See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/252063675

Advanced Sensing Techniques of Energy Detection in Cognitive Radios

(Invited Paper)

Article in Journal of Communications and Networks · February 2010

DOI: 10.1109/JCN.2010.5710555

CITATIONS READS

66 1,621

5 authors, including:

Sungtae Kim
Ajou University
93 PUBLICATIONS 1,429 CITATIONS

SEE PROFILE

All content following this page was uploaded by Dongkyu Kim on 18 September 2014.

The user has requested enhancement of the downloaded file.

JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 12, NO. 1, FEBRUARY 2010 19

Advanced Sensing Techniques of Energy Detection in

Cognitive Radios
Hano Wang, Gosan Noh, Dongkyu Kim, Sungtae Kim, and Daesik Hong

(Invited Paper)

Abstract: Recently, spectrum sensing has been intensively studied tion can operate with no prior information about primary sig-
as a key technology in realizing the cognitive radio. There have nals. Accordingly, only the energy detection technique corre-
been advances in the performance of spectrum sensing through sponds to the general purpose of spectrum sensing for hetero-
both multi-antenna and cooperative sensing schemes. In this pa- geneous wireless communication systems. That is why energy
per, the performances and complicated scenarios of the latest spec- detection is the most intensively investigated sensing technique
trum sensing schemes are analytically compared and arranged into and is also the focus of this paper.
a technical tree while considering practical concerns. This paper
In general, for the purpose of protecting primary users from
will give a macroscopic view of spectrum sensing and will also pro-
vide insight into future spectrum sensing works.
the interference caused by secondary communication, cogni-
tive radios are operated in a geographical far distant from
Index Terms: Cognitive radio, cooperative sensing, multi-antenna the primary system. Hence, the primary signal is received
sensing, spectrum sensing. by the secondary sensing node in a low signal-to-noise-ratio
(SNR) region below zero decibel where energy detection is very
poor [14], [18]. A failure in spectrum sensing means a missed
opportunity for secondary users to utilize the white space of the
I. INTRODUCTION
spectrum or harmful interference to the primary users. There-
Research has been performed on efficient spectrum usage fore, sensing performance enhancement should be required for
since it was reported that considerable licensed spectra exclu- both increasing the throughput of the secondary users and also
sively allocated to conventional wireless communication sys- for protecting the primary users from unintended interference.
tems have been under-utilized [1]. For efficient spectrum utiliza- The sensing performance enhancement of energy detection
tion, the cognitive radio will mostly likely be the most promising can be achieved by using multi-antennas at the sensing node
technology due to its inherent spectrum sensing capability and [14]–[17] or by cooperation between sensing nodes [18], [19].
frequency-agile radio functions [2]. Spectrum sensing has the Advances in multi-antenna and cooperative sensing are re-
especially important missions of finding the white space of li- viewed and in this paper. The aim of this study was to provide
censed spectra and protecting the primary licensed users from a macroscopic view of spectrum sensing, especially with energy
interference caused by cognitive radio communications. Ac- detection, in the cognitive radio. In order to do so, a general-
cordingly, spectrum sensing has been widely researched as a key ized sensing performance evaluation is given first, which allows
technology for allowing cognitive radio communication within for greater understanding of the multi-antenna and the coop-
the real world. erative sensing performances. For multi-antenna sensing tech-
Spectrum sensing can be performed by various detection tech- niques, the performances were analyzed and compared in con-
niques using a matched-filter [3], [4], a statistical feature of sideration to practical problems. For cooperative sensing, com-
the primary signal [5], [6], called feature detection, and a sim- plicated scenarios and practical considerations were arranged
ple energy measurement [7]–[34]. Although the first two de- into a technical tree in order to describe the technique in gen-
tection techniques outperform the energy detection technique, eral. The relationship between branches is also discussed in de-
they require prior information about the primary signals, and tail. Finally, a summary describes the overall structure of the
have a primary system-dependent performance. Heterogeneous research performed on the performance enhancement of energy
wireless communication systems licensed to different primary detection. Also, technical challenges for spectrum sensing are
spectra may overlap within a geographical region. In such cir- discussed for future consideration.
cumstances, matched-filter detection or feature detection are too The rest of this paper is organized as follows: Section II de-
costly for sensing multiple primary spectra, while energy detec- fines common terminologies and introduces scenarios within
general spectrum sharing; Section III provides a general perfor-
Manuscript received October 28, 2009. mance evaluation of energy detection, Section IV analyzes and
This research was supported by the MKE (Ministry of Knowledge Economy),
Korea, under the ITRC(Information Technology Research Center) support pro- compares the first methodology, or multi-antenna sensing tech-
gram supervised by the NIPA (National IT Industry Promotion Agency) (NIPA- niques, that is used for enhancing the energy detection’s sens-
2009-(C1090-0902-0005)) and by Korea Science and Engineering Foundation ing performance, Section V provides a technical tree expressing
through the NRL Program (Grant R0A-2007-000-20043-0).
The authors are with the Information and Telecommunication Lab., Depart-
various cooperative sensing techniques, and, lastly, Section VI
ment of Electrical and Electronic Engineering, Yonsei University 134 Shinchon- summarizes the study.
dong Seodaemungu, Seoul, Korea, 120-749, email: {hano97, gsnoh, dongkyu,
sungtae, daesikh}@yonsei.ac.kr.
1229-2370/10/$10.00 
c 2010 KICS
20 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 12, NO. 1, FEBRUARY 2010

