You are on page 1of 5

Fairness-Constrained Rate Enhancing Superposition Coding Scheme for a Cellular Relay System

Megumi Kaneko, Kazunori Hayashi, Petar Popovski# and Hideaki Sakai


Graduate School of Informatics, Kyoto University, Yoshida Honmachi Sakyoku, Kyoto, 606-8501, Japan Email: {meg,kazunori,hsakai}@i.kyoto-u.ac.jp # Dept. of Electronic Systems, Aalborg University, Aalborg, DK-9220, Denmark. Email: petarp@es.aau.dk
Abstract We consider the Downlink (DL) of a cellular relay system where in each sector, multiple users are served by one Base Station (BS) and one Relay Station (RS). We rst design a scheme based on Superposition Coding (SC) that serves two users simultaneously over the same spectral resource. By superposing the messages to the two users into three layers, this scheme allows both users to benet from the relayed links, thereby improving sum rate. The power allocation parameters of each SC layer are optimized to maximize sum rate while guaranteeing equal rates to each user. Then, this scheme is integrated in the proposed scheduler that exploits simultaneous user allocation. Compared to conventional schedulers based on orthogonal user allocation, the proposed scheduler outperforms the overall sum-rate under the fairness constraint. 1

I. I NTRODUCTION The benets of cooperative diversity in wireless relay systems have been shown in a number of works such as [1] [2] [3]. As a signal sent by a source node is received both at the relay and at the destination node, signicant performance gains can be obtained by combining these signals at destination. Recently, [4] introduced a wireless relaying scheme based on Superposition Coding (SC) that outperformed conventional cooperative diversity schemes, in a single user system with one Base Station (BS) and one Relay Station (RS) based on Decode-and-Forward (DF) half-duplex relaying. SC was rst introduced as the capacity achieving scheme in the Gaussian broadcast channel, where messages to all nodes are superposed and decoded by Successive Interference Cancelation (SIC) [5]. In [4], two messages termed basic and superposed are created, both destined to the single user, and are transmitted by the BS in the rst step. In the second step, after decoding both messages, the RS only forwards the superposed message, thereby reducing the RS transmission time for a given amount of data and increasing the overall rate. Using the superposed message and the direct signal received in the rst step, the destination retrieves the basic message. This scheme was applied to the two-user case in [6], where the basic and superposed messages were each destined to a different user. However, the achievable rate of this scheme under the constraint of allocating equal rates to each user is very limited, especially when the direct link quality of one of the users drops.
1 This work was supported by the Grants-in-Aid for JSPS Fellow no. 204205, for Scientic Research no. 21760289 and no. 10595739 from the Ministry of Education, Science, Sports, and Culture of Japan.

In this work, we address the Downlink (DL) transmissions in a two-user relaying system, where the two users MS1 and MS2 are served by a BS and a half-duplex RS as shown in Fig. 1. We design the 3-Layer 2-User SC (3L-2USC) for this system, where the worst direct link is not used. As fairness is an essential metric in a multi-user system, we aim at maximizing the achievable sum rate under the assumption of serving equal rates to each user. By superposing the messages to the two users into three SC layers and applying the optimal power allocation among layers derived by analysis, the proposed scheme allows both users to benet from their high RS-(MS1 ,MS2 ) relayed links qualities, unlike in [6]. The results in the two-user setting show that, compared to a conventional scheme that achieved the best sum rate so far, our schemes further improve sum rate under the considered fairness constraint. In addition, we evaluate the impact of the proposed 3-Layer 2-User SC scheme in the general multi-user case by integrating it into a scheduling algorithm that allocates the best user pair among all users. Results show that the proposed scheduler outperforms the sum-rates of conventional schedulers based on orthogonal user allocation as in [7], thereby indicating the potential benets of simultaneous user allocation through SC in a cellular relay system. II. S YSTEM M ODEL We rst consider the DL transmissions in a single-antenna, single-frequency system with a BS, a RS and two MSs, MS1 and MS2 as in Fig. 1, where only the direct link with higher quality is used (MS1 here). The BS transmits a vector x of M complex baseband symbols x[m], m {1, .., M }. The received signals at the RS and MS1 are yR = hR x + zR , yD1 = hD1 x + z1 . (1) RS transmits a vector xR of MR complex baseband symbols and the received signal at MSi , i {1, 2}, is given by yRi = hRi xR + z i . (2) hR , hD1 , hRi are the complex channel coefcients of the BSRS, BS-MS1 and RS-MSi channels. zR , z1 , z i are vectors of complex additive white Gaussian noise with a circularsymmetric distribution CN (0, 2 ). The transmitted symbols have mean zero, E{x[m]} = 0, and power normalized to one,

