SkyTel provides a discussion herein as to its concept to use High Accuracy Location ("HALO") of the radios in a Cognitive Radio

Network, to
substantially improve and extend the Available Resoure Maps (AMRs) that are the focus of this paper on moving towards Cogntive Radio
Systems or Networks (as opposed to simply Cognitive Radios).
There is no question that Cogntive Radio Newtorks are the future of wireless and are especially needed, and cost justified even now, for:
Smart Transport, Energy, Environment Radio (STEER) and complimentary wireless (e.g., precision agriculture and resouce manangemnet use
and protection) for core infrastructure and wide-georgraphy industries for the US and all nations.

Distributed Spectrum Sensing for Cognitive Radio

Systems
Claudio R. C. M. da Silva, Brian Choi, and Kyouwoong Kim
Wireless @ Virginia Tech
Bradley Department of Electrical and Computer Engineering
Virginia Polytechnic Institute and State University Blacksburg, VA USA 24061
email: {cdasilva, choib, kyouwoong}@vt.edu
Also, radios in a mesh net can receive and send
more effectively, when their receivers-antennas are
synchronized and coordinated, using mesh nets.

Abstract Cognitive radio is a candidate technology for more

efficient spectrum utilization systems based on opportunistic
spectrum sharing. Because this new technology does not rely
on traditional license-based spectrum allocation policies, it could
disrupt existing systems if the spectrum utilization decision
is based on unreliable spectral estimation. Distributed sensing
methods have the potential to increase the spectral estimation
reliability and decrease the probability of interference of cognitive
radios to existing radio systems. In this paper, we consider
different aspects of the processing and fusion of spectrum sensing
information of cognitive radio systems. The use of cyclic featurebased methods for distributed signal detection and classification
is discussed and recent results are presented.

spectrum sensing. In such a distributed approach, the spectrum occupancy is determined by the joint work of cognitive
radios, as opposed to being determined individually by each
cognitive radio. In this paper, we consider different aspects
of the processing and fusion of spectrum sensing information
of cognitive radio systems. A new system architecture that
combines cognitive radios and available resource maps is
also discussed. A major focus of this paper is on the use of
cyclic feature-based methods for distributed signal detection
and classification.

I. I NTRODUCTION

In application scenarios involving geographically distributed

radios, such as a wireless communication system, distributed
spectrum sensing approaches are worth considering due to
the variability of the radio signal, as suggested in [5]-[7].
Such methods may significantly increase the reliability of the
spectrum estimation process, at the expense of computational
complexity and power/bandwidth usage for the transmission
of spectrum sensing information.
In this paper, we model the cognitive radio system with a
standard parallel fusion network commonly used in decentralized detection problems, shown in Fig. 1. In this model, each
cognitive radio (CR node) obtains some relevant information
yi , i = 1, . . . , N , on the spectrum occupancy. Each CR
node processes this information and then sends a summary
of its own observations to a fusion center, in the form of a
message ui , i = 1, . . . , N , taking values in a finite alphabet.
The fusion center then generates a global spectrum usage
decision u0 based on the messages it has received [8]. The
objective in this Bayesian hypothesis testing problem is to
obtain the set of decision rules that minimize the average
cost of making a decision of the overall system. Taking a
person-by-person optimization methodology [9], and assuming
that the observations at the local detectors are conditionally
independent and that the local decisions are binary, the local
decision rules reduce to threshold tests given by

Aiming at more efficient spectrum utilization, the FCC is

currently revisiting traditional licensed-based policies and
moving toward the adoption of spectrum sharing strategies
such as ultra-wideband (UWB) and cognitive radio. While
UWB systems achieve a more efficient spectrum utilization by
overlaying existing narrowband systems, cognitive radios opportunistically find and use empty frequency bands. Cognitive
radios rely on the fact that a significant portion of the spectrum
allocated to licensed services show little usage over time. A
recent spectrum occupancy measurement project shows that
the average spectrum occupancy taken over multiple locations
is 5.2%, with a maximum occupancy of 13.1% [1].
Originally introduced by Mitola [2]-[3], cognitive radios
are capable of sensing their environment, learning about
their radio resources and user/application requirements, and
adapting behavior by optimizing their own performance in
response to user requests [4]. Cognitive radios are therefore
a powerful tool for solving the spectrum usage problem.
Such radios are capable of sensing spectrum occupancy, and,
in conformity with the rules of the FCC, opportunistically
adapting transmission to utilize empty frequency bands without disrupting other systems. However, this departure from
traditional license-based spectrum allocation policies could
disrupt existing systems if the spectrum utilization decision
is based on unreliable spectral estimation.
One possible approach to increase the spectral estimation
reliability and decrease the probability of interference of cognitive radios to existing radio systems is by using distributed
This work was supported in part by a gift from Texas Instruments Inc.

