IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 60, NO. 12...

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 60, NO. 12, DECEMBER 2012 6509

Interference MIMO Relay Channel: Joint PowerControl and Transceiver-Relay BeamformingMuhammad R. A. Khandaker, Student Member, IEEE, and Yue Rong, Senior Member, IEEE

Abstract—In this paper, we consider an interference mul-tiple-input multiple-output (MIMO) relay system where multiplesource nodes communicate with their desired destination nodesconcurrently with the aid of distributed relay nodes all equippedwith multiple antennas. We aim at minimizing the total sourceand relay transmit power such that a minimum signal-to-inter-ference-plus-noise ratio (SINR) threshold is maintained at eachreceiver. An iterative joint power control and beamforming algo-rithm is developed to achieve this goal. The proposed algorithmexploits transmit-relay-receive beamforming technique to mitigatethe interferences from the unintended sources in conjunction withtransmit power control. In particular, we apply the semidefiniterelaxation technique to transform the relay transmission powerminimization problem into a semidefinite programming (SDP)problem which can be efficiently solved by interior point-basedmethods. Numerical simulations are performed to demonstratethe effectiveness of the proposed iterative algorithm.

Index Terms—Beamforming, interference channel, MIMOrelay, power control.

I. INTRODUCTION

I N a large wireless network with many nodes, multiplesource-destination links must share a common frequency

band concurrently to ensure a high spectral efficiency of thewhole network [1]. In such network, cochannel interference(CCI) is one of the main impairments that degrades the systemperformance. Developing schemes that mitigate the CCI istherefore important.By exploiting the spatial diversity, multi-antenna technique

provides an efficient approach to CCI minimization [1], [2].When each source node has a single antenna and the des-tination nodes are equipped with multiple antennas, a jointpower control and receiver beamforming scheme is devel-oped in [3] to meet the signal-to-interference-plus-noise ratio(SINR) threshold with the minimal transmission power. A jointtransmit-receive beamforming and power control algorithmis proposed in [4], when the source nodes also have multipleantennas. Due to the transmit diversity, the total transmit powerrequired in [4] is less than that in [3].In addition to the transmit and/or receive beamforming

considered in [3] and [4], distributed/network beamforming

Manuscript received November 04, 2011; revised March 22, 2012 and June11, 2012; accepted August 23, 2012. Date of publication September 07, 2012;date of current version November 20, 2012. The associate editor coordinatingthe review of this manuscript and approving it for publication was Prof. DavidLove. This work was supported by the Australian Research Council’s DiscoveryProjects funding scheme (Project Numbers DP110100736, DP110102076).The authors are with the Department of Electrical and Computer Engineering,

Curtin University, Bentley, WA 6102, Australia (e-mail: [email protected]; [email protected]).Digital Object Identifier 10.1109/TSP.2012.2217135

technique [5] can further increase the reliability of the com-munication link even if the direct path between the transmitterand the receiver is subject to serious degradation, especiallyfor long-distance communication. The network beamformingscheme stems from the idea of cooperative diversity [6]–[8],where users share their communication resources such asbandwidth and transmit power to assist each other in datatransmission. The optimal relay matrix design has been re-cently studied for the multiple-input multiple-output (MIMO)broadcast channel [9] and the point-to-point MIMO relaychannel [10], [11]. In [12], a decentralized relay beamformingtechnique has been developed considering a network of onetransmitter, one receiver, and several relay nodes each havinga single antenna. In [13], a wireless ad hoc network consistingof multiple source-destination pairs and multiple relay nodes,each having a single antenna, is considered, where the networkbeamforming scheme is used to meet the SINR threshold at alllinks with the minimal total transmission power consumed byall relay nodes. Relay beamformers are designed in [14] formultiple-antenna relay nodes with single-antenna source-desti-nation pairs. The non-regenerative MIMO relay technique hasbeen applied to multi-cellular (interference) systems in [15]where transceiver beamformers are designed using the partialzero-forcing (PZF) technique.However, it is assumed in [13]–[15] that each source node

uses its maximum available transmit power. Such assumptionnot only raises the system transmit power consumption, but alsoincreases the interference from one user to all other users. Thisindicates that the beamforming and the power control problemshould be considered jointly as in [3] and [4].In this paper, we consider a two-hop interferenceMIMO relay

system consisting of source-destination pairs communicatingwith the aid of relay nodes to enable successful communi-cation over a long distance. Each of the source, relay and des-tination nodes is equipped with (possibly different number of)multiple antennas. The amplify-and-forward scheme is used ateach relay node due to its practical implementation simplicity.In fact, these relay nodes assist in CCI mitigation by performingdistributed network beamforming1.We aim at developing a joint power control and beamforming

algorithm such that the total transmission power consumed byall source nodes and relay nodes are minimized while main-taining the SINR at each receiver above a minimum thresholdvalue. Compared with [12]–[14], we not only use the network

