1

Outage Probability of Cognitive Relay Networks with Interference Constraints

Jemin Lee, Hano Wang, Jeffrey G. Andrews, Senior Member, IEEE, and Daesik Hong, Senior Member, IEEE

AbstractThis paper evaluates the outage probability of cognitive relay networks with cooperation between secondary users based on the underlay approach, while adhering to the interference constraint on the primary user, i.e., the limited amount of interference which the primary user can tolerate. A relay selection criterion, suitable for cognitive relay networks, is provided, and using it, we derive the outage probability. It is shown that the outage probability of cognitive relay networks is higher than that of conventional relay networks due to the interference constraint, and we quantify the increase. In addition, the outage probability is affected by the distance ratio of the interference link (between the secondary transmitter and the primary receiver) to the relaying link (between the secondary transmitter and the secondary receiver). We also prove that cognitive relay networks achieve the same full selection diversity order as conventional relay networks, and that the decrease in outage probability achieved by increasing the selection diversity (the number of relays) is not less than that in conventional relay networks. Index Termscognitive relay networks, outage probability, selection diversity.

I. I NTRODUCTION In cognitive radio networks, unlicensed users (secondary users) are permitted to use the licensed band so long as they protect the data transmission of the licensed user (primary user) using spectrum underlay, overlay and interweave approaches [1]. In the underlay approach, the secondary user is allowed to use the spectrum of the primary user only when the interference from the secondary user is less than the interference level which the primary user can tolerate. Hence, the transmission power of the secondary user is constrained not to exceed the interference level. In the overlay approach, the secondary user uses the same spectrum concurrently with the primary user while maintaining or improving the transmission of the primary user by applying sophisticated signal processing and coding [1]. Otherwise, using the interweave approach, the secondary user utilizes the spectrum not currently being used by the primary user, known as a spectrum hole, after performing detection on the spectrum. Relay networks have been proposed as a way to enhance the total throughput and coverage of wireless networks [2]. The advantage of relay networks lies in reducing the overall path loss achieved by using a relay between a source and a destination. Inspired by cognitive radio and relay networks, cognitive relay networks (CRN) have recently been investigated as a potential way to improve secondary user throughput using one of two approaches: cooperation between primary

and secondary users [3], and cooperation between secondary users [4]-[7]. For cooperation between secondary users, approximate [4] and exact [5] outage probabilities of cognitive relay networks have been presented considering the impact of the spectrumsensing accuracy in overlay coexistence. Also, it is shown that full diversity cannot be achieved under imperfect spectrum sensing [4]. Decentralized schemes for transmit power allocation for secondary relays have been proposed to minimize the overall transmit power or to maximize the received signal to noise-interference ratio (SINR) in [6]. In addition, the performance of cognitive relay networks in licensed and unlicensed bands is compared to that of conventional relay networks [7]. Although the location of the relay and appropriate relay selection are important [8], no prior work has considered an appropriate relay selection procedure for cognitive relay networks. Our goal in this paper is to evaluate the performance of cognitive relay networks using cooperation between secondary users based on the underlay approach while adhering to the interference constraint on the primary user. We rst provide a relay selection criterion suitable for cognitive relay networks and then derive the outage probability of cognitive relay networks. Our main contributions are to show that: 1) the outage performance of cognitive relay networks is affected by the distance ratio of the interference link (from the secondary transmitter to the primary receiver) to the relaying link (between the secondary transmitter and the secondary receiver); 2) the outage probability of cognitive relay networks consists of (or can be divided into) the conventional relay network outage probability and the increase in outage probability resulting from the interference constraint; and 3) cognitive relay networks achieve the same relay selection diversity order as conventional relay networks, but the decrease in outage probability of cognitive relay networks achieved by increasing the selection diversity is equal to or greater than that in conventional relay networks. II. S YSTEM M ODEL The cognitive relay network model of interest to us is described in this section. In this model, a primary user coexists with secondary users as shown in Fig. 1. In the gure, P URX , SUS , and SUD represent a primary receiver, a secondary source, and a secondary destination, respectively. In addition, there are K potential secondary user relays which are denoted by SUk (k = 1, , K ). The relay mode is regenerative mode,