S| E{|x[m]|2 } = 1. The link SNRs are given by S = |h 2 , S {D1, R, R1, R2}. The capacity of each link is given by C (S ) = log2 (1 + S ) [bits/s] for a normalized bandwidth of 1 Hz. We analyze the case where link SNRs are ordered as

D1 < R2 < R1 < R ,

(3)

as the derivations for all other channel orderings are obtained similarly and used in the general scheduling algorithm as shown in the simulation results.
%6

56 06 06

Fig. 2.
6WHS  %6
[

Cellular Relay System


6WHS  %6

Fig. 1.

System Model

In the general multi-user case, we consider a three-sectored cell as depicted in Fig. 2 with 1 km radius and three relays placed 600 m away from the BS in an equidistant manner. Users are generated with a higher density towards the cell edge following the model in [8], which is a typical scenario where relays are mostly of use. We assume that the relays are xed and deployed in a manner that ensures high and constant BS-RS link quality, xed to 40 dB [9]. The SNR values are assumed to remain constant during each scheduling time frame but undergo Rayleigh fading from frame to frame, and path loss decays with distance with exponent three. III. P ROPOSED S CHEME II: 3-Layer 2-USC S CHEME Compared to the Single-User SC (SU-SC) scheme in [4] and the scheme in [6] which will be referred as 2-Layer 2-USC (2L-2USC), the proposed 3-Layer 2-USC (3L-2USC) scheme adopts an innovative approach by enabling both users to take full advantage of the three available links (direct and relayed links for MS1 and relayed link for MS2 ) by the three SC layers, viewing the RS to (MS1 ,MS2 ) channel as a broadcast channel. As this scheme enables to share the relayed links capacity between both users, we can expect sum rate enhancement compared to 2L-2USC, especially when the BS-MS1 link is bad. A. Description of the Steps Step 1: The BS sends a message x = b1 xb1 + s1 xs1 + 2 x2 , (4)

\5 \'

56
\5 [5 \5

56 06

06

06

06

Fig. 3.

Proposed Scheme

Then, the RS receives yR = hR ( b1 xb1 + s1 xs1 + 2 x2 ) + zR .

(6)

As explained in [10] for the broadcast channel, allocating higher power to the links with worst SNR provides the best throughput-fairness trade-off, as signals with lower link SNRs such as D1 , benet from higher allocated power, while receiving only low interference from signals with higher link SNRs as R1 , R2 . Thus, the decoding order should follow the order of increasing link SNRs, i.e., xb1 x2 xs1 , given (3). This ordering entails the optimality of the condition b1 R > 2 R > s1 R , as shown in Section V. RS rst decodes xb1 and cancels hR b1 xb1 , obtaining y R = hR ( s1 xs1 + 2 x2 )+ zR . Then, RS decodes x2 and cancels its contribution, obtaining y R = hR s1 xs1 + zR , from which xs1 is decoded. On the other hand, MS1 receives yD1 = hD1 ( b1 xb1 + s1 xs1 + 2 x2 ) + z1 (7) and keeps it in memory. Step 2: RS sends a message with superposed xs1 and x2 , xR = 1 xR1 + xR2 , (8) where xR1 , xR2 are the decoded and remodulated signals of xs1 and x2 , respectively, and [0, 1] denotes the power allocation ratio. Then, MS1 receives yR1 = hR1 ( 1 xR1 + xR2 ) + z 1 . (9)

composed of 3 superposed messages: xb1 , xs1 destined to MS1 and x2 destined to MS2 with power allocation ratio b1 , s1 and 2 [0, 1], respectively. We refer to message xb1 as the basic message for MS1 and xs1 as the superposed message for MS1 . The sum of power allocation ratios is equal to one, b1 + s1 + 2 = 1. (5)

