Wireless Pers Commun (2014) 78:1879–1890

DOI 10.1007/s11277-014-2050-2

Cooperative Spectrum Sensing with Adaptive Node

Selection for Cognitive Radio Networks

Taiping Cui · Kyung Sup Kwak

Published online: 3 September 2014

© Springer Science+Business Media New York 2014

Abstract In cooperative spectrum sensing, secondary users can cooperate based on their
local sensing observations so as to detect primary signals in a more reliable manner. However,
because of the low strength of received primary signals, observations from some secondary
users may contribute little to global decisions at the fusion center. To reduce energy con-
sumption and sensing overhead, cooperation is recommended only with a subset of nodes
receiving high-strength primary signals. This paper proposes a selection probability based
cooperative spectrum sensing scheme that exploits historical observations to exclude nodes
receiving low-strength primary signals. The selection probability is maintained based on local
sensing information and global decisions. Simulation results demonstrate that the proposed
scheme can exploit location advantages and shows almost the same detection performance
as cooperative spectrum sensing with accurate node selection.

Keywords Cognitive radio · Cooperative spectrum sensing · Node selection · Selection

probability

1 Introduction

Cognitive radio (CR) is an agile radio technology viewed to have the potential to improve
spectrum utilization through negotiated or opportunistic spectrum sharing. When a specific
licensed channel is not occupied by primary users (PUs), the CR or secondary users (SUs)
can reuse the licensed band temporarily without causing any substantial interference for
PUs [1,2]. Because CR is considered low-priority user in accessing a license channel, a
fundamental requirement is to avoid or minimize any interference for a potential PU in its

T. Cui· K. S. Kwak (B)

Telecommunication Engineering Laboratory, Inha University, 100 Inha-ro, Nam-gu,
Incheon 402-751, Korea
e-mail: cuitaiping@[Link]
K. S. Kwak
e-mail: kskwak@[Link]

123
1880 T. Cui, K. S. Kwak

vicinity. A key challenge in this requirement is the reliable detection of PU signals. Therefore,
prior to accessing a licensed spectrum band, the SU should employ sensing techniques to
make a reliable decision on whether the PU signal is present.
Many studies have examined spectrum sensing, including non-coherent energy detec-
tion [3,4] applicable to any signal type and coherent pilot detection that can optimally detects
known PU signals [5]. Although coherent detection outperforms non-coherent detection at
the expense of perfect synchronization circuitry and a priori knowledge of the PU signal
structure [6], energy detectors are widely used for signal detection because of their sim-
plicity. It is an optimal approach when the detector only knows the power of the received
signal [6], which is often encountered in the CR scenarios.
For a single SU, spectrum sensing is a challenging task, and detection performance cannot
be guaranteed when the strength of received PU signals is low. Cooperative spectrum sensing
schemes [7–12] have been proposed to exploit the spatial diversity. With multiple SUs dis-
tributed across various locations, it is possible for them to cooperate to increase the diversity
by providing multiple PU signal measurements and thus show better detection performance.
To reduce energy consumption and sensing overhead, cooperation is recommended only
with a subset of SUs. Previous studies have considered various node selection schemes
[14–16]. In [14], with knowledge of sensor position, sensor nodes sufficiently separated
spatially are selected for spectrum sensing. These nodes with best detection performance are
selected for spectrum sensing in [15]. In [16], a distributed sensor selection mechanism is
proposed to maximize the spectrum utilization under Nash equilibrium and PU protection.
A probability based node selection scheme for reducing the number of SUs involved in
detection is proposed in [13]. Based on received PU signal strength (RPSS), this paper assigns
a nonidentical selection probability to each SU in the CR network. For a low-RPSS SU, its
selection probability is reduced adaptively; whereas for a reliable sensing node, that is, a high-
RPSS node, a high selection probability is assigned. The simulation results demonstrate that
the proposed scheme can exploit location advantages and shows almost the same detection
performance as cooperative spectrum sensing with accurate node selection.
The rest of this paper is organized as follows: Sect. 2 describes the system model of PU
signal detection in a CR network. Section 3 proposes a cooperative spectrum sensing with
adaptive node selection. Section 4 analyzes the performance of the proposed scheme. Section
5 presents the simulation results, and Sect. 6 concludes.