II. D ISTRIBUTED S PECTRUM S ENSING : OVERVIEW

p (yk |H1 ) uk =1 P (H0 )

k ,
p (yk |H0 ) uk =0 P (H1 )

(1)

where
P

uk

k = P

uk

u0 = 1|uk1 P
`

u0 = 1|uk1 P

`
QN
u0 = 1|uk0
i=1,i!=k P (ui |H0 )
,
`
QN
u0 = 1|uk0
i=1,i!=k P (ui |H1 )

needed in
mesh net
anyway,
especially
mobile ad
hoc mesh
net

CR fusion center

0.9
0.8
0.7
0.6
1, 3, 10 sensors

0.5
0.4
0.3
0.2

u
)0
Fig. 1.

Probability of Detection

#
Phenomenon
"
%%$
!! #
!
%
y1 !
y2 #
yN%
!
%%
!
#
"
!
$
#
&
%
...
CR node 1
CR node 2
CR node N
%
!
%
'
u1%
u2
uN !!
%
'
!
%% '
!
(
& '
%
"!
!

Block diagram of a parallel fusion network.

T

uk
=
(u1 , . . . , uk1 , uk+1 , . . . , uN )
and ukj
=
T
(u1 , . . . , uk1 , uk = j, uk+1 , . . . , uN ) , j = 0, 1. The
fusion rule is also a threshold test, and is given by
"
#
"
##
N "
!
1 P Mi
P Mi
ui log
+ (1 ui ) log
PFi
1 PFi
i=1
"
#
u0 =1
P (H0 )
log
. (2)
P (H1 )
u0 =0
Thus, the person-by-person optimization solution to the binary
decentralized Bayesian hypothesis testing problem is given by
a system of nonlinear coupled equations. It is well-known
that the computational effort required to solve a system of
nonlinear coupled equations increases rapidly with the number of detectors. Tsitsiklis and Athans show in their classic
paper [10] that even the simplest problems of decentralized
decision making are hard from an algorithmic viewpoint, and
that it becomes an NP-complete problem if the measurements
at each sensor are not independent.
As the testing functions at both the local detectors and
at the data fusion center have the form of a likelihood
ratio, the decision thresholds are the only free parameters.
Therefore, the distributed detection problem reduces to the
search of the optimal threshold. One possible way to find
these optimal thresholds is by using iterative computational
algorithms. For example, a nonlinear Gauss-Seidel iterative
algorithm derived in [11] allows for the solution of reasonably
large-sized problems, at the expense of requiring messages
to be transmitted among fusion center and CR nodes. Other
possible iterative algorithms can be found in [9].
III. D ISTRIBUTED S IGNAL D ETECTION U SING
S INGLE -C YCLE D ETECTORS
Cognitive radios must be able to detect spectrum usage with
no a priori knowledge of modulation format and characteristics, such as the bandwidth, carrier frequency, and chip-rate,
of primary systems. The most conventional approach for the
detection of an unknown deterministic signal in AWGN is
the radiometer, which is simply a measure of received energy
in time and frequency. However, it is well-known that such
a method is highly susceptible to unknown and changing
noise levels and interference [12]. Cyclic-feature detection

0.1
0

0.2

0.4
0.6
Probability of False Alarm

0.8

Fig. 2. Distributed detection of a known signal in AWGN using single-cycle

detectors (Eb /No = 3 dB).