SUS

hSkP

hkP
SUk PURX

is called the interference constraint. In addition, there is also where the maximum transmission power constraint, Pk P P is the maximum transmission power [6]. By those two constraints, the transmission power constraint of SUk becomes as follows [6], [9]: } { I . (1) Pk min 2, P |hkP | If the secondary source transmits data with the help of the k th potential relay in dual-hop, known to be the most efcient multi-hop transmission with respect to system capacity [8], the capacity of the secondary user based on a unit bandwidth becomes as follows [8]: { ( ) ( )} 2 2 PSk |hSk | 1 Pk |hkD | Ck = min log2 1 + , log2 1 + . 2 No No (2) In (2), 1/2 is from the dual-hop transmission in two time slots and No is the noise variance. In addition, any interference from the primary transmitter is assumed to be neglected3 . III. O UTAGE A NALYSIS OF C OGNITIVE R ELAY N ETWORKS This section present an appropriate relay selection criterion for cognitive relay networks. Also, we analyze the relay transmission performance in terms of the outage probability, and compare it to that of conventional relay networks. A. Cognitive Relay Networks We begin this subsection by reviewing an existing relay selection criterion and conventional relay network outage performance. A variety of criteria for relay selection have been proposed as a way of maximizing capacity or minimizing outage probability in conventional relay networks [15]-[17]. The max-min criterion, which maximizes the minimum of signal-to-noise ratios (SNRs) of the source-relay link and relay-destination link, has proven optimal for this purpose [16]. Hence, a relay selection criterion in conventional relay networks is dened as follows [16], [17]: { } 2 2 l = arg max min |hSk | , |hkD | , (3)
k

hkD
Relaying links Interference links

SU D

Fig. 1. System model of cognitive relay networks; P URX , SUS , SUD , and SUk represent a primary receiver, a secondary source, a secondary destination, and the kth potential relay, respectively.

so a relay decodes the received data and then forwards it to a secondary destination. Let M = {S, D, 1, 2, , K } be the set of secondary user indexes. We assume that a primary receiver occupies multiple primary spectra (frequency channels) spaced with greater frequency separation than coherence bandwidth2 , and each relay uses different frequency channel among the primary spectra. On the ith frequency channel used by SUi , the instantaneous channel of the link between SUi and SUj , the relaying link in Fig. 1, is represented by hij . The channels of the links from SUS and SUi to P URX , the interference links in Fig. 1, are denoted by hSi P and hiP , respectively. The channels are assumed to consist of path loss /2 and an independent fading effect as hij = ij dij (i, j M) where and dij denote the pathloss exponent and the distance between two users, respectively. The fading coefcient, ij , is a complex Gaussian random variable with unit variance. 2 Hence, the channel gain |hij | (i, j M) is an exponential distributed random variable with the mean value 1 and [/ij , ] 2 the average channel power is dened as 1/ij = E |hij | =
d ij where E [] denotes expectation. Since the link distance of hSi P is dSP for all i M, Si P is the same as SP . The 2 perfect channel information such as |hiP | at the secondary 1 users is also assumed . In the underlay approach of this paper, the secondary user can utilize the primary users spectrum so long as the interference it generates on the primary receiver remains below ), which is the maximum tolerable the interference threshold (I interference level at which the primary user can still maintain reliable communication [1]. For this reason, the secondary |hkP |2 , users transmission power is constrained as Pk I/ where Pk is the transmission power of SUk ; this constraint
1 The channel information of |h |2 can be obtained at secondary users by iP direct reception from a primary receiver using some pilot signals [9], [10] or by using the band manager, which can exchange the channel information between primary and secondary users [10]. The channel information estimator without feedback could be also used by the methodology in [11]. 2 This can be compatible with the assumption that multiple primary receivers use different frequency channels, and each channel is occupied by a primary receiver [12].

where l is the selected relay index in conventional relay networks. Thus, the relay is chosen based on the channel gains of the relaying links. Hence, under the assumption that all users use the maximum transmission power, the outage o probability based on the k th relay is dened as PConv,k = ( ) (Sk +kD )UT /P 2C T 1e where U = N 2 1 and C
T o T

is the target bandwidth efciency [16]. Hence, the outage probability in conventional relay networks when the relay selection is applied becomes the following:
o PConv = K { } 1 e(Sk +kD )UT /P . k=1
3 This can be possible if the primary transmitter is located far away from the secondary users [13], or the interference is represented by the noise term under an assumption that the primary transmitters signal is generated by random Gaussian codebooks [14].