2 System Model

A centralized CR network consisting of N SUs and one fusion center, as shown in Fig. 1, is
now investigated. Each SU can perform spectrum sensing during its sensing time slot τ and
reuse the licensed channel during the data transmission time slot T − τ if the PU signal is
considered absent. Local sensing information is sent to the fusion center periodically through
a common reporting channel (CRC). The rest of this paper assumes that an error-free CRC is
used for transmitting control and sensing information between SUs and the fusion center. For
a PU network, assume that it consists one primary transmitter and several primary receivers,
e.g., a TV radio station with many TV radio receivers.
Spectrum sensing involves determining whether a PU signal is present based on an SU’s
observation. Energy detection is an optimal approach to the detection of any unknown zero-
mean constellation signal [6]. For implementation simplicity, the analysis is restricted to
energy detection in context of cooperative spectrum sensing for a CR network. Suppose that
the received PU signal at each SU is sampled at the sampling frequency f and let τ be the

123
Cooperative Spectrum Sensing with Adaptive Node Selection 1881

Fig. 1 Cooperative spectrum sensing in CR networks

identical sensing duration. Then the number of samples M is the maximum integer not great
than τ f , that is, M = τ f , where · is the floor function. The accumulated energy Yi of
SUi can be expressed as [3,4]:
 M
m=1 |n i (m)| , H0 ,
2
Yi =  (1)
M
m=1 |h i s(m) + n i (m)| , H1 ,
2

where n i (m) is complex Gaussian noise with mean zero and variance σn2 ; s(m) is the PU
signal; and h i is the complex channel coefficient between PU transmitter and SUi . H0 and
H1 denote the hypotheses corresponding to the absence and presence of the PU signal,
respectively. The purpose of energy detection is to make a reliable decision between H0 and
H1 based on the observation Yi .
Without loss of generality, assume that the PU signal s(m) is complex-valued PSK sig-
nal [7]. According to the Proposition 1 and 2 in [7], for a high value of M, Yi can be
approximated by a Gaussian random variable with mean and variance given respectively by

Mσn2 , H0 ,
E[Yi ] = (2)
|h i | E s + Mσn , H1 ,
2 2

Mσn4 , H0 ,
V ar [Yi ] = (3)
2|h i | E s σn + Mσn , H1 ,
2 2 4

M
where E s = m=1 |s(m)|2 represents the transmitted signal energy over M samples. Denote
γi as the received signal-to-noise ratio (SNR) of the PU signal measured at SUi under the
hypothesis H1 . Then γi = |hMσ
i | Es
2
2 , and the probability of a false alarm P f,i and the probability
n
of detection Pd,i can be approximated by [7]:

123
1882 T. Cui, K. S. Kwak

  
η √
P f,i = Q −1 M , (4)
σn2
   
η M
Pd,i =Q − γi − 1 , (5)
σn2 1 + 2γi
2
∞ −t
where η is the corresponding decision threshold and Q(x) = √1
x e
2 dt.
2π

3 Cooperative Spectrum Sensing with Adaptive Node Selection

As shown in Fig. 1, SU1 , for instance, is deeply shadowed by a high building over the
sensing channel from the primary transmitter to SU1 . In addition, the RPSS of SU3 is low
because the location of SU3 is far away from the primary transmitter. Consequently, local
sensing information from low-RPSS nodes when the PU signal is present contributes little to
the global decision, burdening sensing overhead and consuming more energy. On the other
hand, these low-RPSS nodes may interfere with primary receivers in their vicinity when they
reuse the PU channel based on their local decisions.

3.1 Node Selection Probability

Let piS (k) denote the probability of the event that SUi is selected for the k-th spectrum
sensing. In addition to employing the threshold η to determine the presence of a PU signal,
another threshold , called selection threshold, is used. Based on η and η + , the local
sensing information Yi (k) of SUi can be divided into three regions if there is a PU signal.
Then the (k + 1)-th selection probability of SUi can be determined accordingly in three cases
when there is a PU signal.
Case 1 Yi (k) < η: SUi fails to detect a PU signal. The (k + 1)-th selection probability of
SUi is calculated as
l−1
piS (k + 1) = α j piS (k − j), if Yi (k) < η, (6)
j=0

