You are on page 1of 4

1002 IEEE COMMUNICATIONS LETTERS, VOL. 16, NO.

7, JULY 2012

On the Performance of
Spectrum Sharing Cognitive Relay Networks with Imperfect CSI
Junjie Chen, Jiangbo Si, Member, IEEE, Zan Li, Member, IEEE, and Haiyan Huang

Abstract—In a spectrum sharing scenario, when the channel V PU receiver


Relay link
state information (CSI) of interference links is imperfect, the hr v Interference link
1
transmission of the secondary user (SU) may cause a harmful hr v
interference on the primary user (PU). To quantify the impact hsv N
r1
of imperfect CSI on the PU, we present a performance metric hsr
hr d
1

termed as interference probability and derive the exact closed- hsr


2
r2 2

s d
form expression for interference probability of the PU. It is shown
that the PU’s interference probability is always equal to 0.75
SU networks
hsr N
# Partial relay selection

when the CSI of interference links is imperfect. In order to rN


guarantee the PU’s interference probability below an acceptable
value, a back-off power control mechanism is adopted. Then,
under the PU’s interference probability constraints, the exact Fig. 1. System model.
outage probability of partial amplify-and-forward (AF) relay
selection scheme in the cognitive relay networks is derived.
Finally, numerical results are provided to validate our analysis.
it was shown that the outage probability of cognitive relay
Index Terms—Cognitive relay networks, spectrum sharing, networks is higher than that of conventional relay networks.
imperfect CSI, outage probability. However, all of the above-mentioned works assume that the
perfect channel state information (CSI) of interference links
I. I NTRODUCTION is available. Under this assumption, the PU’s QoS constraint
is translated into a receiver power constraint and the PU’s
C OGNITIVE radio is proposed as a promising approach
to improve the utilization of scarce radio frequency
spectrum. In the cognitive networks, the secondary users (SUs)
performance analysis is commonly neglected. When the CSI
of interference links is imperfect, the interference from the SU
are allowed to access the primary user’s (PU’s) spectrum using can be higher than the interference constraints. In this case,
spectrum underlay, overlay or interweave approaches [1]. In the PU’s QoS cannot be guaranteed anymore. Therefore, it is
particular, for underlay spectrum sharing cognitive networks, necessary to investigate the impact of imperfect CSI on the
the SU is allowed to share the PU’s spectrum as long as PU. On the other hand, since the SU must operate in low
the interference caused by the SU is below the maximum transmit power regime without causing a harmful interference
allowable interference level at the PU receiver. on the PU, the asymptotic analysis for partial amplify-and-
Recently, relay-assisted cognitive networks have been in- forward (AF) relay selection scheme utilizing the minimum
vestigated as a potential way to improve SU’s performance signal to noise ratio (SNR) among two hops as end-to-end
and the performance analysis of cognitive relay networks have SNR is inappropriate for underlay cognitive relay networks,
attracted significant interests in the research community. In [2], especially when the CSI of interference links is imperfect.
the authors obtained a tight lower bound on the outage prob- The exact outage performance of partial AF relay selection in
ability for underlay cognitive relay networks. With maximum underlay cognitive networks has not been derived yet. Hence,
transmit power limits, the exact outage probability of selection in this letter, we investigate the PU’s interference probability
decode-and-forward (DF) in cognitive relay networks was with imperfect CSI of the interference links. Then, under
derived in [3]. In [4], the authors investigated the capacity of the PU’s interference probability constraints, the exact outage
the SU in cognitive relay networks and proved that the reactive probability of partial AF relay selection scheme in underlay
DF scheme achieves the similar capacity of the proactive cognitive relay networks is derived.
DF scheme. In [5], the difference between cognitive relay
networks and conventional relay networks was studied and II. S YSTEM AND C HANNEL M ODEL
Manuscript received January 12, 2012. The associate editor coordinating Consider an underlay spectrum sharing cognitive relay net-
the review of this letter and approving it for publication was C. Sacchi. work, as shown in Fig. 1. The solid lines denote the relay links
This work was supported by the Major National Science and Technol-
ogy Projects (2010ZX0300600204), the National Natural Science Foun- and the dash dot lines denote the interference links. In the SU
dation of China (61072070), the Doctorial Programs Foundation of the network, source s communicates with destination d by one of
Ministry of Education (20110203110011), the 111 Project (B08038), the the relays R = {r1 , r2 , · · · , rN }. We assume there is no direct
Program for Changjiang Scholars and Innovative Research Team in University
(IRT0852), the Science and Technology New Star Program of Shaanxi link between s and d. The channels between any two nodes
Province (2011KJXX14), and the Fundamental Research Funds for the Central are Rayleigh fading and the instantaneous channel coefficient
Universities (K50510010029). between node i and node j is denoted by hij . Therefore, the
The authors are with the Integrated Service Networks Lab of Xidian 2
University, Xi’an, China, 710071 (e-mail: jjchen2009xd@163.com). channel gain |hij | is an exponentially distributed random
Digital Object Identifier 10.1109/LCOMM.2012.042512.120100 variable with mean λij . For analysis simplicity, we assume
1089-7798/12$31.00 
c 2012 IEEE
CHEN et al.: ON THE PERFORMANCE OF SPECTRUM SHARING COGNITIVE RELAY NETWORKS WITH IMPERFECT CSI 1003