(4)

o From (4), we can see that PConv is solely a function of the distances of relaying links. Moreover, minimizing the outage probability is equal to minimizing Sk + kD , and the outage probability has a symmetric form per Sk and kD . Since the selection criterion in (3) is to minimize the outage probability, we can deduce from (4) that the relay located at exactly the midpoint between the source and destination would be the one most likely to be selected in average. In contrast to conventional relay networks, an additional transmission power constraint, the interference constraint, exists in cognitive relay networks, making the conventional relay selection criterion inappropriate. For instance, a relay which has good channel conditions on its source-relay and relay-destination links is likely to be selected in conventional relay networks, but may be an inappropriate relay in cognitive relay networks if the relay generates a lot of interference at the primary receiver. Hence, as a max-min criterion for maximizing relay transmission performance, the relay selection criterion must be redened while considering both the interference constraint and the maximum power constraint as follows:

probability, as follows: Pr {u UT } =

xUT /I

/P x=I

fxy (x, y ) dy dx
y =0 UT /P

I /P

+ ( =
x=0 y =0 P y I/

fxy (x, y ) dy dx )

(8)

e y UT /P / +1 UT + 1 1 e I

=gu (x , y ), where fxy (x, y ) = x y e(x x+y y) and = x /y . Using (8), the outage probability based on the k th relay, o PCR,k = Pr {Ck CT } = Pr {Uk UT }, can be deo ned as PCR,k = 1 Pr {u1,k UT } Pr {u2,k UT } = 1 (1 gu (SP , Sk )) (1 gu (kP , kD )). Hence, the outage probability in cognitive relay networks is dened as o PCR = Pr {Uk UT }, and presented as { ( ) K P Sk 1,k I/ e o PCR = 1 1 T +1 1,k I/U k=1 ( ) } ekD 2,k I/P (Sk +kD )UT /P 1 . T +1 e 2,k I/U (9) From Lemma 1, we can see that the outage probability when the k th relay is used in cognitive relay networks consists of the outage probability in conventional relay networks, o PConv,k , and an increase due to the interference constraint, goes to zero. Moreover, if P k . k becomes zero as P goes (to innity, then becomes { ( + ) I/U + 1,k 2,k T )( k ) T + 1 2,k I/U T + 1 } so that k becomes 1}/{ 1,k I/U goes to innity, and one as I goes to zero. Hence, zero as I k has a range of 0 k 1. Therefore, it can be said that the outage probability in cognitive relay networks is always equal to or greater than that in conventional relay networks. Moreover, from (9), we can see that the outage probability depends not on the distances of the interference links, but on the distance ratio of the interference link to the relaying link. This means that the relative distance of the interference link based on the relaying link distance has more of an impact on the outage probability than the absolute distance of the interference link. B. Comparison of Relay Selection Diversity This subsection compares the relay selection diversity in cognitive relay networks to that in conventional relay networks. For simplicity, it is assumed that each link has an equal mean of channel gain, Sk = SR , kD = RD , and kP = RP for all k [17], [18], since this assumption provides a simple form for the analysis without changing the

{ k = arg max min min

k

{ { min

I |h S k P | I |hkP |

}
2,

P }

|hSk | , } |hkD |
k 2

2, P

(5)

= arg max min {u1,k , u2,k } = arg max Uk ,

k

where k is the relay index selected for cognitive relay networks. Lemma 1. Using the outage probability of conventional relay o networks, PConv,k = 1 e(Sk +kD )UT /P , the outage probability for cognitive relay networks after applying the relay selection described in (5) can be dened as follows:
K k=1

{
o PConv,k

} + k , (6)

o PCR

where k is an increase in outage probability due to the interference constraint, dened as (7), shown at the bottom of P , 1,k = SP /Sk = (dSP /dSk ) , the page, where n = I/ and 2,k = kP /kD = (dkP /dkD ) . Proof: In (5), due to the interference constraint, Uk = min {u1,k , u2,k } is no longer an exponential random variable. Hence, the rst step { / } is to dene the distribution of 2 2 u = min I |hx | , P |hy | so as to derive the outage { k = e
(Sk +kD )UT /P

