i n v i extending the reach of cognitive radio
Post on 18-Jan-2023
0 Views
Preview:
TRANSCRIPT
INV ITEDP A P E R
Extending the Reach ofCognitive RadioResearch into cognitive radio systems, aware of their environment and capable of
self-adjustment, has shown promising results; programs for large-scale cognitive
wireless systems are being explored.
By Preston F. Marshall, Member IEEE
ABSTRACT | There has been significant research progress
reported over the last few years in the development of
cognitive radio technologies. This paper reviews some of these
results, describes several critical issues to integrate these
results in usable products, and describes the large-scale
Defense Advanced Research Projects Agency efforts to further
develop and exploit dynamic spectrum access, and initiate
development and operational use of cognitive networking and
a new generation of affordable wireless technology. This paper
also describes future research needs to fully exploit cognitive
radio technology and addresses the challenges that will arise
with its large-scale deployment.
KEYWORDS | Cognitive networks; cognitive radio; Defense
Advanced Research Projects Agency (DARPA) XG program;
delay-tolerant networking; mobile ad-hoc ontologies; network-
ing; policy reasoning; spectrum measurements; spectrum
policy
I . INTRODUCTION
The recent upsurge in interest in cognitive radio technol-
ogy attests to the rapid growth in understanding and
appreciation of its potential to revolutionize the interac-
tion between a radio, its environment, and its user. While
it has been known that the environment has strong (andusually deleterious) effects on the operation of a radio,
only recently have we become able to exploit awareness of
this environment. Traditionally, we have considered how a
fixed mode of operation is degraded in a variety of less than
ideal situations. In cognitive radio, there is now the
opportunity to investigate how the use of adaptive modes
maintains and optimizes performance across a range of
actual environments.
Previously, design studies stated an assumed environ-
ment for which performance was optimized; a cognitive
radio design instead can argue for a range of environments,
each of which will have individually and continually
optimized performance. Radio operation thus transitionsfrom being designed by the radio implementer based on a
projected environment to being dynamically determined
based on the radio’s continually updated perception of its
environment. We can approach this as an opportunity to
achieve incremental gains in radio performance or as an
opportunity to fundamentally change the relationship of
radios, their environment, and the networks and applica-
tions that run over them.
II . RECENT PROGRESS
The transition of cognitive radio from concept to reality
has made considerable progress in the last several years,
driven down two fundamental paths.
The first is the drive to mature, demonstrate, and
ultimately deploy dynamic spectrum access (DSA) systems.A number of commercial and governmental organizations
have completed prototype implementations and submitted
them for varying degrees of technical or operational
scrutiny. This effort broadly clusters into two applications,
with much of industry focused on television white space,
such as contemplated by the IEEE 802.22 Wireless Regional
Area Networks [1], and research into more broad-based
spectrum sharing, generally focused on military or publicsafety as the initial applications.
Progress specific to 802.22 applications has been
driven by the opportunities provided by technical achieve-
ments and potential regulatory acceptance of television
Bwhite space[ utilization [2]. Proposed approaches vary
from access to central databases to fully distributed and
fused sensing. Of particular interest is the reliance on
Manuscript received November 11, 2008. Current version published April 15, 2009.
The author is with the Defense Advanced Research Projects Agency, Arlington, VA
22203 USA and the Centre for Telecommunications Value-Chain Research, Trinity
College, Dublin, Ireland (e-mail: preston.marshall@darpa.mil; pmarshal@tcd.ie).
Digital Object Identifier: 10.1109/JPROC.2009.2013008
612 Proceedings of the IEEE | Vol. 97, No. 4, April 2009 0018-9219/$25.00 �2009 IEEE
multiple resolutions of spectrum sensing to control theoperation of the device, with both fast (1 ms) and slow
sensing cycles (25 ms) driving the adaptation of the access
point and customer equipment. Implementations of
receivers supporting this architecture have been reported
[3], supporting the technical feasibility of this approach.
One approach relies upon analog spectral analysis and
temporal correlation, with only the correlation difference
requiring digitization at a very low rate, a departure fromthe assumption of all digital sensing [4]. Increased
processing in the analog domain and processing of
temporal signatures promises much lower energy con-
sumption and potentially lower cost than high-speed
digital approaches. Scheduling of cognitive radio sensing
within and across communities has been explored and
shown to be a feasible approach to the implementation of
multidomain cognitive radio operation [5].While much research and discussion has focused on
the Bwhite space[ arguments, important and relatively
neglected effects of adjacent channel energy are begin-
ning to be discussed within the framework of cognitive
radio. A simple Boccupied/not-occupied[ categorization
has been the basis of much of the discussion of spectrum
availability. Consideration of adjacent band effects [6]
showed that secondary use of television frequencies inadjacent bands has minimal effect on television recep-
tion and that reasonable performance is achievable at
relatively close distance to an adjacent channel television
transmitter.
Perhaps in recognition of both the likely deployment of
sensing-based systems and the history of malicious
applications of technology, methods of detection and
mitigation of falsified incumbent signals and trust ofdistributed cognitive radios sensing data have now been
proposed. These certainly need to be on the technology
agenda if the sensing scope of cognitive radios is to
transcend individual trust domains [7]. Without these
features, it is hard to imagine how any user or operator
would be willing to trust the operation of a network to a
technology so vulnerable to denial of service.
The second major research topic is the increasedunderstanding and initial prototyping of learning-based
cognitive radio algorithms. Reported work by a number of
researchers [8], [9] shows levels of performance that
should be sufficient to make the argument for incorpora-
tion into operational radios in the near future.
Early work on the representation of knowledge within
a cognitive radio has also been extended. Yarkin and
Arslan [10] described a radio knowledge representationlanguage that provides a mechanism to organize multiple
sources of environmental and location awareness into an
integrated representation. Of particular importance, this
work provided a canonical method to organize the channel
awareness into a structure upon which a cognitive radio
can make operating decisions. In a similar vein, other
researchers have extended knowledge representation and
encapsulation to describe the state of knowledge of theradio components of the transceiver [11].
Not only has research in DSA advanced the technology
over the last several years; the regulatory and spectrum
community also has become increasingly involved in the
process, evidenced by the success of the policy program
at the First, Second, and Third IEEE Conferences on
New Frontiers in Dynamic Spectrum Access Networks
(DYSPAN). Corresponding interest in industry andregulatory standards is reflected in transition of the
IEEE P1900 to a Standards Coordinating Committee
(SCC41) for Dynamic Spectrum Access Networks [12].
The breadth and international scope of membership offer
the opportunity to advance DSA as an integrated package
of technologies rather than as individual and standalone
efforts addressed on a country-by-country basis.
Several complete purpose-built cognitive radio exper-imental platforms have been demonstrated or are under
development. The Kansas University Agile Radio spon-
sored by National Science Foundation appeared as one of
the platform alternatives on which cognitive radio wave-
forms and algorithms can be constructed and tested
without being constrained to the characteristics of the
current commercial waveforms [13].