From yR1 , MS1 , having the higher link SNR, decodes xR2 (i.e., x ), treating 1 xR1 as interference. Canceling 2 hR1 xR2 from yR1 , we get y R1 = hR1 1 xR1 + z 1 , from which xR1 (i.e., xs1 ) is decoded. From yD1 in (7), MS1 cancels x2 and xs1 , giving y D1 = hD1 b1 xb1 + z1 , from which xb1 is decoded. In the same way, MS2 receives yR2 = hR2 ( 1 xR1 + xR2 ) + z 2 , (10)

C. Optimizing the Power Allocation We rst determine the optimal power ratio between xR1 and xR2 . Once Rs1 and R2 are xed, max(M1 , M2 ) is minimized for a , denoted , such that M1 = M2 , as M1 and M2 are monotonically increasing and decreasing function of , respectively. We prove that there always exists a in [0, 1]. We dene f ( ) = Rs1 RR2 R2 RR1 , and f ( ) = 0 is equivalent to M1 = M2 . We can show that this is a monotonically increasing function of , and since f (0) < 0 and f (1) > 0, there always exists a unique in [0, 1] such that M1 = M2 . Due to the difculty to obtain a closed-form expression of , we use Newtons method to determine A B numerically. This gives, with Rb1 = min(Rb1 , Rb1 ), R3L2USC = 2(Rb1 + Rs1 ) . b1 +Rs1 1+ R RR2 ( ) (20)

from which xR2 is decoded. B. Achievable Rate We denote Rb1 and Rs1 the transmission rates of the basic and superposed messages to MS1 and R2 the rate of the message to MS2 in Step 1, and RRi , i = {1, 2}, denotes the rate of the relayed messages xRi on the RS-MSi links. As all messages are coded with Gaussian codebooks, we get the following rate constraints Step 1: For the decoding at RS, Rb1 R2 Rs1 C b1 R 1 + (s1 + 2 )R 2 R , C 1 + s1 R C (s1 R ) ,
A = Rb1 ,

Next, we consider the different cases following Rb1 . A B Case 1: Rb1 Rb1 b1 b1 , where we dene
b1 =1

1 D1

(11) (12) (13)

1 . R

(21)

which ensure that RS can decode xb1 , x2 and xs1 respectively. Step 2: For the decoding at MS1 , we get RR2 RR1 Rb1 C R1 1 + (1 )R1 C ((1 )R1 ) , C (b1 D1 ) , =
B Rb1 ,

A . In addition, we need to distinguish In this case, Rb1 = Rb1 different cases for b1 1 or b1 > 1. R (a) 0 b1 b1 1: Here, b1 0 D1 1+ R and b1 1 D1 R . Then, condition (19) becomes

Rb1 + Rs1 = C

b1 R 1 + (1 b1 )R

+ C (s1 R ) = R2 . (22)

(14) (15) (16)

As R3L2USC is an increasing function of Rb1 +Rs1 as derived from Eq. (20), b1 , s1 should be maximized. Thus, we set b1 = b1 . From (22), we get s1 = s1 , where
= s1

which ensure that MS1 can decode xR2 (x2 ), xR1 and xb1 . For the decoding at MS2 , we obtain RR2 C R2 1 + (1 )R2 , (17)