} ( ) ( ) T + 1 eSk 1,k n + 1,k I/U T + 1 ekD 2,k n e(Sk 1,k +kD 2,k )n + 2,k I/U ( )( ) . T + 1 2,k I/U T +1 1,k I/U

(7)

o diversity order. With this assumption, PConv,k and k become equal for all k as PConv and . Thus, the outage probabilities in conventional relay and cognitive [ networks ] [ relay networks ] o K = Conv K and P o = P Conv + become PConv = P CR [ ] CR K . Using binomial theorem [19], P o can be redened P CR as follows:

1.0 0.9 0.8 0.7 0.6

SUS Cognitive Relay Networks

0.2 0.3 0.4 0.5 0.6 1.8 1.6 1.2 0.8

PURX Conventional Relay Networks 0.4

] [ o Conv K + gIC (K ), PCR = P where gIC (K ) =

K ( i=1

(10)
y

0.5 0.4

Relays

K i

i , K i P Conv

0.3

(11)

0.2 0.1 0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1, 0 P 1. Conv + 0 Hence, we can see that the outage probability of cognitive relay networks with relay selection is divided into the outage o probability in conventional relay networks, PConv , and the increased outage probability arising from the interference constraint, gIC (K ). If the outage probability is proportional to m as approaches innity, the selection diversity order is m where is the SNR [20]. Then, the following lemma can be obtained. Lemma 2. Cognitive relay networks achieve the full relay selection diversity order of K . Proof: In cognitive relay networks, we need to consider two cases of SNR from the transmission power constraint in (i.e., I/ |hkP |2 > P ), SNR is 1 = (1). First, when Pk = P /No . In (6), P CR is proportional to 1 as 1 goes to innity P 1 = (22CT 1)1 . So, P o scales no slower than since UT /P 1 CR 2 2 K O( 1 ). Second, when Pk = I/|hkP | (i.e., I/|hkP | P ), o . From k in Lemma 1, the average SNR is 2 = kP I/N ( ) CR is proportional to I/N o 1 as I/N o we can see that P ( ) T = (22CT 1)1 I/N o . So, approaches innity since I/U K o PCR also scales no slower than O( 2 ). Hence, the selection diversity order is K for all cases of transmission power. Note that this is the same as the selection diversity order in conventional relay networks [18]. Hence, with full selection diversity order, the outage probabilities for both cognitive and conventional relay networks decrease as the number of relays increases. However, the reduction in the amount of outage probability achieved by adding a relay in cognitive relay networks is different from what occurs in conventional relay networks. We designate this the diversity gain. In the generally CR 0.5), considered range of the outage probability (i.e., P the following lemma on the diversity gain can be obtained. Lemma 3. The diversity gain in cognitive relay networks is always equal to or greater than that in conventional relay CR 0.5. networks when P Proof: The diversity gains achieved by increasing the total number of relays from K 1 to K in cognitive and conventional relay networks are dened as GCR (K ]K [ ]K 1 [ and GConv (K 1, K ) = P 1, K ) = P ]K CR [ ]K 1CR[ Conv , respectively. From (10), GCR (K Conv P P

SUD

1.0

Fig. 2. Percentages of relay selection in conventional relay networks (dotted line), and in cognitive relay networks (solid line) (400 candidate relays are distributed equidistantly over a square area and the secondary source, the secondary destination and the primary receiver are located at the coordinates (0, 1), (1, 0) and (1, 1), respectively).

1, K ) and GConv (K 1, K ) have the following relation: GCR (K 1, K ) GConv (K 1, K ) = gIC (K 1) gIC (K ). (12) Assuming gIC (t), t R+ , is a continuous function, then is dened as gIC (t) = ( ) (the gradient )t P Conv + Conv h(t) ln P where h(t) = { 1 ( )}t ( ){ ( )}1 PConv P . (Conv + ) ln PConv ln PConv + Conv + Since ln P 0 and h(t) 0 when 0.5 for t > 0, gIC (t) 0. Hence, Conv + P gIC (k ) is a monotonically decreasing function of k R+ .1 Accordingly, gIC (K 1) gIC (K ) is not less than zero, which implies GCR (K 1, K ) GConv (K 1, K ) 0 in (12). From Lemma 2 and Lemma 3, it can be seen that cognitive relay networks have the same relay selection diversity order as conventional relay networks, but that the decrease in outage probability achieved by increasing the selection diversity is larger. Hence, the outage probability of cognitive relay networks, generally larger than that of conventional relay networks, approaches that of conventional relay networks as the number of relays increases. IV. N UMERICAL R ESULTS In this section, we examine the performance of cognitive relay networks based on the outage probability. Simulations were conducted to verify the derived outage probabilities in (9) and (10), and the results closely match the analysis, as shown in Figs. 3-5.
1g IC (t) 0 means gIC (t) is monotonically decreasing at any point t R+ , so gIC (k) is also a monotonically decreasing function with respect to k, k N, since N R+ .