II. PRELIMINARY OF SPECTRUM SHARING defined, hierarchical spectrum sharing is divided into two cate-
gories: Underlay and overlay spectrum sharing.
Before we investigate the advances in spectrum sensing, spec-
trum sharing environments are categorized into several groups, B.1 Underlay Spectrum Sharing
and the terminology used for each scenario is summarized. The
aim of this section is to give a general overview of spectrum In underlay spectrum sharing, if the interference caused by
sharing environments. the secondary communication is received by the primary re-
ceiver under a predetermined threshold, the interference is
A. Open Spectrum Sharing treated as harmless. Therefore, while the harmless interference
condition is maintained, a secondary transmitter is permitted to
The spectrum sharing environment is classified first due to transmit its signal even if the primary link is communicating.
the existence of access priorities among heterogeneous systems This category is interestingly termed ‘spectrum sharing.’
within a spectrum. If every system has the same priority for ac- In order to satisfy the interference constraint condition, the
cessing the spectrum resource, it is referred to open spectrum secondary transmitter must possess information about the inter-
sharing (OSS) [35]–[37]. ference channel gain between the secondary transmitter and the
In OSS, heterogeneous systems with different channel band- primary receiver [40]. Hence, channel estimation using a known
width sizes co-exist in a common spectrum without any cen- signal and a feedback process between a primary receiver and a
tralized coordinations. Hence, a distributed coordination used secondary transmitter should be required with an extremely high
in managing the interference and fair access opportunities is re- accuracy for the interference channel measurement1 . In order to
quired and is called spectrum access etiquette in OSS [38]. realize this, the secondary user should equip a dual-mode sys-
For example, if a system with a large bandwidth channel too tem as follows: one is for the secondary communication and the
frequently accesses the open spectrum or occupies it for a long other for the interference measurement and feedback between
time, it is difficult for other systems to get an opportunity to the secondary transmitter and the primary receiver.
communicate in the spectrum. For fair spectrum sharing, the Although indirect interference channel measurement schemes
traffic arrival rates of systems with different channel bandwidths are presented in [41] and [42], they cannot provide an accu-
should be differentiated [35], [36]. For the same purpose, the rate interference channel measurement for fading environments.
spectrum sensing threshold control is proposed in [37]. In OSS, Therefore, developing effective schemes for interference chan-
a system should check the spectrum availability through spec- nel measurement and feedback may be a bottle neck for the prac-
trum sensing before it start to transmit a signal. In this literature, tical implementation of the underlay spectrum sharing scenario.
the sensing threshold value to detect the vacancy of a frequency
channel is set at higher value for a system with a wider channel B.2 Overlay Spectrum Sharing
bandwidth. Therefore, the access opportunities between systems Different from the underlay spectrum sharing scenario, the
with different channel bandwidths are balanced. secondary transmitter may send a signal to a secondary receiver
Open spectrum sharing scenarios have been developed pri- only for a time period called the idle period, when the primary
marily for the industrial, scientific and medical (ISM) radio communication is inactive. In order to detect the idle period of a
bands [35]–[37]. Recently, the OSS-operated radio frequency primary spectrum, spectrum sensing is the most important func-
has been extended to include licensed spectra for the purpose tionality in realizing overlay spectrum sharing. According to the
of utilizing multi-mode terminals and inter-operability between amount of required prior information about the primary systems,
legacy wireless communication systems [39]. In such circum- spectrum sensing techniques are divided into three types, as de-
stances, frequency resources belonging to various systems com- picted in Fig. 2.
pose a spectrum pool. An end-user terminal can dynamically Matched filter detection: It is widely known that the detec-
access the spectrum pool with different radio resource units de- tor using a matched filter is able to achieve the optimum perfor-
pending on its application QoS and traffic condition of each sys- mance when a secondary sensing node can perform a coherent
tem. Accordingly, the term ‘dynamic spectrum access’ is used detection of the primary signal [3], [4]. However, in order to use
for this category of spectrum sharing. the matched filter within spectrum sensing, the secondary sens-
ing node must be synchronized to the primary system and must
B. Hierarchical Spectrum Sharing even be able to demodulate the primary signal. Accordingly, the
secondary sensing node has to have prior information about the
The most differentiated feature of hierarchical spectrum shar- primary system such as the preamble signaling for synchroniza-
ing from OSS is that there is an access priority between the pri- tion, pilot patterns for channel estimation, and even modulation
mary and secondary systems. What we call the primary system orders of the transmitted signal, et cetera.
is a legacy system operating in a licensed spectrum and its end- Nowadays, heterogeneous wireless communication systems
user terminals. Although the licensed spectrum is exclusively licensed to different primary spectra may overlay one another in
allocated to a primary system, secondary systems are allowed to a geographical region. In such circumstances, idle periods may
use the spectrum because of the considerable amount of unused
licensed spectra within time and space [1]. 1 Under-estimated channel gain value increases the secondary transmission

In order to share the primary spectrum, a secondary system power, resulting in harmful interference to a primary receiver. An over-estimated
channel gain value decreases the secondary transmitter power, which satisfies
should not impart any harmful interference upon the primary the interference constraint condition, but decreases the secondary link through-
communication. According to how the harmful interference is put.
WANG et al.: ADVANCED SENSING TECHNIQUES OF ENERGY DETECTION... 21

Fig. 2. Sensing and transmission structure for energy detector.

tion between secondary sensing nodes have been developed [7]–

[34], and it is challenging to provide a macroscopic view of the
conventional works.
Fig. 1. Three types of spectrum sensing techniques: Only energy detec-
tion requires no prior information about the primary signal.
III. GENERALIZED PERFORMANCE OF ENERGY
DETECTOR
dynamically occur in multiple primary spectra. Therefore, in This section firstly provides a generalized system model for
order to sense the multiple primary spectra, it is necessary to spectrum sensing as well as evaluating the energy detector’s per-
require that the secondary sensing node have receiver functions formance in terms of detection and false alarm probabilities.
for use in multiple primary systems and to be prepared to suffer The generalized energy detection performance dealt with in this
from demodulation complexity. Because of this, the matched- section is the basis for the technical comprehension of the ad-
filter sensing solution is too costly for sensing multiple primary vanced energy detection techniques such as multi-antenna spec-
spectra. trum sensing and cooperative sensing as discussed in Sections
Feature detection: The feature detection technique exploits IV and V, respectively.
the statistical feature built into a primary signal. Generally, the
background noise and interference do not correlate to time or A. System Model for Spectrum Sensing
frequency domains. Hence, if the secondary user has statistical
information about the correlation feature of the primary signal, According to the basic concept of overlay spectrum sharing,
it can increase sensing accuracy [5], [6]. For instance, the Gaus- the secondary user should sense the existence of the primary
sian minimum shift keying (GMSK) used in the Global System signal within the licensed spectrum prior to the secondary trans-
for Mobile Communications (GSM) network has inherent cy- mission. Based on this principle, each secondary transmission
clostationarity, so that the secondary user can effectively detect frame duration is divided into a sensing period and a data trans-
the GSM signal by using it. However, this feature detection can mission period, as shown in Fig. 2. Sensing duration N is de-
only be applicable for few primary signals with such characteris- fined as the length of the sensing period in which the secondary
tics and requires an increase in cost and complexity for the time user ceases transmission and then senses the primary spectrum.
shift correlation process and frequency-domain transformation, If any primary user signal is detected during the sensing period,
respectively [44]. the secondary user stops the transmission until the primary user
signal is again undetectable. Otherwise, if no primary user sig-
Energy detection: While the matched filter and feature de-
nal is detected, the secondary user continues transmission during
tection capabilities require prior information about primary sig-
the data transmission period.
nals, no primary signal information is required for the energy
Apparently, longer and frequent sensing periods improve
detection technique. As depicted in Fig. 2, the only process re-
sensing performance but shorten the data transmission period.
quired for the energy detector is that the primary signal energy is
Conversely, shorter and more sporadic sensing periods degrade
able to be measured within a specified duration. Next, the detec-
sensing performance but lengthen the data transmission period.
tor simply determines whether or not the measured signal energy
Hence, there is a tradeoff between sensing performance and sec-
is over the predetermined threshold level. When considering the
ondary user throughput. Using this tradeoff, throughput opti-
general purpose of spectrum sensing with low complexity, the
mization of sensing duration [7] or secondary frame duration [8]
energy detection technique is decidedly the most feasible spec-
has been attempted. However, this study concentrates on the
trum sensing scheme for detecting the white space of multiple
ways of improve sensing performance under the assumption of
primary spectra licensed to heterogeneous wireless communi-
a fixed sensing duration and period by introducing a variety of
cation systems. This explains why this paper focuses on spec-
traditional and newly developed detection techniques.
trum sensing using energy detection. However, because energy
detection has poor performance in comparison to matched fil-
ter and feature detections, research conducted on enhancing the B. Sensing Performance
performance of energy detection has been the most intensively Two hypotheses are related to the detection of primary user
studied research topic in the spectrum overlay scenario. In or- signals: The null hypothesis H0 and the alternative hypothesis
der to enhance the performance of energy detection, a number H1 . H0 describes a situation in which a primary user signal does
of spectrum sensing schemes using multi-antenna and coopera- not exist in the primary spectrum, and H1 expresses the case
22 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 12, NO. 1, FEBRUARY 2010