The Defense Advanced Research Projects Agency(DARPA) has initiated a Bsecond generation[ purpose-
built cognitive radio program [14]. This program, Wireless
Network after Next (WNaN), uses DSA as its fundamental
operating principle and trades high-performance individ-
ual transceivers for replicated but lower performance
transceivers. An example of this philosophy would be
adapting around situations that would otherwise cause
front-end overload, as will be discussed later in this paper.One of the stated objectives of this program is to
demonstrate that cognitive radio can produce at least the
performance of a conventional radio but accomplish this
with reduced performance components through use of
adaptation to mitigate the performance stressing environ-
ments that would otherwise drive energy consumption and
cost. The WNaN radio has four independent transceivers,
covers 900 MHz to 6 GHz, has high-quality front-endfilters to enable the DSA functions to identify and utilize
low-energy pre-selector bands, and uses commercial
quality radio-frequency integrated circuits for the trans-
ceiver functionality. This platform is intended to provide
not only the DSA functionality but cognitive topology
management and content management as well.
III . THE NEED FOR MULTIPLECOGNITIVE RADIO REASONINGTECHNOLOGIES
It is important to recognize that the various focus areas in
cognitive radio research implicitly have specific assump-
tions about the definition. There appear to be two that we
should consider. When Mitola [15] first referred to the
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 613
term, he was clearly referring to a device that had thecharacteristics of a conscious intellectual processVa
learning system, such as envisioned by decades of artificial
intelligence research. A (perhaps oversimplified) litmus
test of such systems might be that they derive their
behaviors from experiential learning. Consistent with that
definition of cognitive radio, researchers have reported
progress in applying learning engines to the selection of
channels and waveforms [16].In usage, the term cognitive radio has taken on a
significantly broader connotation and less technologically
challenging definition. The DARPA Next Generation
Communications (XG) program reported success in DSA
using fixed algorithms and policies [17]. The U.S. Federal
Communications Commission undertook a cognitive radio
access proceeding [18] focusing specifically on dynamic
spectrum behavior and was silent on how such behaviorwould be implemented. Several industry organizations
have announced progress on cognitive radios, although
there is little published information on their designs,
algorithms, behaviors, and, most important, performance.
Although the strict definition of cognitive might not
seem to include these technologies, if we wish to assert
that cognitive radio implements Bcognition,[ a fundamen-
tal distinction between a cognitive radio and a non–cognitive radio would be the degree to which the device
makes decisions based on awareness of the environment.
Such a definition has advantages since it is focused on
Bwhat[ the device does, not Bhow[ it does it.
The necessity to be open to both definitions is apparent
in the area of frequency selection, where two different and
competing concerns are being addressed through cognitive
radio: noninterfering and optimized operation. An insis-tence that a cognitive radio must derive its algorithms
through learned behavior provides no mechanism to
address the inherent constraint that the effect of some
decisions is only known external to the learning system
itself. Here the necessary feedback to reinforce behavior is
not available. The impressive results reported for physical
layer learning are achievable because the two link partners
constitute a closed system that can use itself to providereinforcement to the learning. In contrast, when we
consider a system comprising many heterogeneous users of
spectrum, with incompatible waveforms, protocols, and
even operational concepts (a broadcast, receive-only node
has no ability to report interference events), then it is
much harder to apply feedback as the sole mechanism to
discover and correct Baberrant[ behavior.
Some other practical issues also arise. Given the poorstate of information assurance technology, few system
owners or implementers would be willing to have their
system exchange information with any external party. The
growth of Internet malware demonstrates that it is
unrealistic to require participants to cooperate and
exchange information that will impact behavior and
performance. Engineers might argue that the exchanges
were harmless, but the seemingly endless discovery ofexploits and unintentional behaviors in commercial soft-
ware products would undoubtedly hinder any solution that
required nodes to Btrust[ actions or information provided
by other nodes external to the nodes own trust domain.
Regulators tightly control spectrum, due to the extreme
social and economic consequences of interference to key
infrastructure and services. It is unlikely that they would
accept the principle that, although a node’s behavior willinitially be very poor, it will eventually learn to Bplay nice[in the spectrum (after causing any number of interference
events while learning). Similarly, they might not be
pleased to learn a device tested and accepted as behaving
well and not interfering would have the capability to later
learn Bbad habits[ that would have the very consequences
the technology was intended to avoid.
There is a strong argument that when improperbehavior has negative impacts on other, external spectrum
users, than the use of declarative techniques is advanta-
geous. These techniques offer an approach that can be
analytically proven, experimented with to any desired level
of confidence, and then Bfrozen.[ Admittedly, this is not as
intellectually pleasing as a radio that Bprograms itself,[ but
it is testable, verifiable, predictable, and perhaps even
provable.This provides a clear partitioning of the cognitive radio
reasoning infrastructure. On one side, there are functional
behaviors such as sharing and etiquette that focus on the
device’s impacts on other users in the environment.
Presumably, this behavior is externally mandated, as the
radio does not benefit from the constraints and might have
its performance reduced. On the other side, one or more
adaptation mechanisms can optimize performance whileconstrained by external policies. Not surprisingly, these
two research areas have unique approaches to exploiting
environmental awareness.
One of the DARPA XG program’s objectives is
developing a declarative language and processing tool set
for external policy control, such as required by spectrum
sharing. The program has developed two radio-borne
reasoning engines that provide inferential processing ofpredicate calculus policies based on a spectrum manage-
ment ontology and rule set [19], [20].
IV. AN INCLUSIVE MODEL OFCOGNITIVE RADIO
It is appropriate to partition the intelligence model of a
cognitive radio into endogenous and exogenous compo-nents. Previously, the author had proposed the partition-
ing of DSA aspects of cognitive radio into two essential
elements: a system strategy reasoner (SSR) that optimizes
the spectrum decisions and performance of the device
and a policy conformance reasoner (PCR) that executes
and enforces externally provided policies, as shown in
Fig. 1 [21].
Marshall: Extending the Reach of Cognitive Radio
614 Proceedings of the IEEE | Vol. 97, No. 4, April 2009
The PCR provides the external (exogenous) policy
component(s) that address the device’s impact on theexternal environmentVprimarily, avoidance of spectrum
interference. The SSR provides the internal optimizing
behaviors (endogenous) to maximize performance through
selection of operating mode and other parameters. This
partition reflects the fact that the external effects of the
device are difficult and often impossible to detect from the
reference frame of the individual radio. It is therefore
difficult to argue for an exclusively learning model of acognitive radio or one based solely on performance
optimizing algorithms.
An example of this partitioning is apparent in the three
principles of cognitive radio, which the author proposed as
the standard for evaluating the performance of the XG
DSA technology [22]. These were:
1) do no harm (to other users of the spectrum);
2) add value (to the user, operator, or owner whoinvested in the technology);
3) perform (robustly and reliably in a range of
environments and user mission needs).
