You are on page 1of 6

Collaborative Spectrum Sensing in Real Cognitive

Radio Network
Tanumay Manna, Iti Saha Misra
Department of Electronics and Telecommunication Engineering
Jadavpur University
Kolkata, India
tanumay_manna@yahoo.co.in, itisahamisra@yahoo.co.in

Abstract—Collaborative spectrum sensing through coalition sensing [3, 4] is an attractive and effective approach to combat
games among secondary users (SU) in cognitive environment is multipath fading [5], shadowing and the receiver uncertainty
implemented to provide better performance than traditional problem.
fusion center based centralized cooperative spectrum sensing. In
this paper, collaborative spectrum sensing have been designed A traditional approach to cooperative spectrum sensing is
and implemented on Wireless Open-Access Research Platform by relying on a centralized fusion centre [6] (FC). However,
(WARP) to study its performance on real wireless cognitive SUs belonging to different service providers might need to
environment. The WARP has been used for energy detection interact with each other for collaboration without relying on
based spectrum sensing of primary user (PU), and a separate FC. Also a centralized approach can lead to significant
reporting channel has been designed to transfer the local sensing overhead and increased complexity. In [7], collaborative
bits to a cluster head (CH) selected through distributed coalition
spectrum sensing through coalitional games has been shown to
games. In addition, through WARP implementation the inherent
tradeoff has been shown between the gains in terms of provide a distributed alternative to cooperative sensing. In this
probability of detection of PU and the costs in terms of false model, CR users autonomously collaborate and self-organize
alarm probability. into disjoint independent coalitions. However, to the best of
our knowledge, collaborative spectrum sensing has not been
Keywords—Game Theory, Cooperative Spectrum Sensing, implemented on real wireless cognitive scenario usingWireless
Energy Detection, Fusion Centre, WARP Open-Access Research Platform (WARP) for experimental
verification. In this paper, we have implemented the
I. INTRODUCTION collaborative spectrum sensing in WARP board designing a
In recent years there has been an increasing demand of cognitive radio transmission.WARP is a scalable and
wireless services. However, traditionally spectrum is licensed extensible programmable wireless platform, to prototype
to operators and each system has to operate within a limited advanced wireless networks [17]. It provides an ideal platform
frequency band. As the unlicensed spectrum is already to setup real cognitive environment. Performance analysis of
congested there is a need to find new ways to exploit licensed the system has been done and results show that the probability
radio spectrum. Recent studies have showed that the actual of miss detection decreases at the cost of increased probability
licensed spectrum remains unoccupied for large periods of time of false alarm.
[1] termed as ‘white spaces’. Cognitive Radios(CRs) can be
used to sense these ‘white spaces’ and the Secondary Users Collaborative spectrum sensing consists of three-stages-
(SUs) can share the spectrum with the licensed users without local sensing, adaptive coalition formation and coalitional
causing harmful interference to the Primary User (PU), thereby sensing. The rest of the paper discusses these steps along with
providing higher data throughput with respect to different the implementation details in wireless scenario on WARP. In
applications [2]. Section II, the system model is illustrated, while local spectrum
sensing through energy detection is discussed in Section III. In
The detection performance in spectrum sensing [3] is Section IV, collaborative spectrum sensing on WARP is
crucial to the performance of both PUs and SUs. Detection outlined, followed by the discussion of WARP implementation
performance can be evaluated with the help of probability of in Section V. Section VI gives the results followed by some
false alarm (PF), and probability of miss detection (PM). It is concluding remarks in Section VII.
required to minimize PM subject to the constraint of PF.
CRs experience multipath and shadow fading and may not II. SYSTEM MODEL
detect the PU’s signal. So, a CR which failed the detection of A simple system has been considered where there is one
PU may begin transmission and interfere with the PU receiver. PU and two SUs. Both SUs perform energy detection for local
However, it is unlikely that all CR users experience the same spectrum sensing. As shown in Fig. 1, SU2 is subjected to
amount of fading; most of them would detect the correct PU shadow fading. A reporting channel using BPSK modulation
signal. By cooperating with each other, the sensing results may has been designed for reliable communication, which enables
be communicated to other users. So, cooperative spectrum passing of detection bits and SNR information among the SUs.