in which a primary signal does exist. The objective of energy

detection is to estimate which hypothesis is more credible and
to decide whether H0 or H1 is correct by measuring the energy.
The sensing performance of the energy detector can be evalu-
ated via two general performance metrics: The false alarm prob-
ability PF A and the detection probability PD . A false alarm
event occurs when the energy detector decides upon H1 when
the correct decision is H0 , the probability of which is defined
as the false alarm probability. When a false alarm happens, the
secondary user does not exploit the spectrum which is actually
empty, and loses an opportunity to transmit its data. Hence, the
lower the false alarm probability, the higher is the throughput of
the secondary user. On the other hand, the detection probabil-
ity is defined as the probability of the energy detector making
a correct decision for H1 . If the detection fails, or a “miss de-
tection” occurs, the secondary user starts an undesirable trans-
mission in the primary spectrum where the primary user is also
transmitting, and, therefore causes a strong interference with the
Fig. 3. PDF of test statistics. Miss detection and false alarm cannot be
primary user signal. This degrades the signal quality of the pri- reduced simultaneously.
mary communication and violates the fundamental doctrine of
overlay spectrum sharing. Therefore, when the detection proba-
bility is higher, it is then possible to provide enhanced protection spatial domain. As a simple and basic sensing scenario, it is
for the primary user. assumed that M nodes are located at spatially independent po-
However, there is a tradeoff relationship between the spec- sitions, so M degree-of-freedom can be fully exploited during
trum usage efficiency (i.e., the false alarm probability) and the the energy detection process, which will provide a performance
sensing performance (i.e., the detection probability). As illus- basis to compare with the sensing performance in the correlated
trated in Fig. 3, although the proportion between the false alarm channel between antennas or adjacent sensing nodes in multi-
probability and the detection probability can be adjusted via antenna sensing and cooperative sensing, respectively.
threshold control, it is not possible to simultaneously attain a For a given degree-of-freedom obtained from M nodes and a
low false alarm probability and a high detection probability or a sensing duration of N samples, suppose xm (n) is the received
low miss detection probability, as long as the probabilistic char- signal at the m-th sensing node and s(n) is the common primary
acteristics of the received signal are fixed. In order to enhance user signal that we wish to detect. The signal s(n) is transmitted
the sensing performance, the probability density function (PDF) over a fading channel whose gain is hm (n) and then corrupted
of each hypothesis should be steepened or the distance between by the noise wm (n). In this case, the detection problem is for-
two PDFs needs to be lengthened. This can be accomplished via mulated as:
two methods as follows: 1) Intensifying the received SNR of the
primary user measured at the secondary user or 2) increasing the H0 : xm (n) = wm (n)
dimension or degree-of-freedom of the received signal space. H1 : xm (n) = hm (n)s(n) + wm (n) (1)
Intensifying the received SNR is very challenging within a
practical situation, due to noise uncertainty, shadowing, and for m = 1, 2, · · ·, M and n = 1, 2, · · ·, N . The signal s(n) is
multi-path fading, whose effects are neither predictable nor able assumed to be phase-shift keying (PSK) modulated with the re-
to be compensated for [9]. Thus, we focus upon increasing the ceived signal power P . The channel gain hm (n) follows an in-
degree-of-freedom of the received signal space. If the secondary dependent and identically distributed (i.i.d.) Rayleigh fading
user receives an increased number of observation samples, they channel, i.e., hm (n) ∼ CN (0, σh2 ). The noise wm (n) is an
are combined into an aggregated observation and the final deci- i.i.d. zero-mean, complex-valued additive white Gaussian noise
2
sion can be made with more reliability. The sources for degree- (AWGN), i.e., wm (n) ∼ CN (0, σw ).
of-freedom are mainly time and space. If the secondary user As mentioned before, we employed an energy detector, which
sums N samples of received energy within a sensing period, as combines the measured energy during the sensing duration
depicted in Fig. 2, we obtain N degree-of-freedom in the time along the sensing nodes. Assuming each node and each sam-
domain. However, as N increases, the time fraction that the sec- ple is independent, the energy is combined with equal gain. The
ondary user can effectively use for data transmission decreases decision rule can be written as:
and, hence, restricts the increased use of the degree-of-freedom
N 
 M
within the time domain. H1
T = |xm (n)|2 ≷ η (2)
Therefore, the degree-of-freedom of the sensing signal space n=1 m=1
H0
should be expanded into the spatial domain. For practical sit-
uations, if the secondary user is equipped with M antennas or where T is the test statistic for the binary hypothesis test and
M secondary users are engaged in cooperative sensing, we have η is the threshold [10]. In order to derive the false alarm and
an opportunity for exploiting M degree-of-freedom within the detection probabilities, the probabilistic characteristics such as
WANG et al.: ADVANCED SENSING TECHNIQUES OF ENERGY DETECTION... 23

in spectrum sensing, accomplished by equipping multiple an-

tennas with a secondary user or cooperating with neighboring
nodes by exchanging sensing information.

IV. SENSING WITH MULTI-ANTENNA

Spectrum sensing schemes exploiting multi-antenna setups
are investigated in [13], [14]–[16]. Ideally, it can be assumed
that the channels for each antenna are faded independently, and
the performance of the multi-antenna-aided spectrum sensing is
largely identical to the result in (4).
Taking advantage of different fading channels for multiple
antennas, the maximum-ratio-combining (MRC) or antenna se-
lection increases the spectrum sensing performance [14]. How-
ever, the secondary user utilizing energy detection cannot co-
herently receive the primary signal due to unavailability of the
primary signal information, including the modulation technique,
Fig. 4. Relation between detection probability and false alarm probability
(γ = −5 dB, N = 10). pilot signaling and so on. Hence, unfortunately, any schemes re-
quiring the channel measurements cannot be practically imple-
mented with the energy detector scheme.
the PDF of the test statistic are first developed for both H0 and In order to evaluate the practical performance of multi-
H1 . T is the sum of the squared Gaussian random variables and antenna-aided spectrum sensing, one more thing that we should
follows a Chi-squared distribution with a degree of freedom of discuss is the correlation between antennas. As intuitively ex-
2M N . When assuming sufficiently high degree of freedom, the pected, the sensing performance becomes degraded as the corre-
Chi-squared distribution approaches the Gaussian distribution lation between antennas increases. In IEEE 802.22 wireless ru-
by the central limit theorem (CLT) [11]. According to the CLT, ral area network (WRAN), the secondary WRAN system should
the detection problem (1) can be rewritten as: be located outside of the keep-out region which is set for pro-
  tecting the primary users. The radius of the keep-out region is
2 4
H0 : T ∼ N M N σ w , M N σw generally assumed to be over one hundred kilometers. Accord-
 