These principles reflect the role of the cognitive radio
intelligence mechanisms in protecting other users from
the effect of device’s behavior and providing a measurable
return on the recurring and nonrecurring investment in its
creation and manufacture. These first two principlesaddress unique and nonoverlapping objectives and respon-
sibilities. They would appear to be difficult to accommo-
date in any single abstraction (learning versus declarative
knowledge.
Looking beyond cognitive spectrum management,
there is a need to extend the concepts of cognitive radio
from a spectrum and link focus to an integrated abstraction
that is inclusive of aspects of wireless device decisionmaking. This includes multiple decision domains, such as
network routing, topology, user interaction, security, and
content management, in a structure capable of expressing
both constraining and optimizing policies, such as thoseshown in Fig. 2.
Two issues must be addressed to make this vision a
reality: 1) an abstraction is needed of how these different
cognitive adaptations interact and 2) a computer science
implementation. Without these, the cognitive radio
community will be building hundreds of stand-alone
experiments but few usable products that can benefit
from more than one breakthrough due to lack of anintegrated mechanism that can unify multiple technologies
for use in constructing cognitive radios. As a user of
artificial intelligence technology, the cognitive radio
community may be one of the first communities able to
articulate the need for a unified theory of cognitive
automata that can integrate learning and declarative
knowledge into a single logical framework rather than
distributed across two disparate mechanisms.As an example, consider spectrum-related decision
making. Traditionally, other users of the spectrum are
represented by national regulatory administrations. In the
cognitive radio model, these same protective interests are
represented by the constraining exogenous policy compo-
nent. The internal optimization of performance is provided
by the endogenous component. The exogenous compo-
nent(s) ensures that the decisions of the endogenouscomponent(s) do not have unacceptably detrimental
effects on other users of the environment with which the
cognitive device interacts, such as the spectrum, network,
or device users. This partitioning allows the endogenous
component(s) to operate as an unconstrained Bgreedy[algorithm, since the Bsocial[ values of the community
(other spectrum users, network users, and device manu-
facturers) are protected by the exogenous component(s).A reasonable assumption would be that the permissive
decisions of the exogenous component must be made
acceptable to the interests of the other environment users,
Fig. 2. Candidate scope of cognitive radio policies.
Fig. 1. PCR and SSR model of a DSA cognitive radio.
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 615
typically through a certification or licensing process. This
implies its behavior should be demonstrably deterministic.
It would be hard to obtain approval for a device that could
later learn bad behaviors.
This discussion points to the necessity that the advance-
ment and deployment of cognitive radio not be viewed assolely a technical challenge. Almost all descriptions of
cognitive radio include some degree of spectrum selection
and control; environmentally aware selection among
spectral and frequency options is clearly one of the defining
aspects of the technology. To achieve this, it is desirable
(inevitable) that the regulatory community be highly
involved in the development of the technology and that it
include features that address their considerations to at leastthe rigor of current technical and regulatory methods.
The operation of the exogenous processes can be
grouped into one or more policy conformance reasoners,
and the operation of the endogenous ones can be grouped
into a system strategy reasoner, as depicted in Fig. 3.
Potentially, the actual logic of these reasoners could be
provided by an external party, such as a spectrum
regulator, network owner, or hardware developer, througha common policy language. In fact, in a heterogeneous
network, a desirable attribute of the policy process is that it
specifically enable operation using policies derived from
multiple sources. Equally important, these sources of
policy should be able to develop policies that areindependent of the policy context (i.e., other policies a
user may have loaded) within the device. A spectrum
owner might provide policies for sharing a band, but these
policies would have to coexist harmoniously with policies
independently provided by national regulators, and
perhaps device limitations provided by the manufacturer.
Similarly, a device manufacturer might have products that
could be running a variety of cognitive radio software, andwould provide policies to ensure that the operating limits
of its specific device were not exceeded. Fig. 2 illustrates
the necessity of fitting policies together dynamically into
the architecture depicted in Fig. 3.
This figure depicts the control cycle of a cognitive radio:
sensing the environment(s); creating responsive strategies;
ensuring their compliance with exogenous constraints; and
(implicitly) implementing them within the device. Someauthors have treated the spectrum environment as unique
from the user behavior and device status; however, a more
general framework treats these as symmetric domains that
coexist and are integrated into the behavior of the cognitive
device through mechanisms within the system strategy
reasoner. Intuitively, it is reasonable that the more these
domains can be integrated within the SSR, the more
globally optimal the decisions should be. Although thefigure shows policy conformance reasoners for each of the
environmental domains, this one-to-one relationship is not
necessary if the policy language is general enough. The list
of environmental domains is not intended to be exhaustive
but to show the scope of a symmetric solution to cognitive
radio conformance policies.
Some illustrative spectrum domain considerations are
provided in Table 1 and are partitioned between theexogenous and endogenous reasoning processes. In
general, it is apparent that the exogenous concerns are
those typically associated with communications device
regulation, while those of the endogenous component are
typically associated with conventional communications
link design. This reflects that cognitive radio strategy
development is really an analog of the manual design
process, different in that it is interactive with itsenvironment, based on measured rather than predicted
data, and continually adjusted. However, the considera-
tions are the same as for a noncognitive device.
DSA may be the first domain in which we argue for
adaptation but clearly is not the only one and perhaps, in
retrospect, may not even be the most important.
Table 1 Considerations for Endogenous and Exogenous Spectrum Reasoning
Fig. 3. Relationship of endogenous and exogenous reasoning in
cognitive radio.
Marshall: Extending the Reach of Cognitive Radio
616 Proceedings of the IEEE | Vol. 97, No. 4, April 2009
Most consideration of the exogenous PCR has been in
the spectrum domain and has been focused on the issues
associated with spectrum access [19]. The concept is
equally applicable to the other environments of the cog-
nitive radio. For example, a network owner might requirethe device to validate its admission and packet traffic to
ensure that the actions of the device were consistent with
the network’s operating rules, or a device manufacturer
might include a validation engine at the hardware interface
to ensure that the operation of cognitive radio algorithms
did not violate any of the operating limitations of the
device (e.g., thermal, duty cycle, amplifier bias, etc.). A
conventional network firewall is an obvious specializedimplementation of this structure’s PCR construct. Simi-
larly, the V-chip (television programming content control)
provides rudimentary video policy enforcement; Internet
content filters are also examples of PCRs that operate at
the user interface to ensure that information over this
interface complies with a externally provided policy. The
PCR is thus not really a new concept; it is a generalization
of functionality that experience has shown it is needed tobuffer the actions of a technology and the environments on
which it has an impact.