978-1-4673-6217-7/13/$31.00 2013
c IEEE 137
Reporting Channel Primary User
Secondary User-1 Spectrum
Spectrum
(SU 1) |X(f)|

Reporting Channel
(BPSK)
Primary 30 KHz 9.5 MHz
Frequency
Transmitter
Secondary User-2 5 KHz 18MHz
(SU 2)
Shadow Element Fig. 2. Frequency Spectrum

Fig. 1. System Setup Ȝi

The frequency spectrum for the coexistence of primary T yi(t)


BPF ()2 Decide
signal and reporting channel in case of WARP implementation xi(t) ³ H0 or H1
is shown in Fig. 2. 0

The SUs calculate the SNR of their received primary Fig. 3. Block diagram of energy detector
signals, and use them to calculate their respective energy
detection threshold levels and obtain their corresponding The energy collected by SU i during the detection time is,
probabilities of detection (PD), miss-detection (PM) and false
alarm (PF) as explained in the next section. 2TW
Yi = ¦ | xi [n] | 2 (2)
The SUs calculate the SNR of the reporting channel to n =1
determine the probability of error for BPSK modulation. The
SU that has the higher SNR determine whether the total utility where, TW is the Time-Bandwidth product of the energy
of the system is increased due to coalition. The utility function detector.
[7] represents the gain in terms of increase in PD taking the cost Following the work of Urkowitz [4], Yi may be shown to
of increase in probability of false alarm into account. If there is have the following distribution,
an increase in utility value then the SU with higher SNR serves
as the cluster head (CH) and inform the other SU that the ­° χ 2 , H 0
reporting channel is available for transfer of local detection Yi ~ ® 2 2 TW (3)
bits. According to our system model SU1 serves as CH and °̄ χ 2 TW ( 2γ ) , H 1
SU2 sends its detection bit to SU1. The SU 1 applies a fusion
rule to determine whether the PU is occupying the spectrum. where, χ 2TW
2
and χ 22TW (2γ ) denote central and non-central chi-
SU 1 then informs SU 2 of its decision via the reporting square distributions respectively with a non-centrality
channel. parameter of 2γ for the latter. Let 2TW be denoted by N .
III. BACKGROUND KNOWLEDGE: ENERGY DETECTION According to central limits theorem, if TW > 125 , Yi can
Energy detection is the most common detection mechanism be approximated as a Gaussian random variable with
in cooperative sensing [8]. The goal of spectrum sensing is to distribution
decide between the following two hypotheses,
­ N ( N σ 02i , 2 N σ 04i ) , H 0
­ ni (t ), H 0 Yi ~ ® (4)
xi (t ) = ® (1)
¯ N ( N σ 1i , 2 N σ 1i ) , H 1
2 4

¯hi (t ) s (t ) + n (t ), H 1
Here, xi(t) is the SU i received signal, s(t) is the primary where, σ 0i2 denotes the variance of white noise, σ 1i2 denotes
signal, ni(t) is the Gaussian white noise and hi(t) is the channel the variance of yi(t) under H 1 . It can be shown that,
gain of a wireless fading channel between SU i and the PU.
hi(t) varies over time, hence before the start of each iteration σ 12i = σ 02i (1 + γ i ) (5)
the channel gain needs to measured, and then assumed constant
for the duration of the iteration, which in this paper is later For a zero mean white noise, noise power is
shown to be equal to 0.384ms.
1 N 2
The principle of energy detector is that the input signal xi(t) N 0i = ¦ ni (t ) =E [ni 2 (t )] = Var [ni (t )] = σ 02i
N n =1
(6)
is firstly passed through a band pass filter, having bandwidth
W, and then sampled. The energy of the signal is calculated and
Similarly received power, S 0i = Yi − N 0 i (7)
compared to the set threshold as shown in Fig. 3.