H1 : T ∼ N M N (P σh2 + σw 2
), M N (P σh2 + σw
2 2
) . (3) ingly, such a large distance between the primary transmitter and
the secondary user generates a small received-channel angular
Using the probabilistic model in (3), we can obtain detection spread value, which results in a highly correlated channel be-
probability PD and the false alarm probability PF A . PD and tween antennas at the secondary receiver [17].
PF A have the following relationships [12], [13]:
  A. Simple Energy Detection with Multi-Antenna
1  −1 √ 
PD (M ) = Q Q (PF A ) − M N γ The performance of a simple energy detection that consid-
1+γ ers the correlated antennas is investigated in [13]. The detection
√ 
PF (M ) = Q M N γ + (1 + γ)Q−1 (PD ) (4) problem (2) that considers the antenna correlation can be written
as:
where Q(·) is the tail probability of the Gaussian distribution  2 4

H0 : T ∼ N M N σ w , M N σw
and γ = P σh2 /σw 2
is the SNR of the primary user measured at  
M

the secondary sensing node.
The above relationship between the detection and false alarm H1 : T ∼ N M N (P σh2 + σw 2
), N (P σh2 λm + σw2 2
)
m=1
probabilities is depicted in Fig. 4. We assume a very low SNR
(5)
for the primary user: γ = −5 dB, which is necessary in order to
detect the primary user even when the signal experiences deep where λm , m ∈ 1, 2, · · ·, M are the eigen-values of the follow-
fade. The sensing duration is fixed to N = 10 samples. Due ing correlation matrix:
to the intrinsic tradeoff between the two performance metrics,
it was already shown that achieving higher detection probability ρi−j , i≤j
Rij = ∗
and low false alarm probability is difficult. If the detection prob- Rji , i>j
ability increases, then the false alarm probability also increases.
i, j = 1, 2, · · ·, M , and ρ (0 ≤ ρ ≤ 1) is the correlation between
The sensing performance curve is shifted in the direction of
adjacent antennas.
improving both performance metrics only when the additional
From the detection problem in (5), we can calculate the de-
degree of freedom is achieved. In Fig. 4, as the degree of free-
tection and false alarm probabilities as follows:
dom M increases, the detection probability increases while the ⎡ ⎤
false alarm probability decreases without sacrificing any other  M
 
metrics such as secondary user throughput. With this in mind, PDe (M ) = Q ⎣(ηe − M N (γ + 1)) N (γλm + 1)2 ⎦
we investigate how to achieve an additional degree of freedom m=1
24 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 12, NO. 1, FEBRUARY 2010

⎡   M ⎤
M
  
PDo (M ) = Q ⎣ ηo − N γλm N 2
(γλm ) ⎦
m=1 m=1
⎡   M  ⎤
M   2
γλ m N γλ m
PF o (M ) = Q ⎣ ηo − N ⎦ (9)
m=1
γλ m + 1 m=1
γλ m +1

⎡  ⎤
 1 M
√ channels between the antennas are highly correlated. There is
PF e (M ) = Q ⎣Q−1 (PD ) (γλm + 1)2 + γ M N ⎦ another observation of (10) in that the optimum LRT detection
M m=1 outperforms simple energy detection in both the detection and
(6) false alarm probabilities in the correlated antenna case. The per-
formance difference between them comes from the differently
where the decision threshold is determined as: weighted matrices. In the optimum LRT detection, the weighted
 matrix is determined based on the antenna correlation while all
 M
spatial channels are equally weighted for the simple energy de-
ηe = Q −1
(PDe )N (γλm + 1)2 + M N (γ + 1). (7) tection. However, additional complexity is required in order to
m=1
calculate the weighted matrix in (8) via the optimum LRT de-
tection.
B. Optimum Energy Detection with Multi-Antennas
For the case of ρ = 0, another relationship among the sensing
In Section IV-A, the sensing performance of the simple en- performances in (4), (6), and (9) can be made as follows:
ergy detection was shown in a correlated channel. However, in
a correlated channel, the optimum sensing performance can be lim PDo (M ) = lim PDe (M ) = PD (M )
ρ→0 ρ→0
achieved by the likelihood ratio test (LRT) [10]. In this case, the
lim PF o (M ) = lim PF e (M ) = PF (M ). (11)
decision rule can be expressed as: ρ→0 ρ→0

N 
 M This result show that if the correlation is very low, the sensing
H1
T = YmH (n)Ym (n) ≷ ∗
ηM (8) performance of the energy detection and the optimum LRT de-
H0 tection closely approaches the generalized sensing performance
n=1 m=1
 in (4). Therefore, in such a case, the energy detection is nearly
where Ym (n) = σ1w γλ the optimum. And, its sensing performance can be improved
γλm um X(n), m ∈ {1, 2, ..., M } when
m

X(n) = [x1 (n), x2 (n), ..., xM (n)], and λm and um are the m- upon continuously as the number of antennas is increased.
th eigenvalue and eigenvector of the antenna correlation matrix,
respectively. V. COOPERATIVE SENSING
Applying the CLT to (8) for the same reason as in (3), the
The energy detector is generally operated in a very low
detection and false alarm probabilities can be calculated in (9)
SNR region. Hence, if a signal from the primary transmitter is
where the decision threshold is given as
severely shadowed as well as faded, a secondary sensing node
 should experience difficulty in deciding whether the primary
 M M
 spectrum is unused or occupied by the primary system. From
(PDo )N
−1 2
ηo = Q (γλm ) + N γλm .
the small scale point of view, a spatially faded primary signal
m=1 m=1
can be effectively sensed by using a multi-antenna. However, it
cannot be the solution to the secondary sensing node which is
located in a deeply shadowed geographical region from the pri-
C. Performance Comparison of Multi-Antenna Sensing Schemes mary transmitter, which can be overcome by cooperative sensing
The sensing performances of the energy detection in (6) and techniques.
the optimum detection in (9) are compared with that of single Cooperative sensing takes advantage of geographical varieties
antenna case (M =1) in (4). For any values of M and γ, the of secondary sensing nodes which experience different chan-
following relationship can be determined: nel conditions. As depicted in Fig. 5, spatially distributed sens-
ing nodes measure the signal from the primary transmitter, and
lim PDo (M ) > lim PDe (M ) > PD (1) report the measurement results to the fusion center. The fu-
ρ→1 ρ→1
sion center makes the final decision about the primary spectrum
lim PF o (M ) < lim PF e (M ) < PF (1) (10) availability based on the collected measurement results. There-
ρ→1 ρ→1
fore, even if some of the sensing nodes are shadowed from the
where the number of antennas M is larger than one. primary transmitter, the sensing performance can be improved
The first observation of (10) is that sensing with multiple an- upon via the primary signal measurements of other unshadowed
tennas always outperforms sensing with a single antenna even if sensing nodes.
WANG et al.: ADVANCED SENSING TECHNIQUES OF ENERGY DETECTION... 25

Fig. 5. Cooperative sensing. Fig. 6. Technical tree in performance enhancement of energy detector.