10

Cognitive Relay Networks: PURX=(1.0,1.0)

Outage Probability of Secondary User

10

Cog-Relay (Analysis)

Cog-Relay (Simulation) Conv-Relay (Analysis) 10

-1

Outage Probability of Secondary User

Conv-Relay (Simulation)

10

-1

(P / N )
o

Cognitive Relay Networks: PURX=(1.5,1.5)

10

-2

K=2

10

-2

Cog-Direct (Simulation) Cog-Relay (Analysis) Cog-Relay (Simualtion) Conv-Relay (Analysis) Conv-Relay (Simualtion)

10

-3

K=4 A

Conventional Relay Networks

Ratio of interference threshold to maximum transmission power I / P

10

-1

10

10

-4

6 8 10 12 14 16 Ratio of maximum transmission power to noise

P / N o (dB)

18

20

Fig. 3. Comparison of outage probabilities of cognitive relay networks (CogRelay), conventional relay networks (Conv-Relay), and direct transmission in cognitive radio networks (Cog-Direct) according to the interference threshold P (In this simulation, = compared to the maximum transmission power, I/ /No = 30dB, CT = 5 bps/Hz, and 20 potential relays are distributed P uniformly within a rectangular area described by the coordinates (0, 0), (0, 1), (1, 0), and (1, 1)).

Fig. 4. Comparison of outage probabilities of cognitive relay networks (CogRelay) and conventional relay networks (Conv-Relay) according to the ratio /No (I/N o = 12dB, of the maximum transmission power to the noise, P and CT = 1.5 bps/Hz).

As discussed above, the relay selection criterion in cognitive relay networks considers the interference constraint as well as the maximum power constraint. Hence, the relay selected is different from the one selected in conventional relay networks, even in the same environment. Fig. 2 shows the distribution of probabilities that a relay in a specic location will be selected in cognitive relay and conventional relay networks. To obtain more precise distributions, 400 potential relays are located equidistantly (marked with x in Fig. 2). The numbers on the contour lines represent the probability that a specic relay will be selected, and the relay that is most likely to be selected is indicated by a square. In conventional relay networks, the relay located exactly at the midpoint between source and destination (near the coordinates (0.5, 0.5)) is the one most likely to be selected. However, in cognitive relay networks, the relay located farthest from the primary receiver (near the coordinates (0.35, 0.35)), is the one most apt to be selected. Accordingly, the conventional relay selection criterion is not suitable in cognitive relay networks. P . For Fig. 3 depicts the outage performances versus I/ this simulation, we vary the interference threshold while P is xed, so that the outage performance of the conventional P . In relay network is not affected by the variation of I/ Fig. 3, if we compare the results when the primary receivers locations are (1, 1) (dotted lines) and (1.5, 1.5) (solid lines), we see that the outage performance of cognitive relay networks improves when the primary receiver is located farther away from the secondary users. In addition, the outage probability of cognitive relay networks approaches that of conventional relay networks as the interference threshold increases. In general, we can see that the relaying transmission achieves better outage performance than direct transmission. However,

