IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 865 Performance Analysis of Cognitive Relay Networks Over Nakagami-m Fading Channels Xing Zhang, Senior Member, IEEE, Yan Zhang, Senior Member, IEEE, Zhi Yan, Jia Xing, and Wenbo Wang, Member, IEEE Abstract—In this paper, we present performance analysis for underlay cognitive decode-and-forward relay networks with the N th best relay selection scheme over Nakagami-m fading channels. Both the maximum tolerated interference power constraint and the maximum transmit power limit are considered. Specifically, exact and asymptotic closed-form expressions are derived for the outage probability of the secondary system with the N th best relay selection scheme. The selection probability of the N th best relay under limited feedback is discussed. In addition, we also obtain the closed-form expression for the ergodic capacity of the secondary system with a single relay. These expressions facilitate in effectively evaluating the network performance in key operation parameters and in optimizing system parameters. The theoretical derivations are extensively validated through Monte Carlo simulations. Both theoretical and simulation results show that the fading severity of the secondary transmission links has more impact on the outage performance and the capacity than that of the interference links does. Through asymptotic analysis, we show that the diversity order for the N th best relay selection scheme is min(m1 , m3 ) × (M − N ) + m3 , where M denotes the number of cognitive relays, and m1 and m3 represent the fading severity parameters of the first-hop transmission link and the second-hop transmission link, respectively. Index Terms—Cognitive relay networks, N th best relay selection, Nakagami-m fading, outage probability, ergodic capacity. I. I NTRODUCTION R ADIO spectrum is an important and scarce resource which is increasingly demanded by many kinds of users. Cognitive radio is an efficient technology to improve the spectrum resources utilization and has gained much attention in recent years [1]. There are three main cognitive radio paradigms: underlay, overlay and interweave [2]. The underlay paradigm allows cognitive (secondary) users to utilize the licensed specManuscript received January 4, 2014; revised May 8, 2014 and July 15, 2014; accepted August 23, 2014. Date of publication September 30, 2014; date of current version April 21, 2015. This work was supported in part by the National Natural Science Foundation of China (NSFC) under Grant 61372114, by the National 973 Program of China under Grant 2012CB316005, by the Joint Funds of NSFC-Guangdong under Grant U1035001, and by Beijing Higher Education Young Elite Teacher Project under Grant YETP0434. X. Zhang, J. Xing, and W. Wang are with the Wireless Signal Processing and Network Laboratory, Key Laboratory of Universal Wireless Communication, Ministry of Education, Beijing University of Posts and Telecommunications, Beijing 100876, China (e-mail: zhangx@ieee.org; xjbupt@gmail.com; wbwang@bupt.edu.cn). Y. Zhang is with Simula Research Laboratory, Fornebu 1364, Norway (e-mail: yanzhang@ieee.org). Z. Yan is with the School of Electrical and Information Engineering, Hunan University, Changsha 410082, China (e-mail: yanzhi@hnu.edu.cn). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSAC.2014.2361081 trum if the interference caused to primary users is below a given interference threshold. In overlay systems, both primary and secondary users can utilize the licensed spectrum simultaneously through sophisticated signal processing and coding. In interweave systems, secondary users opportunistically utilize spectrum holes to communicate without interfering the transmission of primary users. In this paper, we focus on the underlay systems. On the other hand, relay communication has emerged as a powerful spatial diversity technology for effectively combating channel fading and greatly improving the transmission performance of wireless communication systems [3]. There are two kinds of classical relay communication protocols, i.e., amplify-and-forward (AF) and decode-and-forward (DF) [4]. In multiple-relay systems, the relay-selection-based transmission protocol can get higher spectrum efficiency compared with the traditional relay communication protocol. Best relay selection [5], [6] is an ideal protocol to achieve the best performance. In practice, however, the best relay might not be available due to some scheduling, load balancing or channel side information (CSI) imperfect feedback conditions. In this case, the second best relay or more generally the N th best relay might be selected. Therefore, the study of the N th best relay selection is of great need. Inspired by cognitive radio and relay communication, the cognitive relay network which combines these two techniques is proposed. In cognitive relay networks, both the spectrum efficiency and the transmission performance can be improved. Recently, the research of cognitive relay networks has attracted much attention, especially on the underlay paradigm [7]–[19]. In [7], the outage probability of a cognitive dual-hop network with a single AF relay under the interference power constraint was derived. The outage probability of a cognitive DF relay network without a direct transmission link and with best relay selection was evaluated in [8]. In [9], a rough upper bound on outage probability for cognitive DF relay networks with a direct transmission link and with best relay selection over independent and identically distributed (i.i.d) Rayleigh fading channels was obtained. However, these works can be improved by considering the dependence among the received signal-tonoise ratios (SNRs) in the first hop. Considering this kind of dependence, the accurate upper and lower bound on outage probability for such systems were respectively obtained in [10] and [11]. In [12], the exact outage probability expression for cognitive DF relay network was derived. The authors in [13] investigated the diversity performance of cognitive relay networks with 0733-8716 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. 