Você está na página 1de 6

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO.

6, JUNE 2009

2807

Average SNR and Ergodic Capacity Analysis for Opportunistic DF Relaying with Outage over Rayleigh Fading Channels
Sungeun Lee, Student Member, IEEE, Myeongsu Han, Student Member, IEEE, and Daesik Hong Senior Member, IEEE

AbstractIn the paper, we deal with a single-selection opportunistic relaying with the decode-and-forward (DF) protocol over Rayleigh fading channels. The exact end-to-end average signalto-noise ratios (SNR) and ergodic capacities of both proactive and reactive opportunistic relaying are derived as a closed-form for arbitrary link SNR. In addition, the effective ergodic capacity satisfying the minimum required data rate without outage is also identied for both relaying schemes. The analysis results are used to demonstrate which relaying scheme outperforms the other for given system parameters. Index TermsOpportunistic relaying, decode-and-forward, ergodic capacity.

I. I NTRODUCTION OOPERATIVE relaying techniques have recently achieved a great deal of popularity as an efcient way to mitigate fading in wireless networks [1]. Generally, cooperative relaying has focused on simultaneous transmission from multiple relays [2]. However, recent research has shown that carefully selected relay transmission incurs no performance loss compared to multiple-relay transmission in terms of the diversity-multiplexing gain tradeoff and outage probability [3]. Cooperative opportunistic relaying describes a situation where a single relay among several relay nodes is selected depending on which relay provides the best end-to-end path between the source and destination in a distributed wireless network [3]. In the opportunistic decode-and-forward (DF) relaying scheme, there are two variant techniques which are outage-optimal: proactive (P-DF) and reactive (R-DF) [3]. In P-DF mode, a specied relay selected prior to the source transmission participates in the cooperation, whereas in R-DF mode, relays that successfully decode the message take part in the cooperation. Bletsas et al. studied an outage-based comparison of the two methods where they showed that the outage probability for PDF is exactly the same as that for R-DF [3]. Michalopoulos et
Manuscript received April 25, 2008; revised July 30, 2008 and November 5, 2008; accepted December 19, 2008. The associate editor coordinating the review of this letter and approving it for publication was K. B. Lee. S. Lee and D. Hong are with the Department of Electrical and Electronic Engineering, Yonsei University, Seoul, Korea (e-mail: {softmind, daesikh}@yonsei.ac.kr). M. Han is with the Republic of Korea Marine Corps, Korea. This research was supported in part by the Ministry of Knowledge Economy, Korea, under the ITRC (Information Technology Research Center) support program supervised by the IITA (Institute of Information Technology Assessment) (IITA-2008-C1090-0803-0002), and in part by Korea Science and Engineering Foundation through the NRL Program (Grant R0A-2007000-20043-0). Digital Object Identier 10.1109/TWC.2009.080574

al. extended this result over the environment where the sourcedestination channel is taken into account [4], and showed that the R-DF scheme is slightly better than P-DF in terms of outage, since R-DF seems to take better advantage of the direct source-destination channel. In addition, an approximate closed-form of the bit-error probability (BEP) for both the PDF and the R-DF modes is derived in [4]. However, evaluating the advantages and disadvantages of the two methods more precisely will require investigating the capacity in order to assess which opportunistic relaying method is superior to the other. Recently, the authors observe the end-to-end average signal-to-noise ratios (SNR) and ergodic capacities of two opportunistic DF relay schemes based on the assumption of a high SNR [5]. From [5], it is shown that the ergodic capacity of P-DF scheme outperforms that of R-DF scheme under high SNR regime where channel qualities of all relays satisfy the predetermined threshold. However, the impact of predetermined threshold on ergodic capacity is much more signicant in low to medium SNR than high SNR since only a part of relays can participate in the cooperation by the threshold in low to medium SNR. Therefore, it is very important to evaluate the exact performance in low to medium SNR regime to compare these two relaying schemes. In this paper, we analyze the exact end-to-end average SNR and the ergodic capacities of two opportunistic DF relaying schemes for arbitrary link SNRs over dual-hop Rayleigh fading channels. In addition, we introduce a useful notion, effective ergodic capacity, which corresponds to the spectral efciency of the successful transmissions excluding the outage event. This concept is inferred from the threshold-based DF relaying schemes [6][7]. This effective ergodic capacity is derived by discarding data packets that do not support a required spectral efciency. A potential advantage of this approach is that our analytic results can serve as a guide when examining the practical throughput of each relaying scheme, which reecting in the outage events, and this can be used to determine the superiority of one scheme over another for given system parameters. This paper is organized as follows: Section II presents the characteristics of the two opportunistic DF relaying schemes. Section III derives the end-to-end average SNRs of the two methods in order to explain the properties of the two relaying schemes. Section IV presents and analyzes the typical and effective end-to-end ergodic capacities for both relaying types, and Section V provides the simulation and numerical results verifying the analysis as well as a discussion of the capacity trends of the two schemes according to system parameters. Finally, our conclusions are presented in Section VI.