Due to the simple system model of multi-antenna sensing, A.2 Weighted-Fusion

the performances of the multi-antenna sensing schemes can be Sensing nodes should be located in geographically indepen-
clearly analyzed and compared using definite mathematical ex- dent positions in order to avoid correlated channels. When con-
pressions from the previous section. However, for cooperative sidering such a case, the distance between the primary transmit-
sensing, system models cannot be unified because there are too ter and each sensing node is different. Hence, as depicted in Fig.
many variations in sensing scenarios. Correspondingly, a direct 5, the average SNR values γi of the measured primary signal at
performance comparison between different schemes is usually each sensing node are different. Intuitively, sensing informa-
unavailable since one cooperative sensing scheme solves a prob- tion reported by a sensing node with a higher SNR from that
lem, but still requires additional problems to be solved by other of the primary transmitter provides more credits in determin-
proposed schemes. ing the existence of the primary signal. Accordingly, in order
In this situation, it is effective to attain insights within a to enhance the cooperative sensing performance, the sensing in-
technical area by arranging the conventional works as a tech- formation from different sensing nodes are properly weighted
nical tree, as is shown in Fig. 6. In this technical tree, coop- and fused based on the SNR values of the received primary sig-
erative sensing is categorized into two main branches: Coop- nal [20]–[22].
erative sensing using soft-information and cooperative sensing In [20], the performance of the optimum LRT detector is pro-
using hard information. Each issue branches into performance posed in the same manner as the optimum LRT detector for the
evaluations and has its own practical problems and solutions. multi-antenna in the previous section, and evaluated. Using the
LRT detector, the fusion center can reflect the different detec-
tion probabilities of sensing nodes with different average SNR
A. Soft-Information Decision Fusion
values, and fuse the sensing information from a sensing node
In cooperative sensing techniques using soft-information, with a higher SNR, thus indicating more importance.
each sensing node reports its raw primary signal measurement in In [21], the average performance of cooperative sensing is
(1) to the fusion center. In this case, the index m in (1) denotes evaluated with respect to location probability for uniformly dis-
a secondary sensing node. If the local sensing information of M tributed secondary sensing nodes. The LRT-based cooperative
secondary users is perfectly delivered to the fusion center, the sensing performance should be evaluated differently according
spectrum sensing performance of the secondary network is also to channel model because the LRT detector directly uses the
exactly identical to the result shown in (4). However, there are a PDFs of the received primary signal at the sensing nodes. While
number of techniques that can enhance the cooperative sensing the Chi-square distribution is assumed in [20] and [21], the case
performance, as well as issues that should be considered for a considering the log-normal fading model is dealt with in [22].
practical implementation of the cooperative spectrum sensing in In [20], it is shown that the LRT-based cooperative detector
the real world. outperforms the equal-gain-combining (EGC) detector in (1).
However, there is a practical implementation difficulty of the
A.1 Performance in a Correlated Channel cooperative sensing techniques using the LRT. In order to per-
form the LRT, the fusion center must have information about
Similarly to the multi-antenna sensing performance with a the channel model and the average SNR values between sens-
correlated channel, the performance of cooperative sensing is ing nodes and the primary transmitter. Therefore, if the SNR
degraded by secondary sensing nodes experiencing correlated values are not accurately estimated or the statistical character-
shadowing [19]. For instance, if the distance between two ad- istics of channel models are different from the actual values,
jacent secondary sensing nodes is between 40 and 80 meters, the performance of the LRT-based cooperative sensing will be
the correlation value between the two nodes is over 0.5 at 1.9 degraded. Accordingly, the performance of the LRT-based co-
Ghz [45]. If cognitive radios intend to involve a small coverage operative sensing needs to address the SNR-estimation errors.
network, cooperative sensing should be designed to reflect the
correlation.
26 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 12, NO. 1, FEBRUARY 2010