where l is the span of historical selection probabilities and α j is the selection weight for the
j-th historical selection probability. Note that the selection probability of SUi for subsequent
sensing is a linear combination of l historical selection probabilities and is proportional to
its previous detection performance. To reduce the number of low-RPSS nodes  involved in
detection, their selection probability is reduced. Therefore, α j should satisfy l−1 j=0 α j < 1.
Form (6), it is clear that each SU should store l historical selection probabilities. To reduce
sensing burden, each SU considers only its most recent selection probability. Then (6) can
be rewritten as
piS (k + 1) = αpiS (k), if Yi (k) < η, (7)
where α is the selection weight and 0 < α < 1. It is obvious that a low value of α sharply
reduces the selection probability of low-RPSS nodes. This strategy is adopted to exclude
low-RPSS nodes from participating in the detection process.
Case 2 Yi (k) ≥ η + : SUi detects a PU signal with high reliability. The (k + 1)-th
selection probability of SUi is presented by
piS (k + 1) = Ph1 , if Yi (k) ≥ η + , (8)

123
Cooperative Spectrum Sensing with Adaptive Node Selection 1883

Fig. 2 Flow chart of cooperative

sensing with adaptive node
selection

where Ph1 is a high selection probability.

Case 3 η ≤ Yi (k) < η + : SUi detects a PU signal with moderate reliability. The
(k + 1)-th selection probability of SUi is given by
piS (k + 1) = Ph2 , if η ≤ Yi (k) < η + , (9)
where Ph2 is a moderate selection probability.
It is observed that the selection probability of an SU that can detect the presence of a PU
signal (i.e., Yi ≥ η) does not depend on its historical selection probability and has a constant
value Ph1 or Ph2 for subsequent spectrum sensing. A higher-RPSS node should be assigned
a higher selection probability. Therefore, Ph1 and Ph2 should satisfy 0 < Ph2 < Ph1 ≤ 1.
Combining (7), (8) and (9) yields
⎧
⎨ Ph1 , if Yi ≥ η + ,
piS (k + 1) = Ph2 , if η ≤ Yi < η + , (10)
⎩ i
αp S (k), if Yi < η.

3.2 Adaptive Node Selection Scheme

Figure 2 shows a flow chart of cooperative spectrum sensing with adaptive node selection.
For conventional cooperative spectrum sensing schemes, all SUs conduct spectrum sensing
and send their local sensing results to the fusion center. In the proposed scheme, low-RPSS
nodes are excluded from cooperation by reducing their selection probability.
Because the set of low-RPSS nodes is unknown until the initial sensing, initialize the
selection probability to be equal to Ph1 for all SUs, that is, p 1S (1) = p 2S (1) = · · · = p SN (1) =
Ph1 , which ensures that all SUs can conduct spectrum sensing in the initial stages. For the
k-th spectrum sensing, each node is selected based on its selection probability.
Let Di (k) denote the k-th local spectrum sensing decision for SUi . The local spectrum
sensing decision rule if SUi is selected is given by [7,10]:

1, if Yi (k) ≥ η,
Di (k) = (11)
0, if Yi (k) < η,
where η is the corresponding decision threshold, and {1} indicates the state of the channel
occupied by PUs and {0} implies the absence of a PU signal.

123
1884 T. Cui, K. S. Kwak

All selected SUs send their sensing decisions to the fusion center through the CRC. At
the fusion center, the global decision G is obtained by comparing the sum the local decisions
from all selected SUs with a threshold. For a given threshold λ at the fusion center, the k-th
global decision can be obtained by [7,10]:
 
1, if Di (k) ≥ λ,
G(k) =  (12)
0, if Di (k) < λ.
Then the fusion center broadcasts the global decision to SUs in the CR network through
the CRC. The selection probability is maintained based on local sensing information and the
global decision. There are two cases:
1) If G(k) = 0, then the selection probability of each SU has no change. The licensed
channel can be reused for SUs to transmit data for the rest of the k-th frame.
2) If G(k) = 1, then those nodes selected for cooperation execute this step based on (10).
The selection probability of unselected nodes retains the same value. All SUs keep silent
for the rest of the k-th frame to avoid interference for PUs.
Algorithm 1 outlines the steps of the proposed scheme for the k-th cooperative spectrum
sensing.

Algorithm 1 Cooperative spectrum sensing with adaptive node selection for CR networks
1: Node selection based on piS (k), i = 1, 2, . . . , N
2: Selected nodes perform spectrum sensing
3: Make local sensing results Di (k) and send sensing results to the fusion center
4: Obtain a global decision G(k) and broadcast to SUs through the CRC
5: if G = 1 then
6: selected nodes: maintain selection probability piS (k + 1) based on (10)
7: unselected nodes: piS (k + 1) = piS (k)
8: else
9: piS (k + 1) = piS (k)
10: end if

4 Performance Analysis

This section first provides formulations for the cooperative detection probability and the
cooperative false alarm probability for the proposed scheme and then analyzes the average
selection probability for a given SNR.