1536-1276/09$25.00 c 2009 IEEE

Authorized licensed use limited to: Yonsei University. Downloaded on July 5, 2009 at 22:22 from IEEE Xplore. Restrictions apply.

2808

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 6, JUNE 2009

II. C HARACTERISTICS OF T WO O PPORTUNISTIC DF R ELAY S YSTEMS Let us consider a half-duplex dual-hop scenario where a single source (S ) communicates with a single destination (D) through a total of M DF relays (Rm for the mth relay with m V0 = {1, 2, . . . , M }). We focus on the environment the direct path between the source and destination is blocked by an intermediate wall, while relays are located at the periphery of the obstacle, i.e., the direct path from S to D is not taken into account [3]. The instantaneous SNRs of the S -Rm and Rm -D channels are represented as 1,m and 2,m , which yield to a statistically independent exponential distribution with a mean of 1/1 and 1/2 , respectively [3]. The cumulative distribution functions (CDF) of 1,m and 2,m can be given by Fi,m (t) = 1 ei t for i = 1, 2, and t 0. As noted above, there are two kinds of opportunistic relay schemes, P-DF and R-DF. In P-DF, the best relay Rm is chosen to maximize the minimum of the channel strengths between the links S -Rm and Rm -D for all M relays, i.e., m = arg maxmV0 [min(1,m , 2,m )]. In R-DF, on the other hand, all relays listen to the signal from S , and only those with 1,m > T decode the received signal at the relay, where T denotes a predetermined SNR threshold indicating successful decoding at the receiver. The single relay with the best 2,m among them then transmits to D. Let us dene VT as the index set of the relay for which the instantaneous SNR of the S -Rm link satises the threshold, i.e., VT = {m | 1,m T for m V0 }. In this case, the end-to-end instantaneous SNRs for P-DF and R-DF can then be represented as P-DF = min(1,m , 2,m ) (1)

Both P-DF & R-DF paths

1,1 = 1.5
S

R1 R2 R3

2,1 = 2 2,2 = 3
D
S

1,1 = 1 1,2 = 2

R1 R2 R3
.

2,1 = 2 2,2 = 2.2


D

1,2 = 1

1,3 = 1.8

2,3 = 1.6

1,3 = 2.5
P-DF:

2,3 = 1.8
R-DF:

P-DF . path only


(a) low SNR, P-DF > R-DF = 0

(b) medium SNR, P-DF = R-DF

R-DF path

1,1 = 2
S

R1 R2 R3
P-DF path

2,1 = 4 2,2 = 2
D

1,2 = 3

1,3 = 3

2,3 = 3

(c) high SNR, P-DF > R-DF Fig. 1. Schematic comparisons of end-to-end SNR when the required SNR (threshold) is 2, i.e., T = 2. In case (a), the signal in the R-DF scheme is discarded whereas the signal in the P-DF is transmitted regardless of the outage. Both schemes select the same path to communicate in case (b). In case (c), in contrast, the end-to-end SNR for P-DF is better than that for R-DF.

where m = arg maxmV0 [min(1,m , 2,m )] as described, and 0 , if VT = , (2) R-DF = min(1,m , 2,m ) , otherwise, where m = arg maxm VT [2,m ]. Fig. 1 shows the schematic end-to-end SNR comparisons for both modes when the required SNR (threshold) is T = 2. Fig. 1(a) describes the low SNR case where the signal in the R-DF scheme is discarded at the relay since there is no decodable relay during the rst hop. The R-DF scheme drops the signal in advance, before the outage event happens at the destination. However, the P-DF scheme rst transmits the signal regardless of the outage, and then the outage occurs at the destination after the relay retransmits the signal. In Fig. 1(b), both PDF and R-DF select the same relay to communicate, so the end-to-end SNRs for both schemes are identical to each other. For high SNR regions, on the other hand, the SNR of the PDF outperforms that of the R-DF since the P-DF considers the SNRs of both links when choosing the best relay. This would seem to indicate that the behavior of the end-to-end average SNRs for two relaying schemes is different from the trend of the outage probabilities [3]. From the perspective of outage probability, both P-DF and R-DF have exactly the same results, regardless of the mean channel gain [3]. On the contrary, the end-to-end average SNR can be different according to the mean channel gain for given threshold. At