A.3 Consideration of SNR-Estimation Errors of a secondary sensing node is quantized into two bits. In (1),
the original detection problem results in binary states. However,
SNR-estimation errors can be generated by two main causes:
in [26], the presence of the primary signal is expressed via four
The inherent estimator error and the inaccurate source sam-
states as follows: Strong empty, weak empty, weak presence and
ples for the estimator. When considering both causes, the per-
strong presence. Because thresholds for the four states are given
formance degradation of cooperative sensing due to the SNR-
heuristically, this work shows that cooperative sensing with
estimation errors is analyzed in [23] assuming the finest SNR es-
two-bit-quantized soft information can almost achieve sensing
timator with the Cramer-Rao-lower-bound performance, which
performance using perfect soft information. A more advanced
shows that the required number of sensing nodes to satisfy a pre-
soft information quantization scheme is proposed in [27]. This
determined sensing accuracy should be large depending on the
scheme also uses two bits for the soft information quantization.
SNR-estimation errors.
Different from [26], thresholds to divide the primary signal in-
However, from a practical implementation point of view, most
formation into four states are analytically proposed consider-
of the conventional SNR estimators might not be applicable to
ing probability distribution of the fading channel between the
secondary sensing nodes performing the energy detection be-
primary transmitter and the secondary receiver. Results of this
cause they need some prior information about the primary sig-
work confirm that the sensing performance from the perfect soft
nals such as coherent received signal sampling, the PDFs of the
information fusion can almost be achieved by only a two-bit-
primary signals, or the Doppler shift of the primary signal’s
quantized primary signal strength level.
spectrum [46], [47]. As we know, the most favorable charac-
teristic of the energy detector is that it requires no prior infor- Another concern to the feedback problem is how to deliver
mation about primary signals. Therefore, investigations on SNR the sensing information to the fusion center. So far there have
estimators taking advantage of the statistical characteristics of been few scenarios for realizing the feedback information de-
the background noise floor2 should become a key for practical livery. In [28], wireless local area network (WLAN) delivers
implementations of cooperative sensing. the sensing information to the fusion center. However, WLAN
has a very small coverage area with a radius of less than 15 me-
As an example for overcoming the SNR-estimation errors in
ters. If cooperative sensing is operated in this small area, the
cooperative sensing, a cooperative sensing scheme utilizing ran-
sensing information between sensing nodes will experience a
dom matrix theory is proposed in [24]. In this approach, only the
highly correlated shadow fading. Accordingly, in this case, it
maximum and minimum eigenvalues of the covariance matrix,
is difficult for the cooperative sensing to experience the gain of
composed by collecting sensing information, are used to deter-
sensing performance from the geographical diversity of the sec-
mine the existence of the primary signal. Therefore, the SNR-
ondary sensing nodes. In addition, the feedback can be transmit-
estimation process is not needed. Although this scheme requires
ted using a spread spectrum transmission methodology without
additional computational complexity for calculating the eigen-
harmful interference to the primary system [29]. This kind of
values, its performance is better than that of the EGC detector.
secondary transmission is known as underlay spectrum sharing,
A.4 Sensing Information Feedback Problem as classified in Section II. Above all, although the cognitive ra-
dio identifies and utilizes an empty spectrum for the secondary
Since the sensing information from the sensing nodes is re- usage, it seems to be a paradox that we are able to tell that the
ported to the fusion center, additional radio resource consumed legacy licensed systems needs to be used for sensing informa-
for reporting should be considered. Although the performance is tion reporting.
generally improved upon as the number of cooperative sensing
nodes increases, the amount of the sensing information feed-
back burden is proportional to the number of cooperative sens- B. Hard-Information Decision Fusion
ing nodes [25]. Therefore, the tradeoff between the overhead In cooperative sensing using hard-information decision fu-
reduction for the sensing information reporting and the cooper- sion, a sensing node reports only binary state information to
ative sensing performance needs to be an important design con- the fusion center. The binary state information is generated by
sideration. each sensing node, which has its own local energy detector. Ac-
In [25], the performance optimization in consideration of the cordingly, the hard-information decision fusion requires mini-
tradeoff is analyzed. However, the objective function defined by mized radio resource consumption for sensing information feed-
a linear combination of the sensing performance and the sensing back. Generally, the performance of cooperative sensing using
information feedback burden generates an ambiguous quantity3 , hard-information is worse than that using soft-information [20].
hence it is difficult to apply when evaluating the performance However, from a practical implementation point of view, coop-
of the cooperative sensing scheme. Accordingly, a more general erative sensing with hard-information is worth considering due
frame work is required in evaluating this tradeoff frame. to its minimized feedback burden.
Alternatively, feedback information reduction schemes are in-
vestigated in [26] and [27]. In those papers, the soft information B.1 Basic Fusion Rules
2 Statistical characteristics of the background noise floor are the only informa- There are three decision fusion rules in cooperative sensing
tion used by the general energy detector. using hard-information: The AND fusion rule, the OR fusion
3 Generally, the sensing performance metrics such as detection or false-alarm
rule and the majority fusion rule. The AND fusion rule declares
probabilities can be expressed in percentile, and the amount of feedback infor-
mation can be measured in bits per second. However, the linear combination of the existence of the primary signal H1 if all sensing nodes re-
those two quantities is hard to express in a quantity unit. port the decision state H1 . Using the OR fusion rule, H1 is true
WANG et al.: ADVANCED SENSING TECHNIQUES OF ENERGY DETECTION... 27

B.3 Cluster-based Cooperative Sensing

Cluster-based sensing schemes have been proposed in order
to utilize selection diversity in cooperative sensing using hard-
information decision fusion. In cluster-based cooperative sens-
ing, a whole sensing node is divided into several clusters. A clus-
ter head exists in a cluster for collecting sensing information, as
well as for reporting the collected information to the fusion cen-
ter, as depicted in Fig. 7 [33], [34].
It is generally assumed that the cluster head and sensing nodes
within a cluster are located in close proximity to each other.
Hence, the wireless link between a cluster head and its sensing
nodes is reliable enough for exchanging the sensing informa-
Fig. 7. Cluster-based cooperative sensing.
tion without error. For the same reason, the radio resources con-
sumed for exchanging sensing information within a cluster are
minimized. Accordingly, the feedback burden for the sensing in-
formation delivery or the feedback error within a cluster are not
if H1 is reported by at least one sensing node. The fusion center
research issues. However, we have to consider the same feed-
adopting the majority rule decides upon H1 when the number of
back error problem between the cluster head and the fusion cen-
H1 results is larger than the number of H0 results. The major-
ter. Therefore, for enhanced performance of the cluster-based
ity fusion rule is considered as a simple suboptimal fusion rule
sensing scheme, we should be able to choose a cluster head that
[31]. The OR fusion rule performs best for detection probabil-
is able to make the most credible local decision and then re-
ity, and the worst for false alarm probability. The AND fusion
port the local decision to the fusion center with the most reliable
rule performs the worst in detection and false alarm probabili-
channel [33], [34].
ties, opposite that of the OR fusion rule.
Just as in [28], it is shown that the sensing information re-
porting error dominantly limits the performance of cluster-based
B.2 Local Decision and Feedback Errors sensing. In order to minimize the reporting error, a sensing
node with the maximum SNR between the sensing node and
In order to perform hard-decision fusion, each sensing node the fusion center is elected as a cluster head. It is pointed out
has a local energy detector for making its own decision. For the in [34] that a larger number of clusters or sensing nodes does
cooperative sensing, each sensing node has a different sensing not guarantee a more improved sensing performance. Based on
accuracy because of the different geographical position of each the tradeoff between the sensing information reporting overhead
sensing nodes with respect to the primary transmitter. Hence, and sensing accuracy, a cluster head selection scheme is pro-
local decision errors from sensing nodes with low SNR values posed to enhance sensing performance. Subsequently, the op-
from the primary transmitter degrade the sensing performance. timum cluster number is analyzed in terms of the number of
It is pointed out in [32] that the sensing performance is not sensing nodes and the average SNR for them.
always improved upon when the number of sensing nodes in-
creases in cooperative sensing using hard-information fusion.
Therefore, only selected sensing nodes with SNR values over VI. SUMMARY AND DISCUSSION
a predetermined threshold have to report their decisions while This paper studied sensing performance enhancement tech-
the unselected sensing nodes remain silent. From this fact, we niques including multi-antenna sensing and cooperative sensing
can deduce that hard-information decision fusion is also not free within cognitive radios. In order to clarify the overall relation-
from the SNR estimation between the primary transmitter and ship between them, a summary of this paper is given in Fig. 8.
the sensing nodes, and the estimated SNR values should be de- The whole spectrum sensing techniques introduced in this paper
livered to the fusion center. are divided into two categories: Sensing performance enhance-
The feedback error of the sensing information may occur ments available in a single sensing node, and those achieved via
within cooperative sensing using hard information. In contrast cooperation between sensing nodes. Each category has elemen-
to the soft information feedback error, the hard information may tal sensing techniques dealt with in detail as part of this paper.
cause a totally opposite decision at the fusion center. Hence, Through reviewing a number of advanced sensing techniques,
the feedback error of hard-information degrades the sensing we were able to reach a conclusion that further investigations
performance more severely in comparison with that of soft- are needed for ‘Intermediate Solutions’ in the multiple sensing
information using cooperative sensing. nodes cooperation category of Fig. 8. When developing more
In hard-information decision fusion, the effect of feedback er- practical parameter estimation techniques, it might be possible
ror on the sensing performance is investigated of [28]. Accord- to create a realizable sensing scheme which can closely achieve
ing to the results in [28], the feedback error limits the sensing the optimum LRT performance. Based on our survey, one prac-
performance of hard-information decision fusion. Accordingly, tical concern is raised for the practical implementation of co-
it is suggested that diversity techniques, such as space time or operative sensing: The methodologies for sensing information
frequency coding schemes, should be adopted for mpre reliable exchange. Conventional scenarios for this seem unrealistic. In
hard sensing information feedback. order to realize cooperative sensing, a multiplexing scheme and
28 JOURNAL OF COMMUNICATIONS AND NETWORKS, VOL. 12, NO. 1, FEBRUARY 2010

spectrum sensing under correlated log-normal shadowing,” IEEE Commun.