4.1 Cooperative Detection Performance

Assume that all SUs in the CR network adopt identical sensing duration τ and threshold η
such that the individual false alarm probability is equal to P f under hypothesis H0 . Suppose
that n ∗ SUs based on their selection probability are selected to perform spectrum sensing and
let λ = 1 (that is, the OR rule at the fusion center). Then the cooperative detection probability
Q d and the cooperative false alarm probability Q f can be written as [7,10]:
∗

n
Qd = 1 − (1 − Pdi ), (13)
i=1

123
Cooperative Spectrum Sensing with Adaptive Node Selection 1885

and
∗
Q f = 1 − (1 − P f )n . (14)

4.2 An Analysis of Selection Probability

Note that a conservative strategy is employed to adaptively reduce the selection probability
when a node fails to detect a PU signal. If the selection weight α is reduced, then the selection
probability of a low-RPSS node decreases sharply. The selection probability of a low-RPSS
node decreases and approaches zero such that the node is excluded from cooperation.
The selection threshold  controls the average number of selected nodes with reliable
detection. An increase in the value of  reduces the number of nodes with an opportunity
to be assigned with high selection probability. Denote  piS as the average probability to be
selected to perform spectrum sensing for SU with SNR γ . Then 
i i pi can be computed as
S

piS = Ph1 P{Yi ≥ η + } + Ph2 P{η ≤ Yi < η + } + Pl P{Yi < η}
    
η +  − M(1 + γi )σn2 η − M(1 + γi )σn2
= Ph1 Q  + Ph2 Q 
M(1 + 2γi )σn4 M(1 + 2γi )σn4
    
η +  − M(1 + γi )σn2 η − M(1 + γi )σn2
−Q  + Pl 1 − Q  , (15)
M(1 + 2γi )σn4 M(1 + 2γi )σn4
where Pl stands for the selection probability if SUi fails to detect a PU signal. From (7) and
(8), it can be seen that Pl ≤ α Ph1 . For the RPSS of a reliable node is high such that the
accumulated energy is not likely to be less than η. Therefore, the third part of  piS can be
ignored if the SNR γi is sufficiently high.
Proposition 1 If α ≤ Ph2 /Ph1 , then the average selection probability of an SU to perform
spectrum sensing is proportional to the strength of the received PU signal.
Proof Differentiating 
pi with respect to γ gives
S i
  
∂
piS Ph1 √ Mσ 4 (η+−A)2 Ph2 √
= √ Mσn2 B + (η +  − A) √ n e− 2B +√ Mσn2 B
∂γi 2π B B 2π B
  4
Mσn4 − (η−A)2 √ Mσ
+(η − A) √ e 2B − Mσn2 B + (η +  − A) √ n
B B
  4
(η+−A)2 P √ Mσ (η−A)2
×e− 2B Mσn2 B + (η − A) √ n e− 2B
l
−√
2π B B

Mσ 2   (η+−A)2
= √ n (Ph1 − Ph2 ) B + (η +  − A)σn2 e− 2B
2π B 3

(η−A)2
+(Ph2 − Pl )(B + (η − A)σn2 )e− 2B , (16)

where A = M(1 + γi )σn2 and B = M(1 + 2γi )σn4 . From the definitions of Ph1 and Ph2 in (8)
and (9), it is clear that Ph1 − Ph2 > 0. It can be easily obtained that B + (η +  − A)σn2 ≥
B + (η − A)σn2 = (η + Mγi σn2 )σn2 > 0, and if α ≤ Ph2 /Ph1 , then, together with (7),
∂
P − P ≥ P − α P > 0. Therefore, S > 0 if α ≤ P /P , and 
pi
h2 l h2 h1 ∂γi h2 h1pi is monotonically
S
increasing in γi if α ≤ Ph2 /Ph1 , which leads to Proposition 1.