that λsri = λsr , λri d = λrd and λri v = λrv for ∀ri ∈ R. relay selection protocol, the interference event occurs if and
Moreover, the background noise at the relays and the receivers only if one of the two cases below occurs:
are zero-mean Gaussian random variables with variance N0 . Case One: In the first time slot, SU source adjusts the
In the relay selection, we adopt the two time-slots AF partial transmit power Ps incorrectly and causes an interference
relay selection protocol. During the first time slot, all the relays higher than the interference threshold.
listen to the source and the relay rk with the best link s − rk Case Two: In the first time slot, SU source does not interfere
is selected according to rk = arg max:ri ∈R {γsri }, where γsri with the PU; however, in the second time slot, the selected
denotes the received SNR at the relay. In the second time slot, relay adjusts the transmit power Prk incorrectly and causes
the selected relay amplifies the received signal with the gain an interference higher than the interference threshold.
P
G2k = P |h rk|2 +N and forwards it to the destination, where Therefore, according to the total probability law, the inter-
s srk 0
Ps and Prk denote the transmit power of the source and the ference probability can be expressed as
selected relay, respectively.
PI = Pr(Ps |hsv |2 > Q)
In the underlay spectrum sharing cognitive relay networks,
the interference caused by the SU should be regulated below + Pr(Ps |hsv |2 ≤ Q) Pr(Prk |hrk v |2 > Q) (3)
the maximum allowable interference level at the PU. In
order to meet the interference constraints, the source and where rk denotes the selected relay. When the CSI of inter-
the selected relay adaptively adjust their transmit powers so ference links is perfect, the transmit powers of the SU source
that the interference constraint Q at the PU is satisfied, i.e. and the selected relay satisfy the interference constraints,
Ps |hsv |2 ≤ Q and Prk |hrk v |2 ≤ Q. However, in practice, i.e. Ps = |hQ |2 and Prk = |h Q |2 , so it is obvious that
sv rk v

obtaining the perfect CSI of interference links is difficult due PI is always equal to zero. When the CSI of interference
to a variety of reasons, such as channel estimation error, links is imperfect, Ps = Q 2 and Prk = Q
2 . Since
|ĥsv | |ĥrk v |
mobility and feedback delay. It is shown in [6] that the |hij |2 = |ĥij |2 , the PU’s interference probability is not always
true channel coefficient hij and the channel estimate ĥij can equal to zero anymore. In (3), the term Pr(Ps |hsv |2 > Q) can
be expressed as ĥij = ρhij + 1 − ρ2 ε, where ρ is the be calculated as [8]
correlation coefficient between ĥij and hij . ε is a circular  
symmetric complex Gaussian random variable with mean zero 2 Q 2
Pr(Ps |hsv | > Q) = Pr 2 |hsv | > Q
and variance λij /2. Therefore, the joint probability density |ĥsv |
function (PDF) of |hij |2 and |ĥij |2 can be given by [6]  ∞  ∞ − (1−ρx+y  √ 
e 2 )λsv
2ρ xy
− x+y
 √  = I0 dxdy
e (1−ρ )λij
2
2ρ xy 0 y (1 − ρ2 ) λ2sv (1 − ρ2 ) λsv
f|hij |2 ,|ĥij |2 (x, y) = I0 (1)  ∞ − λy    
(1 − ρ2 ) λ2ij (1 − ρ2 ) λij e sv 2y 2y
= Q1 ρ , dy
where I0 (·) is the zeroth-order modified Bessel function of 0 λsv (1 − ρ2 ) λsv (1 − ρ2 ) λsv
the first kind. Then, by performing integration in (1), the PDF = 0.5 (4)
2
of |ĥij | can be calculated as [7]
where Q1 (α, β) is the first order Marcum Q-function. Simi-
 ∞ − x+y
 √  larly, according to the partial relay selection protocol, we have
e (1−ρ2 )λij 2ρ xy
f|ĥ 2 (y) = I0 dx
ij |
0 (1 − ρ2 ) λ2ij (1 − ρ2 ) λij Pr(Prk |hrk v |2 > Q)
   