Lett., vol. 11, pp. 34–36, Jan. 2007.
[20] J. Ma, G. Zhao, and Y. Li, “Soft combination and detection for cooper-
ative spectrum sensing in cognitive radio networks,” IEEE Trans. Wireless
Commun., vol. 7, no. 11, pp. 4502–4507, Nov. 2008.
[21] S. S. Jeong, W. S. Jeon, and D. G. Jeong, “Collaborative spectrum sensing
for multiuser cognitive radio systems,” IEEE Trans. Veh. Technol., vol. 58,
pp. 2564–2569, June 2009.
[22] J. Unnikrishnan and V. V. Veeravalli, “Cooperative sensing for primary
detection in cognitive radio,” IEEE J. Sel. Areas Commun., vol. 2, no. 1, pp.
18–27, Feb. 2008.
[23] Y. Chen and N. C. Beaulieu, “Performance of collaborative spectrum sens-
ing for cognitive radio in the presence of Gaussian channel estimation er-
rors,” IEEE Trans. Commun., vol. 57, no. 7, pp. 1944–1947, July 2009.
[24] F. Penna, R. Garello, and M. A. Spirito, “Cooperative spectrum sensing
based on the limiting eigenvalue ratio distribution in Wishart matrices,”
IEEE Commun. Lett., vol. 13, pp. 507–509, July 2009.
Fig. 8. Summary of spectrum sensing techniques. [25] Y. Chen, “Optimum number of secondary users in collaborative spectrum
sensing considering resource usage efficiency,” IEEE Commun. Lett., vol.
12, no. 12, pp. 877–879, Dec. 2008.
[26] M. Mustonen, M. Matinmikko, and A. Mammela, “Cooperative spectrum
protocols for the sensing information, reporting of plural sens- sensing using quantized soft decision combining,” in Proc. 4th international
ing nodes should be investigated as part of future works. conference on CROWNCOM 2009.
[27] Y. Chen, “Optimum number of secondary users in collaborative spectrum
sensing considering resource usage efficiency,” IEEE Commun. Lett., vol.
12, pp. 877–879, Dec. 2008.
REFERENCES [28] W. Zhang and K. B. Letaief, “Cooperative spectrum sensing with trans-
mit and relay diversity in cognitive radio networks,” IEEE Trans. Wireless
[1] Federal Communications Commission, “Spectrum policy task force re- Commun., vol. 7, no. 12, pp. 4761–4766, Dec. 2008.
port, (ET Docket No. 02-135),” Nov. 2002. [Online]. Available: hraun- [29] Q. Pan, Y. Chang, R. Zheng, X. Zhang, Y. Wang, and D. Yang, “Solution of
foss.fcc.gov/edocs public/attachmatch/DOC-228542A1.pdf information exchange for cooperative sensing in cognitive radios,” in Proc.
[2] J. Mitola and G. Q. Maguire, “Cognitive radio: Making software radios IEEE WCNC, Apr. 2009, pp. 1–4.
more personal,” IEEE Person. Commun., vol. 6, no. 4, pp. 13–18, Aug. 1999. [30] N. Nguyen-Thanh and I. Koo, “An enhanced cooperative spectrum sensing
[3] D. Cabric, S. M. Mishra, and R. W. Brodersen, “Implementation issues in scheme based on evidence theory and reliability source evaluation in cogni-
spectrum sensing for cognitive radios,” in Proc. Asilomar Conference on tive radio context,” IEEE Commun. Lett., vol. 13, no. 7, pp. 492–494, July
Signal, Systems and Computers, Nov. 2004. 2009.
[4] J. G. Proakis, Digital Communications, 4th ed., McGraw-Hill. [31] V. Aalo and R. Viswanathan, “Asymptotic performance of a distributed de-
[5] M. Oner and F. Jondral, “On the extraction of the channel allocation infor- tection system in correlated Gaussian noise,” IEEE Trans. Signal Process.,
mation in spectrum pooling systems,” IEEE J. Sel. Areas Commun., vol. 25, vol. 40, no. 1, pp. 211–213, Feb. 1992.
no. 3, pp. 558–565, Apr. 2007. [32] E. Peh and Y. C. Liang, “Optimization for cooperative sensing in cognitive
[6] J. Lunden, V. Koivunen, A. Huttenen, and H. V. Poor, “Collaborative cy- radio networks,” in Proc. IEEE WCNC, Mar. 2007, pp. 27–32.
clostationary spectrum sensing for cognitive radio systems,” IEEE Trans. [33] C. Sun, W. Zhang, and K. B. Letaief, “Cluster-based cooperative spec-
Signal Process., accepted for publication. trum sensing in cognitive radio systems,” in Proc. IEEE ICC, June 2007, pp.
[7] Y. -C. Liang, Y. Zeng, E. Peh, and A. T. Hoang, “Sensing-throughput trade- 2511–2515.
off for cognitive radio networks,” IEEE Trans. Wireless Commun., vol. 7, [34] C. Guo, T. Peng, S. Xu, H. Wang, and W. Wang, “Cooperative spec-
no. 4, pp. 1326–1337, Apr. 2008. trum sensing with cluster-based architecture in cognitive radio networks,”
[8] Y. Pei, A. T. Hoang, and Y.-C. Liang, “Sensing-throughput tradeoff in cog- in Proc. IEEE VTC-Spring, Apr. 2009, pp. 1-5.
nitive radio networks: How frequently should spectrum sensing be carried [35] Y. Xing, R. Chandramouli, S. Mangold, and S. Shankar, “Dynamic spec-
out?” in Proc. IEEE PIMRC, Sept. 2007, pp. 1–5. trum access in open spectrum wireless networks,” IEEE J. Sel. Areas Com-
[9] A. Sahai, N. Hoven, and R. Tandra, “Some fundamental limits on cognitive mun., vol. 24, no. 3, Mar. 2006.
radio,” in Proc. Allerton Conf. Communication, Control, Computing, Oct. [36] S. Mangold and K. Challapali, “Coexistence of wireless networks in un-
2004. licensed frequency bands,” Wireless World Research Forum #9, Zurich,
[10] S. M. Kay, Fundamentals of Statistical Signal Processing: Detection The- Switzerland, July 2003.
ory, Prentice-Hall, 1998. [37] J. Lee, H. Wang, S. Kim, and D. Hong, “Sensing threshold control for
[11] A. Papoulis and S. U. Pillai, Probability, Random Variables and Stochastic fair coexistence of heterogeneous systems in open spectrum,” IEEE Trans.
Processes, 4th ed., McGraw-Hill, 2002. Wireless Commun., accepted for publication.
[12] G. Noh, J. Lee, H. Wang, S. You, and D. Hong, “A new spectrum sensing [38] D. P. Satapathy and J. M. Peha, “Etiquette modification for unlicensed
scheme using cyclic prefix for OFDM-based cognitive radio systems,” in spectrum: Approach and impact,” in Proc. IEEE VTC, vol. 1, 1998, pp.
Proc. IEEE VTC-Spring, May 2008, pp. 1891–1895. 272–276.
[13] S. Kim, J. Lee, H. Wang, and D. Hong, “Sensing performance of energy [39] S. Buljore, H. Harada, S. Filin, P. Houze, K. Tsagkaris, O. Holland, K.
detector with correlated multiple antennas,” IEEE Signal Process. Lett., vol. Nolte, T. Farnham, and V. Ivanov, “Architecture and enablers for optimized
16, no. 8, pp. 671–674, Aug. 2009. radio resource usage in heterogeneous wireless access networks: The IEEE
[14] A. Pandharipande and J.-P. M. G. Linnartz, “Performance analysis of pri- 1900.4 Working Group,” IEEE Commun. Mag., vol. 47, no. 1, pp. 122–129,
mary user detection in a multiple antenna cognitive radio,” in Proc. IEEE Jan. 2009.
ICC, July 2007, pp. 6482–6496. [40] A. Ghasemi and E. S. Sousa, “Fundamental limits of spectrum-sharing in
[15] V. Kuppusamy and R. Mahapatra, “Primary user detection in OFDM based fading environments,” IEEE Trans. Wireless Commun., vol. 6, no. 2, pp.
MIMO cognitive radio,” in Proc. 3rd Int. Conf. Cognitive Radio Oriented 649–658, Feb. 2007.
Wireless Networks and Communications, May 2008, pp. 1–5. [41] K. Hamdi, W. Zhang, and K. B. Letaief, “Power control in cognitive radio
[16] J.-H. Lee, J.-H. Baek, and S.-H. Hwang, “Collaborative spectrum sensing systems based on spectrum sensing side information,” in Proc. IEEE ICC,
using energy detector in multiple antenna system,” in Proc. 10th Int. Conf. June 2007, pp.5161–5165.
Advanced Communication Technology, Feb. 2008, pp. 427–430. [42] R. Zhang and Y. C. Liang, “Exploiting hidden power-feedback loops for
[17] D.-S. Shiu, G. Foschini, M. Gans, and J. Kahn, “Fading coorelation and cognitive radio”, in Proc. 3rd IEEE Int. Symp. on New Frontiers in Dynamic
its effect on the capacity of multielememt antenna systems,” IEEE Trans. Spectrum Aceess Networks (DySPAN), Oct. 2008, pp. 1–5.
Commun., vol. 48, pp. 502–315, Mar. 2000. [43] IEEE P802.22/D1.0 Draft Standard for Wireless Regional Area Networks
[18] E. Visotsky, S. Kuffner, and R. Peterson, “On collaborative detection of Part 22: Cognitive Wireless RAN Medium Access Control (MAC) and
TV transmissions in support of dynamic spectrum sharing,” in Proc. IEEE Physical Layer (PHY) Specifications: Policies and Procedures for Opera-
1st Symposium on Dynamic Spectrum Access Networks, 2005. tion in the TV Bands,” Apr. 2008.
[19] A. Ghasemi and E. S. Sousa, “Asymptotic performance of collaborative [44] T. Zhang, G. Yu, and C. Sun, “Performance of cyclostationary features
WANG et al.: ADVANCED SENSING TECHNIQUES OF ENERGY DETECTION... 29