Another consideration is the location of these compo-
nents. Although there is a tendency to think of cognitive
radios as peers and self-contained; there is no necessity
that implementation of these components be symmetric or
local. Spectrum selection and conformance checking can
certainly be performed at a central node that isadvantaged; network policies can be provided at boundary
devices, etc. An illustrative list of these environments, the
characteristics that are dependent on them, and their
significance is provided in Table 2.
The partitioning of the reasoning into these two
categories provides a simplifying abstraction. At best, the
policy conformance process cannot improve performance;
its effect is only to preclude certain actions. Only the
endogenous component can contribute to performance,
and therefore it is the primary contributor to the
performance benefits achieved from cognitive radio.
The line between cognitive and noncognitive radio isnot absolute. Many low-cost consumer products have
adaptive behaviors, such as selecting the type of cellular
service to connect to, adapting WiFi modes, selecting ISM
channels for cordless phones, and selecting between WiFi
and cellular connectivity. These clearly meet at least the
letter if not the spirit of the cognitive radio definition.
The intuitive distinction is that a cognitive radio must
adapt a significant number of these parameters andmodes, and do so in a way that recognizes the interaction
of these parameters across multiple domains, as depicted
in Figs. 2 and 3.
V. A NEW STRUCTURE OF THEARGUMENT FOR COGNITIVE RADIO
The tendency has been to view cognitive radio as a newer,and ultimately more complex, device than the radios that
have preceded it in the evolution of wireless devices. This
relegates the use of cognitive radio to relatively cost-
insensitive applications. An alternative view is that
cognitive radio provides the opportunity to mitigate
shortfalls in hardware performance and, by doing so,
make less capable hardware perform reliably and effec-
tively in environments in which a conventional devicewould fail.
This argument would lead to the conclusion that the
community can present the benefits from cognitive radio
operation in two ways: improve the performance of similar
equipment or reduce component performance needs of the
implementation while not sacrificing performance. In
practice, an application might elect a mix of both of these.
Table 2 Illustrative Cognitive Radio Operating Environments
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 617
Table 3 illustrates a summary of how cognitive radio
benefits can be reflected.
The first measure (row 1) is the effect of cognitive radio
on the performance and reliability of the radio and
network. There is a growing literature providing both
analysis and methodology addressing this metric. The
second row of this table has received scant attention. Asmentioned, the DARPA WNaN program has the explicit
goal of demonstrating that a reduced performance
cognitive radio can achieve a high level of performance
due to the effectiveness of the adaptation methodology.
Fig. 4 illustrates two perspectives on cognitive radio
performance.
Line A illustrates the improvement in performance
achieved by implementing cognitive radio technology witha fixed device performance level (row 1 of Table 3).
Achieving this increased performance has been one of the
motivators for cognitive radio research.
What should be of equal interest to the cognitive radio
community is the opportunity permitted by line B (row 2
of Table 3). Line B represents the reduction in component-
level performance that can be tolerated (through cognitive
radio technology) while maintaining a consistent and/oradequate level of network performance from the device.
The reduction in component performance provides the
opportunity to either greatly decrease the cost of the
networks or greatly increase the density of the network.
An example of this argument is the mitigation of front-
end overload effects caused by high levels of energy within
the receiver preselector passband. A difficult challenge for
spectrum management is to ensure that radios do not
attempt to operate simultaneously on the same frequency.
However, it is even more difficult to also ensure that a
strong signal is not present near the assigned channel. This
is particularly true in peer-to-peer applications, such as
military or public safety operations. A noncognitive radiomust have sufficient front-end linearity to operate
effectively in any environment to which it might be
assigned a frequency. A cognitive radio, on the other hand,
need only have sufficient front-end linearity to ensure that
at least one of its front-end preselector bands (covering
candidate frequencies) has less than the overload thresh-
old energy. By applying DSA and operating band selection,
the radio’s front-end linearity performance requirements[such as third-order intercept point (IIP3)] can be reduced
from a level that can operate in all of the filter selections
that might contain an assignment to a level that is
sufficient to ensure reliable operation in at least one of the
filter settings. Since front-end energy consumption is
typically proportional to front-end dynamic range, a
corresponding savings in energy consumption is obtained.
Work by the author has shown that there are strongarguments that orders of magnitude reduced IIP3 levels, or
similar increases in reliability, can be achieved through
such adaptations [23].
This argument reverses some perceptions of cognitive
radio cost. One of the objectives of the WNaN program is
to demonstrate that a highly reliable, military mission
capable radio can be significantly reduced in cost (to $500
in quantity) through cognitive radio technology whileproviding more flexibility at physical and network layers.
WiFi, cellular, and MP3 players all demonstrate that
significant reductions in cost can create uses for technol-
ogy that would have been unimaginable at higher cost
points. Cognitive radio research should strive to enable
affordability as well as to enhance performance.
VI. THE NETWORK AND UPPER LAYER’SBEHAVIORS IN COGNITIVE RADIO
While much of the argument for cognitive radio has
focused on the characteristics of physical layer adaptation;
it can be argued that the network layer, leveraging the
flexibility enabled by the DSA and cognitive radio model,
Table 3 Metrics of Cognitive Radio Performance Benefits
Fig. 4. Benefits of cognitive radio operation.
Marshall: Extending the Reach of Cognitive Radio
618 Proceedings of the IEEE | Vol. 97, No. 4, April 2009
provides equally significant benefits. DSA, by eliminating
the specific assignment of individual links to specific
frequencies and bandwidths, enables adaptation not only
of the physical layer but also in the organization of the
network itself. Some of the adaptations that WNaN will
perform to address and mitigate the effects of a range of
environments, device limitations, and network delivery
requirements are shown in Fig. 5. These adaptations have aprerequisite that the network be able to unilaterally select
frequency and bandwidth for each link.
If bandwidth in one region of the network is
inadequate, the network can Bmove[ spectrum through
locating or reassigning spectrum to supplement the
throughput. If a link is interfered with by other signals
or overloaded by adjacent channel energy, the cognitive
radio locates spectrum more suitable to its operation and isfree to implement that decision locally and unilaterally. If
there is no way to locate enough spectrum in a congested
region of the network, the network changes its topology to
route out of and around the congested region.
The DSA community has advocated the acceptance of
DSA technology to address many of the shortfalls in
manual frequency planning and management. The WNaN
approach additionally argues to adopt DSA for theflexibility it affords:
• to the device for self-management of the environ-
ment and the resulting reduction in component
performance requirements, and thus cost;
• to the spectrum manager to not have to separate
receivers from strong sources of adjacent band
energy;
• to the network and applications to reconfigure andreprovision wireless bandwidth dynamically
Simply put, it argues that a DSA cognitive radio has
advantages, even if the access to spectrum is neutral.
In initiating the DARPA Wireless Network after Next
WNaN cognitive radio program, the author postulated
11 theses for the program; seven of which are shown in
Table 4 [14]. Although many of these theses relate to
the network layer operation, they are enabled solely bythe adaptation that is provided by the DSA mechanism
at the physical layer.