this point, let us analyze the end-to-end average SNRs for the P-DF and R-DF modes for a given threshold T , and then investigate how these SNRs are different by comparison. III. A NALYSIS OF E ND - TO -E ND AVERAGE SNR A. End-to-end Average SNRs for P-DF, P-DF To calculate the CDF of the selected relay in the P-DF relaying scheme, let min,m = min(1,m , 2,m ) and P-DF = max (min,m ). By order statistics [8], the CDF of min,m is

1 e(1 +2 )t , and the CDF of the instantaneous SNR for the selected relay, P-DF , can be described as FP-DF (t) = P P-DF t = P = 1e
(1 +2 )t M mV0

mV0

max (min,m ) t

(3)

In order to calculate the end-to-end average SNR and ergodic capacity, the following rule of integration by parts is used for the random variable x: x = E {x}a =
b b a

xfx (x) dx = [x Fx (x)]a

b a

Fx (x) dx,

(4) where E{x}b is the expectation of the x over the interval [ a, b], a and fx (x) and Fx (x) denote the probability density function (pdf) and cdf of the variable x, respectively. Using the binomial theorem (x + y )n =
n

(3) and applying the rule of integration by parts in (4), the

k=0

n k

xnk y k for

Authorized licensed use limited to: Yonsei University. Downloaded on July 5, 2009 at 22:22 from IEEE Xplore. Restrictions apply.

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 6, JUNE 2009

2809

end-to-end average SNR for the P-DF relaying scheme can be derived via some mathematical calculation from [9, Eq. 0.155.4]:
P-DF

Consequently, using (6) and (8), the total end-to-end average SNR of the R-DF schemes is R-DF =
M k=1

=E =

P-DF 0 M

=
m=1

M (1) m m (1 + 2 )

m+1

PT,k R-DF . k

(9)

1 1 + 2

HM 1 = , m 1 + 2 m=1

(5)

If the quality of the Rm -D link is improved continuously, i.e., 2 0, R-DF converges to 1/1 , meaning that R-DF k performance is bounded to the average SNR of S -Rm link if the average SNR of Rm -D is high enough [5]. IV. A NALYSIS OF E ND - TO - END E RGODIC C APACITY A. Typical and Effective Ergodic Capacities for P-DF, C P-DF and C P-DF eff Let us dene the end-to-end capacity as C = 1 2 log2 (1 + ), where is the instantaneous end-to-end SNR. This can then be transformed into = 2 2C 1. If the CDF of C denotes FC (y ), this can also be converted into F (22y 1). As a result, using (3), the CDF of the capacity for the P-DF mode, FC P-DF (y ), can be represented as FC P-DF (y ) = P 22C
P-DF

where HM is the M -th harmonic number. From (5), it can be easily inferred that the average SNR becomes greater as the number of relays M increases. Actually, HM can be interpreted as the SNR gain achieved by relay selection among 1 M relays since 1 + 2 is the end-to-end average SNR for a single relay. Note that this growth rate decelerates as the number of relays increases. B. End-to-end Average SNR for R-DF, R-DF In contrast to the P-DF method, the R-DF scheme rst chooses the relays {Rn } whose S -Rn link quality satises the threshold, i.e., n VT . Then, among these selected relays, only that one relay with the best Rn -D link quality transmits the signal. In the R-DF method, the probability PT,k that k relays will pass the threshold test (1,m T ) can be expressed by M k1 T M k PT,k = 1 e1 T e . (6) k Let us dene m k as the best relay index in R-DF schemes when there are k relay candidates that satisfy the threshold is condition (1,m T ). In this case, the CDF of 1,m k < t ) = derived as 1 e1 (tT ) u(t T ) since P (1,m k P (1,m < t|1,m T ), and the CDF of 2,m is [1 e2 t ]k . k Then, the CDF of the instantaneous SNR for k relay candiR-DF dates, k , is given by

1 22y 1
2y

= 1 e(1 +2 )(2

1)

(10) .