866 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 three different relaying protocols, i.e., selective AF, selective DF, and AF with partial relay selection. In [14], the authors studied the outage performance of a cognitive network adopting incremental DF protocol. In [15], the outage probability, symbol error probability and ergodic capacity were derived for cognitive AF relay networks with best relay selection. References [7]–[15] focused on Rayleigh fading channels while references [16]–[19] studied the more general fading environment, i.e., Nakagami-m fading channels. In [16]–[18], the authors studied the outage performance of cognitive relay networks with a single relay over Nakagami-m fading channels. In [19], the outage performance of cognitive DF relay network with best relay selection over independent and non-identically distributed Nakagami-m fading channels was studied. The above mentioned references all focused on the best relay selection scheme. Only a few studies involved the N th best relay selection scheme. As we have mentioned, the N th relay selection is more of practical significance. In [20], the performance for conventional AF and DF relay networks with the N th best relay selection over Rayleigh fading channels was studied. The asymptotic symbol error rate for a conventional AF relay network with the N th best relay selection over Nakagami-m fading channels was derived in [21]. In [22], the authors investigated the outage behavior of a conventional dual-hop N th-best DF relay system in the presence of cochannel interference over Rayleigh fading channels. The outage performance for cognitive relay networks with the N th best relay selection over Rayleigh fading channels were studied in [23]. In summary, it is observed that there have been no prior works on the performance of cognitive relay network with the N th best relay selection scheme over Nakagami-m fading channels. While in practical networks, the channels will not always be simply Rayleigh-distributed. Thus, a comprehensive study of cognitive relay network with N th best relay selection over the general Nakagami-m fading channel will be beneficial for the design in practical cognitive relay systems. In this paper, we investigate the performance of an underlay cognitive DF relay network over Nakagami-m fading channels. Our main contributions are as follows: • The exact outage probability of the secondary system with the N th best relay selection is derived over Nakagami-m fading channels, which build the relationship between the outage performance and the related system parameters. In addition, the selection of the N th best relay in the limited feedback scenario is discussed. • An asymptotic analysis is carried out to get the asymptotic outage probability of the secondary system with the N th best relay selection. The diversity order is also obtained. • The closed-form expression for the ergodic capacity of the secondary system with single relay is derived over Nakagami-m fading channels. • The results show that the fading severity of the secondary transmission links has more impact on the outage performance and the ergodic capacity than the fading severity of the interference links. The rest of this paper is organized as follows. Section II describes the system model. Sections III and IV present the de- Fig. 1. System model. tailed analysis of exact and asymptotic outage performance of the secondary system with the N th best relay selection scheme, respectively. The selection probability of the N th best relay under limited feedback is discussed in Section V. In Section VI, the exact ergodic capacity is derived and analyzed. Numerical results are shown in Section VII. Finally, conclusions are given in Section VIII. coefficient and n! Notation: Cnk represents the binomial ∞ α−1 t e−t dt, Γ(α, x) = represents the factorial of n. Γ(α) = x α−1 0−t ∞ α−1 −t t e dt and γ(α, x) = t e dt denote the gamma x 0 function [24, eq. (8.310.1)], the upper incomplete gamma function [24, eq. (8.350.2)] and the lower incomplete ∞gamma −t function [24, eq. (8.350.1)], respectively. Ei(x) = − −x e t dt represents the exponential integral function [24, eq. (8.211.1)]. The cumulative distributed function (CDF) and the probability density function (PDF) of random variable X are expressed as FX (·) and fX (·), respectively. II. S YSTEM M ODEL We consider an underlay cognitive DF relay network, as illustrated in Fig. 1. It involves one primary user receiver (P U ) and a secondary system. The secondary system is a dual-hop relay communication system which consists of one secondary source (SS), one secondary destination (SD) and M secondary relays (SRi , i = 1, . . . , M ). All nodes are equipped with a single antenna and operate in half-duplex mode. The interference from the primary transmitter is assumed to be neglected as in [7]–[19]. This can be possible if the primary transmitter is located far away from the secondary users, or the interference is modeled as the noise term [8]. Like [13], [15], [16], [18], etc., we assume that there is no direct link between SS and SD due to the severe shadowing and path loss. We employ the CSI-assisted DF relaying protocol and the N th best relay selection scheme. A whole transmission process of the secondary system consists of two phases. In the first phase, the SS broadcasts messages to M relays under a transmit power constraint which guarantees that the interference on the primary user receiver does not exceed a threshold. In the second phase, the N th best relay that is selected from the successful decoding relay set based on the channel quality of the secondhop links forwards source messages to the SD. Finally, the SD ZHANG et al.: PERFORMANCE OF COGNITIVE RELAY NETWORKS OVER NAKAGAMI-m FADING CHANNELS 867 TABLE I PARAMETERS N OTATIONS decodes source messages. The transmitters of the secondary system are with the maximum transmit power constraint Pmax , and the maximum interference power constraint of the primary user receiver is Q. All links are assumed to be independent Nakagami-m flat fading channels with integer values of the fading severity parameters and unit average power. The channel gain of SS-SRi , SS-P U , SRi -SD, and SRi -P U are denoted as gsi , gsp , gid , and gip , whose fading severity parameters are m1 , m2 , m3 , and m4 , respectively. The thermal noise at each receiver is modeled as additive white Gaussian noise (AWGN) with variance σ 2 . More details of the parameters used in this paper are given in Table I. III. E XACT O UTAGE P ERFORMANCE A NALYSIS In this part, we derive the exact outage probability