138 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)
Therefore, received SNR, QmS = ∏[ PM , i × (1 − Pe , i , r ) + (1 − PM , i ) × Pe, i , r ] (15)
i∈S
S 0i
γi = (8) Q f S = 1 − ∏ [(1 − PF , i ) × (1 − Pe , i , r ) + PF , i × Pe , i , r ] (16)
N 0i i∈S

Probability of false alarm is given by where, PF , i , PM , i and Pe , i , r are the probabilities of false-alarm,
λi − Nσ 02i miss-detection and reporting channel error for user i, i∈S.
PF i = P (Yi > λi / H 0 ) = Q ( ) (9)
2 Nσ 04i The utility of a coalition S is defined as [7]

Probability of detection is given by v ( S ) = Q dS − C (Q fS ) (17)

λi − Nσ 12i where, QdS and Q fS are the detection and false alarm
PD i = P(Yi > λi / H 1 ) = Q ( ) (10)
2 Nσ 4 probability respectively of coalition S. QdS is given by,
1i
QdS = 1 − QmS (18)
Probability of miss-detection is given by
and C (Q f S ) is defined by a logarithmic barrier penalty function
PM i = 1 − PD i (11) given by [13]. The utility function is such that it will increase
for an increase in detection probability at the cost of increasing
The detection threshold can be calculated following Likelihood false alarm probability.
Ratio Test (LRT) given by
­ § § Q ·2 ·
H1 ° − α 2 ⋅ ln¨1 − ¨ f S ¸ ¸, if Q f S < α (19)
C (Q f S ) = ® ¨ ¨© α ¸¹ ¸
P (Yi / H 1 ) > P ( H 0 ) . (12) ° © ¹
P (Yi / H 0 ) < P ( H 1 ) ¯+ ∝ , if Q f S ≥ α
H0 where, Į is a false alarm constraint per coalition (per SU). In
In this paper, the primary traffic distribution is assumed to this paper Į is taken to be equal to 0.1 since false alarm
be exponential ON-OFF signal with the primary occupancy probability greater than 0.1 in unacceptable for system design.
probability of 20%. So, P ( H 0 ) = 4 and P ( H 1 ) = 1 . A. Merge and Split Rule:
5 5 In adaptive coalition formation phase, the SUs through an
Equation (12) can be solved resulting in the following iteration of merge and split rules [7] assess whether to share
threshold, their sensing results with nearby coalition.
When SU 2 requests a merge to the cluster head, SU 1, in
(1 + γ i ) N 0 [ Nγ i + N 2 γ i + 4 N γ i (2 + γ i ) ln{4(1 + γ i )}]
2

λi = (13) order to form a cluster; SU 1 evaluates the missing and false


γ i (2 + γ i ) alarm probabilities of the coalition. Simplifying (15) and (16)
we get,
where, Ȝi is calculated by every SU i individually, using this Ȝi
the corresponding PFi, PDi and PMi can be obtained using (9), QmS = [ PM 2 × (1 − Pe ,r ) + (1 − PM 2 ) × Pe ,r ] × PM 1 (20)
(10), (11) respectively.
Q f S = 1 − (1 − PF 2 ) × (1 − Pe ,r ) × (1 − PF 1 ) (21)
IV. COLLABORATIVE SPECTRUM SENSING
Since this is a non-transferable utility game, the utility of
Collaborative spectrum sensing in [7] has been modeled as each SU is equal to the utility of the coalition to which SU i
a non-transferrable (M, v) coalitional game, where M is the set belongs [14].
of cooperating CR users and v is the utility function. The SUs
share their SNR information via the reporting channel. The SU Initially there are two coalitions each consisting of one SU.
having the lowest miss-detection probability and hence best The collection of these two coalitions is denoted by
SNR is chosen as the cluster head for the other SUs in the R={Ri|i=1,2}. The merging of the two SUs results in a new
coalition. With reference to Fig. 1, here SU1 becomes the collection S={Si|i=1,2}. Ri and Si denote the SUi in collections
cluster head (CH) as it does not experience shadow fading. R and S respectively. Now the merge will take place if [15]
2 2
The probability of error in the reporting channel, inside a
coalition S, between a SU i∈S and the cluster head is given by, ¦ v ( R ) <¦ v ( S )
i =1
i
i =1
i
(22)