For cognitive radio to fully achieve its maximum bene-
fit, the networking community needs to reframe the ques-
tion from Bhow do we route over a topology?[ to Bwhat is
the right topology to form in order to perform this
mission?[ This is a fundamental change in the direction of
networking and is an opportunity that is not present in theconventional terrestrial or more static wireless technolo-
gies. For the core network, it is inconceivable that new
fiber-optic lines could be determined, buried, terminated,
and provisioned to a network in a subsecond timeline, but
that is exactly the opportunity that cognitive radio pro-
vides. Similarly, the connectivity of new satellite trans-
ponders or cellular towers can ot be created by a network
Bon demand,[ such as a DSA architecture can inherentlyprovide.
For 30 years, topology has been treated as a given, and
the routing as the consequence. With cognitive radio, this
relationship can be reversed. It determines the routing that
is needed, then uses DSA to create the right topology to
implement it. The cognitive radio and network community
must accelerate the research to develop these mechanisms.
Table 4 Selected WNaN Cognitive Radio Theses
Fig. 5. Upper layer exploitation of DSA flexibility.
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 619
One of the initial investigations being performed by theWNaN program is the integration of content management
with the operation of the cognitive radio algorithms. The
Delay Tolerant Networking bundle protocol agent [24] is
the underlying protocol for this experimentation, which is
being performed as an element of the Disruption Tolerant
Networking program. Unlike a packet, a bundle is a unitary
set of payload and metadata information that can be
cached, routed, or stored by intermediate nodes. Itprovides the cognitive network the facility to organize
and optimize the placement of information within the
network. Integrated with the DSA facility, it now enables
the network to balance the constraints on topology
(derived from spectrum awareness and link path char-
acteristics) and the user behaviors in accessing content.
Content can be located as dictated by actual needs and
real-time assessment of physical constraints rather thanstatically preplanned. The cognitive radio network can
consider descriptions of information content as an address
rather than descriptions of fixed server node locations. In
fact, if we look at the process of Internet access today, the
search engines create essentially a content-based network,
just overlaid on a packet and address-based network.
A use case for this technology is instructive. Although
backhaul bandwidth in much of the world is plentiful, inother regions backhaul is not available, or highly con-
strained. A cognitive radio network would manage the
location of content distributed across end-user devices,
without discrete servers or any fixed address structure.
Many areas of the world would benefit from even localized
services that could provide information on locally critical
topics, such as medical assistance, food and water
availability, and education. Without the requirement forbackhaul to fixed services, a content-based cognitive radio
network is a feasible and highly desirable outcome of the
current research. Although wireless may not be the ideal
backhaul to distant services, it is an effective means to set
up networks whose users are concerned with resources
within walking distance more than those on worldwide
social networking sites.
An additional consideration is that the cognitivenetwork does not make demands on what is perhaps the
scarcest resource in many environments: the intellectual
infrastructure to assign frequencies and addresses, define
servers, manage indexes, caches and backhaul networks,
etc. An information network that operates Bout of the box[certainly has appeal and unimagined utility.
VII. WHAT IF WE SUCCEED? WHAT NEWRESEARCH IS NEEDED?
Much of the cognitive radio research is focused on current
problems in wireless technology, and clearly it is essential
that cognitive radio offer useful solutions to these
problems. However, we should also be cognizant that
when cognitive radio deployment occurs, it will introduce
new consequences, opportunities, and challenges of itsown. The deployment of tens of cognitive radios will have
little impact; the deployment of hundreds of thousands
will change the nature of the spectrum environment,
create new issues in network operation, and require policy
control solutions that are beyond those envisioned today.
In particular, the inclusion of dynamic and opportunistic
spectrum within the cognitive radio will fundamentally
change the nature of a radio deployment, due throughsignificant reduction in the constraints of spectrum avail-
ability and management. It is important that a portion of
our research look ahead to this epoch. It might not be that
far away.
A. Density and ScalingA fundamental change that cognitive radio will induce
is in network density. Prior work in self-forming networksand related technologies focused on discovering and
exploiting all possible connectivity. At some level of
network density, the situation reverses; the network needs
to organize its topology to reduce the complexity and
interactivity of the connectivity. Fig. 6 illustrates the
conventional view of mobile ad-hoc networks.
In this model, the network’s strategy is to maximize
connectivity by rendezvousing all of the nodes onto asingle frequency [14]. Unfortunately, this architecture has
been shown to have severe scaling limitations [25] due to
(among other reasons) mutual interference between the
nodes. Even nodes that cannot demodulate a signal may be
interfered with by its presence. The interference range is
well in excess of the communications range. A simple
measure of this network is to note that only one node can
transmit at any one time, despite the number oftransceivers that are present. This is a strong intuitive
argument against its scalability.
In WNaN, the author proposed [14] that providing
multiple separate transceivers, each forming small local
networks, would be one mechanism by which the network
could avoid performance constrained by interference.
Small, local networks on separate frequencies (made
possible by DSA) would decouple the effect of changes in
Fig. 6. Typical MANET network.
Marshall: Extending the Reach of Cognitive Radio
620 Proceedings of the IEEE | Vol. 97, No. 4, April 2009
one portion of the network to reduce thrashing. Fig. 7
illustrates the network being developed by the WNaN
program. In this architecture, each node connects to a
different set of neighboring nodes through multiple
transceivers, frequencies, and autonomously managedsubnetworks.
Actual experimentation with this architecture will not
begin until later this year, and dense quantities of devices
will not be available for another year after that, so the
experimental results of this approach will not be available
for several years.
This architecture is enabled by DSA and cognitive radio
technologies. The management of this many frequencies ina complex and constantly changing network would be
impractical for any static spectrum assignment strategy
and is probably only practical with automated management
and interference control.
The second enabling consideration is the use of the
adaptation of a cognitive radio to use lower cost and
performance transceiver components in each of the nodes,
making the multitransceiver configuration affordable at alower cost than if a single, high-quality transceiver was
utilized. Not only does the cognitive radio functionality
improve the link-level performance; it is the key factor in
replication in the physical layer.
This approach may provide a mechanism to address
the scaling limits imposed by node-to-node interference
that are otherwise inherent in the MANET network struc-
ture. The issue of how to route and determine appropriatetopologies with this many degrees of freedom (frequency,
interference environment, waveform, power, connectivi-
ty, routing, etc.) is a fundamental research question that
arises with the integration of DSA and advanced
networking.
B. Expressions of Cognitive Radio Algorithmsand Reasoning
Cognitive radio research is being undertaken by a
significant number of researchers, attacking a large number
of problems and opportunities in the field. The resulting
concern is that as individual techniques are developed andvalidated to meet cognitive radio challenges, such as
channel selection, waveform selection, power control, and
filter settings, there is no common or community accepted
framework to integrate these technologies together into
products. A potential result of this will be islands of tech-
nology capabilities, each targeting specific problems but
not readily integrated with other solutions to other
problems. Even if the range of policies and controls shownin Fig. 2 are developed, how will their operation be in-
tegrated onto a single device, or within a single networking
product?