expression for the previously described underlay cognitive relay network with the N th best relay selection, which can be used to evaluate the impact of the related parameters on the outage performance, which include the maximum interference power constraint Q, the maximum transmit power constraint Pmax , the fading severity parameters mi (i = 1, 2, 3, 4), the number of relays M and the order of the selected relay N . Considering the maximum transmit power constraint Pmax of the secondary transmitters and the interference power constraint Q of the primary user, the transmit power of the SS and the ith relay SRi should be no more than min(Pmax , Q/gsp ) and min(Pmax , Q/gip ), respectively. In order to maximize the transmission performance of the secondary system, the secondary transmitters transmit signals with the maximum allowable transmit power. Hence, the transmit power at SS can be written as PS = min(Pmax , Q/gsp ), where gsp denotes the channel coefficient of the link between SS and P U . Similarly, the transmit power at SRi is given by PRi = min(Pmax , Q/gip ). In the first-hop transmission, the SS broadcasts messages to relays. As a result, the received SNR at the ith relay SRi is written as min(Pmax , Q/gsp )gsi = γsi = σ2 P max gsi , σ2 Qgsi σ 2 gsp , for Pmax < gQsp for Pmax ≥ gQsp . (1) From the above equation, it is worth noting that γsi for each i is related with gsp while Pmax > Q/gsp . Hence, the received SNRs at relays are correlated in the first-hop transmission while Pmax > Q/gsp , but they are independent while Pmax ≤ Q/gsp . We denote the target transmission rate of secondary system as R. Then the received SNR at the ith relay should meet the following inequality if the ith relay can successfully decode source messages. R≤ 1 log2 (1 + γsi ). 2 (2) We define γth = 22R − 1, so the successful decoding constraint of the ith relay can be simplified to γsi ≥ γth . We denote the successful decoding relay set as R(s) in the first-hop transmission. Lemma 1: The probability of the successful decoding relay set R(s) is given by the expression (3), shown at the bottom of the next page, where n = |R(s)| denotes the number of relays m 1 −1 jwj . in R(s), γPmax = Pmax /σ 2 , γQ = Q/σ 2 , and H = j=0 Proof: See Appendix A. In the second-hop transmission, the N th best relay selected from the successful decoding relay set R(s) forwards the source messages to the SD. N should be less than or equal to |R(s)|. It is worth noting that the N th best relay selection makes sense only when R(s) is not empty, since N ≥ 1. With the N th best relay selection scheme, the received SNR at the SD is γrd = N th max (γid ) = N th max (γid ), i∈R(s) γsi ≥γth (4) 868 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 where N th max(·) denotes the N th maximum item, γid = min(Pmax ,Q/gip )gid denotes the received SNR at the SD if the σ2 ith relay of R(s) forwards source messages to the SD in the second phase. Lemma 2: The CDF of γrd is given as (5), shown at the bottom of the page. Proof: See Appendix B. Finally, the SD decodes source messages. Hence, the equivalent end-to-end received SNR is γrd in the transmission procedure, and the mutual information of secondary system is given as C= 1 log2 (1 + γrd ). 2 (6) According to the Shannon information theory, the outage occurs when C < R. We denote the event that the N th best relay is selected as SN . Therefore, the conditional outage probability of the secondary system given SN and R(s) (|R(s)| ≥ N ) is calculated as Pr (outage|SN , R(s)) = Fγrd (γth ). (7) Considering all the possibilities of R(s), the outage probability of the secondary system with the N th best relay selection scheme can be written as the following expression according to the law of total probability. Pr(outage|SN ) = M n CM Pr (R(s)) Fγrd (γth ). (8) n=N Substituting (3) and (5) into (8), the exact outage probability for the N th best relay selection can be obtained. IV. A SYMPTOTIC O UTAGE P ERFORMANCE A NALYSIS In this part, we derive the asymptotic outage probability expression in high SNR regions to reveal the diversity performance of the secondary system with the N th best relay selection scheme. For a relay selection diversity communication system, the diversity order is an important performance metric. It is defined as d = − limγ→∞ (log Pout (γ)/ log(γ)), where γ denotes the SNR of systems. The diversity order essentially indicates the number of received independent fading signals at the receiver. To derive the diversity order of the secondary system with the N th best relay selection scheme, the asymptotic outage probability in high SNR regions should be obtained firstly. For the simplicity of analysis, as [13] and [25], we set γ = 1/σ 2 to represent the SNR of the secondary system in the subsequent discussions. Hence, the high SNR region arises while σ 2 → 0. According to the asymptotic behavior of γ(n, x), we have lim γ(n, x) = x→0 n=N In high SNR regions, the higher order terms of 1/γ can be omitted. From (13), the n = N term (i.e., (1/γ)m1 (M −N )+m3 ) ⎤n ⎡ ⎤M −n m1 γth m2 Q 1 γth n M −n+i γ m γ m1 , m , γ m , 1 2 γPmax γPmax Pmax ⎦ ⎦ ⎣ ⎣ + Pr (R(s)) = 1− Γ(m1 ) Γ(m1 ) Γ(m2 ) i=0 × Fγrd (x) = N k−1 2 l!mm 2 Γ m2 + H, m2 γQ +m1 γth l γPmax Γ(m2 ) u Cnk−1 Ck−1 (−1)k−1−u k=1 u=0 − l=0 m1 γth γQ H ⎧ ⎨ γ m3 , γm3 x γ m4 , Pm4 Q P max max ⎩ 4 mm 4 Γ(m4 ) Γ(m3 ) m 3 −1 i=0 (9) Note that the n in (9) is not necessarily an integer. Therefore, the asymptotic outage performance analysis in this section is applicable to the cases of arbitrary Nakagami fading parameters, including the non-integer ones. Lemma 3: The probability of the successful decoding relay set R(s) can be asymptotically approximated as (10), shown at the bottom of the next page,where ∝ represents “proportional to.” Proof: See Appendix C. Lemma 4: The asymptotic expression for the CDF of γrd is written as (11), shown at the bottom of the next page. Proof: See Appendix D. By utilizing Lemma 3 and Lemma 4, and substituting (10) and (11) into (8), the asymptotic outage probability of the secondary system with the N th best relay selection scheme can be obtained as (12), shown at the bottom of the next page. Meanwhile, we can see from (12) that M m1 (M −n)+m3 (n−N +1) 1 . (13) Pr(outage|SN ) ∝ γ ⎡ xn . n Γ(m4 ) l i+l Cni CM −n+i (−1) w0 +...