direct transmission can outperform relay transmission when the primary receiver is located close to the secondary users or the interference threshold is low, as shown in Fig. 3. This results from the fact that, in those cases, the loss from using twice the resources (e.g., 1/2 term in (2)) is more dominant than the improved capacity achieved via relay transmission. In addition, we note that the relay transmission is more sensitive to any variation in the interference threshold or the location of the primary receiver than direct transmission. Figs. 4-6 are to show the diversity order and the gain when the primary receiver is located at the coordinates (1.5, 1.5), and each channels mean is assumed to be equal, Sk = SR = 2/2 , kD = RD = 2/2 , and kP = RP = 2/2 for all k , as described in Section III.B. ( ) /No K are included as In Fig. 4, the curves obtained by P references to explicitly verify the diversity order. Comparing the slopes of the results conrms that cognitive relay networks indeed achieve full diversity order equivalent to conventional relay networks. However, due to the transmission power constraint in (1), the outage probability of cognitive relay is high, e.g., in networks converges to a constant when P the area A in Fig. 4. In this case, the transmission power is |hkP |2 with high probability. So, the diversity order needs I/ o as Fig. 5, which shows the to be veried according to I/N is high, i.e., P /No = 40dB. In Fig. outage probability when P 5, the same slopes of the results also conrm the full diversity order of cognitive relay networks. We can thus see that the interference constraint in cognitive relay networks does not change the diversity order. In Fig. 6, both the outage probabilities of cognitive relay networks and conventional relay networks decrease and become similar to each other as the number of relays increases. On comparing GCR (K 1, K ) and GConv (K 1, K ), we can see that the diversity gain achieved by increasing the number of relays is greater in cognitive relay networks than in

10

Cog-Relay (Analysis)

0.25

Cog-Relay (Simulation) Outage Probability of Secondary User 10

-1

Outage Probability of Secondary User

(I / N )
o

0.2

Cog-Direct (Simulation) Cog-Relay (Analysis) Cog-Relay (Simulation) Conv-Relay (Analysis) Conv-Relay (Simulation)

K=1
10
-2

0.15 G (K-1, K)

K=2
10
-3

0.1

CR

K=4

0.05

Conv

(K-1,K)

10

-4

8 10 12 14 Ratio of interference threshold to noise I

/ N o (dB)

16

18

4 5 6 Number of Potential Relays K

Fig. 5. Outage probability of cognitive relay networks (Cog-Relay) according o (P /No = 40dB, to the ratio of the interference threshold to the noise, I/N and CT = 2 bps/Hz).

conventional relay networks, as is discussed in Section III.B. For instance, if the number of relays is increased from 3 to 4, the outage probability reduction amounts are 0.05 in cognitive relay networks and 0.01 in conventional relay networks. We can therefore achieve a greater reduction in outage probability in cognitive relay networks than in conventional relay networks by providing additional relays. V. C ONCLUSIONS The overall contribution of this paper is the evaluation of the outage probability of cognitive relay networks when a suitable relay selection criterion is applied. As opposed to conventional relay networks, the outage probability in cognitive relay networks is affected by the distance-ratio of the interference link to the relaying link, and not the absolute distances. Moreover, it is always higher than that of conventional relay networks due to the interference constraint, and the difference between them, depending on the interference threshold and the maximum transmission power, is quantied. The outage probability of cognitive relay networks decreases with full selection diversity order the same as conventional relay networks, but the decrease in outage probability achieved by increasing the number of relays is greater than that of conventional relay networks. These observations give quantitative insight into the effect of the interference constraint and the number of potential relays on the outage probability of cognitive relay networks. R EFERENCES
[1] A. Goldsmith, S. A. Jafar, I. Maric, and S. Srinivasa, Breaking Spectrum Gridlock With Cognitive Radios: An Information Theoretic Perspective, Proceedings of the IEEE, Vol. 97, No. 5, pp. 894 - 914, May 2009. [2] C. Chang and J. Chang, Optimal Design Parameters in a Multihop Packet Radio Network using Random Access Techniques, IEEE Global Commun. Conf., Vol. 1, pp. 493-497, June 1984. [3] O. Simeone, U. Spagnolini, and Y. Bar-Ness, Stable Throughput of Cognitive Radios With and Without Relaying Capability, IEEE Trans. on Commun., Vol. 55, Issue. 12, pp. 2351-2360, Dec. 2007.

