEEE 802.11af Working Group,IEEE 802.11-12/0866r2 (PHY Overview),2012. [Online]. Available: https://mentor.ieee.org/802.11/dcn/12/11-12-0866-02-00af-phy-overview.pptx.

[148] J. D. Deaton, R. E. Irwin, and L. A. DaSilva,‘‘The effects of a dynamic spectrum accessoverlay in LTE-Advanced networks,’’ in Proc.IEEE Symp. New Frontiers Dyn. SpectrumAccess Netw., May 2011, pp. 488–497.

[149] S. Haykin, ‘‘Cognitive radar: A way of thefuture,’’ IEEE Signal Process. Mag., vol. 23,no. 1, pp. 30–40, Jan. 2006.

[150] S. Haykin, X. Yanbo, and P. Setoodeh,‘‘Cognitive radar: Step toward bridging thegap between neuroscience and engineering,’’

Proc. IEEE, vol. 100, no. 11, pp. 3102–3130,Nov. 2012.

[151] X. Li, Z. Hu, R. C. Qiu, and M. C. Wicks,‘‘Experimental demonstration of cognitiveradar for target localization under stronginterference,’’ in Proc. Military Commun.Conf., Oct. 2012, DOI: 10.1109/MILCOM.2012.6415695.

[152] M. Wicks, ‘‘Spectrum crowding andcognitive radar,’’ in Proc. 2nd Int. WorkshopCogn. Inf. Process., Elba Island, Italy,Jun. 2012, pp. 452–457.

ABOUT T HE AUTHO RS

Stephen M. Dudley (Member, IEEE) received the

B.Sc. degree in electrical engineering from the

University of Alberta, Edmonton, AB, Canada, in

1980 and the M.S. degree in electrical engineering

from Georgia Institute of Technology, Atlanta, GA,

USA, in 1995. Currently, he is working toward the

Ph.D. degree in electrical engineering at Virginia

Polytechnic Institute and State University,

Blacksburg, VA, USA, studying wireless networks

for disaster recovery.

William Christopher Headley received the B.S.

and M.S. degrees in electrical engineering (with

first class honors) from Virginia Polytechnic Insti-

tute and State University, Blacksburg, VA, USA, in

2006 and 2009, respectively. He is a Virginia Tech

Bradley Fellow alumnus, and is currently working

toward the Ph.D. degree in electrical engineering

with Wireless @ Virginia Tech.

His research interests include wireless com-

munications, signal detection and classification,

signal parameter estimation, and collaborative spectrum sensing.

Marc Lichtman (Student Member, IEEE) received

the B.S. and M.S. degrees in electrical engineering

from Virginia Polytechnic Institute and State

University, Blacksburg, VA, USA, in 2011 and

2012, respectively, where he is currently working

toward the Ph.D. degree, under the advisement of

Dr. J. H. Reed.

His research is focused on designing anti-jam

approaches against sophisticated jammers, using

machine learning techniques. He is also interested

in analyzing the vulnerability of LTE to jamming.

Eyosias Yoseph Imana received the B.S. degree

in electrical engineering from Bahir Dar Universi-

ty, Bahir Dar, Ethiopia, in 2007 and the M.S. degree

in electrical engineering from Florida Institute of

Technology, Melbourne, FL, USA, in 2009. Cur-

rently, he is working toward the Ph.D. degree in

electrical engineering at Virginia Polytechnic In-

stitute and State University, Blacksburg, VA, USA.

His research interest is radio-frequency (RF)

front–ends of flexible radios and implementation

of wireless devices. His paper on cognitive RF front–end control was

named as the best R&D track paper in the 2012 software-defined radio

forum. He is also a recipient of the Virginia innovation partnership I6

grant to develop mobile software-defined radios.

Xiaofu Ma (Student Member, IEEE) received the

B.S. degree in electronics from Northwest Uni-

versity, Xi’an, China, in 2008 and the M.S. degree

in computer science from Tongji University,

Shanghai, China, in 2011. Currently, he is working

toward the Ph.D. degree in electrical engineering

at Virginia Polytechnic Institute and State Univer-

sity, Blacksburg, VA, USA.

His research interests include network optimi-

zation, cognitive radio, dynamic spectrum access,

and wireless healthcare.