+wm1 −1 =l γQ m2 γQ + m1 γth l + m2 +H m 1 −1 j=0 1 wj ! wj 1 j! (3) 4Q Γ m4 , Pmmax Γ(m4 ) ⎡ ⎤⎫n−u xm +m γ −(m4 +i) ⎬ Γ m4 + i, γ3P 4 Q xm i xm 3 3 max ⎣ ⎦ + m4 ⎭ i! γQ γQ (5) ZHANG et al.: PERFORMANCE OF COGNITIVE RELAY NETWORKS OVER NAKAGAMI-m FADING CHANNELS is left when m1 ≤ m3 , and the n = M term (i.e., (1/γ)m3 (M −N +1) ) is left when m1 > m3 . To sum up, the asymptotic outage probability of secondary system in high SNR regions is proportional to (1/γ)m1 (M −N )+m3 when m1 ≤ m3 , but it is proportional to (1/γ)m3 (M −N +1) when m1 > m3 . Hence, the diversity order of the secondary system is min(m1 , m3 ) × (M − N ) + m3 . It is indicated that the diversity performance of the secondary system with the N th best relay selection scheme is affected by the channel fading severity parameters of the transmission links, as well as the difference between the number of relays M and the order of the selected relay N . The channel fading severity parameters of the interference links have no impact on the diversity order. We assume that L bits are used to feedback the SNRs, so there are q = 2L quantization intervals. Given the quantized codebook {γ̂1 , γ̂2 , · · · , γ̂q }, the quantized value of γid (i ∈ R(s)) is determined by γ̂id = arg The relay selection process is mainly based on the obtained channel knowledge. In practice, due to imperfect CSI feedback, the chosen relay may not be the best one. Limited feedback is often used to perform relay selection [26]. In this section, we discuss the impact of limited feedback on the relay selection process. The relay is selected according to the SNR of SRi -SD γid (i ∈ R(s)). Considering channel reciprocity, we assume relay SRi can acquire the channel fading coefficient of SRi -SD. Therefore, γid is available at SRi . In order to select a relay, SRi should feedback its SNR γid to the decision-making node who performs relay selection. Pr (R(s)) ≈ N γ→∞ Fγrd (x) ≈ k=1 1 Γ(m1 + 1) m3 x γ γ→∞ Pr(outage|SN ) ≈ m1 γth γ m1 M −n m3 (n−k+1) 1 Γ(m2 ) min γ̂∈{γ̂1 ,γ̂2 ,···,γ̂q } |γid − γ̂|. (14) Then each relay transmits the index of its quantized value to the decision-making node. We assume that γid falls into each quantization interval with equal probability p = 1/q through some non-uniform quantizer. With limited feedback, the SNRs of different relays may fall into the same quantization interval. The “best” relay selection depends on the best quantization interval, i.e., the quantization interval with the largest quantized value that contains at least one relay. We denote the set of relays whose SNR falls into the best quantization interval as R(b). R(b) is a subset of the successful decoding relay set R(s). For a given R(s) with n = |R(s)|, when n = 0, i.e., R(s) is empty, no relay would be selected. When n > 0, i.e., R(s) is not empty, the probability that there are nb relays in the best quantization interval for the case of nb < n can be calculated as V. S ELECTION P ROBABILITY OF THE N TH B EST R ELAY U NDER L IMITED F EEDBACK γ→∞ 869 Pr (|R(b)| = nb |R(s)) = Cnnb pnb q−1 (1 − ap)n−nb . (15) a=1 For the case of nb = n, i.e., all relays are in the same quantization interval, the probability that there are nb relays in R(b) for m1 (M −n) m2 Q γ m2 , Pmax Pmax m1 (M −n) 1 m2 Q Γ m1 (M − n) + m2 , + m2 Q Pmax 1 (10) n−k+1 1 Γ(m3 + 1)Γ(m4 ) m 3 n−k+1 m3 1 1 m4 Q m4 Q × γ m4 , Γ m3 + m4 , + Pmax Pmax m4 Q Pmax Cnk−1 (11) m1 (M −n)+m3 (n−k+1) n k−1 M −n n−k+1 N M 1 3 γth mm mm CM Cn 1 3 γ Γ(m2 ) Γ(m1 + 1) Γ(m3 + 1)Γ(m4 ) n=N k=1 m 3 n−k+1 m3 1 1 m4 Q m4 Q × γ m4 , Γ m3 + m4 , + Pmax Pmax m4 Q Pmax m1 (M −n) m1 (M −n) 1 1 m2 Q m2 Q × γ m2 , Γ m1 (M − n) + m2 , + (12) Pmax Pmax m2 Q Pmax 870 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 the given R(s) can be expressed as = 1 − Pr{γsr > x} Pr{γrd > x} Pr (|R(b)| = nb |R(s)) = qpn . The “best” relay is selected from R(b) randomly, i.e., each relay in R(b) is chosen with probability n1b . We denote the event that the N th best relay is chosen as SN (N = 1, 2, · · · , n). Therefore, the conditional selection probability of the N th best relay given R(s) is expressed as Pr (SN |R(s)) = n 1 Pr (|R(b)| = nb |R(s)) . nb (17) nb =N According to the law of total probability, the selection probability of the N th best relay under limited feedback is given by Pr(SN ) = M n CM Pr (R(s)) Pr (SN |R(s)) , where Pr(R(s)) can be calculated as (3). Combining this with the results in Section III, we can obtain the outage probability of the secondary system with limited feedback as Pout = Pr (R(s) = ∅)+ Pr(SN ) Pr(outage|SN ). (19) N =1 VI. E XACT E RGODIC C APACITY A NALYSIS In this section, we derive the exact ergodic capacity of the cognitive DF relay network over Nakagami-m fading channels. Specifically, we study the special case where there is only one single relay in the secondary system, i.e., M = 1 and N = 1. Additionally, we consider symmetric channel conditions where the channels associated with the first hop and the second hop have the same parameters. The fading severity parameters of the secondary transmission links are denoted as m1 = m3 = ms while those of the interference links are denoted as m2=m4=mp . For the case of M = 1 and N = 1, the end-to-end SNR of the secondary system is given by γe2e = min(γsr , γrd ). Since there is only one relay node, the CDF of γrd is the same as the CDF of γid in (42). Due to the symmetric channel conditions, γsr has the same distribution as γrd . We rewrite it as mp Q mp Q sx Γ m γ ms , γm , , γ m p p Pmax Pmax Pmax + Fγsr(x) = Fγrd(x)= Γ(ms ) Γ(mp ) Γ(mp ) i −(mp +i) ms −1 m ms x 1 ms x mp p + mp − Γ(mp ) i=0 i! γQ γQ m s x + mp γ Q × Γ mp + i, . (22) γPmax According to the definition, the ergodic capacity of the secondary system can be expressed as 1 ∞ C̄ = log2 (1 + x)fγe2e (x) dx. (23) 2 0 By using the same method as [27], we can rewrite the expression for the ergodic capacity as ∞ 1 1 C̄ = (1 − Fγe2e (x)) dx 2 ln 2 0 1 + x ∞ 1 1 (1 − Fγsr (x)) (1 − Fγrd (x)) dx. = 2 ln 2 0 1 + x (24) By substituting (22) into (24) and utilizing the series representation of the incomplete Gamma function [24, eq. (8.352.6)], we can get (25), shown at the bottom of the page, where A and B(v) are defined as ∞ i+j x − 2ms x e γPmax dx (26) A= 1+x 0 and (20) ∞ B(v) = Thus the CDF of γe2e can be written as 0 Fγe2e (x) = Pr {min(γsr , γrd ) ≤ x} xi+j − 2ms x e γPmax dx, v (1 + x)(ms x + mp γQ ) + (mp γQ ) m Q p mp − Pmax e 2 l=0 k=0 mp +i−1 mp +j−1 Γ(mp + i)Γ(mp + j) (27) respectively. 