There has been progress reported in this area. Among
other initiatives, SCC41 has established a policy language
as one of its goals. The DARPA XG program has reported
development of two policy reasoners for spectrum policy
conformance reasoning [19], [20]. This is a reasonablestart, but a broader base of the community needs to be
involved in contributing their research insight into to these
ontologies, semantics, and necessary syntactic structures;
and a more extensive set of reasoning domains needs to be
considered during their formulation.
Additionally, the reasoning engine is only one-half of
the tool set that must integrate into a cognitive radio.
There is an equal need for learning-based engines thatoffer so much promise for the endogenous portion of the
decision process. Even with these two components, there
is still the challenge of integrating these two very different
structures into a cohesive theory and implementation for
network devices. There are a large number of promising
results being reported but little in the way of an abstract or
conceptual structure which they can populate.
C. StabilityThe logistics of acquiring, programming, and operating
wireless nodes has generally limited the size and extent of
the networks that have been used to prototype cognitive
and ad-hoc wireless technology. When the node count is in
the hundreds, we can certainly treat each node as an
individual, and the dynamics of the system is sufficiently
limited that it can be analyzed by discrete tools. As wecontemplate networks that have large membershipVnot
in the hundreds, but in the hundreds of thousands of
nodesVthen explicit understanding of the behavior of
individual nodes becomes unachievable and a transition to
a more systemic treatment of network behavior is
mandatory.
Large-scale simulation tools have given us the ability to
achieve insight into the operation of individual nodeswithin routing networks. What will make cognitive radio
even more challenging is that the complex networks
envisioned not only interact through the network mechan-
isms but also interact through the spectrum they share at
the physical layer and have the necessity and capability to
continually morph the connectivity and content locations.
Conventional radios assume (and generally receive)
Fig. 7. Cognitive radio–enabled WNaN network.
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 621
spectrum that is intended to make node operationorthogonal; a cognitive radio will share spectrum, so deci-
sions that are made on one node will ripple through
neighboring nodes and, through them, potentially of all the
nodes within a network. A cognitive radio network will
consist of at least the physical, media access, and network
layer processes, each making decisions that ripple across
and between layers.
These stimuli transit the network at different rates, aredamped or initiate cascading changes at different points,
and have different responses at each layer. The problem
becomes a system exhibiting characteristics of transient
response, coupling, damping, and oscillation.
An example of such behavior could be as simple as just
one radio starting to transmit on a frequency, forcing ten
radios to move to new frequencies, each of which in turn
force ten more to do the same. At the same time as thisripple of frequency changes moves through the network,
the network layer would be attempting to update routing,
even as the network connectivity was changing. Some of
these effects are shown in Fig. 8, providing change stimuli,
coupling, and damping effects on a potentially oscillatory
system. Stability analysis of layer 3 has become an accepted
technique, but a system that has this scale of interaction
with and through the environment, users, devices, andapplications of a cognitive radio network is inherently
much more complex.
Neel [26] provides an argument that the behavior of
spectrum interaction can become stable, even given locally
optimizing behaviors, but this analysis must be extended to
include the interaction of these processes with the other
layers and the effects between them.
The change in focus from node to system behavior isimperative to adoption of cognitive radio technology. One
appeal of cognitive radio is that it can manage dense
environments; the problem is that it may become so
complex we cannot prove it will have an acceptable range
of behavior and performance.
The critical consideration of density is inherent in any
of the concepts of spectrum sharing. One effect of current
spectrum management has been to protect most radiosfrom detrimental effects of their environments, including
adjacent band signals. Where these protections have failed,
the effects have been significant, as demonstrated in the
case of interference between adjacent cellular services and
public safety radio systems in the United States [27]. Since
one assumption of cognitive radio is that it will share
spectrum, it is reasonable to assume that it will both be
required to locate in the proximity of strong signals, suchas television broadcasts, and will increase the density of
band usage where it is permitted.
A quantitative sense of one of these effects is
considering that if a radio had just enough IIP3 perfor-
mance to not have its noise floor impacted in a given
environment, and DSA enables ten times the emitter
density (a typical claim for increased spectral usage) [17],
then energy provided to the front end would be expectedto increase proportionately by ten and the third-order
intermodulation products to correspondingly increase by
20–30 dB. Density creates its own issues at both the
network and the physical layers of the device!
To some extent, the dense cognitive radio network will
function like an ecosystem, similar to a population model
that makes no assertions about the future state of
individual organisms but can predict the aggregatedynamics and end state of the population as a whole.
The need for showing the stability of an extremely large
number of nodes would argue for a transition from a
discrete simulation to a formal analytic model of the
dynamics of a system of nodes. This transition might have
significant implications for how the design of such large-
scale and highly coupled systems is approached. Proof of a
formal stability may have to become a fundamental driverin media access control, network, and applications layers.
Proof that they have appropriate damping of impulsive
modes, resist excitation, randomize responses, vary
resonant modes, and the kind of design decisions normally
associated with a control process may need to become
integral in the design of these layers, and of equal or
greater importance than today’s concerns with throughput
and latency. The understanding of these complex interac-tions needs to move from Bif, then, else[ to differential
equations, modes, and generalized expressions of stability
and equilibrium conditions that can be proven for all cases
rather than shown for a single or set of discrete cases.
D. Cognitive Radio Environmental AbstractionsThe environment of a cognitive radio is complex and
unique to each location on the earth. It is significant thatthere are now a number of concerted efforts to capture and
disseminate the specifics of representative environments.
However, this leaves the researcher with the profound
problem of how to show that a given cognitive radio
technique is effective in a range of environments All
researchers cannot be expected to show that all techniques
work against all of the (increasing number of) collectionsFig. 8. Illustrative stimuli, coupling, and damping modes.
Marshall: Extending the Reach of Cognitive Radio
622 Proceedings of the IEEE | Vol. 97, No. 4, April 2009
of spectrum and network data. An automated threshold
determination algorithm has been presented by research-
ers at the University of Kansas [28], which adds rigor and
quantitative confidence to what other researchers have
assumed from Bvisual[ inspection and points to the need
for a universal and community accepted way to both
exchange and characterize spectrum measurements. Thequestion of spectrum characterization (beyond occupancy
metrics) has not been approached generally.