1Although the relay beamforming matrices are optimized by a central pro-cessing unit in our algorithm, the relay beamforming operation is indeed dis-tributed in the sense that the relays are geographically distributed and they per-form beamforming only using their own received signal without exploiting theinformation on the received signals at other relay nodes.

1053-587X/$31.00 © 2012 IEEE

6510 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 60, NO. 12, DECEMBER 2012

Fig. 1. Block diagram of an interference MIMO relay system.

beamforming technique at the relay nodes, but also apply thejoint transmit-receive beamforming technique for multiple-an-tenna users to mitigate the CCI. In contrast to [15], we developan iterative technique to solve the total power minimizationproblem rather than using the suboptimal PZF approach. More-over, transmit power control is used in our algorithm to mini-mize the total transmit power and the interference to other users,which is not considered in [12]–[15].A two-tier iterative algorithm is proposed to jointly optimize

the source, relay and receive beamformers, and the source trans-mission power. We update the relay beamformer in the outerloop using fixed source power, transmit beamformers, and re-ceive beamformers. Since the relay beamforming optimizationproblem is nonconvex, we use the semidefinite relaxation (SDR)technique to transform the problem into a semidefinite program-ming (SDP) problem which can be efficiently solved by interiorpoint-based methods. Then in each iteration of the inner loop,we optimize the receive beamformers first with fixed transmitand relay beamformers and source power. Next, we update thesource power such that the target SINR is just met with giventransmit, relay and receive beamformers. Finally in the innerloop, we update the transmit beamformers with known transmitpower, relay beamformers, and receive beamformers. Numer-ical simulations are carried out to evaluate the performance ofthe proposed algorithm.The rest of this paper is organized as follows. In Section II,

the system model of an interference MIMO relay network isintroduced. The joint power control and beamforming algorithmis developed in Section III. Section IV shows the simulationresults which justify the significance of the proposed algorithmunder various scenarios. Conclusions are drawn in Section V.

II. SYSTEM MODEL

We consider a two-hop interference MIMO relay system withsource-destination pairs as illustrated in Fig. 1. Each source

node communicateswith its correspondingdestinationnodewiththe aid of a network of distributed relays in order to enablesuccessful communication over a long distance. The direct linksbetween the source nodes and the destination nodes are not con-sidered as they undergomuch larger path attenuations comparedwith the links via relays. The source and destination nodes of theth link are equipped with and antennas, respectively,whereas the th relay node is mounted with antennas.We assume that all relay nodes work in half-duplex mode

as in [12]–[14]. Thus the communication between the source-destination pairs is completed in two time slots. In the first timeslot, the th source node transmits an signal vector

, where is the information-carrying symbol and is thetransmit beamforming vector. The received signal vector at theth relay node is given by

where is the MIMO channel matrix betweenthe th transmitting node and the th relay node and is the

additive Gaussian noise vector at the th relay node.In the second time slot, the th relay node multiplies its re-

ceived signal vector by an complex matrix andtransmits the amplitude- and phase-adjusted version of its re-ceived signal. Thus the signal vector transmittedby the th relay node is given by

(1)

The received signal at the th destination node is obtained as theweighted sum of the received signals at each antenna elementof that node, and is given by

(2)

where is the MIMO channel matrix betweenthe th relay node and the th destination node, andare the receive beamforming weight vector and theadditive Gaussian noise vector at the th destination node, re-spectively, and denotesmatrix (or vector) Hermitian trans-pose. We assume that all noises are independent and identicallydistributed (i.i.d.) complex Gaussian noise with zero mean andvariance .Let us introduce the following definitions

where denotes matrix (or vector) transpose,stands for a block-diagonal matrix, and . Herecan be viewed as the effective first-hop channel vector be-

tween and all relay nodes, is the MIMO channel matrixbetween all relay nodes and the th receiver, is the effectiveblock-diagonal relay precoding matrix, and is a vector con-taining the noises at all relay nodes. Using these definitions, (2)can be rewritten as