1
Pe , i , r = erfc ( γ i , r ) (14) It means that the merge process takes place if coalition set
2 S is preferred to coalition set R. Merging will take place if the
total utility of the system increases as a result of coalition
where, Ȗi,r is the SNR of the reporting channel between cluster formation. However, if the condition of (22) is not satisfied
head and SU i. The missing and false alarm probabilities of a then the CH will not go ahead with the merge process and the
coalition S, as given in [12] are,

2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI) 139
reporting channel will not be used to transfer of individual SU RADIO_3:= Primary transmitter
detection bits. RADIO_1:= SU 1 Trans receiver
RADIO_2:= SU 2 Trans receiver
Even after coalition formation through merging, the CH Define primary & reporting
periodically checks the validity of (22). This periodic checking channel by appropriate BPF.
also takes care of the mobility aspect of SUs. If it is not
satisfied then the coalition will split into individual SUs, as
then coalition set R will be preferred to coalition set S.
After a merge is formed, the SU i sends its detection bit to RADIO_3 generates primary
cluster head, and the cluster head uses OR rule [16] to perform signal
decision fusion. Let, ui be the local decision of CR user i and u
be the cooperative decision by the cluster head, ui, u ∈ {0,1},
RADIO_1 and Radio_2 obtain
and a ‘1’and a ‘0’ indicates PU’s presence (H1) and absence
SNRs γ 1 and γ 2 then calculate
(H0), respectively. According to the OR rule u=1, if ui=1, for
any i. The cooperative decision is then communicated to other thresholds λ1 and λ 2 respectively.
SUs of the coalition by the cluster head by BPSK modulation
via reporting channel.

V. WARP IMPLEMENTATION RADIO_1 and RADIO_2


A single WARP v2 kit has been used in order to calculate individual PF, PD, PM.
implement collaborative spectrum sensing. A single kit
consists of four trans-receiver radios, out of which 3 has been
used. RADIO_3 serves as the primary transmitter (Tx mode),
RADIO_1 as the SU 1 trans-receiver and RADIO_2 as the SU Local Sensing by Energy
2 trans-receiver. Detection

RADIO_3 primary signal is generated using exponential


ON_OFF distribution. The ON signal has a mean time of 1ms RADIO_1 calculates γ r and Pe ,r
and the OFF signal a mean time of 4ms. At first both
RADIO_1 (SU 1) and RADIO_2 (SU 2) are configured to
operate as receivers (Rx mode) and they calculate their
primary received SNR γ 1 and γ 2 respectively using (6), (7)
Equation (22) No
and (8). With the help of these SNR values RADIO_1 and use_reporting_channel=
RADIO_2 evaluate their respective energy detection threshold satisfied? 0
levels λ1 and λ 2 using (13), and perform local sensing by
energy detection in Rx mode and obtain their respective
detection bits. The SUs then calculate their local detection Yes
probabilities of false alarm, detection and miss-detection using use_reporting_channel = 1
(9), (10) and (11) respectively. RADIO_2 is switched to Tx
mode and sends signal to RADIO_1 via reporting channel and
RADIO_1 calculates the SNR of the reporting channel, γ r and
RADIO_2 sends detection bits to
the corresponding probability of error ( Pe ,r ) using (14). RADIO_1
RADIO_1 then implements merge and split rule by checking
the condition of (22), if it is satisfied then the reporting
channel will be used to send detection bits from RADIO_2 to
RADIO_1. RADIO_1 informs this by sending a control bit RADIO_1 uses OR rule to obtain
use_reporting_channel=’1’. If (22) is not satisfied then fusion result.
reporting channel is not used. When
use_reporting_channel=’1’, RADIO_2 will send its detection
bit by BPSK modulation to RADIO_1 through the reporting RADIO_1 transmits the fusion
channel in Tx mode. RADIO_1 then employs OR rule to result to RADIO_2
obtain the fusion results. It switches to Tx mode and transmits
the fusion bit to RADIO_2, which changes to Rx mode to
receive it.Fig.4 gives a flow chart of the sequence of events.