The integration of the exponential form in (10) can be rearranged as


0

ez(2

2y

1)

dy =

ez E1 (z ), 2 ln 2

(11)

where E1 (z ) is the exponential integral function dened as t E1 (z ) = z e t dt [9, ch.8.21]. Hence, exploiting (4) and (10), the end-to-end ergodic capacity, C P-DF , is obtained in (12) at the bottom of the page. In the P-DF scheme, the best relay, Rm , decodes the received message, and always re-transmits a re-encoded message to D regardless of the outage. Therefore, communication through this relay can fail due to outage when either of the (t) = 1 P min(1,m , 2,m )>t FR-DF k k two hops (S -Rm or Rm -D link) fails; see Fig. 1(a). Although k 2 t k no reliable information is delivered to D for this case, the = 1e ergodic capacity C P-DF still includes the case in the calculation. 1 (tT ) 2 t k 1 1e u(t T ). Consequently, C P-DF is not a good measure to evaluate the + 1e (7) realistic system performance in terms of practical throughput. Therefore, to measure the practical end-to-end performance Applying the rule in (4) for (7), the end-to-end average SNR while taking the outage into account, we use the effective for R-DF can be given by ergodic capacity C P-DF eff , which as dened here corresponds to R-DF R-DF the average spectral efciency of the successful transmissions k = E k 0 excluding the outage event. C P-DF is derived in (13) at the eff k m2 T m2 T 1 k 1 e e m+1 log bottom of the page where R = 2 (1 + T ) is the spec= + (1) . 2 m m + m 2 1 2 tral ef ciency required to decode the message successfully, m=1 z 2y E ( z ( T + 1)). This effective (8) and R ez(2 1) dy = 2 e ln 2 1
M

C P-DF = E C P-DF C P-DF = E C P-DF eff


R

M 1 (1)m+1 em(1 +2 ) E1 (m(1 + 2 )) 2 ln 2 m=1 m


M M

(12)

= R 1 1 e(1 +2 )T

+
m=1

M E1 (m (1 + 2 ) (T + 1)) (1)m+1 em(1 +2 ) m 2 ln 2

(13)

Authorized licensed use limited to: Yonsei University. Downloaded on July 5, 2009 at 22:22 from IEEE Xplore. Restrictions apply.

2810

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 6, JUNE 2009

EndtoEnd Ergodic Capacity [bits/s/Hz]

ergodic capacity is evaluated within the interval [R, ] to include only successful decoding cases. The rst term in (13) represents the average capacity limited by the threshold, and for low mean SNR regions below this term dominates C P-DF eff the threshold T . This term in the effective ergodic capacity enables us to observe the achievable data rate in the low SNR region properly, instead of using the impractical ergodic capacity. On the other hand, for high mean SNR values, i.e., 1/i T , C P-DF converges to C P-DF since the SNR quality of eff the selected relay link is good enough to exceed the threshold. Consequently, C P-DF can be an upper bound for C P-DF eff . B. Typical and Effective Ergodic Capacities for R-DF, C and C R-DF eff
R-DF

EndtoEnd Typical and Effective Ergodic Capacity 5 P/RDF relaying [ Typical, analysis] P/RDF relaying [Effective, analysis] PDF relaying [ Typical, simulation] RDF relaying [ Typical, simulation] PDF relaying [Effective, simulation] RDF relaying [Effective, simulation]

4.5

3.5

2.5

R=1.5 & 3.0 (T=7 & 63)


1.5

R=1.5 (T=7)
1

R=3.0 (T=63)

As with the average SNR analysis in Section III-B, we rst evaluate the ergodic capacity of the R-DF scheme for each k relay candidates to derive the total ergodic capacity. Then, using this capacity in conjunction with the probability that k relays will be selected, we obtain the total end-to-end ergodic capacity for R-DF. The CDF of the capacity for the R-DF R-DF (y ), can be represented mode with k relay candidates, FCk as
R-DF (y ) = 1 e2 (2 FCk 2y

0.5

10

15 Mean SNR of the Hop [dB]

20

25

30

1)

+ 1 e1 (2
1) k

2y

22R )

Fig. 2. Typical and effective ergodic capacity vs. mean channel gain for the proactive and reactive DF relaying schemes. The number of relays and the required spectral efciency are M = 4 and R = 1.5, and 3 bps/Hz, respectively, with the assumption that 1 = 2 . The lines represent the analysis, and the markers denote the simulation results.