123
1886 T. Cui, K. S. Kwak

Fig. 3 The average number of selected nodes versus the percentage of low PRSS nodes for different selection
weights α with N = 10, P f = 0.01, η = 15.4 dB, Ph1 = 1, Ph2 = 0.8 and  = 0

5 Simulation Results

This section presents the simulation results for the proposed cooperative sensing scheme with
adaptive node selection for N = 10, M = 10, P f = 0.01, Ph1 = 1 and Ph2 = 0.8. From
(4), the corresponding threshold η = 15.4 dB can be found for P f = 0.01. There are two
reasons why Ph1 = 1 is set. First, this is to ensure that all the SUs can participate in spectrum
sensing in the initial stages. Second, all nodes with best detection performance are involved
in cooperation. Here it is assumed that the average received SNRs of SUs at good locations
are uniformly distributed from −5 to 5 dB. The SNR of low-RPSS nodes is −20 dB. In the
proposed scheme, the selection weight indicates a decreasing rate of the selection probability
of low-RPSS nodes, whereas the selection threshold controls the average number of selected
nodes with high RPSS. Here it is shown how these two parameters affect the average number
of selected nodes and detection performance.
Figures 3 and 4 show the average number of selected nodes versus the percentage of low-
PRSS nodes. The results indicate that more nodes are excluded from performing spectrum
sensing with an increasing number of low-RPSS nodes in the CR network. In Fig. 3,  = 0 is
set to focus the analysis on the affect of the selection weight α. With the decreasing selection
weight α, the selection probability of low-PRSS nodes decrease rapidly, which reduces the
average number of nodes to be selected for spectrum sensing. In Fig. 4, an increasing in 
reduces the number of nodes assigned with high selection probability and thus reduces the
average number of SUs involved in cooperation. Note that even if  = 0 and there are no
low-RPSS nodes in the CR network, the average number of selected nodes cannot reach 10
based on the proposed scheme. This is reasonable because the detection probability of an SU
is less than 1. Therefore, if the SU fails to detect a PU signal when the global decision is the
presence of a PU signal, then its selection probability is less than 1 based on the proposed
scheme.

123
Cooperative Spectrum Sensing with Adaptive Node Selection 1887

Fig. 4 The average number of selected nodes versus the percentage of low-PRSS nodes for different selection
thresholds  with N = 10, M = 10, P f = 0.01, η = 15.4 dB, Ph1 = 1, Ph2 = 0.8 and α = 0.5

Fig. 5 The cooperative detection probability versus the percentage of low-PRSS nodes for different selection
weights α with N = 10, M = 10, P f = 0.01, η = 15.4 dB, Ph1 = 1, Ph2 = 0.8 and  = 0

Figures 5 and 6 illustrate the detection performance of the proposed schemes versus the
percentage of low-PRSS nodes. The cases of random selection and accurate node selection
are plotted as references. If the percentage of low-RPSS nodes is low, then the proposed
scheme shows almost the same detection performance as the cooperative spectrum sensing
with accurate node selection. As shown in Fig. 5, a decrease in the selection weight α produces
detection performance slightly worse than that of accurate node selection. As shown in Fig. 6,

123
1888 T. Cui, K. S. Kwak

Fig. 6 The cooperative detection probability versus the percentage of low-PRSS nodes for different selection
thresholds  with N = 10, M = 10, P f = 0.01, η = 15.4 dB, Ph1 = 1, Ph2 = 0.8 and α = 0.5

the performance of cooperative detection decreases with an increase in the selection threshold
. This is because fewer nodes are involved in detection with an increase in the selection
threshold . It is clear that the proposed scheme shows better detection performance than
conventional cooperative sensing schemes with random node selection, particularly when
there are more percentage of low-RPSS nodes in the CR network. This is because low-RPSS
nodes are less likely to have opportunities to cooperate based on a decrease in their selection
probability.

6 Conclusions

This study proposes a selection probability based cooperative spectrum sensing scheme with
adaptive node selection in a CR network. Low-RPSS nodes are excluded from cooperation
by reducing their selection probability. The proposed scheme shows almost the same sensing
performance as the cooperative spectrum sensing with accurate node selection.

Acknowledgments This research was supported by the MSIP (Ministry of Science, ICT and Future Plan-
ning), Korea, under the ITRC (Information Technology Research Center) support program (NIPA-2014-
H0301-14-1042) supervised by the NIPA (National IT Industry Promotion Agency).