(3)

KHANDAKER AND RONG: INTERFERENCE MIMO RELAY CHANNEL 6511

where is the equivalent vector channel responsebetween the th source node and the th destination node, and

is the equivalent noise vector at the threceiver.From (3), the total power of the received signal at the desti-

nation node of the th link is given by

(4)

where stands for statistical expectation, denotes com-plex conjugate, is the covari-ance matrix of , and is an identity matrix. Here weassume that is the transmit power of the th in-formation-carrying symbol. Based on (4), the SINR at the thdestination node is given by

(5)

Using (1), the transmission power consumed by the th relaynode can be expressed as

(6)

where denotes matrix trace,

is the covariance matrix ofthe received signal vector at the th relay node. Using (6), thetotal transmit power consumed by the whole network can be ex-pressed as

(7)

III. JOINT POWER CONTROL AND BEAMFORMING

Let us define the relay beamforming vector from the relayamplifying matrices as

... (8)

where , and stands for a vectorobtained by stacking all column vectors of a matrix ontop of each other. In this section, we design the sourcetransmit power vector , the relaybeamforming vector , transmit beamforming vectors

, and receive beamformingvectors , such that a targetSINR threshold , is maintained at theth destination node with the minimal . The optimizationproblem can be written as

(9)

(10)

The problem (9)–(10) is nonconvex due to the constraints in(10). We propose a two-tier iterative algorithm to efficientlysolve the problem (9)–(10). In the following, we solve corre-sponding subproblems to optimize each variable.

A. Receive Beamforming

The optimal receive beamforming vectors ,for fixed , and can be obtained such that it minimizesthe noise-plus-interference power at the receiver under the con-dition of unity gain for the signal of interest, which can bewritten as

(11)

(12)

The unity gain condition ensures that the desired signal isunaffected by beamforming. Using the Lagrangian multipliermethod, the solution to the problem (11)–(12) is given by

(13)

where is the interference-plus-noise covariance matrix at the th receiver, and denotesmatrix inversion.

B. Transmit Power Allocation

To obtain optimal with given beamforming vectors ,and , we reformulate the problem (9)–(10) as

(14)

(15)

where is an covariance matrix such thatand . Here for a matrix

indicates the th element of . In the optimal power allo-cation, the transmit power of each user is set to the minimumrequired level such that the target SINR is just met [3], [4]. Thatis, the constraints in (15) should hold with equality as

(16)

which can be equivalently rewritten as

(17)

Equation (17) can be written in matrix form as

(18)

where , and is an

vector whose th element is given by .From (18), it can be seen that the optimal solution to the problem(14)–(15) is given by

(19)


C. Transmit Beamforming

With given and , the optimal can be obtainedsimply by swapping the roles of the transmitters and the re-ceivers as in [16]. First we rewrite the objective function by sub-stituting in (6) into (7) as

(20)

where . Let us now denote

and . Thus (20)can be equivalently written as

(21)

Since the equivalent noise at the th destination node is non-white,we need to perform the pre-whitening operation beforeweswap the roles of the transmitting and the receiving nodes. Afterthe pre-whitening and receive beamforming operations, the re-ceived signal at the th destination node can be expressed as

(22)

where is the equivalent MIMOchannel between the th source node and all relay nodes and

.It can be seen from (22) that the equivalent noise is nowwhite,

and the received SINR in the th virtual link (where is the re-ceive beamforming vector and becomes the transmit beam-forming vector) can be expressed as

(23)

Here is the transmit power in the th vir-tual link. Note that since the noise in the original link is pre-whitened before we swap the roles of transmitters and receivers,the equivalent virtual link noise is also white with unit-variance.Thus, the corresponding noise power after the receive beam-forming is given by in (23).The optimal can be obtained from (23) by solving the

following problem for each

(24)

(25)

The solution to this problem is given by

(26)

where is the noise-plus-inter-ference covariance matrix at the th receiver of the virtual link.The transmit power of the virtual link can be obtained as

(27)

where ,

and . Here for a vector

stands for the th element of .