based spectrum sensing method in a multiple antenna cognitive radio sys- Sungtae Kim received B.S., M.S., and Ph.D. degrees
tem,” in Proc. IEEE WCNC, Apr. 2009, pp. 1–5. from the Department of Electronics Engineering at
[45] J. Weitzen and T. J. Lowe, “Measurement of angular and distance correla- Yonsei University in 2002, 2004, and 2009, respec-
tion properties of log-Normal shadowing at 1900 MHz and its application to tively. He is currently a senior engineer with Samsung
design of PCS systems,” IEEE Trans. Veh. Technol., vol. 51, pp. 265–274, Electronics Co., LTD. His current research interests
Mar. 2002. are spectrum sensing and MIMO systems.
[46] S. Kay, Fundamentals of Statistical Signal Processing: Estimation Theory.
Englewood Cliffs, NJ: Prentice-Hall, 1993.
[47] S. Kim, H. You, J. Lee, and D. Hong, “Low bias frequency domain SNR
estimator using DCT in mobile fading channels,” IEEE Trans. Wireless
Commun., vol. 8, no. 1, pp. 45–50, Jan. 2009.
Daesik Hong received B.S. and M.S. degrees from the
Department of Electronics Engineering at Yonsei Uni-
versity in 1983 and 1985, respectively, and a Ph.D.
degree from Purdue University, West Lafayette, Indi-
Hano Wang received his B.E. degree from the De- ana, in 1990. Since 1991 he has been a Professor with
partment of Electronics Engineering at Yonsei Univer- the School of Electrical and Electronic Engineering
sity in 2004. He is working toward the Ph.D. degree at Yonsei University. He has been serving as chair of
in Electrical and Electronic Engineering at the same the Center for Electronic and Informative Telecommu-
university. His research interests are spectrum sharing, nication of Yonsei University since March 2002, and
cognitive radios and resource management of mobile also as chair of the Samsung-Yonsei Research Center
networks. for Mobile Intelligent Terminals. Currently, he is a di-
vision editor of the Journal of Communications and Networks and an editor of
IEEE Transactions on Wireless Communications. His current research activities
are in 4G wireless communication systems, orthogonal frequency-division mul-
tiplexing and multicarrier systems, multiple antenna and relay-based wireless
Gosan Noh received the B.S. degree in Electrical communication systems, and cross-layer techniques, cognitive radio, machine
and Electronic Engineering from Yonsei University, to machine communication.
in 2007. He is working toward the Ph.D. degree in
Electrical and Electronic Engineering at the same uni-
versity. His current interests include physical layer in
wireless communications, such as cognitive radio, hi-
erarchical cell structure, and optimization techniques.

Dongkyu Kim received the B.S. degree in Electrical

Engineering from Konkuk University, in 2006 and re-
ceived the M.S. degree in Electrical Engineering from
Yonsei University, in 2008. He is working toward the
Ph.D. degree in Electrical Engineering at the Yon-
sei University. His current research interests are in
physical layer in wireless communication, cooperative
communications and cognitive radio networks.

View publication stats