2 m s −1 m s −1 1 mi+j 1 1 mp Q s C̄ = γ mp , A 2 i+j 2 ln 2 (Γ(mp )) i=0 j=0 i!j! Pmax γPmax m+j−1 1 1 mp Q 1 mp Q − Pmax mp (m γ ) Γ(m + j)e B(mp + j − k) + γ mp , p Q p Pmax γPi max k! γPkmax k=0 mp +i−1 1 1 mp Q 1 mp Q − Pmax mp + γ mp , (mp γQ ) Γ(mp + i)e B(mp + i − l) j Pmax γP l! γPl max l=0 max (21) (18) n=N M = 1 − (1 − Fγsr (x)) (1 − Fγrd (x)) . (16) 1 1 B(2mp + i + j − l − k) l!k! γPl+k max (25) ZHANG et al.: PERFORMANCE OF COGNITIVE RELAY NETWORKS OVER NAKAGAMI-m FADING CHANNELS By using [24, eq. (3.353.5)], we obtain 2ms 2ms A = (−1)i+j−1 e γPmax Ei − γPmax h i+j γPmax + (h − 1)!(−1)i+j−h . 2ms 871 (28) h=1 To calculate the function B(v), we should consider two cases. For the case of ms = mp γQ , by splitting the term 1/(1 + x)(ms x + mp γQ )v , we get ∞ i+j 1 x − γ2ms x Pmax dx e B(v) = (mp γQ − ms )v 0 1 + x − v ms a (m γ p Q −ms ) a=1 Φ1 ∞ 0 xi+j − γ2ms x Pmax dx . e (ms x+mp γQ )v+1−a Φ2 (29) The integral term Φ1 is the same as A. The integral term Φ2 can be calculated with the help of the Meijer’s G-function [28]. Specifically, the term 1/(ms x + mp γQ )v+1−a in Φ2 can be expressed as 1 (ms x + mp γQ )v+1−a ! ms x !! a − v 1 1 1,1 G . = (mp γQ )v+1−a Γ(v + 1 − a) 1,1 mp γQ ! 0 (30) By substituting (30) into Φ2 and using [24, eq. (7.813.1)], we have i+j+1 γPmax 1 1 Φ2 = (mp γQ )v+1−a Γ(v + 1 − a) 2ms ! Pmax !! −i − j, a − v . (31) × G1,2 2,1 0 2mp Q ! Thus B(v) is obtained for the case of ms = mp γQ . For the case of ms = mp γQ , B(v) can be written as ∞ xi+j 1 − 2ms x B(v) = v e γPmax dx. v+1 ms 0 (1 + x) (32) By the same way we get Φ2 , we can obtain the closed-form expression for (32). In conclusion, we obtain B(v) as (33), shown at the bottom of the page. By substituting A and B(v) which are given by (28) and (33), respectively, into (25), we can get the closed-form expression for the system ergodic capacity. Fig. 2. Outage probability versus interference power constraint to noise ratio for different channel fading severity parameters with M = 6, N = 2, Pmax /σ 2 = 10 dB and R = 1 bit/s/Hz. VII. N UMERICAL R ESULTS AND D ISCUSSIONS In this section, we present numerical results to validate our theoretical analysis in Sections III–VI. A detailed investigation is given on the impact of the interference power constraint, the maximum transmit power constraint, the fading severity parameters, the number of relays and the relay selection scheme on the outage and diversity performance of the secondary system. The effect of limited feedback on the relay selection probability is also studied. Fig. 2 illustrates the exact outage probability of the secondary system versus the interference power constraint to noise ratio Q/σ 2 for various channel fading severity parameters. The number of relays M and the order of the selected relay N are set to 6 and 2, respectively. The solid lines represent our analytical results, and the square symbols represent the Monte Carlo simulation results. From Fig. 2, we can see that our analytical results match well with the simulation results, which validates our theoretical analysis. The outage performance improves with the increase of the fading severity parameters. It is worth noting that the outage probability has a decline in the low Q region. This is because when Q gets smaller, there are fewer relays available in the successful decoding relay set, so the chance that the N th best relay makes sense (N ≤ |R(s)|) gets smaller. Besides, the outage performance improves with the increase of Q in the median Q region, but will reach saturation when Q is large enough due to the existence of Pmax . Fig. 3 plots the outage probability of the secondary system versus the maximum transmit power constraint to noise ratio Pmax /σ 2 . For the similar reason of the case in Fig. 2, the outage ⎧ ! i+j+1 v γPmax ms Pmax ! −i−j,a−v 1 1 ⎪ ⎨ (mp γQ1−ms )v A− , G1,2 2,1 2mp Q ! 0 (mp γQ −ms )a (mp γQ )v+1−a Γ(v+1−a) 2ms a=1 B(v) = ! i+j+1 ⎪ ! γPmax γ ⎩ 1v 1 G1,2 Pmax ! −i−j,−v , ms Γ(v+1) 2ms 2,1 2ms 0 for ms = mp γQ for ms = mp γQ (33) 872 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 Fig. 3. Outage probability versus maximum transmit power constraint to noise ratio for different channel fading severity parameters with M = 6, N = 2, Q/σ 2 = 10 dB and R = 1 bit/s/Hz. Fig. 5. Outage probability versus interference power constraint to noise ratio for different orders of selected relays (N ) with M = 6, m1 = m2 = m3 = m4 = 2, Pmax /σ 2 = 10 dB and R = 1 bit/s/Hz. Fig. 4. Outage probability versus interference power constraint to noise ratio for different numbers of relays (M ) with N = 2, m1 = m2 = m3 = m4 = 3, Pmax /σ 2 = 10 dB and R = 1 bit/s/Hz. Fig. 6. Outage probability versus interference power constraint to noise ratio for different fading severity of the transmission links and the interference links with M = 6, N = 2, Pmax /σ 2 = 10 dB and R = 1 bit/s/Hz. probability has a decline in the low Pmax region and reaches saturation in the high Pmax region. Fig. 4 illustrates the impact of the number of relays M on the outage performance of the secondary system with the N th best relay selection (N = 2). In the median and high Q regions, more relays can provide better outage performance. It is also observed that the turning point between the low Q region and the median Q region shifts left with the increase of M . Fig. 5 gives the outage performance of the secondary system for different relay selection schemes. The number of relays M is set to 6. It shows that in the median and high Q regions, the outage performance decreases with the increase of the order of the selected relay (i.e., N ) since the performance of the second hop is worsened. In addition, we can observe that the turning point between the low Q region and the median Q region shifts right with the increase of N . The impact of the fading severity of the transmission links and the interference links on the outage performance is illustrated in Fig. 6. It is observed that the fading severity of the transmission links has great impact on the outage performance, but the fading severity of the interference links has little impact on the outage performance. This is in compliance with the results in [29] which investigates the imperfect CSI of the transmission links and the interference links. Fig. 7 plots the impact of the fading severity of the firsthop links and the second-hop