techniques are an alternative approach for the unknown signal

detection problem that have many advantages, including signal
classification capabilities and reduced sensitivity to unknown
and changing background noise. Such techniques exploit timing or phase properties of digitally modulated signals, and have
been receiving a great deal of attention by the IEEE 802.22
work group [13].
Assume, for example, a basic cyclic-feature detector known
as a single-cycle detector. The test statistic of such a detector
is given by [14]
$%
$
$
$ H1

$
SxT (t, f ) SS (f ) df $$ ,
(3)
$
H0

where SxT (t, f ) is the cyclic periodogram of the received

(cycloergodic) signal x (t), given by
'
&% T
t+ 2
1
u
j2 (f +

2 ) du
x (u) e
SxT (t, f ) =
T
t T2
&% T
'
t+ 2
j2 (f
u

)
2

x (u) e
du ,(4)
t T2

and SS (f ) is the ideal spectral correlation function.

The performance of a single-cycle detector in a distributed
architecture, using the Gauss-Seidel algorithm for data fusion,
is shown in Fig. 2. It is seen that when signal detection is
performed using 10 sensors instead of using a single sensor,
the probability of detection increases from approximately 30%
to 60%, for a probability of false alarm equal to 10%.
IV. S IGNAL C LASSIFICATION U SING C YCLOSTATIONARITY
In addition to frequency occupancy estimation, cognitive
radio systems may also need to classify the primary system
that occupies a given frequency band. For example, the
protection (in terms of allowable interference level) that a
cognitive radio system may provide to a primary user may
be dependent on the primary system. By taking advantage of

BPSK Cycle Freq. Profile with SNR=0 dB Obs. Length=100

1

Max. Ampltitude of Spectral Coherence

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Cycle frequency !/Fs

Fig. 3.

Fig. 5.

Spectral coherence function of a BPSK signal.

Cycle frequency domain profile of a BPSK signal.

FSK Cycle Freq. Profile with SNR=0 dB Obs. Length=100

Max. Ampltitude of Spectral Coherence

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Cycle frequency !/F

Fig. 4.

Fig. 6.

Spectral coherence function of a BFSK signal.

the inherent cyclostationarity existent in digital signals, cyclicfeature algorithms have the potential to provide reliable signal
classification even at low signal-to-noise ratio scenarios.
Cyclic-feature algorithms for signal classification typically
use the spectral coherence function of the received signal,
defined as [15]

Cx (f ) =

[Sx0 (f

Sx (f )
.
+ /2)Sx0 (f /2)]1/2

lim SxT (t, f )f ,

f 0 T

(6)

where

SxT (t, f )f =

1
f

f +f /2

f f /2

SxT (t, v)dv,

signal classification. It can be shown that most digital signals

and some analog signals have distinctive spectral coherence
functions.
It is shown in [16] that in order to ease the design of the
signal classification algorithm, it is convenient to define the
following cycle frequency domain profile function

(5)

This function is of particular interest as it gives a normalized

measure of the cross-correlation between signal components
at frequencies f /2 and f + /2. The magnitude of the
spectral coherence function ranges from 0 to 1, and is invariant
to linear transformations to the incoming signal.
In (5), the spectral correlation function Sx (f ) is defined as
Sx (f ) = lim

Cycle frequency domain profile of a BFSK signal.

(7)

and the cyclic periodogram SxT (t, f ) is given by (4).

The spectral coherence functions of BPSK and BFSK signals are shown in Figs. 3 and 4, respectively. It is seen that the
spectral coherence function corresponding to these modulation
techniques have distinct features that ultimately allow for

I() = max |Cx (f )| .

f

(8)

The profile functions of BPSK and BFSK signals are shown

in Figs. 5 and 6, respectively. As described in [16], an efficient
signal classification algorithm is obtained by matching the
profile function I() of the received signal with a database
of profile functions of possible digital and analog modulation
schemes. The probability of correct classification for such an
algorithm is shown in Fig. 7, assuming a low signal-to-noise
ratio environment and a collection of five modulation schemes.
It is observed that except for the QPSK modulation (QPSK
has only cyclic feature for the symbol rate, and the strength
of cyclic feature accounting for the symbol rate is less than
the one due to carrier frequency), all other schemes have very
good probability of correct classification [16].
V. C OGNITIVE R ADIO S YSTEMS BASED ON AVAILABLE
R ESOURCE M APS
An unlicensed wireless WAN based on the combination
of cognitive radio and available resource maps (ARM) was