Fig. 4. Flow chart for WARP implementation

140 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)
VI. RESULTS
Each trans-receiver node of WARP board consists of
buffers each having size of 214 samples. The WARP board is
clocked at 40MHz. The detection time is such that all the
samples of the received signal in one buffer are used for
sensing the primary spectrum. The received signal in the
buffers of the WARP board trans-receiver nodes are transferred
and plotted in MATLAB.
Some of the experimental test-bed parameters used are
described in Table I.

TABLE I. EXPERIMENTAL TEST-BED PARAMETERS


Features Values
Detection time 0.384 ms
Sampling frequency 40 MHz
Primary carrier frequency 9.5 MHz
Primary signal bandwidth 18 MHz Fig. 6. Primary user transmission and SU 2 received signal with SU 2
detection bit and cluster head decision
Reporting channel 30 KHz
centre frequency
Fig. 7 and Fig. 8 show the variation of PM, PF with
Reporting channel bandwidth 5 KHz
received SNR of SU 2, respectively.The SNR is varied by
varying the PU transmission power. Due to shadowing,the
Fig.5(a) represents the PU transmission on CR received SNR of SU 2 is very low. Fig. 7 shows the distinct
environment while Fig.5(b) shows received signal by SU 1 advantage of collaborative spectrum sensing as opposed to
during PU transmission. Since SU 1 does not suffer individual local sensing. It is seen that the probability of miss
shadowing, it can successfully detect PU presence. The PU detection of SU 2 decreases in case of cooperative sensing.
signals are un-deterministicin nature, the received signal is However, there is a tradeoff, the probability of false alarm
detected with the help of energy detection threshold, and the increases but it is always less than 0.1 as evident from Fig. 8.
detection bits are shown in Fig.5(c) which shows the presence
of PU transmission correctly.
In Fig.6(b) it can be seen that, due to shadowing the
received signal of SU 2 is severely attenuated. Since the
received energy is less than the detection threshold, SU 2
cannot detect the primary signal, so the sensing bit of SU 2 in
Fig.6(c) is always ‘0’. Fig.6(d) shows SU 2 using the result of
collaboration via reporting channel from SU 1. SU 2 becomes
aware of PU presence, thus providing correct decision of PU
presence as depicted in Fig.6(d).

Fig. 7. PM vs SNR of SU 2 for individual Spectrum sensing and collaborative


spectrum sensing

Fig. 8. PF vs SNR of SU 2 for individual spectrum sensing and collaborative


spectrum sensing
Fig. 5. Primary user transmission and SU 1 received signal with detection bit