links on the outage performance. From this figure, we can see that when the number of relays M is relatively small, the fading severity of the second-hop links has more influence on the outage performance than that of the first-hop links. But when M is relatively large, these two hops have nearly equal impact on the outage performance of the secondary system. Figs. 8 and 9 present the exact and asymptotic outage probability curves according to Sections III and IV. From these two figures, we can observe that the asymptotic outage probability is very close to the exact one in high SNR regions. It is indicated that our asymptotic outage probability expression can be used to effectively evaluate the outage performance ZHANG et al.: PERFORMANCE OF COGNITIVE RELAY NETWORKS OVER NAKAGAMI-m FADING CHANNELS Fig. 7. Outage probability versus interference power constraint to noise ratio for different fading severity of first-hop links and second-hop links with N = 2, Pmax /σ 2 = 10 dB and R = 1 bit/s/Hz. Fig. 8. The exact and asymptotic outage probability versus system SNR (γ = 1/σ 2 ) for different fading severity of the transmission links with M = 4, N = 2 and m2 = m4 = 2, Q = 10 dB, Pmax = 10 dB and R = 1 bit/s/Hz. of the secondary system in high SNR regions, even for noninteger fading severity parameters. According to our analysis in Section IV, the diversity order of the secondary system is min(m1 , m3 ) × (M − N ) + m3 , which coincides with the slope of the curves in these figures. In Fig. 10, the selection probability of the N th best relay with limited feedback is illustrated. From this figure, we can observe that the selection probability of the relay decreases with the increase of N . This means the probability that a better relay is selected is larger than the probability that a worse relay is selected in the limited feedback scenario. For the best relay, as the number of feedback bits L increases, its selection probability gets larger. For the worse relays (N ≥ 3), the selection probability decreases with the increase of L. Fig. 11 depicts the ergodic capacity of the secondary system with one single relay. It can be observed that the ergodic capacity improves with the increase of the interference power constraint Q and will reach saturation when Q is large enough 873 Fig. 9. The exact and asymptotic outage probability versus system SNR (γ = 1/σ 2 ) for different relay selection scheme with m1 = m2 = m3 = m4 = 2, Q = 10 dB, Pmax = 10 dB, σ 2 = 1 and R = 1 bit/s/Hz. Fig. 10. Selection probability of the N th best relay versus the number of feedback bits (L) with M = 10, m1 = m2 = 3, Q/σ 2 = 10 dB, Pmax /σ 2 = 10 dB and R = 1 bit/s/Hz. Fig. 11. Ergodic capacity versus interference power constraint to noise ratio for different fading severity of the transmission links and the interference links with M = 1, N = 1 and Pmax /σ 2 = 10 dB. 874 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 due to the maximum transmit power constraint Pmax . It can also be seen that the fading severity of the transmission links has much greater impact on the ergodic capacity than that of the interference links. As the fading of the transmission links gets severer, the capacity of the secondary system will decrease greatly. However, as the fading of the interference links gets severer, the capacity will increase when Q is small and will decrease when Q is large. VIII. C ONCLUSION AND F UTURE W ORKS In this paper, we study the performance of an underlay cognitive DF relay network with the N th best relay selection over Nakagami-m fading channels, considering both the maximum transmit power limit and the interference power constraint. The exact and asymptotic outage probability expressions for such system are derived. Through asymptotic analysis, we obtain the diversity order of the secondary system, which is min(m1 , m3 ) × (M − N ) + m3 , where M represents the number of relays and m1 , m3 denote the fading severity parameters of the first-hop transmission link and the secondhop transmission link. It is indicated that the fading severity of channels, the number of relays and the relay selection scheme have great impact on the outage performance of the secondary system. The selection probability of the N th best relay is given in the limited feedback scenario. Besides, we obtain the exact ergodic capacity for the special case where there is one single relay in the secondary system. The theoretical analysis is validated by simulations. The results show that the fading severity of the transmission links has more impact on the outage performance as well as the ergodic capacity than that of the interference links. In this paper the direct link for the secondary system is not considered. We believe that considering the direct link can be an interesting topic in our future works. The first summand in (34) can be calculated as n M −n Pmax gsi Pmax gsi ≥ γ < γ Pr I1 = Pr th th σ2 σ2 Q × Pr Pmax < gsp n M −n γth σ 2 γth σ 2 Q Fgsp = 1−Fgsi Fgsi Pmax Pmax Pmax ⎤n ⎡ ⎤M −n ⎡ 1 γth 1 γth γ m1 , m γ m1 , m γPmax γPmax ⎦ ⎣ ⎦ = ⎣1 − Γ(m1 ) Γ(m1 ) × 2Q γ m2 , Pmmax Γ(m2 ) max 2 m2 −1 −m2 t e mm 2 t dt Γ(m2 ) ⎤M −n+i ⎡ ∞ n γ m1 , m1γγQth t ⎦ ⎣ = Cni (−1)i Q Γ(m1 ) P i=0 × max mm2 tm2 −1 e−m2 t dt × 2 Γ(m2 ) According to the definition of R(s), the probability of the set R(s) can be written as ⎡ ⎤ # # Pr(R(s)) = Pr⎣ (γsi ≥ γth ), (γsi < γth )⎦ = Pr⎣ i∈R(s) σ2 (36) Then, utilizing the following expansion for an incomplete gamma function [24, eq. (8.352.6)]: % $ n−1 i x −x . (37) γ(n, x) = Γ(n) 1 − e i! i=0 ⎤ # Pmax gsi Q⎦ ≥ γth , < γth , Pmax< σ2 gsp i ∈R(s) I2 can be transformed into I2 = where the first summand denotes that all relays can successfully decode in the set R(s), and the second summand denotes that the other relays decode failed. Cni (−1)i ∞ Q Pmax ⎤M −n+i ⎡ γ m1 , m1γγQth t ⎦ ⎣ Γ(m1 ) mm2 tm2 −1 e−m2 t dt × 2 Γ(m2 ) I2 (34) n i=0 I1 ⎤ # # Qgsi Qg Q si ⎦, + Pr⎣ ≥ γth , < γth , Pmax ≥ σ 2 gsp σ 2 gsp gsp i∈R(s) i ∈R(s) ⎡ i ∈R(s) i∈R(s) # Pmax gsi (35) In the second summand, it is found that all parts are correlated with the variable gsp . Hence, the second summand can be written as n M −n ∞ Qgsi Qgsi ≥ γth < γth I2 = fgsp (t)dt Pr Pr Q σ2 t σ2 t Pmax ⎤n ⎡ ⎤M −n ⎡ ∞ γ m1 , m1γγQth t γ m1 , m1γγQth t ⎦ ⎣ ⎦ ⎣1 − = Q Γ(m ) Γ(m ) 1 1 P A PPENDIX A P ROOF OF L EMMA 1 ⎡ . = n Cni (−1)i i=0 × ⎡ ∞ Q Pmax ⎢ ⎣1−e j ⎤M −n+i m γ t − 1γ th Q 2 m2 −1 −m2 t e mm 2 t dt Γ(m2 ) m1 γth t m 1 −1 γQ j=0 j! ⎥ ⎦ ZHANG et al.: PERFORMANCE OF COGNITIVE RELAY NETWORKS OVER NAKAGAMI-m FADING CHANNELS = n M −n+i i=0 l i+l Cni CM −n+i (−1) l=0 ×e − ⎡ m1 γth l+m2 γQ γQ t ⎣ m 1 −1 j=0 1 j! 2 mm 2 Γ(m2 ) m1 γth t γQ ∞ tm2 −1 = Q Pmax j ⎤l ⎦ dt = (38) N k=1 N 875 Cnk−1 [Pr(γid ≤ x)]n−k+1 [Pr(γid < x)]k−1 Cnk−1 [Fγid (x)]n−k+1 [1−Fγid (x)]k−1 , (41) k=1 where Fγid (x) is expressed as According m 1 −1 ⎡ ⎣ 1 j! j=0 m 1 −1 1 j! j=0 to the multinomial theorem, the j l m1 γth t in (38) can be expanded as γQ m1 γth t γQ j ⎤l ⎦ m 1 −1 = l! w0 +w1 +···+wm1 −1 =l j=0 j wj 1 m1 γth t 1 wj ! j! γQ ⎧ ⎨ = l! term w0 +w1 +···+wm1 −1 =l m1 γth t ⎩ γQ H m 1 −1 j=0 ⎫ wj⎬ 1 1 , ⎭ wj ! j! (39) where H = m 1 −1 jwj . So I2 can be rewritten as j=0 I2 = n M −n+i i=0 l=0 × × = l i+l Cni CM −n+i (−1) w0 +w1 +...