D. Relay Beamforming

In this subsection we optimize the relay amplifying matricessuch that the total relay transmit power is minimized while sat-isfying the SINR constraints in (10). First, (4) can be rewrittenas

(28)

where , and. Using (28), the SINR of the th link

in (5) can be expressed as

(29)

Applying the fact that[17], where denotes the matrix Kronecker product,

(29) can be expressed as

(30)

where . Let us now introduce thelink between in (8) and as , where

is a matrix of ones and zeros and is constructedby observing the nonzero entries of . Note that doesnot depend on the exact numeric value of , instead itdepends on the way the entries of are taken to form . Asan example, for a system with two relay nodes each having two

antennas, there is with

and , where , are 2 1 vectors


and denotes an matrix with all zero elements. Inthis case, we have

Therefore, to obtain , matrix should be con-structed as

Now (30) can be rewritten as

(31)

From (8), we have , withdefined as ,

where and

. By using the identity offor [17], the

transmit power of the th relay node in (6) can be expressed as

(32)

Using (31) and (32), with given and , the problem(9)–(10) can be reformulated as the following nonconvexquadratically constrained quadratic program (QCQP) problem

(33)

(34)

where we introduce

(35)

The problem (33)–(34) is non-convex, since in (35) canbe indefinite. In the following, we resort to the SDR technique[18]–[21] to solve the problem (33)–(34). By introducing

, the problem (33)–(34) can be equivalently rewritten as

(36)

(37)

(38)

(39)

where means that is a positive semidefinite (PSD)matrix, and denotes the rank of a matrix. Note that

TABLE IRANDOMIZATION TECHNIQUE FOR SEMIDEFINITE RELAXATION APPROACH

in the problem (36)–(39), the cost function is linear in , thetrace constraints are linear inequalities in , and the PSDmatrixconstraint is convex. However, the rank constraint on is notconvex. Interestingly, the problem (36)–(39) can be solved bythe SDR technique [18]–[21] as explained in the following. Firstwe drop the rank constraint (39) to obtain the following relaxedSDP problem which is convex in .

(40)

(41)

(42)

SDP problems like (40)–(42) can be conveniently solved byusing interior point methods at a complexity order that is atmost [19]. One can use, for example, the CVXMATLAB toolbox for disciplined convex programming [22] toobtain the optimal . Due to the relaxation, obtained bysolving the problem (40)–(42) is not necessarily rank one in gen-eral. If it is, then its principal eigenvector (scaled by the squareroot of the principal eigenvalue of ) is the optimal solution

to the original problem (33)–(34). If and, the recent results on Hermitian matrix rank-one decom-

position in [23] can be used to generate the exact optimalfor the problem (33)–(34) based on . Otherwise, we mayresort to alternative techniques such as randomization [18]–[21]to obtain a (suboptimal) from . Different randomizationtechniques have been studied in the literature [18]–[21]. Theone we choose is summarized in Table I. Note that using thisapproach, some of the constraints in (10) may be violated afterthe randomization operation. However, a feasible relay beam-forming vector can be obtained by simply scaling so that allthe constraints are satisfied.Now the original total transmit power minimization problem

(9)–(10) can be solved by an iterative algorithm as shown inTable II. Here , are small positive numbers close tozero up to which convergence is acceptable, stands for themaximal element of a vector, and the superscript anddenotes the number of iterations at the outer loop and the innerloop, respectively. It can be seen from Table II that the pro-posed algorithm iteratively optimizes two blocks of variables:(i) The relay weighting coefficients ; (ii) The transmit beam-former vectors , the receive beamformer vectors , andthe transmit power vector . With fixed , we solve the problemof optimizing , and through step (3) in Table II.In fact, this problem is similar to the joint transceiver designproblem in a single-hop MIMO interference channel [4]. There-fore, it can be shown similar to [4] that the inner iteration in step(3) converges to the optimal solution of , and for