1 1 , R D1 1 + R

(23)

which ensures that MS2 can decode xR2 . (14) is satised a fortiori given that R2 < R1 , so only (17) is considered. A B Rb1 , Rb1 denote the two constraints on Rb1 to be satised. In the next section, we determine the optimal power allocation A B variables that satisfy Rb1 = min(Rb1 , Rb1 ) and equality in A B (12)-(14), (15) and (17). BS transmits M (min(Rb1 , Rb1 )+ Rs1 + R2 ) bits in Step 1. In Step 2, RS forwards M (Rs1 + R2 ) bits. The transmission symbol time MR at Step 2 is the larger time between the times required to transmit to MS1 Rs1 M R2 and MS2 , e.g., M1 = M RR1 and M2 = RR2 . Thus, MR = max(M1 , M2 ). The sum rate is now R3L2USC =
A B , Rb1 ) + Rs1 + R2 min(Rb1

R < 1. We have s1 0 D1 1+ , in which if 0 s1 R R , case b1 + s1 < 1. On the other hand, if D1 > 1+ R then s1 0, so we choose s1 = 0. This boils down to the 2-Layer 2-USC scheme, which is thus a special case of the 2-Layer 2-USC scheme. Thus, we obtain R R , i) if 1+ R D1 1+R

R3L2USC =

(1ai)

log2 (1 + R ) 1+

log2 (1+R ) U ( ) 2RR2 with b1 = b1 , s1 = s1 , 2 = 1 b1 s1 .

(24)

ii) if

R 1+R

< D1 R , R3L2USC = R2L2USC with b1 = 1 2 , s1 = 0, 1 + R 1 2 = . R


(1aii) (1)

1 + max

Rs1 R2 RR1 , RR2

(18)

(25)

The condition for equal rates to MS1 and MS2 is given by


A B , Rb1 ) + Rs1 = R2 . min(Rb1

(19)

: Here, b1 1 implies D1 > R , (b) 0 b1 1 < b1 so only the direct link is used for MS1 , i.e., s1 = 0. As the order of link SNRs is reversed, RS should rst decode x2 ,

followed by xb1 . RS doesnt need to decode xb1 as only x2 A is forwarded so Rb1 disappears. After calculations, we nd if D1 > R , R3L2USC = with b1 = (1 +
D1 R ) (1b)

2 log2 (1 + b1 D1 ) 1+
log2 (1+b1 D1 ) log2 (1+R2 ) D1 2 R )

In each algorithm, the achievable sum-rate is determined for each user pair, each user being allocated equal rates by orthogonal time division. The user pair with the best sum-rate is scheduled. B. Proposed SC Scheduling Algorithms Unlike the previous algorithms, the proposed schedulers simultaneously allocate the two users over the whole frame through SC. The Proposed 2L-SC Scheduler and Proposed 3LSC Scheduler allocate the user pair that achieves the maximum sum-rate among all pairs, given the equal rates constraint. In each scheduler, each pair is allocated based on the 2L-2USC scheme and the proposed 3L-2USC scheme in Section III, respectively, or by SC for broadcast channel [5]. In particular, the Proposed 3L-SC Scheduler works as follows: 1. Given a user pair (j, k ), the algorithm determines the best two-user scheme: a. If Dj Rj and Dk Rk , i.e., the direct link of each user is better than their relayed link, SC for the broadcast channel from BS to (MSj ,MSk ) is used. That is, if Dj Dk , the individual rates are Cj = log2 (1 + Dj ) and Ck = )Dk log2 (1 + (1 1+Dk ), where the power allocation ratio is determined by the equal rates constraint. b. Otherwise, the proposed 3L-2USC scheme is used where, given the link SNRs ordering, the decoding orders are adapted and optimal power ratios are derived as in Section III, giving the two cases below (and their symmetric inverting j and k ) since Rk > Dk , Rj > Dj : i. if Rk > Rj > Dk > Dj : BS sends x = bk xbk + sk xsk + j xj and the decoding order is xbk xj xsk , ii. if Rk > Rj > Dj > Dk : BS sends x = bj xbj + sj xsj + k xk and the decoding order is xbj xsj xk . 2. Iterating over all user pairs, the one with the highest sum rate is allocated in the frame. V. N UMERICAL R ESULTS

(1 + 2D1

+ 4D1

, (26)

s1 = 0, 2 = 1 b1 .