2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI) 141
Fig. 9 and Fig. 10 show the variation of PM, PF with Results also show that miss detection probability decreases due
received SNR of SU 1. For SU 1 the effect of collaborative to collaborative sensing while keeping probability of false
sensing is less pronounced as it does not experience shadow alarm within tolerable limit of 0.1.
fading, hence individual and collaborative spectrum sensing
ACKNOWLEDGMENT
performances almost have the same value. The received SNR
of SU 1 is much higher than SU 2, however for comparison The authors deeply acknowledge the support of DST, Govt.
purposes the SNR of SU 1 is kept below 1.5 dB by reducing of India for infrastructure facility in the form of FIST 2007
PU transmission power. Since SU 1 does not suffer Project, ETCE, Jadavpur University and CSIR, Govt. of India
shadowing, for higher PU transmission, SNR increases and PM for financial assistance.
for SU 1 falls to less than 0.1.
REFERENCES
The decrease in PM in both SU 1 and SU 2 can be [1] Federal Communications Commission, “Spectrum Policy Task Force,”
explained from (15). In a coalition, Qms decreases as the Rep. ET Docket no. 02-135, Nov. 2002.
number of members in a coalition increases. In our scenario [2] T. Chakraborty, I. S. Misra and S. K. Sanyal, “Proactive QoS
when a merge is formed the number of members in a coalition Enhancement Technique for Efficient VoIP Performance over Wireless
increases from one to two. Similarly, in (16) the probability of LAN and Cognitive Radio Network,” in Journal of Networks, vol. 7,
false alarm increases as the members in a coalition increase. no. 12, pp. 1925-1942, Dec 2012.
The results can be extended to design the system with higher [3] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms
number of PUs and SUs by incorporating more WARP boards forcognitive radio applications”, Communications Surveys
Tutorials,IEEE 11 vol. 1, 2009, pp. 116–130.
in the test-bed.
[4] D. Cabric, S. Mishra and R. Brodersen, “Implementation issues
inspectrum sensing for cognitive radios”, Proc. of AsilomarConf.on
Signals, Systems, and Computers, vol. 1, 2004, pp. 772–776.
[5] B. Bhattacharyya, I.S. Misra and S.K. Sanyal, “Analysis and
Optimization of Error Performance of a WiMAX Transceiver using
Novel Adaptive Cyclic Prefix Strategy under AWGN and Rayleigh
Fading Scenario,” Journal of Networks, vol. 7, no. 12, pp. 1952-1966,
Dec. 2012 .
[6] A. Ghasemi and E. Sousa, “Collaborative spectrum sensing for
opportunistic access in fading environments”, Proc. of IEEE
DySPAN2005, 2005, pp. 131–136.
[7] W. Saad, Z. Han, M. Debbah, A. Hjorungnes and T. Basar, “Coalitional
games for distributed collaborative spectrum sensing in cognitive radio
networks”, Proc. of IEEE INFOCOM 2009, 2009, pp. 2114–2122.
Fig. 9. PM vs SNR of SU 1 for individual spectrum sensing and collaborative [8] I.F. Akyildiz, W.-Y. Lee and K.R. Chowdhury, “CRAHNs: cognitive
spectrum sensing radioad hoc networks”, Ad Hoc Networks, vol. 7 (5), 2009, pp. 810–
836.
[9] H. Urkowitz, “Energy detection of unknown deterministic signals,”
Proceedings of IEEE, vol. 55, pp. 523–231, April 1967.
[10] F.F. Digham, M.-S. Alouini and M.K. Simon, “On the energy detection
of unknown signals over fading channels,” IEEE Transactions on
Communications, vol. 55 (1), 2007, pp. 21–24.
[11] E. Visotsky, S. Kuffner and R. Peterson, “On collaborative detection of
tv transmissions in support of dynamic spectrum sharing,” Proc. Of
IEEE DySPAN 2005, 2005, pp. 338–345.
[12] W. Zhang and K. B. Letaief, “Cooperative spectrum sensing with
transmit and relay diversity in cognitive networks,” IEEE Trans.
Wireless Commun., vol. 7, Dec. 2008, pp. 4761–4766.
[13] S. Boyd and L. Vandenberghe, “Convex Optimization,” New York,
USA: Cambridge University Press, Sep. 2004.
Fig. 10. PF vs SNR of SU 1 for individual spectrum sensing and collaborative
[14] R. B. Myerson, “Game Theory, Analysis of Conflict,” Cambridge, MA,
spectrum sensing
USA: Harvard University Press, Sep. 1991.
[15] K.R. Apt and A. Witzel,"A generic approach to coalition formation
VII. CONCLUSION
(extended version),” arXiv:0709.0435v1[cs.GT], Sep.2007.
In this paper, collaborative spectrum sensing has been [16] W. Zhang, R. Mallik and K. Letaief, “Optimization of cooperative
tested in wireless environment using WARP nodes as trans- spectrum sensing with energy detection in cognitive radio networks”,
receivers. The reporting channel has been designed without IEEE Transactions on Wireless Communications, vol. 8 (12),2009, pp.
compromising the PU transmission. Implementation results 5761–5766.
show that SUs under shadow fading obtain correct PU [17] (2012) Rice University WARP homepage. [Online]. Available:
spectrum occupancy data by collaborating with other SUs. warp.rice.edu

142 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)

You might also like