TABLE IIPROCEDURE OF SOLVING THE PROBLEM (9)–(10) BY THE PROPOSED

ITERATIVE ALGORITHM

a given . With fixed , and , we optimize throughstep (2) in Table II.In numerical simulations we observe that the outer loop con-

verges typically within 3 to 5 iterations, while the inner loopconverges usually within 3 iterations. However, a rigorous anal-ysis on whether the outer loop converges to a locally optimalsolution is difficult, due to the coupling between the optimiza-tion variables in (10). We also observe that the proposed algo-rithm requires less iterations till convergence for lower targetSINR thresholds. Moreover, it can be seen from Table II thatthe amount of computations required for the convergence ofthe inner loop is much smaller than the computation involvedin solving the SDP problem in the outer loop. Therefore, theoverall computational complexity of the proposed algorithm canbe estimated as with between 3 and 5.Before moving to the next section, we would like to comment

on several issues related to the implementation of the proposedalgorithm in practice.Remark 1: The channel state information (CSI) on

andis required in the proposed algorithm.

Since the perfect CSI is not available in a real communicationsystem due to limited feedback and/or inaccurate channel esti-mation, robust designs can be considered in case of imperfectCSI. A worst-case based robust relay matrices design for in-terference relay system has been proposed in [14] where each

source and destination node has a single antenna (i.e., onlyneeds to be optimized). However, when all source and desti-nation nodes have multiple antennas, the worst-case based ro-bust design becomes extremely challenging since the worst-caseSINR is a very complicated function of , and. Alternatively, we can try the statistically robust design [24],where we average over the mismatch between the true and theestimated CSI. However, the statistical expectation of in (5)with respect to all channel matrices turns out to be an extremelycomplicated expression of the design variables , and

. This makes the statistically robust design problem everydifficult to solve. The impact of imperfect CSI on the perfor-mance of the proposed algorithm will be studied through nu-merical simulation in Section IV.Remark 2: The procedure in Table II needs to be carried out

by a central processing unit due to the requirement of the globalCSI. With the advancement of modern chip design, the amountof computation can be handled by the centralprocessing unit. Nevertheless, it is interesting to investigate dis-tributed algorithms that can solve the problem (9)–(10). In fact,the inner loop in step (3) of Table II is easier than step (2) fora distributed implementation. The reason is that in step (2), anSDP problem needs to be solved, which is difficult to be imple-mented in a distributed manner.Remark 3: In practical applications, to meet the SINR re-

quirements (10), some nodes may require larger transmissionpower that exceeds their available limit. A possible way out tothis problem is to identify the SINR constraints that producethe largest increase in terms of transmit power first, and thenrelax those constraints in order to reduce the required powerusing a perturbation analysis [25]. Alternatively, one may applyan admission control algorithm first to maximize the number oflinks possibly served, and then perform optimal power alloca-tion [26].

IV. NUMERICAL EXAMPLES

In this section, we study the performance of the proposed jointpower control and beamforming algorithm for an interferenceMIMO relay system through numerical simulations where allnodes are equipped with multiple antennas. For simplicity, weassume , and

, in all simulations. All noises arei.i.d. complex circularly symmetric Gaussian noise with zeromean and unit variance (i.e., ). The channel matriceshave entries generated as i.i.d. complex Gaussian random vari-ables with zero mean and variances and for and

, respectively. All simulation results are averaged over500 independent channel realizations.For the proposed algorithm, the procedure in Table II is car-

ried out in each simulation to obtain the power vector , transmitbeamforming vectors , relay beamforming vector , and re-ceive beamforming vectors . To initialize the algorithm inTable II, we randomly generate the transmit and receive beam-forming vectors and , respectively, along with arbi-trary transmit power vector and virtual power vector .In the first example, we compare the performance of the

proposed joint power control and beamforming algorithm


Fig. 2. Total power versus target SINR., and .

(Proposed TxRxBF) with the relay-only beamforming withoutpower control (RoBF-NPC) scheme studied in [13], [14] andthe conventional singular-value decomposition (SVD)-basedtransmit beamforming approach (SVD-based TxBF). For theSVD-based TxBF scheme, we choose as the principal rightsingular vector of . Then we update the transmit powervector , relay beamforming vector and receive beamformers

based on the proposed structure. We plot the total powerconsumed by all source nodes and relay nodes versus the targetSINR threshold (dB). Two channel fading environments aresimulated: (i) Both and have Rayleigh fading;(ii) Only has Rayleigh fading while has Riceanfading with a Ricean factor of 5. Fig. 2 shows the performanceof all three algorithms for