SkyTel believes this sort of AMR-agumented Cognitive Radio (CR) network idea can be much fruther ehanced by the CR radios and network employing near-constant
sub-foot-level High Accuracy Location (HALO) in the participating CR radios. HALO determination can be used, with time-matched, near-constant spectrum-use
sensing of each radio in all times of day and season. These two correlated measurements can be used to build a map of the wireless network geographic region and
how, on a cubic-foot to cubic-foot basis, the radio signals propagate and interacte (including co-channel [CC] and adjacent-channel [AC] RF multipath. With known
fixed-station tranceivers, and the mobile received moving in various patterns, this will of course result in a massive database, but those are not cheap to store, build on,
and utilze. The resolution / accuracy would improve over time, the more the above paried measurements take place over the whole Network region. (Continued below.)

recently proposed in [17], [18]. ARM-based cognitive radio

systems are based on the same operational principles of
conventional cellular networks, but with the following fundamental peculiarities:
Spectrum is shared and a database (ARM) provides
spectrum availability,
A public radio control channel (RCC) is used for session
setup, and
Base transceiver stations (BTSs) report their spectrum
usage to the ARM through a wired control channel.
In this architecture, the system infrastructure provides the
framework for unlicensed spectrum access and spectrum bartering. The ARM is a real-time map of all spectrum usage
updated and maintained by user equipment (UE) and BTSs.
Local frequency allocation is managed exclusively by an
ARM-based base station. ARM-based cognitive radio networks may operate as a shared resource (secondary user)
system in spectrum licensed to a primary system. In this case,
primary systems may be connected to the ARM to allow for
spectrum sharing and coordination.
In the ARM-based architecture, a RCC is used to coordinate
the spectrum access of all the UE in the system. Before any
radio transmission takes place, each UE and BTS announces its
intentions over the RCC. All UE and BTSs monitor the RCC
to ensure that all intended new communication links will not
interfere with them. In the case that a new session will create
harmful interference, objections are raised over the RCC to
signal that the new session may not use the resources it intends
to [17], [18].
It should be noted that, as opposed to conventional cognitive
radio systems, this new architecture has a central entity responsible for spectrum allocation and management. Therefore,
we believe that such an approach could, at least partially,
address existing concerns on the possible interference between
cognitive (secondary) and primary systems (compared, for
example, to ad-hoc cognitive radio networks). As an additional
interference protection layer, we envision that spectral sensing
can be incorporated into the original concept of ARM-based
systems as a method that would allow for (1) validation of the
data transmitted through the RCC, and (2) gathering of spectrum usage information from systems that are not connected
to the ARM (for example, other unlicensed systems). In this
configuration, the ARM could serve as the entity responsible
for spectrum sensing data fusion.
VI. C ONCLUSIONS
Cognitive radio is a new and exciting technology that,
among other applications, has the potential to unlock the
spectrum necessary for the deployment of next generation high
data rate systems. However, for this concept to become a practical technology, research into the processing, transmission,
and fusion of spectrum sensing information is still necessary.
In this paper, different aspects of a distributed approach to
spectrum sensing were addressed and preliminary results were
presented. The use of cyclic feature-based methods for signal
detection and classification was discussed, and it was shown

Fig. 7. Probability of correct classification using a cyclic feature algorithm

described in [16].