1 y N
= exp − (2) Q 2 2 2
λij λij = Pr 2 |hrk v | > Q ∩ |hsrk | > r ∈R,r max |hsri |
i =rk
k=1 |ĥrk v | i
   
N 2
Q|hrk v | 2 2
= Pr 2 > Q Pr |hsrk | > r ∈R,r max |hsri |
i =rk
III. I NTERFERENCE PROBABILITY OF THE PRIMARY USER k=1 |ĥrk v | i

In the underlay cognitive relay networks, the SU should (5)


adjust the transmit powers (Ps and Prk ) in the two time-  
2
slots relay transmission without causing a harmful interference where the first term Pr Q
2 |hrk v | > Q = 0.5. In
|ĥrk v |
on the PU. When the perfect CSI of interference links is  
2 2
obtained at the SU, the interference from the SU is regulated addition, note that the term Pr |hsrk | > max |hsri |
ri ∈R,ri =rk
below the interference threshold Q. However, when the CSI can be calculated as
of interference links is imperfect, the SU may adjust the  
2 2
transmit power incorrectly and cause an interference higher Pr |hsrk | > max |hsri |
ri ∈R,ri =rk
than the interference threshold Q. Thus, in order to quantify  ∞  
the impact of imperfect CSI of the interference links on the 2
= Pr z > max |hsri | f|h | (z) dz
2
PU, we present a performance metric termed as interference 0 ri ∈R,ri =rk srk

probability, PI , which is defined as the probability that the N −1  


N − 1 (−1)n 1
interference from the SU is higher than the interference = = (6)
n n+1 N
threshold Q. Since we adopt the two time-slots AF partial n=0
1004 IEEE COMMUNICATIONS LETTERS, VOL. 16, NO. 7, JULY 2012