, and . It can be seen from Fig. 2 thatthe proposed algorithm requires significantly less total powercompared with the other two schemes in both Rayleigh andRicean fading environments.Note that the RoBF-NPC scheme performs better in Ricean

fading channel whereas the performance of the other two ap-proaches degrades under Ricean fading environment. This canbe explained as follows. In the RoBF-NPC scheme, each trans-mitter and receiver has a single antenna as in [13] and [14],which indicates that the relay-destination channels arein fact multiple-input single-output (MISO) channels. Thereforethe line-of-sight (LOS) path component improves the systemperformance. For the other two schemes, the relay-destinationchannels are MIMO channels. In MIMO Ricean channels, thebenefit of scattering environment reduces due to the LOS com-ponent. This weaker scattering component causes the perfor-mance degradation. Similar phenomenon has been observed in[27] for point-to-point MISO and MIMO Ricean channels.In the second example, we vary the number of transmit an-

tennas to show the effect of transmit diversity with, and . Fig. 3 indi-

cates the significance of transmit beamforming in the proposedalgorithm. It is obvious from Fig. 3 that with the increase in the

Fig. 3. Total power versus target SINR for different number of transmit an-tennas. , and .

Fig. 4. Total power versus target SINR for different number of relays., and .

spatial dimension of the transmit beamformers the performanceof the proposed algorithm keeps improving.In the next example, we study the performance of the pro-

posed algorithm for different number of relays with, and . The total

power required for , and 15 versus (dB) is dis-played in Fig. 4. As expected, if we increase the number of re-lays the proposed algorithm requires less power since more re-lays provide more spatial diversity. We also show the impactof the number of relay antennas in Fig. 5. This time, we set

, andand the total power required for and 3 versus (dB) isdisplayed. Note that with the increase in the number of relay an-tennas, the performance of the proposed scheme improves butat the same time, the computational complexity of solving theproblem (40)–(42) significantly increases. Therefore, it is im-portant to make a tradeoff between the performance and com-plexity based on the system requirements and the available re-sources.


Fig. 5. Total power versus target SINR for different number of relay antennas., and .

Fig. 6. Effect of the first-hop channel quality., and .

In the next two examples, we study the impact of channelquality on the proposed algorithm.We assume that a larger vari-ance of channel coefficients indicates a better channel. The im-pact of different and on the proposed algorithm is shownin Figs. 6 and 7, for and , respectively. In theseexamples, we set , and .A careful inspection of Figs. 6 and 7 reveals that the effect ofchannel variance of either hop is not homogeneous in general,but the results clearly demonstrate that the proposed algorithmperforms better as the channel quality improves.Next, we study the effect of channel interferences on the pro-

posed algorithm. By increasing the number of source-destina-tion pairs , the interfering signal received at each destinationnode is also increased. The performance of the algorithm fordifferent is illustrated in Fig. 8 for

, and . From this figure it is clear thatif there are more active users communicating simultaneously

Fig. 7. Effect of the second-hop channel quality., and .

Fig. 8. Total power versus target SINR for different number of users., and .

in the system, we need more power to achieve the same targetSINR threshold .In the last example, we study the impact of imperfect CSI on

the performance of the proposed algorithm. The mismatch be-tween the true CSI and the estimated CSI is modelled as com-plex Gaussian matrices with zero-mean and unit-variance en-tries. Fig. 9 shows the performance of all three algorithms for

, and. Clearly, the proposed algorithm outperforms the ex-

isting techniques with both perfect and imperfect CSI.

V. CONCLUSION

We considered a two-hop interference MIMO relay systemwith distributed relay nodes and developed an iterative tech-nique to minimize the total transmit power consumed by allsource and relay nodes such that a minimum SINR thresholdis maintained at each receiver. The proposed algorithm exploitsbeamforming techniques at the source, relay, and destination


Fig. 9. The impact of the CSI mismatch on the tested algorithms., and .

nodes in conjunction with transmit power control. Simulationresults demonstrate that the proposed power control and beam-forming algorithm outperforms the existing techniques.

ACKNOWLEDGMENT

The authors would like to thank the anonymous reviewersfor their valuable comments and suggestions that improved thequality of the paper.

REFERENCES

[1] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.New York: Cambridge Univ. Press, 2005.