that such methods provide reliable detection/classification even

at low signal-to-noise ratio scenarios.
R EFERENCES
[1] M. A. McHenry, NSF Spectrum Occupancy Measurements Project
Summary, Shared Spectrum Company Report, Aug. 2005.
[2] J. Mitola III and G. Q. Maguire Jr., Cognitive radio: Making software
radios more personal, IEEE Pers. Commun., vol. 6, pp. 13-18, Aug.
1999.
[3] J. Mitola III, Cognitive radio: An integrated agent architecture for
software defined radio, Doctor of Technology dissertation, Royal Inst.
Technol. (KTH), Stockholm, Sweden, 2000.
[4] D. Maldonado, B. Le, A. Hugine, T. W. Rondeau, and C. W. Bostian,
Cognitive radio applications to dynamic spectrum allocation: A discussion and an illustrative example, in Proc. Dynamic Spectrum Access
Nets., Baltimore, MD, 2005, pp. 597-600.
[5] D. Cabric, S. M. Mishra, and R. W. Brodersen, Implementation issues in
spectrum sensing for cognitive radio, in Proc. Asilomar Conf. on Signals,
Syst., and Comput., Pacific Grove, CA, 2004, pp. 772-776.
[6] S. Shankar N., C. Cordeiro, and K. Challapali, Spectrum agile radios:
Utilization and sensing architectures, in Proc. Dynamic Spectrum Access
Nets., Baltimore, MD, 2005, pp. 160-169.
[7] G. Ganesan and Y. Li, Cooperative spectrum sensing in cognitive radio
networks, in Proc. Dynamic Spectrum Access Nets., Baltimore, MD,
2005, pp. 137-143.
[8] J. N. Tsitsiklis, Decentralized detection, in Advances in Statistical
Signal Processing, Signal Detection, vol. 2., Greenwich, CT: JAI Press,
pp. 297-344, 1993.
[9] P. K. Varshney, Distributed Detection and Data Fusion, New York, NY:
Springer-Verlag, 1997.
[10] J. N. Tsitsiklis and M. Athans, On the complexity of decentralized
decision making and detection problems, IEEE Trans. Autom. Control,
vol. 30, pp. 440-446, May 1985.
[11] Z.-B. Tang, K. R. Pattipati, and D. L. Kleinman, An algorithm for
determining the decision threshold in a distributed detection problem,
IEEE Trans. Syst. Sci. Cybern., vol. 21, pp. 231-237, Jan./Feb. 1991.
[12] G. K. Yeung and W. A. Gardner, Search-efficient methods of detection
of cyclostationary signals, IEEE Trans. Signal Process., vol. 44, pp.
1214-1223, May 1996.
[13] M. Bellec, Y. C. Cheong, C. Cordeiro, C.-J. Kim, H.-S. Kim, J. Laskar,
A PHY/MAC proposal for IEEE 802.22 WRAN Systems. Part 1: The
PHY, doc. IEEE802.22-06/0004, Jan. 2006.
[14] W. A. Gardner and C. M. Spooner, Signal interception: Performance
advantages of cyclic-feature detectors, IEEE Trans. Commun., vol. 40,
pp. 149-159, Aug. 1992.
[15] W. A. Gardner, Measurement of Spectral Correlation, IEEE Trans.
Acoust., Speech, Signal Processing, vol. 34, pp. 1111-1123, Oct. 1986.
[16] K. Kim, I. A. Akbar, K. K. Bae, J.-S. Um, C. M. Spooner, and J. H.
Reed, Cyclostationary approaches to signal detection and classification
in cognitive radio, submitted for presentation at IEEE DySpan 2007.
[17] A. Batra, W. Krenik, and C. Panasik, Cognitive radios for unlicensed
WANs, presented at the BWRC Cognitive Radio Workshop, 2004.
[18] W. Krenik and A. Batra, Cognitive radio techniques for wide area
networks, in Proc. Conf. Design Automation, Anaheim, CA, 2005, pp.
409-412.

(Continued from above.) If (as SkyTel Plans for its networks) the Network knew all authorized users, and they were all using this form of CR on the
Network, then, after said Resolution/ Accuracy was sufficiently developed by the increasingly knowledgable Network, it would be able to instruct each
radio at each place and time what spectrum it may use without interference to and from other users-- based upon past experiences. This should
subtantially (1) shorten the time of these determinations, and (2) decrease the resources (power and distributed-mesh radio reporting as described in
this paper) needed by each mobile radio, for the determinations, (3) provide an alternative, historical source of this detemination to the real-time
spectrum-use sensing of the radio, etc. (4) It would be able to track all radios precisely and perform a "traffic cop" role as the radios move around the
Network, and their authorized services and priorities change. Thus, HALO may play a vital role in "Cogintive Radio Networks."