Now, by substituting (6) into (5), we can obtain that addition, making use of the binomial theorem, it follows that
N
2 0.5     N −1
Pr(Prk v |hrk v | > Q) = = 0.5. Finally, substituting
N z
k=1
(4) and (5) into (3) yields the following conclusion Pr max |hsri |2 < z = 1 − exp −
ri ∈R,ri =rk λsr
N −1    
PI =0.75 (7) N −1 n nz
= (−1) exp − (12)
It is important to note that the PU’s interference probability n=0
n λsr
is always equal to 0.75 when the CSI of interference links  
Xγrk d
is imperfect. Therefore, in order to guarantee that the PU’s In (11), Pr X+γrk d +1 < γth can be calculated as
interference probability is below an acceptable value, we adopt  
a back-off transmit power control mechanism, i.e. the SU Xγrk ,d
Pr < γth
selects a reduced transmit power as the actual one: Ps  = ηPs X + γr k d + 1
and Prk  = ηPrk , where η denotes the back-off power control  ∞  
γth (y+1)
coefficient (0 ≤ η ≤ 1). Finally, using the same rationale, the = Fγrk d (γth ) + Pr X < fγrk d (y) dy
γth y − γth
PU’s interference probability by adopting the back-off power (13)
control is given by
1
2
PI  = Pr(Ps  |hsv | ≤ Q) Pr(Prk  |hrk v | > Q)
2 Note that Fγrk d (γth ) = 1 − λrv N0 γth and fγrk d (y) =
λrd ηε Q +1
2
+ Pr(Ps  |hsv | > Q)
λrv N0
λrd ηε Q
⎛ ⎞⎛ ⎞  2 , (13) can be rewritten as
λrv N0 y
λrd ηε Q +1
η − 1 3 η − 1
= ⎝1 + ⎠⎝ − ⎠  
2 4 2 Xγrk ,d 1
(η + 1) − 4ρ2 η 4 (η + 1) − 4ρ2 η Pr < γth = 1 − λrv N0 γth
(8) X + γrk ,d + 1 λrd ηε Q + 1
 ∞   λrv N0
ηε Qz (y − γth ) λrd ηε Q
IV. O UTAGE PROBABILITY OF THE SECONDARY USER + exp −  2 dy (14)
λsv N0 γth (y+1)
In this section, we investigate the exact outage probability
γth λrv N0 y
λrd ηε Q + 1
of the SU under the interference probability constraints. We
assume that the PU will request the SU to select a suitable By substituting (14) and (12) into (11), and after some
power control coefficient ηε so that the interference from the manipulations, the outage probability of the SU is given by
SU is higher than the interference threshold Q with a desired N −1    
probability value ε, i.e. PI  = ε. To meet the interference N −1 n 1 1
Pout = N (−1) + Bn
probability constraints, the SU should operate in the low n=1
n n + 1 λλrdNηεγQ + 1
rv 0 th
transmit power regime, which make the asymptotic analysis (15)
inappropriate. According to the partial AF relay selection
N −1  
protocol, the end-to-end SNR at the destination is given by N −1
where N n (−1)n / (n + 1) = 1 and Bn is given by
γsrk γrk d n=1
γend = (9)
γsrk + γrk d + 1   
∞ ∞ λrv N0
ηε Qz (y − γth ) λrd ηε Q
ηε Q|hsrk |
2
ηε Q|hrk d |
2
Bn = exp −  2 dy
where γsrk = and γrk d = . Then, the exact 0 λsv N0 γth (y+1) λrp N0 y
N0 |ĥsv |
2
N0 |ĥrk v |
2 γth
λrd ηε Q + 1
outage probability of the SU can be expressed as  
  1 (n + 1) z
N × exp − dz (16)
Pout = Pr γend < γth ∩ |hsrk |2 > max |hsri | 2 λsr λsr
k=1 ri ∈R,ri =rk
(10) By changing the order of integration and applying the change
of variable t = y − γth for (16), Bn can be rewritten as
where γth denotes the outage threshold. Since γsrk and |hsrk |2  ∞
are not independent, it is difficult to derive (10) directly. So t + β1 1
Bn = αn 2 dt (17)
we first derive the outage probability of the SU conditioned on 0 t + β 2n (t + β3 )
|hsrk |2 , and then perform the integration, which can be given
by where αn = (λsr ηε Q+λsv N0 γth (n+1))λrv , β1 = γth + 1, β2n =
λrd ηε Qλsv γth

N  ∞   λsv N0 γth (γth +1)(n+1)


λsr ηε Q+λsv N0 γth (n+1) and β3 = γth + λrv N0 . Then, using
λrd ηε Q
2
Pout = Pr max |hsri | < z the partial fraction expansion theorem, we have
0 ri ∈R,ri =rk
k=1
   β1 −β2n β1 −β2n β1 −β3

Xγrk d ∞
(β2n −β3 )2 (β2n −β3 )2 β2n −β3
× Pr < γth f|h |2 (z) dz (11) Bn = αn − 2 dt +
X + γr k d + 1 srk
0 t + β2n t + β3
(t + β3 )
  
where X = ηε Qz
and the cumulative density function
2 β1 − β3 1 β1 − β2n β2n
N0 |ĥsv |   = αn − 2 ln (18)
(CDF) of X can be given by FX (x) = exp − xληεsvQz β2n − β3 β3 (β2n − β3 ) β3
N0 . In
CHEN et al.: ON THE PERFORMANCE OF SPECTRUM SHARING COGNITIVE RELAY NETWORKS WITH IMPERFECT CSI 1005

0.8
0.75 irrespective of correlation coefficient ρ. In other words, the
0.7 interference from the SU will exceed the interference threshold
0.6
with a constant probability PI = 0.75 in the two time-slots
relay transmission as long as ρ = 1. In addition, for η = 1, it
Interference probability

0.5
is shown that the PU’s interference probability decreases with
0.4 the increase of correlation coefficient ρ.
Analytical results: K=1
Analytical results: K=0.8 Fig. 3 presents the exact outage probability of the SU
0.3 Analytical results: K=0.6
Analytical results: K=0.4
against the interference threshold among various number of
0.2
Simulation results: K=1 relays. It is observed that the exact outage probability of
Simulation results: K=0.8
Simulation results: K=0.6 the SU fits the simulation results well in low, medium and
0.1 Simulation results: K=0.4 high Q regime. As expected, the outage probability of the SU
0
decreases as the interference threshold Q grows. In addition,
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
U after careful inspection, it can be seen that an increase of the
Fig. 2. Interference probability versus ρ with N = 6 and Q = 5dB.
number of relays leads to a decrease of the outage probability
0
in low and medium Q regime, but in high Q regime, the outage
10
probability does not change with the increase of the number
of relays. Furthermore, for comparison, we also plot the PU’s
interference probability. It is obvious that an increase of power
10
-1 control coefficient can improve the outage performance of the
K=0.6
Outage probability