+wm1 −1 =l m1 γth γQ ∞ t H m 1 −1 j=0 m2 −1+H − e 1 wj ! wj 1 j! m1 γth l+m2 γQ γQ Q Pmax n M −n+i i=0 l=0 2 mm 2 l! Γ(m2 ) t dt l i+l Cni CM −n+i (−1) w0 +w1 +...+wm1 −1 =l 2 mm 2 l! Γ(m2 ) H m2 +H m1 γth γQ γQ m2 γQ + m1 γth l m wj 1 −1 1 1 m2 γQ + m1 γth l × Γ m2 +H, × . γPmax wj ! j! j=0 × min(Pmax , Q/gip )gid Fγid (x) = Pr(γid ≤ x) = Pr ≤ x σ2 Pmax gid Qgid Q Q ≤ x, Pmax< ≤ x, Pmax ≥ = Pr +Pr 2 σ2 g σ gip gip ip Pmax gid Q = Pr ≤ x Pr Pmax < σ2 gip J1 ∞ Qgid ≤ x fgip (t)dt . + Pr (42) Q σ2 t Pmax J2 Utilizing the CDF of gip and gid , J1 can be easily calculated as m4 Q 3x γ m3 , γm , γ m 4 Pmax Pmax . (43) J1 = Γ(m3 ) Γ(m4 ) For J2 , the integral can be calculated as following by using the expansion expression (37) as follows: ∞ γ m3 , m3 xt 4 m4 −1 −m4 t γQ mm e 4 t dt J2 = Q Γ(m ) Γ(m ) 3 4 Pmax i ∞ m3 −1 4 m x mm − γ 3 t 1 m3 x m4 −1 −m4 t 4 Q = 1−e dt t e t Γ(m4 ) P Q i! γQ i=0 max i 4Q 3 −1 Γ m4 , Pmmax 4 m mm 1 xm3 4 − = Γ(m4 ) Γ(m4 ) i=0 i! γQ −(m4 +i) xm3 xm3 + m4 γQ × + m4 Γ m4 + i, . γQ γPmax (44) Substituting (43) and (44) into (42), we can get Fγid (x). Then from (41), we can obtain Fγrd (x) as (5). (40) A PPENDIX C P ROOF OF L EMMA 3 Substituting (35) and (40) into (34), we can obtain (3). A PPENDIX B P ROOF OF L EMMA 2 According to the expression of γrd (4), the CDF of γrd can be written as Fγrd (x) = Pr N th max γid ≤ x i∈R(s) According to (9), I1 can be written as the following expression in high SNR regions: m1 ⎤n⎡ m1 ⎤M −n ⎡ m1 γth m1 γth m2 Q γ m , 2 γ→∞ Pmax γ Pmax ⎦ ⎣ Pmax γ ⎦ I1 ≈ ⎣1− Γ(m1 +1) Γ(m1 +1) Γ(m2 ) ≈ 1 Γ(m1 + 1) m1 M −n γ m , m2 Q 2 Pmax m1 γth . Pmax γ Γ(m2 ) (45) 876 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 33, NO. 5, MAY 2015 For I2 , it can be approximated written as m 1 ⎤ n ⎡ m1 ⎤M −n ⎡ m1 γth t m1 γth t ∞ γ→∞ Qγ Qγ ⎦ ⎣ ⎦ ⎣1 − I2 ≈ Q Γ(m + 1) Γ(m 1 1 + 1) P max mm2 tm2 −1 e−m2 t dt × 2 Γ(m2 ) m1 ⎤M −n m1(M−n) ⎡ ≈⎣ m1 γth Qγ Γ(m1 +1) ⎦ 1 m2 Γ(m2 ) m2 Q Γ m1 (M −n)+m2 , . Pmax (46) Substituting (45) and (46) into (34), we can obtain (10). A PPENDIX D P ROOF OF L EMMA 4 According to (9), J1 can be written as the following expression in high SNR regions: m 3 γ m , m 4 Q 4 Pmax γ→∞ m3 x 1 . (47) J1 ≈ Γ(m3 + 1) Pmax γ Γ(m4 ) For J2 , it can be written as m3 m4 m4 −1 −m4 t ∞ γ→∞ 1 m4 t e m3 xt dt J2 ≈ Q Γ(m3 + 1) Qγ Γ(m4 ) Pmax ⎞ ⎛ m 3 Γ m + m , m 4 Q 3 4 Pmax m3 x ⎠. ⎝ = (48) m4 Qγ Γ(m3 + 1)Γ(m4 ) Substituting (47) and (48) into (42), we can obtain the asymptotic expression for the CDF of γid as m 3 m 3 γ→∞ m3 x 1 1 Fγid (x) ≈ γ Γ(m3 + 1)Γ(m4 ) Pmax m 3 1 m4 Q m4 Q × γ m4 , Γ m3 +m4 , + . (49) Pmax m4 Q Pmax From (49), we can easily obtain that γ→∞ 1 − Fγid (x) ≈ 1. (50) Substituting (49) and (50) into (41), we can obtain the asymptotic expression for the CDF of γrd as (11). R EFERENCES [1] S. Haykin, “Cognitive radio: Brain-empowered wireless communications,” IEEE J. Sel. Areas Commun., vol. 23, no. 2, pp. 201–220, Feb. 2005. [2] A. Goldsmith, S. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum gridlock with cognitive radios: An information theoretic perspective,” Proc. IEEE, vol. 97, no. 5, pp. 894–914, May 2009. [3] R. Pabst et al., “Relay-based deployment concepts for wireless and mobile broadband radio,” IEEE Commun. Mag., vol. 42, no. 9, pp. 80–89, Sep. 2004. [4] J. Laneman, D. Tse, and G. Wornell, “Cooperative diversity in wireless networks: Efficient protocols and outage behavior,” IEEE Trans. Inf. Theory, vol. 50, no. 12, pp. 3062–3080, Nov. 2004. [5] D. S. Michalopoulos and G. K. Karagiannidis, “Performance analysis of single relay selection in Rayleigh fading,” IEEE Trans. Wireless Commun., vol. 7, no. 10, pp. 3718–3724, Oct. 2008. [6] S. S. Ikki and M. H. Ahmed, “Performance analysis of adaptive decodeand-forward cooperative diversity networks with best-relay selection,” IEEE Trans. Commun., vol. 58, no. 1, pp. 68–72, Jan. 2010. [7] T. Q. Duong, V. N. Q. Bao, and H.-J. Zepernick, “Exact outage probability of cognitive AF relaying with underlay spectrum sharing,” Electron. Lett., vol. 47, no. 14, pp. 1001–1002, Aug. 2011. [8] J. Lee, H. Wang, J. G. Andrews, and D. Hong, “Outage probability of cognitive relay networks with interference constraints,” IEEE Trans. Wireless Commun., vol. 10, no. 2, pp. 390–395, Feb. 2011. [9] Y. Guo, G. Kang, N. Zhang, W. Zhou, and P. Zhang, “Outage performance of relay-assisted cognitive-radio system under spectrum-sharing constraints,” IET Electron. Lett., vol. 46, no. 2, pp. 182–184, Jan. 2010. [10] Z. Yan, X. Zhang, and W. Wang, “Outage performance of relay assisted hybrid overlay/underlay cognitive radio systems,” in Proc. IEEE WCNC, Cancun, Mexico, Mar. 2011, pp. 1920–1925. [11] L. Luo, P. Zhang, G. Zhang, and J. Qin, “Outage performance for cognitive relay networks with underlay spectrum sharing,” IEEE Commun. Lett., vol. 15, no. 7, pp. 710–712, Jul. 2011. [12] Z. Yan, X. Zhang, and W. Wang, “Exact outage performance of cognitive relay networks with maximum transmit power limits,” IEEE Commun. Lett., vol. 15, no. 12, pp. 1317–1319, Dec. 2011. [13] H. Ding, J. Ge, D. Costa, and Z. Jiang, “Asymptotic analysis of cooperative diversity systems with relay selection in a spectrum-sharing scenario,” IEEE Trans. Veh. Technol., vol. 60, no. 2, pp. 457–472, Feb. 2011. [14] K. Tourki, K. A. Qaraqe, and M.