B A Rb1 b1 b1 , so Rb1 = Case 2: Here we have Rb1 C (b1 D1 ). There are two subcases, b1 0 and b1 < 0. Similarly to Case 1, we obtain after some derivations R R (a) if 1+ , R D1 1+D1

R3L2USC = R3L2USC with b1 = b1 , s1 = s1 ,


s1 . 2 = 1 b1

(2a)

(1ai)

(27)

(b) if D1

R 1+R ,

R3L2USC = R3L2USC 1 + R 1 with b1 = 0, s1 = , R 2 = 1 s1 . IV. M ULTI -U SER S CHEDULING A LGORITHMS A. Conventional Orthogonal Scheduling Algorithms

(2b)

(1ai)

(28)

We consider the multi-user scheduling problem in the cellular relay system of Fig. 2. The conventional algorithms as in [7] allocate each user orthogonally, such as different time slots or subcarriers. For performance comparison, we describe two benchmark scheduling schemes, namely the Conventional CD Scheduler and the Conventional SC Scheduler. For each user, the single user scheme giving the best rate is chosen among Direct, MH and CD transmissions for the Conventional CD Scheduler and among Direct, MH and SC transmissions for the Conventional SC Scheduler. The rates achieved by each single user transmission schemes are given by Direct (Dir) transmission: user k achieves rate

In this section, we rst compare the sum rate performance of the proposed 3L-2USC scheme with 2L-2USC scheme and the Conventional scheme where MS1 and MS2 are allocated CDIR,k = log2 (1 + Dk ). (29) orthogonally in time under the equal rates constraint. The single user schemes that gave the best rates are used, namely Multi-Hop (MH) transmission: user k achieves rate SU-SC for transmission to MS1 and MH transmission to MS2 . We can observe in Fig. 4 that the 2L-2USC scheme outperlog2 (1 + R ) log2 (1 + Rk ) CMH,k = . (30) forms the Conventional scheme for D1 = 10 to 30 dBs, but log2 (1 + R ) + log2 (1 + Rk ) performs poorly in the other regions. This is because 2L-2USC Cooperative Diversity (CD) transmission: with Maximum does not take advantage of the relayed link for serving MS1 , Ratio Combining (MRC) of the direct and relayed signals, since there are only two layers, one basic for MS1 and one user k achieves rate superposed for MS2 . Thus, when D1 is low, the achievable 1 rate for MS1 is low, thereby limiting the rate for MS2 due to CCD,k = min{log2 (1 + R ), log2 (1 + Dk + Rk )}. (31) 2 the equal rates constraint. However, the proposed 3L-2USC scheme outperforms the Conventional scheme for all SNR Single-User SC (SU-SC) transmission of [4]: regions, thanks to its efcient power allocation among the log2 (1 + R ) log2 (1 + Rk ) three superposed layers. In particular, the region 30 D1 CSC,k = . (32) R log2 ( ) + log2 (1 + Rk ) Dk corresponds to the case where D1 R (Case 1b), where the

8 7.5 Conv. SC Sched. Conv. CD Sched. Prop. 3LSC Sched. Prop. 2LSC Sched.

7 6.5

Sum Rate [b/s/Hz]

5 Sum Rate [b/s/Hz] 3L2USC 2L2USC Conv. 5 10 15 20 D1 [dB] 25 30 35 40 6 5.5 5 4.5 4 3.5 3 2.5 10 15 20 25 30 35 Number of Users per Sector 40 45 50

1 0

Fig. 4.