K=0.4 SU but meanwhile it leads to the performance degradation


of the PU. Therefore, with the imperfect CSI of interference
Analytical results: N=2 links, the tradeoff between the PU’s interference probability
Analytical results: N=4
10
-2
Analytical results: N=6 and the SU’s outage performance should be taken into con-
Simulation results: N=2 sideration.
Simulation results: N=4
Simulation results: N=6
PU's interference probability
VI. C ONCLUSION
-3
10
-5 0 5 10 15 20 25 In this letter, we present the interference probability and
Q/dB
find that the PU’s interference probability is always equal to
Fig. 3. Outage probability versus Q with ρ = 0.95. 0.75 when the CSI of interference links is imperfect. Thus,
Finally, by substituting (18) into (15), the exact outage prob- in order to guarantee the PU’s interference probability below
ability of the SU can be given by an acceptable value, we adopt the back-off power control
mechanism and derive the exact outage probability of partial
N −1  
1 N −1 AF relay selection in underlay cognitive relay networks.
Pout = λrd ηε Q +N (−1)n
λrv N0 γth + 1 n=0
n
   R EFERENCES
β1 − β3 1 β1 − β2n β2n
× αn − 2 ln (19) [1] A. Goldsmith, S. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum
β2n − β3 β3 (β2n − β3 ) β3 gridlock with cognitive radios: an information theoretic perspective,”
Proc. IEEE, vol. 19, pp. 894–914, May 2009.
[2] L. Luo, P. Zhang, G. Zhang, and J. Qin, “Outage performance for cog-
V. N UMERICAL AND SIMULATION RESULTS nitive relay networks with underlay spectrum sharing,” IEEE Commum.
Lett., vol. 15, no. 7, pp. 710–712, July 2011.
In this section, we present the numerical and simulation [3] Z. Yan, X. Zhang, and W. Wang, “Exact outage performance of cognitive
results to verify the validity of our analysis. The parameters in relay networks with maxinmum transmit power limits,” IEEE Commun.
Lett., vol. 15, no. 12, pp. 1317–1319, Dec. 2011.
the simulations are set as follows: λsr = 10dB, λrd = 10dB, [4] S. Sagong, J. Lee, and D. Hong, “Capacity of reactive DF scheme in
λsv = 2dB, λrv = 3dB, γth = 3 and N0 = 1. cognitive relay networks,” IEEE Trans. Wireless Commun., vol. 10, no.
The PU’s interference probability against correlation coef- 10, Oct. 2011.
[5] J. Lee, H. Wang, J. G. Andrews, and D. Hong, “Outage probability
ficient with different power control coefficient is illustrated of cognitive relays networks with interference constraints,” IEEE Trans.
in Fig. 2. It is observed that the theoretical results match the Wireless Commun., vol. 10, no. 2, Feb. 2011.
simulation ones well. As shown in Fig. 2, the PU’s interference [6] H. A. Suraweera, P. J. Smith, and M. Shafi, “Capacity limits and per-
formance analysis of cognitive radio with imperfect channel knowledge,”
probability is always equal to zero with perfect CSI of the IEEE Trans. Veh. Technol., vol. 59, no. 4, pp. 1811–1822, May 2010.
interference links (i.e., ρ = 1). However, in the presence of [7] I. Gradshteyn, I. Ryzhik, and A. Jeffrey, Table of Integrals, Series, and
imperfect CSI (i.e., ρ = 1), the PU’s interference probability Products. Academic Press, 2007.
[8] M. K. Simon and M. S. Alouini, “Some new results for integrals involving
is not equal to zero anymore. Furthermore, it is worth noting the generalized Marcum Q function and their application to performance
that, for η = 1, the PU’s interference probability is always evaluation over fading channels,” IEEE Trans. Wireless Commun., vol. 2,
equal to 0.75 when the imperfect CSI is obtained at the SU, no. 4, pp. 611–615, July 2003.

You might also like