1 1 e2 (2

2y

u(y R), (14)

exclude decoding failure cases. The overall effective end-toend ergodic capacity of R-DF scheme can then be described as C R-DF = eff
M

by substituting y = 1 2 log2 (1 + t) into (7). After some manipulation with (11), the end-to-end average capacity of , is obtained by the R-DF when k relays are selected, C R-DF k (15) at the bottom of the page. Using (15) in conjunction with the relay selection probability (6), the total end-to-end average capacity of the R-DF relaying scheme can be obtained as C R-DF =
M k=1

PT,k C R-DF k,eff .

(18)

k=1

V. S IMULATION AND N UMERICAL R ESULTS In this section, we provide simulation and numerical results to verify the analytic results in the previous section. In addition, we will use this data to examine the tendencies of the typical and effective ergodic capacities for various environments. Fig. 2 evaluates the typical and effective end-to-end ergodic capacities vs. the mean channel gain per hop for the P-DF and R-DF methods. The number of relays and the pre-determined spectral efciency are M = 4 and R = 1.5 and 3 [bps/Hz], respectively, with 1 = 2 . Recall that R corresponds to a xed value associated with the pre-determined threshold T , and may vary from application to application; in fact, it determines the number of relays that belong to VT in the R-DF scenario, and whether or not the information is delivered to D from the selected relay for measuring the effective capacity. For a given spectral

PT,k C R-DF . k

(16)

As described above, the R-DF scheme investigates the signal quality at the relay in advance to prevent outage events. Nevertheless, it is still possible to fail to communicate at the destination due to outage since this scheme examines the S -Rm link quality only, not both S -Rm and Rm -D link qualities: outage occurs if the quality of the Rm -D link is poor. Therefore, the R-DF scheme also needs an effective measure for examining the practical ergodic capacity. The effective ergodic capacity for R-DF schemes with k relay candidates, C R-DF k,eff , can be derived by (17) expressed at the bottom of the page. This term is also evaluated in the interval [R, ) to
0

R-DF C R-DF = E Ck k k

=
m=1 R-DF C R-DF k,eff = E Ck

2 m k m+1 e E1 (2 m) E1 (2 m (T + 1)) + e1 (T +1) E1 ((1 + 2 m) (T + 1)) (1) m 2 ln 2 R

(15)

= R 1 1 e2 T

+
m=1

2 m k m+1 e (1) e1 (T +1) E1 ((1 + 2 m) (T + 1)) m 2 ln 2

(17)

Authorized licensed use limited to: Yonsei University. Downloaded on July 5, 2009 at 22:22 from IEEE Xplore. Restrictions apply.

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 6, JUNE 2009

2811

efciency R, we can show that both the typical and effective ergodic capacity analysis in (12), (13), (16), and (18) are in exact agreement with the simulation results. Therefore, we can conrm that both ergodic capacities of the P-DF and R-DF can be evaluated by these results for variable circumstances. First, let us focus on the typical ergodic capacities C P-DF and C R-DF which do not take outage events at D into consideration. In the gure, solid lines denote the typical ergodic capacities for both schemes, P-DF and R-DF. Note that, in the low SNR regions below the threshold, C P-DF has only one mono-increasing curve, whereas C R-DF has two curves, corresponding to each required rate R, both of which are almost zero restricted by the threshold. The P-DF scheme does not concern itself with any successful decoding constraint and always retransmits the signal to D irrespective of the outage condition. Consequently, the typical capacity curve for P-DF is independent of the threshold. In contrast, the R-DF scheme may suffer a pause in transmission at the relay if outage is expected at D due to poor S -Rm link quality. Therefore, the typical capacity for R-DF has a tiny value for mean SNR values below the threshold. As a result, there is a big gap between the typical ergodic capacity of P-DF and R-DF in the low SNR regions. In addition, for high mean SNR regions in which all M relays are supposed to be chosen as candidates, C P-DF outperforms C R-DF . This is because the P-DF scheme considers both links when selecting the best relay to transmit, while only the Rm -D link is considered in the R-DF method. Actually, it is an interesting result that the typical ergodic capacities of both schemes differ according to the mean SNR since the outage probabilities of both schemes are exactly the same for the same mean SNR [3][5]. To combine these different results simultaneously, we introduce and evaluate the effective ergodic capacity concerning the decoding failure at D. A remarkable fact of these capacity results is that the effective ergodic capacities of both schemes, P-DF and R-DF, are almost the same for the low-to-medium SNR regions. This means that the practical system throughput for both schemes is very similar over low-to-medium SNR regions, even though two schemes use different methodologies to select the best relay. In addition, the outage tendency of the P-DF scheme can be inferred from the effective ergodic capacity curve, compared to the typical capacity curve. Accordingly, a fairer comparison of the capacities between the P-DF and R-DF relaying schemes can be done, especially for the low-to-medium SNR regime, by applying the effective concept, which takes the outage into consideration. Moreover, in a high mean SNR regime, the effective ergodic and C R-DF approach the general ergodic capacities C P-DF eff eff P-DF capacities C and C R-DF . Therefore, we can state that the effective ergodic capacity is bounded to the typical capacity for high SNRs. Since C P-DF > C R-DF for high SNR, , C P-DF eff is also better than C R-DF for this regime. eff Fig. 3 describes how the effective ergodic capacity can be varied as the number of relays M changes. The thresholds are xed to T = 7 in the analysis. As shown in the gure, the effective ergodic capacity of R-DF becomes saturated very quickly compared to that of the P-DF scheme, and this trend is more prominent for high mean SNR values. For example,