Sum rates (D1 =040 dB, R2 =20 dB, R1 =25 dB, R =30 dB)

Fig. 5. Sum rate of the proposed and conventional scheduling schemes, multi-user case

gain of 3L-2USC over 2L-2USC comes from the additional SC scheme on the RS to MS1 , MS2 links. In the region 15 D1 30, the performance of 3L-2USC and 2L-2USC coincide as it corresponds to Case (1aii) in the analysis. Thus, for D1 = 15 dB, the optimal power ratio for 2L-2USC is = 0.03, whereas the optimal ratios for 3L-2USC are b1 = 0.97, 2 = 0.03 and s1 = 0. Moreover, the optimal power ratios satisfy the assumption made on their ordering in Section III, as they verify b1 R > 2 R > s1 R , validating the chosen decoding order. In the multi-user case, the sum-rate performance of the conventional and proposed SC based scheduling algorithms is given in Fig. 5 for different number of users per sector K . Although the orthogonal Conventional SC Scheduler improves over Conventional CD Scheduler that uses the benchmark CD scheme thanks to the SU-SC scheme, it is largely outperformed by the Proposed 3L-SC Scheduler with simultaneous user allocation. However, we observe that the Proposed 2L-SC Scheduler attains the lowest sum rate despite simultaneous user allocation. As explained above, this is because there is only one layer per user in the 2L-2USC scheme, preventing him to benet from his high relay link quality, while the orthogonal schemes achieve fairly high throughput thanks to the optimal time allocation between the users. The Proposed 2L-SC Scheduler suffers from substantial degradation in the multi-user environment, but it is completely overcome by the Proposed 3L-SC Scheduler. VI. C ONCLUSION We have considered the problem of sum rate optimization under a fairness constraint for the DL transmissions in a cellular relay system. First, we have proposed the 3L-2USC scheme which enables both users to take advantage of the high RS-(MS1 ,MS2 ) quality. The optimal power allocation parameters have been obtained by analysis. Next, the 3L-2USC scheme was integrated in the proposed scheduler for the multiuser case. The simulation results show the supremacy of the

3L-2USC scheme in the two-user relay system, as well as of the proposed scheduler compared to benchmark schedulers in the multi-user system, showing the benet of simultaneous allocation in place of traditional orthogonal allocation. In the future work, we intend to study the implementation of the proposed schemes assuming practical rates and codes, as well as applying them on top of systems using orthogonal resources, such as multi-carrier or multi-antenna systems. R EFERENCES
[1] J.N. Laneman, D.N.C. Tse and G.W. Wornell, Cooperative Diversity in Wireless Networks: Efcient Protocols and Outage Behavior, IEEE Trans. Info. Theory, vol. 50, no. 12, pp. 30623080, December 2004. [2] Y. Zhao, R. Adve and T.J. Lim, Improving Amplify-and-Forward Relay Networks: Optimal Power Allocation versus Selection, in Proc. of the IEEE International Symposium On Information Theory, Seattle, WA, July 2006. [3] M. Kaneko, K. Hayashi, P. Popovski, K. Ikeda, H. Sakai and R. Prasad, Amplify-and-Forward Cooperative Diversity Schemes for Multi-Carrier Systems, IEEE Trans. on Wireless Comm., vol. 7, no. 5, pp. 18451850, May 2008. [4] P. Popovski and E. de Carvalho, Improving the Rates in Wireless Relay Systems through Superposition Coding, IEEE Trans. Wireless Comm., vol. 7, no. 12, pp. 48314836, December 2008. [5] T.M. Cover and J.A. Thomas, Elements of Information Theory. John Wiley & Sons Inc., 2006. [6] P. Popovski, E. de Carvalho et al., Method and apparatus for transmitting and receiving data using multi-user superposition coding in a wireless relay system, US Patent No. U S 2008/0227388A1, Sept. 2008. [7] M. Salem et al., An Overview of Radio Resource Management in Relay-Enhanced OFDMA-Based Networks, IEEE Comm. Surveys & Tutorials, vol. 12, no. 3, pp. 422438, Third Quarter 2010. [8] M. Newton and J. Thompson, Classication and Generation of Non Uniform User Distributions for Cellular MultiHop Networks, in IEEE ICC, Istanbul, Turkey, June 2006, pp. 45494553. [9] R. Pabst, et al., Relay-Based Deployment Concepts for Wireless and Mobile Broadband Radio, IEEE Wireless Comm. Mag., vol. 42, no. 9, pp. 8089, September 2004. [10] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cambridge, UK: Cambridge University Press, 2005.

You might also like