Mahi Abdelbar graduated with honors and

received the M.S. degree from the Electronics

and Communications Department, Engineering

Faculty, Mansoura University, Mansoura, Egypt,

in 2010. Currently, she is working toward the Ph.D.

degree in electrical engineering at the Wireless @

VT laboratory, Electrical Engineering Department,

Virginia Polytechnic Institute and State University,

Blacksburg, VA, USA.

Her research interests include distributed sig-

nal detection and classification, spectrum sensing, and cognitive radio

networks.

Aditya Padaki (Student Member, IEEE) received

the B.S. degree in electronics and communication

engineering from PES Institute of Technology,

Bangalore, India, in 2009 and the M.S. degree in

electrical engineering from Missouri University of

Science and Technology, Rolla, MO, USA, in 2012.

He is currently working toward the Ph.D. degree at

Wireless @ VT, Virginia Polytechnic Institute and

State University, Blacksburg, VA, USA.

His current research is on evolving receiver

performance metrics for efficient cognitive spectrum access. He is

interested in dynamic spectrum access, cognitive radio, multiple-input–

multiple-output (MIMO) systems, radio-frequency (RF) system design, etc.

Abid Ullah is currently working toward the Ph.D. degree in electrical

engineering at Virginia Polytechnic Institute and State University,

Blacksburg, VA, USA.

His research interests include database systems for spectrum

planning.



Munawwar M. Sohul received the B.Sc. degree in

electrical and electronics engineering from

Bangladesh University of Engineering and Tech-

nology (BUET), Dhaka, Bangladesh and the M.Sc.

degree in communication and signal processing

from Imperial College, London, U.K. Currently, he is

working toward the Ph.D. degree in electrical

engineering in the Bradley Department of Electri-

cal and Computer Engineering, Virginia Polytech-

nic and State University, Blacksburg, VA, USA,

working with Dr. J. H. Reed.

His research interest in wireless communication and signal processing

includes carrier aggregation, self-organizing networks, spectrum shar-

ing, and cognitive radios.

Taeyoung Yang (Member, IEEE) received the B.S.

and M.S. degrees in electronic engineering from

Sung-Kyun-Kwan University, Korea, in 1997 and

1999, respectively, and the M.S. and Ph.D. degrees

in electrical engineering from Virginia Polytechnic

Institute and State University, Blacksburg, VA,

USA, in 2003 and 2012, respectively.

He is currently a Research Scientist with the

Bradley Department of Electrical and Computer

Engineering, Virginia Polytechnic Institute and

State University. He has strong theoretical background on radiation

physics, co-site interference, and size-performance limits of wireless

devices. Along with the theoretical background, he also has extensive

hands-on experience in antennas, RF, sensors, and measurements. He

developed fully functional prototypes to prove new technologies and

demonstrate innovative wireless solutions. His current research empha-

sis is on developing core technologies in the area of spectrum sharing

and management, cognitive medical wireless infrastructure, medical

software-defined radio user equipment, and medical cloud services.

Collaborating with academic, nonprofit, and industry partners, he is

leading Health Care of the Future research initiative at the Wireless @

Virginia Tech.

Jeffrey H. Reed (Fellow, IEEE) received theB.S.E.E.,

M.S.E.E., and Ph.D. degrees from the University of

California Davis, Davis, CA, USA, in 1979, 1980, and

1987, respectively.

Currently, he serves as Director of Wireless @

VT, Virginia Polytechnic Institute and State Uni-

versity, Blacksburg, VA, USA. He is the Founding

Faculty member of the Ted and Karyn Hume

Center for National Security and Technology and

served as its interim Director when founded in

2010. His book Software Radio: A Modern Approach to Radio Design was

published by Prentice-Hall. He is cofounder of Cognitive Radio Technologies

(CRT), a company commercializing of the cognitive radio technologies; Allied

Communications, a company developing technologies for 5G systems; and

for Power Fingerprinting, a company specializing in security for embedded

systems.

Dr. Reed became Fellow to the IEEE for contributions to software

radio and communications signal processing and for leadership in

engineering education, in 2005. He is also a Distinguished Lecturer for

the IEEE Vehicular Technology Society. In 2013, he was awarded the

International Achievement Award by the Wireless Innovations Forum. In

2012, he served on the President’s Council of Advisors of Science and

Technology Working Group that examines ways to transition federal

spectrum to allow commercial use and improve economic activity.