EndtoEnd Effective Ergodic Capacity, R=1.5 (T=7), =


1

3.5

3 EndtoEnd Ergodic Capacity [bits/s/Hz]

2.5

1 = 0.01 (20dB)

1.5

= 0.1 (10dB)
1

0.5 PDF relaying [Analysis] RDF relaying [Analysis] 0 2 4 6 8 10 12 14 Number of Relays 16 18 20 22 24

Fig. 3. Effective ergodic capacity analysis vs the number of relays M for the P-DF and R-DF relaying schemes. The threshold and the mean SNR per hop are T = 7, and 1/1 = 1/2 = 10 and 20 dB, respectively.

in the case of R-DF with 1 = 0.01 (20dB), no additional improvement is achieved for M > 4 by increasing the number of relays. This is because the diversity gain achieved through relay selection is limited to consideration only of the Rm -D link in the R-DF mode, as mentioned in (2). If there are |VT | relay candidates1 whose S -Rm link SNRs are better than the threshold, the R-DF method selects the one relay which has the best channel on the Rm -D link from those candidates, i.e., m VT . The selected relay, Rm , therefore, has the best channel gain for the Rm -D link. However, this does not guarantee that the channel gain of the S -Rm link for the selected relay is also the best compared to that of the other relays. The end-to-end link quality through the selected relay may not be good, even though the Rm -D link is the best, which means that the overall link quality is dominated by the quality of the S -Rm link for the R-DF scheme. Consequently, the effective ergodic capacity of the R-DF scheme can be bounded to a certain value even though the number of relays increases. In contrast, the effective ergodic capacity of the P-DF method improves as the number of relays increases, similar to the average SNR in (5). In the P-DF relaying scenario, both the S -Rm and Rm -D links are considered simultaneously when choosing the best relay. Therefore, the selection diversity order can be increased by increasing the number of relays in the PDF. Fig. 4 depicts the effective ergodic capacity depending on the relative channel quality between the S -Rm and Rm -D links. Let us dene the relative channel quality ratio as = 2 /1 for given 1 . This term then determines whether or not the S -Rm link is better than Rm -D: Lower values (below 1) mean better Rm -D link quality; higher values (above 1) indicate better S -Rm link quality. Actually, can be also expressed by the path-loss exponent and distance. When the
1 |X |

stands for the cardinality of a set X : number of elements in the set

Authorized licensed use limited to: Yonsei University. Downloaded on July 5, 2009 at 22:22 from IEEE Xplore. Restrictions apply.