There are a number of techniques that have been
developed to characterize spectrum usage on a frequency
by frequency basis, such as the Spectrum Utilization Index,
which provide a focus on frequencies usage characteristics
[29] and effective use of the spectrum. Measures such as
Bpercentage spectrum occupied[ are useful for illustratingthe opportunity for DSA systems and the effectiveness of
spectrum management but are not insightful into the
dynamics of how devices will operate in the spectrum. For
the community to be able to compare results of alternative
algorithms and technologies, a much more subtle set of
effects must be considered. Table 5 illustrates some of the
considerations that any environmental abstraction should
address, ideally as closed-form expressions.To compare alternative implementations of cognitive
radio, it is necessary to be able to characterize the
environments in ways that are repeatable within a line of
research, and across researchers. Additionally, it is
preferable to be able to characterize these environments
in ways that enable the community to apply the generality
of formal methods rather than rely solely on empirical data
whose general applicability is often unknowable, andcertainly not provable. These closed-form arguments could
become powerful tools for understanding and advocating
cognitive radio technology without sole reliance on
specific sets of empirical data and limited scope of proof.
The author has proposed a structure to describe the
distribution of narrow- and wide-band energy throughout a
spectral environment [30], but the purpose of this
discussion is not to advocate for any specific representa-tion of spectrum characteristics. This is itself a subject for
continued research. Would the community rather argue
that a given technique appears to work well in the Chicago
spectrum collection but seems to fail in the New York one;
or would it prefer to argue that this technique has provably
positive performance gains for any environment in which
some agreed-to parameter �spectrum > 3:0.
E. Decision Theory in Cognitive RadioThere is a considerable body of literature that details the
optimizations that can be performed on individual chan-
nels, and we are seeing the beginning of a similar literaturein the area of more complex adaptations. In the imple-
mentation of DSA functionality, the information required
to make spectrum decisions can be obtained at relatively
low cost in energy and channel time. This class of sensing
can generally be accomplished as unilateral operations
performed by the nodes, without explicit coordination. On
the other hand, channel measures such as delay spread can
only be obtained when the channel is established and ableto exchange information in sufficient quantity to satisfy the
decision needs of the adaptation algorithms. The actual
amount of channel state information may not be excessive,
but the time to perform rendezvous, synchronization, and
measurements may be significant, particularly if the time is
significant compared to the correlation time of the channel
information that is being acquired.
Although it has been shown that the use of this in-formation can improve the performance of devices, prog-
ression of the field requires we now consider the benefits
we obtain from additional channel information compared
to the cost of obtaining it. Designers generally have at least
statistical knowledge of the environment in which their
devices will operate, so the problem is one of decision
theory: how much is a decision improved if uncertainly is
reduced compared to the cost of obtaining additional in-formation. For example, a radio might have both a constant
envelope and high peak average ratio waveform, and would
choose based on the measured delay spread and range of
intersymbol interference that would occur. The question
that the research community must start to address is how
much is a better decision between these waveforms worth,
compared with the cost of initially and periodically mea-
suring the channel characteristics. One can imagine thatsome decisions would be better left with some uncertainly,
but today there is no formal mechanism to investigate these
trades available. The first algorithm needed is to decide
whether to invest in acquiring channel state at all, not the
decision of how to use it.
Table 5 Examples of Desirable Spectrum Distribution Characteristics
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 623
F. Relationship of Content Flow and Cognitive RadioMost research in cognitive radio has followed the con-
ventional abstraction of the application layer operating
independently of the network layer. As previously men-
tioned, current research is just starting to investigate ex-
panding the cognitive management domain to include
determining the location of content in the network. Just as
reaching down to the physical layer greatly increased the
complexity and interaction of the networking dynamics,introduction of dynamic content location does the same for
the upper layer functions of the network. It is certainly too
soon to argue for the benefits of this expansive vision of
cognitive adaptation, but there is a strong intuitive argu-
ment that the control over content location provides radio
and network optimizing options that are not available when
content location is fixed, and managed as an independent
domain. A decade ago, self-forming packet wireless net-works were a research community priority; perhaps self-
forming wireless information networks should be the next.
G. The Practice of Interference Tolerant OperationMuch of the dynamic spectrum access research has
focused on the challenge of assuring noninterfering
operation with respect to incumbent users. A much more
aggressive role for cognitive radio is to ensure that devicescan meet stated reliability, even in the presence of
increasing levels of interference. In this regime, devices
need not ensure noninterfering operation to the levels of
assurance otherwise required to protect interference
intolerant devices. Fig. 9 illustrates the differences from
current DSA to range of technologies that are needed to
provide interference-tolerant operation. Note that each of
these is a very active research area in its own right; thedistinction is integrating their individual contributions to
significantly increase the density of devices. In this model,
the current evolution of DSA is a waypoint to a much more
significant change in communications practice: complete
delegation of both interference avoidance and mitigation
to the devices that populate the spectrum.
VIII. CONCLUSION
Cognitive radio research has now advanced to the point
where a number of techniques and technologies have
shown desirable analytic and experimental results. Some
of the technology has advanced to the point of reporting
meaningful field test results that strongly argue that
cognitive radio can begin to fulfill its promise to
revolutionize wireless communications.
There would appear to be sufficient technology andexperimental platform development under way to enable
early deployments of cognitive radio platforms, probably
focusing on DSA as the initial argument for deployment.
However, to complete this evolution, the research
community needs to focus on establishing unified abstrac-
tions, research taxonomies, and tool sets that can span the
scope of individual research projects. This will enable
results to be aggregated and create the ability to integrateand synthesize products that draw from technology
developed by independent research thrusts.
Additionally, there is a transition from envisioning
cognitive radio as an adjunct to current wireless architec-
tures and services to new environments, opportunities, and
capabilities that can be created uniquely through cognitive
radio. The opportunity for cognitive radio may lie not in
displacing existing wireless network infrastructure but increating wireless information networks that provide new
services and capabilities that current wireless service
architectures are incapable of providing. h
REF ERENCE S
[1] Functional Requirements for the802.22 WRAN Standard, IEEE 802.22Working Group on Wireless Regional AreaNetworks. [Online]. Available: http:www.ieee802.org/22.
[2] FCC, Notice for proposed rulemaking(NPRM 03 322): Facilitating opportunities forflexible, efficient and reliable spectrum useemploying spectrum agile radios technologies,ET Docket 03-108, Dec. 2003.
[3] J. Park, T. Song, J. Hur, S. M. Lee, J. Choi,K. Kim, J. Lee, K. Lim, C. H. Lee, H. Kim, andJ. Lasker, BA fully integrated UHF receiver
with multi-resolution spectrum sensing(MRSS) functionality for IEEE 802.22cognitive radio applications,[ in Proc. 2008IEEE Int. Solid State Circuits Conf., Feb. 2008.
[4] Y. Hur, J. Park, W. Woo, J. S. Lee, K. Lim,C. H. Lee, H. S. Kim, and J. Laskar,BA cognitive radio system employing adual-stage spectrum sensing technique:A multi-resolution spectrum sensingand a temporal signature detectiontechnique,[ in Proc. Global Telecommun.Conf., Nov. 2006, pp. 1–5.