Fig. 6. Comparison of outage probabilities of cognitive relay networks (CogRelay), conventional relay networks (Conv-Relay), and direct transmission in cognitive radio networks (Cog-Direct) according to the number of potential relays, K . GCR (K 1, K ) and GConv (K 1, K ) represent the outage probability reduced by adding a relay from K 1 to K in cognitive relay P = 10dB, networks and in conventional relay networks, respectively (I/ /No = 30dB, and CT = 5 bps/Hz). =P

[4] K. Lee and A. Yener, Outage Performance of Cognitive Wireless Relay Networks, IEEE Global Commun. Conf., pp. 1-5, Nov. 2006. [5] H. A. Suraweera, P. J. Smith, and N. A. Surobhi, Exact Outage Probability of Cooperative Diversity with Opportunistic Spectrum Access, IEEE Intl. Conf. on Commun., pp. 79-84, May 2008. [6] J. Mietzner, L. Lampe, and R. Schober, Distributed Transmit Power Allocation for Multihop Cognitive-radio Systems, IEEE Trans. on Wireless Commun., Vol. 8, No. 10, pp. 5187-5201, Oct. 2009. [7] S. Kim, W. Choi, Y. Choi, J. Lee, Y. Han, and I. Lee, Downlink Performance Analysis of Cognitive Radio based Cellular Relay Networks, Int. Conf. on Cognitive Radio Oriented Wireless Networks and Communications, pp. 1-6, May 2008. [8] J. Lee, H. Wang, W. Seo, and D. Hong, QoS-guaranteed Transmission Mode Selection for Efcient Resource Utilization in Multi-hop Cellular Networks, IEEE Trans. on Wireless Commun., Vol. 7, No. 10, pp. 36973701, Oct. 2008. [9] T. W. Ban, W. Choi, B. C. Jung, and D. K. Sung, Multi-User Diversity in a Spectrum Sharing System, IEEE Trans. on Wireless Commun., Vol. 8, No. 1, pp. 102-106, Jan. 2009. [10] A. Ghasemi and E. S. Sousa, Fundamental Limits of Spectrum-sharing in Fading Environment, IEEE Trans. on Wireless Commun., Vol. 6, Issue 2, pp. 649-658, Feb. 2007. [11] K. Hamdi, W. Zhang, and K. B. Letaief, Power Control in Cognitive Radio Systems based on Spectrum Sensing Side Information, IEEE Intl. Conf. on Commun., pp. 5161-5165, June 2007. [12] H. Wang, J. Lee, S. Kim, and D. Hong, Capacity of Secondary Users Exploiting Multispectrum and Multiuser Diversity in Spectrum-Sharing Environments, IEEE Trans. on Veh. Technol., Vol. 59, Issue 2, pp. 10301036 , Feb. 2010. [13] X. Kang, Y. C. Liang, and A. Nallanathan, Optimal Power Allocation for Fading Channels in Cognitive Radio Networks: Delay-Limited Capacity and Outage Capacity, IEEE Vehicular Technol. Conf., pp. 1544-1548, May 2008. [14] R. Etkin, A. Parekh, and D. Tse, Spectrum Sharing for Unlicensed Bands, IEEE J. Sel. Areas Commun., Vol. 25, Issue 2, pp. 517-528, April. 2007. [15] V. Sreng, H. Yanikomeroglu, and D. D. Falconer, Relayer Selection Strategies in Cellular Networks with Peer-to-peer Relaying, IEEE Vehicular Technol. Conf, Vol. 3, pp. 1949- 1953, Oct. 2003. [16] A. Muller and J. Speidel, Relay Selection in Dual-Hop Transmission Systems: Selection Strategies and Performance Results, IEEE Intl. Conf. on Commun., pp. 4998-5003, May 2008. [17] A. Bletsas, A. Khisti, D. P. Reed, and A. Lippman, A Simple Cooper-

ative Diversity Method Based on Network Path Selection, IEEE J. Sel. Areas Commun., Vol. 24, Issue 3, pp. 659-672, March 2006. [18] Y. Jing and H. Jafarkhani, Single and Multiple Relay Selection Schemes and Their Achievable Diversity Orders, IEEE Trans. on Wireless Commun., Vol. 8, Issue 3, pp. 1414-1423, March 2009. [19] M. Abramowitz and I. A. Stegun (Eds.). Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, 9th printing. New York: Dover, 1972. [20] C. K. Lo, S. Vishwanath, and R. W. Heath, Relay Subset Selection in Wireless Networks Using Partial Decode-and-Forward Transmission, IEEE Trans. on Veh. Technol., Vol. 58, Issue 2, pp. 692-704, Feb. 2009.