2812

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 6, JUNE 2009

EndtoEnd Effective Ergodic Capacity, R=1.5 (T=7), M=4 4 PDF relaying [Analysis] RDF relaying [Analysis] 3.5

3 EndtoEnd Ergodic Capacity [bits/s/Hz]

are different, consideration should be given to which type of relaying scheme is better for that environment. In addition, it is better to use the R-DF scenarios when the SNRs of the S -Rm and Rm -D links vary frequently, as in the case of a mobile relay environment, for example. VI. C ONCLUSION

2.5

1 = 0.01 (20dB) RD link is better

SR link is better

1.5

= 0.1 (10dB)
1

0.5

0 12

6 3 0 3 SR link / RD link mean SNR ratio (10 log

10

6 ( / )) [dB]
2 1

12

Fig. 4. Effective Ergodic capacity analysis vs. relative channel quality ratio , dened as = 2 /1 . The term indicates whether or not the S -Rm link is better than the Rm -D link for given 1 . The threshold and the number of relays are T = 7, and M = 4, respectively.

signal power is reduced by long-term fading with the pathloss exponent , and the link distances of S -Rm and Rm -D are d1 and d2 , respectively, the ratio can be described as d2 = . Therefore, this result can be also interpreted d1 the performance with respect to the relay position. Of interest in the result inferred from the gure is that P-DF is more sensitive to this disparity between the S Rm and Rm -D SNR values. As mentioned above, the P-DF relaying scheme reects both the S -Rm and Rm -D links when choosing the path, whereas the R-DF scheme considers only the Rm -D link when selecting the relay. Therefore, it is clear that the P-DF method is more deeply affected by the SNR difference between the two links. In addition, the effective ergodic capacity of the R-DF scheme maintains almost the same value in the interval < 1 for the high SNR regime. This is because the overall link quality for R-DF is not affected by the increase in Rm -D quality for the xed S Rm SNR value, 1 , since the capacity is primarily dominated by the S -Rm link, as mentioned above; in Section III-B it was converges to 1/1 as the shown that the average SNR R-DF k quality of the Rm -D links improves (2 0). Consequently, when the channel qualities for the S -Rm and Rm -D links

We evaluated the performance of proactive and reactive opportunistic decode-and-forward relaying schemes from the viewpoints of the end-to-end average SNR and the effective ergodic capacity with respect to outage events, and numerical and our simulation results veried the correctness of our analysis for arbitrary link SNRs. The derived average SNR and capacity are expressed by the number of relays, a predetermined threshold to prevent outage, and the arbitrary average SNR of the S -Rm and Rm -D links. From these results, it is possible to determine which opportunistic relaying scheme is better in terms of SNR and effective capacity for all relaying systems with arbitrary parameters. In general, the proactive scheme is proven to outperform the reactive one in terms of general and effective ergodic capacity, whereas the outage probabilities of both schemes are the same. R EFERENCES
[1] J. N. Laneman, D. Tse, and G. W. Wornell, Cooperative diversity in wireless networks: efcient protocols and outage behavior, IEEE Trans. Inform. Theory, vol. 50, no. 12, pp. 30623080, 2004. [2] P. Anghel and M. Kaveh, Exact symbol error probability of a cooperative network in a Rayleigh-fading environment, IEEE Trans. Wireless Commun., vol. 3, no. 5, pp. 14161421, Sept. 2004. [3] A. Bletsas, H. Shin, and M. Win, Cooperative communications with outage-optimal opportunistic relaying, IEEE Trans. Wireless Commun., vol. 6, no. 9, pp. 34503460, Sept. 2007. [4] D. Michalopoulos and G. Karagiannidis, Performance analysis of single relay selection in Rayleigh fading, IEEE Trans. Wireless Commun., vol. 7, no. 10, pp. 37183724, Oct. 2008. [5] S. Lee, M. Han, and D. Hong, Average SNR and ergodic capacity analysis for proactive and reactive DF relaying over Rayleigh fading channels, in Proc. IEEE Vehicular Technology Conference VTC Spring 2009, to be published. [6] A. Adinoyi and H. Yanikomeroglu, Cooperative relaying in multiantenna xed relay networks, IEEE Trans. Wireless Commun., vol. 6, no. 2, pp. 533544, Feb. 2007. [7] N. Yi, Y. Ma, and R. Tafazolli, Rate-adaptive bit and power loading for OFDM based DF relaying, in Proc. IEEE Vehicular Technology Conference VTC Spring 2008, May 2008, pp. 13401344. [8] A. Papoulis, Probability, Random Variables, and Stochastic Processes, 4th. ed. New York: McGraw-Hill, 2002. [9] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and Products, 6th ed. New York: Academic, 2000.

Authorized licensed use limited to: Yonsei University. Downloaded on July 5, 2009 at 22:22 from IEEE Xplore. Restrictions apply.

Você também pode gostar