References

1. Mitola, J., & Maguire, G. Q. (1999). Cognitive radio: Making software radios more personal. IEEE
Personal Communications, 6(4), 13–18.
2. Haykin, S. (2005). Cognitive radio: Brain-impowered wireless communications. IEEE Journal on Selected
Areas in Communications, 23(2), 201–220.
3. Urkowitz, H. (1967). Energy detection of unknown deterministic signals. Proceedings of IEEE, 55(4),
523–531.

123
Cooperative Spectrum Sensing with Adaptive Node Selection 1889

4. Digham, F. F., Alouini, M.-S., & Simon, M. K. (2007). On the energy detection of unknown signals over
fading channels. IEEE Transation on Communications, 55(1), 21–24.
5. Cabric, D., Mishra, S. M., & Brodersen R. W. (2004). Implementation issues in spectrum sensing. In
Proceedings of Asilomar conference on signal, systems and computers, Monterey, USA (pp. 772–776).
6. Sahai, A., Cabric, D., Hoven, N., Tandra, R., Mishra, S., & Brodersen, R. (2005). Spectrum sensing:
Fundamental limits and practical challenges. In Proceedings of the IEEE international symposium on
dynamic spectrum access networks, Baltimore, Maryland.
7. Liang, Y.-C., Zeng, Y. H., Peh, E., & Hoang, A. T. (2008). Sensing throughput tradeoff for cognitive radio
networks. IEEE Transactions on Wireless Communications, 7(4), 1326–1337.
8. Ma, J., Zhao, G., & Li, Y. (2008). Soft combination and detection for cooperative spectrum sensing in
cognitive radio networks. IEEE Transactions on Wireless Communications, 7(11), 4502–4507.
9. Ganesan, G., & Li, Y. (2007). Cooperative spectrum sensing in cognitive radiopart: Multiuser networks.
IEEE Transactions on Wireless Communications, 6(6), 2214–2222.
10. Letaief, K. B., & Zhang, W. (2009). Cooperative communications for cognitive radio. Proceedings of the
IEEE, 97(5), 878–893.
11. Quan, Z., Shuguang, C., & Sayed, A. H. (2008). Optimal linear cooperation for spectrum sensing in
cognitive radio networks. IEEE Journal of Selected Topics in Signal Processing, 2(1), 28–40.
12. Cui, T., & Kwak, K. (2013). Conservative cooperative spectrum sensing without CRC in CR network.
IEICE Communications Express, 2(4), 161–166.
13. Cui, T., & Kwak, K. (2013). Cooperative spectrum sensing with adaptive node selection for cognitive
radio networks. In Proceedings of the fifth international conference on ubiquitous and future networks
(ICUFN), Da Nang, Vietnam (pp. 506–510).
14. Selen, Y., Tullberg, H., & Kronander, J. (2008). Selection for cooperative spectrum sensing. In Proceedings
IEEE DySPAN, Chicago, USA (pp. 1–11).
15. Khan, Z., Lehtomaki, J., Umebayashi, K., & Vartiainen, J. (2010). On the selection of the best detection
performance sensors for cognitive radio networks. IEEE Signal Processing Letters, 17(4), 359–362.
16. Yuan, W., Leung, H., Chen, S., & Cheng, W. (2011). A distributed sensor selection mechanism for
cooperative spectrum sensing. IEEE Transactions on Signal Processing, 59(12), 6033–6044.

Taiping Cui received his M.S. degree in communication system from

Chongqing University of Posts and Telecommunications (CQUPT) in
2008. He is currently working towards his Ph.D degree under the super-
vision of Prof. Kwak in the graduate school of information technology
and telecommunications, Inha University. His research interests include
cognitive radio and cooperative communication.

123
1890 T. Cui, K. S. Kwak

Kyung Sup Kwak received his B.S. degree from Inha University,
Inchon, Korea, in 1977, his M.S. degree from the University of South-
ern California in 1981, and his Ph.D. degree from the University of Cal-
ifornia at San Diego in 1988 under Inha University and Korea Electric
Association Scholarship grants. From 1988 to 1989, he was a member
of the technical staff at Hughes Network Systems, San Diego, CA. He
worked at IBM Network Analysis Center at Research Triangle Park,
NC, from 1989 to 1990. Since then, he has been with Inha University,
Korea, as a professor. His research interests include multiple access
communication systems, mobile and satellite communication systems,
data networks, and wireless multimedia. Dr. Kwak has been a mem-
ber of the Board of Directors of the Korean Institute of Communication
Sciences (KICS) since 1994 and a director of IEEE Seoul Section.

123