[5] W. Hu, D. Willkomm, M. Abusubaih, J. Gross,G. Vlantis, M. Gerla, and A. Wolisz,BDynamic frequency hopping for efficient
IEEE 802.22 operation,[ IEEE Commun. Mag.,May 2007.
[6] V. Petty et al., BFeasibility of dynamicspectrum access in underutilized televisionbands,[ in Proc. 2nd IEEE Int. Symp.New Frontiers Dyn. Spectrum Access Netw.,Apr. 17–20, 2007, pp. 331–339.
[7] R. Chen, J. Park, Y. Hou, and J. Reed,BToward secure distributed spectrum sensingin cognitive radio networks,[ IEEE Commun.Mag., pp. 50–55, Apr. 2008.
[8] T. Rondeau, BApplication of artificialintelligence to wireless communications,[Ph.D. dissertation, Virginia Polytechnic Inst.and State Univ., Blacksburg, Sep. 20, 2007.
Fig. 9. Transition from noninterfering to interference-tolerant
operation.
Marshall: Extending the Reach of Cognitive Radio
624 Proceedings of the IEEE | Vol. 97, No. 4, April 2009
[9] A. Mody, S. Blatt, N. Thammakhoune,T. McElwain, J. Niedzwiecki, D. Mills,M. Sherman, and C. Myers, BMachinelearning based cognitive communications inwhite as well as the gray space,[ in Proc. IEEEMilitary Commun. Conf. (MILCOM 2007),Oct. 2007, vol. 26, no. 1, pp. 2321–2327.
[10] S. Yarkin and H. Arslan, BExploiting locationawareness toward improved wireless systemsdesign in cognitive radio,[ IEEE Commun.Mag., vol. 46, pp. 128–135, Jan. 2008.
[11] K. Nolan, P. Sutton, and L. Doyle,BAn encapsulation for reasoning, learning,knowledge representation, andreconfiguration cognitive radio elements,[ inProc. 1st Int. Conf. Cognitive Radio OrientedWireless Netw. Commun., Jun. 8–10, 2006,pp. 1–5.
[12] IEEE Standards Coordinating Committee 41(Dynamic Spectrum Access Networks).[Online]. Available: http://www.scc41.org/.
[13] G. Minden et al., BCognitive radios fordynamic spectrum access,[ IEEE Commun.Mag., vol. 45, pp. 113–121, May 2007.
[14] DARPA, BWireless network after next,Proposer information pamphlet.[ [Online].Available: http://www.darpa.mil/sto/solicitations/WNaN/ pip.htm.
[15] J. Mitola, BCognitive radio: An integratedagent architecture for software definedradio,[ Ph.D. dissertation, Royal Instituteof Technology, Stockholm, Sweden,May 8, 2000.
[16] D. Scaperoth, B. Le, T. Rondeau,D. Maldonado, C. Bostian, and S. Harrison,BCognitive radio platform development forinteroperability,[ in Proc. IEEE MilitaryCommun. Conf. (MILCOM 2006), Oct. 2006,vol. 25, no. 1, pp. 2640–2645.
[17] P. Marshall, BXG program field test results,[in Proc. Defense Spectrum Summit 2007,Nov. 2007.
[18] FCC, BFacilitating opportunities for flexible,efficient, and reliable spectrum use employingcognitive radio technologies,[ ET Docket03-108, Dec. 17, 2003.
[19] D. Elenius, G. Denker, M. Stehr,R. Senanayake, C. Talcott, and D. Wilkins,BCoRaLVPolicy language and reasoningtechniques for spectrum policies,[ in
Proc. 2007 IEEE Workshop Policies Distrib. Syst.Netw., Jun. 2007.
[20] F. Perich, BPolicy-based network managementfor next generation spectrum access control,[in Proc. IEEE Int. Symp. New Frontiers Dyn.Spectrum Access Netw., Apr. 2007.
[21] P. Marshall, BXG communications program:Information briefing,[ in Proc. Semantic WebApplicat. Nat. Security (SWANS) Conf.,Apr. 2005.
[22] M. McHenry, E. Livsics, T. Nguyen, andN. Majumdar, BXG dynamic spectrum accessfield test results,[ IEEE Commun. Mag.,vol. 45, pp. 51–57, Jun. 2007.
[23] P. Marshall, BDynamic spectrummanagement of front end linearity anddynamic range,[ in Proc. 3rd IEEE Int. Conf.New Frontiers Dyn. Spectrum Access Netw.,Chicago, IL, 2008.
[24] V. Cerf et al., BDelay-tolerant networkarchitecture,[ IETF RFC 4838, Apr. 2007.
[25] P. Gupta and P. R. Kumar, BThe capacity ofwireless networks,[ IEEE Trans. Inf. Theory,vol. 46, pp. 388–404, Mar. 2000.
[26] J. Neel, BAnalysis and design of cognitiveradio networks and distributed resourcemanagement algorithms,[ Dissertation,Virginia Polytechnic Inst. and State Univ,Blacksburg, 2006.
[27] L. Luna, BNextel interference debaterages on,[ Mobile Radio Technol.,Aug. 1, 2003.
[28] F. Weidling, D. Datla, V. Petty, P. Krishnan,and G. Minden, BA framework for R.F.spectrum measurements and analysis,[ inProc. 1st IEEE Int. Symp. New Frontiers Dyn.Spectrum Access Netw., Nov. 8–11, 2005,pp. 573–576.
[29] G. Chan and B. Brown, BA measure ofspectrum utilization in land mobile bands,[ inProc. Can. Conf. Electr. Comput. Eng. 1996,May 1996, vol. 1.
[30] P. Marshall, BClosed-form analysis ofspectrum characteristics for cognitive radioperformance analysis,[ in Proc. 3rd IEEE Int.Conf. New Frontiers Dyn. Spectrum AccessNetw., Chicago, IL, 2008.
ABOUT TH E AUTHOR
Preston F. Marshall (Member, IEEE) has an almost 30 year
background in communications systems and software in both
research and commercial environments. Currently he is with the
Defense Advanced Research Projects Agency (DARPA) Strategic
Technology Office (STO), and serves a Program Manager for many
of the DARPA Wireless and Cognitive Radio and Network
programs. These programs include development of dynamic
spectrum access technology, Wireless Networking after Next
(WNAN) program for low cost wireless networking and the
Disruption and Delay Tolerant Networking Program (DTN). These
programs collectively provide the technology base for high performance and affordable
infrastructureless wireless networking. Mr. Marshall has been Technical Program Chair
for all of the IEEE DYSPAN conferences. Mr. Marshall holds a B.S.E.E and M.S. Information
Science from Lehigh University, and is a Graduate student at Trinity College, Dublin.
Marshall: Extending the Reach of Cognitive Radio
Vol. 97, No. 4, April 2009 | Proceedings of the IEEE 625
top related