-S. Alouini, “Outage analysis for underlay cognitive networks using incremental regenerative relaying,” IEEE Trans. Veh. Technol., vol. 62, no. 2, pp. 721–734, Feb. 2013. [15] V. N. Q. Bao, T. Q. Duong, D. B. Costa, G. C. Alexandropoulos, and A. Nallanathan, “Cognitive amplify-and-forward relaying with best relay selection in non-identical Rayleigh fading,” IEEE Commun. Lett., vol. 17, no. 3, pp. 475–478, Mar. 2013. [16] T. Q. Duong, D. B. Costa, M. Elkashlan, and V. N. Q. Bao, “Cognitive amplify-and-forward relay networks over Nakagami-m fading,” IEEE Trans. Veh. Technol., vol. 61, no. 5, pp. 2368–2374, Jun. 2012. [17] T. Q. Duong, D. B. Costa, T. A. Tsiftsis, C. Zhong, and A. Nallanathan, “Outage and diversity of cognitive relaying systems under spectrum sharing environments in Nakagami-m fading,” IEEE Commun. Lett., vol. 16, no. 12, pp. 2075–2078, Dec. 2012. [18] C. Zhong, T. Ratnarajah, and K. Wong, “Outage analysis of decode-andforward cognitive dual-hop systems with the interference constraint in Nakagami-m fading channels,” IEEE Trans. Veh. Technol., vol. 60, no. 6, pp. 2875–2879, Jul. 2011. [19] T. Q. Duong, K. J. Kim, H. J. Zepernick, and C. Tellambura, “Opportunistic relaying for cognitive network with multiple primary users over Nakagami-m fading,” in Proc. IEEE ICC, Budapest, Hungary, Jun. 2013, pp. 5668–5673. [20] S. S. Ikki and M. H. Ahmed, “On the performance of cooperativediversity networks with the Nth best-relay selection scheme,” IEEE Trans. Commun., vol. 58, no. 11, pp. 3062–3069, Nov. 2010. [21] S-I. Chu, “Performance of amplify-and-forward cooperative communications with the Nth best-relay selection scheme over Nakagami-m fading channels,” IEEE Commun. Lett., vol. 15, no. 2, pp. 172–174, Feb. 2011. [22] A. M. Salhab, F. Al-Qahtani, S. A. Zummo, and H. Alnuweiri, “Outage analysis of Nth-best DF relay systems in the presence of CCI over Rayleigh fading channels,” IEEE Commun. Lett., vol. 17, no. 4, pp. 697– 700, Apr. 2013. [23] X. Zhang, Z. Yan, Y. Gao, and W. Wang, “On the study of outage performance for cognitive relay networks (CRN) with the Nth best-relay selection in Rayleigh-fading channels,” IEEE Wireless Commun. Lett., vol. 2, no. 1, pp. 110–113, Feb. 2013. [24] I. S. Gradshteyn, I. M. Ryzhik, A. Jeffrey, and D. Zwillinger, Table of Integrals, Series and Products., 7th ed. New York, NY, USA: Academic, 2007. [25] Y. Zhao, R. Adve, and T. Lim, “Symbol error rate of selection amplifyand-forward relay systems,” IEEE Commun. Lett., vol. 10, no. 11, pp. 757–759, Nov. 2006. [26] R. Tannious and A. Nosratinia, “Spectrally-efficient relay selection with limited feedback,” IEEE J. Sel. Areas Commun., vol. 26, no. 8, pp. 1419– 1428, Oct. 2008. [27] J. Si, Z. Li, H. Huang, J. Chen, and R. Gao, “Capacity analysis of cognitive relay networks with the PU’s interference,” IEEE Commun. Lett., vol. 16, no. 12, pp. 2020–2023, Dec. 2012. [28] Meijer G-Function, 2013. [Online]. Available: http://functions.wolfram. com/PDF/MeijerG.pdf [29] K. Tourki, K. A. Qaraqe, and M. M. Abdallah, “Outage analysis of spectrum sharing cognitive DF relay networks using outdated CSI,” IEEE Commun. Lett., vol. 17, no. 12, pp. 2272–2275, Dec. 2013. ZHANG et al.: PERFORMANCE OF COGNITIVE RELAY NETWORKS OVER NAKAGAMI-m FADING CHANNELS Xing Zhang (M’10–SM’14) received the Ph.D. degree from Beijing University of Posts and Telecommunications (BUPT), Beijing, China, in 2007. Since July 2007, he has been with the School of Information and Communications Engineering, BUPT, where he is currently an Associate Professor. He is the author/coauthor of two technical books and more than 50 papers in top journals and international conferences and filed more than 30 patents (12 granted). His research interests are mainly in wireless communications and networks, green communications and energy-efficient design, cognitive radio and cooperative communications, traffic modeling, and network optimization. Prof. Zhang has served on the editorial boards of several international journals, including KSII Transactions on Internet and Information Systems and the International Journal of Distributed Sensor Networks, and as a TPC Cochair/ TPC member for a number of major international conferences, including MobiQuitous 2012, IEEE ICC/GLOBECOM/WCNC, CROWNCOM, Chinacom, etc. He received the Best Paper Award in the 9th International Conference on Communications and Networking in China (Chinacom 2014) and the 17th International Symposium on Wireless Personal Multimedia Communications (WPMC 2014). Yan Zhang (SM’10) received the Ph.D. degree from Nanyang Technological University, Singapore. Since August 2006, he has been with Simula Research Laboratory, Fornebu, Norway, where he is currently the Head of the Department of Networks. His recent research interests include wireless networks, machine-to-machine communications, and smart grid communications. He is a Regional Editor or an Associate Editor on the editorial board and a Guest Editor of a number of international journals. 877 Zhi Yan received the B.Sc. degree in mechanical engineering and automation and the Ph.D. degree in communication and information system from Beijing University of Posts and Telecommunications (BUPT), Beijing, China, in 2007 and 2012, respectively. From August 2012 to March 2014, he was a Researcher with the Network Technology Research Center, China Unicom Research Institute. He is currently an Assistant Professor with the School of Electrical and Information Engineering, Hunan University, Changsha, China. His current research interests are in the cognitive radio, cooperative communication, and cellular network traffic analysis and modeling. Jia Xing received the B.S. degree in communication engineering in 2012 from Beijing University of Posts and Telecommunications, Beijing, China, in 2012 where she is currently working toward the M.S. degree in the Key Laboratory of Universal Wireless Communications, School of Information and Communication Engineering. Her research interests include cognitive radio and cooperative communication. Wenbo Wang received the B.S., M.S., and Ph.D. degrees from Beijing University of Posts and Telecommunications (BUPT), Beijing, China, in 1986, 1989, and 1992, respectively. He is currently a Professor with and the Executive Vice Dean of the Graduate School, BUPT. He is also the Deputy Director of the Key Laboratory of Universal Wireless Communication, Ministry of Education. He has published more than 200 journal and international conference papers and 6 books. His current research interests include radio transmission technology, wireless network theory, and software radio technology.