[2] A. Goldsmith, Wireless Communications. New York: CambridgeUniv. Press, 2005.

[3] F. Rashid-Farrokhi, L. Tassiulas, and K. J. R. Liu, “Joint optimal powercontrol and beamforming in wireless networks using antenna arrays,”IEEE Trans. Commun., vol. 46, pp. 1313–1324, Oct. 1998.

[4] J.-H. Chang, L. Tassiulas, and F. Rashid-Farrokhi, “Joint transmitterreceiver diversity for efficient space divisionmultiaccess,” IEEE Trans.Wireless Commun., vol. 1, pp. 16–27, Jan. 2002.

[5] Y. Jing and H. Jafarkhani, “Network beamforming using relays withperfect channel information,” in Proc. IEEE ICASSP, Honolulu, HI,Apr. 2007, vol. 3, pp. 473–476.

[6] A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diver-sity—Part I: System description,” IEEE Trans. Commun., vol. 51, pp.1927–1938, Nov. 2003.

[7] A. Sendonaris, E. Erkip, and B. Aazhang, “User cooperation diver-sity—Part II: Implementation aspects and performance analysis,” IEEETrans. Commun., vol. 51, pp. 1939–1948, Nov. 2003.

[8] J. N. Laneman, D. N. C. Tse, andG.W.Wornell, “Cooperative diversityin wireless networks: Efficient protocols and outage behavior,” IEEETrans. Inf. Theory, vol. 50, pp. 3062–3080, Dec. 2004.

[9] R. Zhang, C. C. Chai, and Y.-C. Liang, “Joint beamforming andpower control for multiantenna relay broadcast channel with QoSconstraints,” IEEE Trans. Signal Process., vol. 57, pp. 726–737, Feb.2009.

[10] X. Tang and Y. Hua, “Optimal design of non-regenerativeMIMOwire-less relays,” IEEE Trans. Wireless Commun., vol. 6, pp. 1398–1407,Apr. 2007.

[11] O.Muñoz-Medina, J. Vidal, and A. Agustı́n, “Linear transceiver designin nonregenerative relays with channel state information,” IEEE Trans.Signal Process., vol. 55, pp. 2593–2604, June 2007.

[12] V. Havary-Nassab, S. Shahbazpanahi, A. Grami, and Z.-Q. Luo, “Dis-tributed beamforming for relay networks based on second-order sta-tistics of the channel state information,” IEEE Trans. Signal Process.,vol. 56, pp. 4306–4316, Sept. 2008.

[13] S. Fazeli-Dehkordy, S. Shahbazpanahi, and S. Gazor, “Multiplepeer-to-peer communications using a network of relays,” IEEE Trans.Signal Process., vol. 57, pp. 3053–3062, Aug. 2009.

[14] B. K. Chalise and L. Vandendorpe, “Optimization of MIMO relaysfor multipoint-to-multipoint communications: Nonrobust and robustdesigns,” IEEE Trans. Signal Process., vol. 58, pp. 6355–6368, Dec.2010.

[15] T. Taniguchi, N. B. Ramli, and Y. Karasawa, “Design of multiuserMIMO AF relay system with interference cancellation,” in Proc. 6thInt. Wireless Commun. Mobile Comput. Conf., Caen, France, June2010, pp. 1075–1080.

[16] F. Rashid-Farrokhi, K. J. R. Liu, and L. Tassiulas, “Transmit beam-forming and power control for cellular wireless systems,” IEEE J. Se-lect. Areas Commun., vol. 16, pp. 1437–1450, Oct. 1998.

[17] R. A. Horn and C. R. Johnson, Topics in Matrix Analysis. Cambridge,U.K.: Cambridge University Press, 1991.

[18] W.-K. Ma, T. N. Davidson, K. M. Wong, Z.-Q. Luo, and P.-C. Ching,“Quasi-ML multiuser detection using semi-definite relaxation with ap-plication to synchronous CDMA,” IEEE Trans. Signal Process., vol.50, pp. 912–922, Apr. 2002.

[19] N. D. Sidiropoulos, T. N. Davidson, and Z.-Q. Luo, “Transmit beam-forming for physical-layer multicasting,” IEEE Trans. Signal Process.,vol. 54, pp. 2239–2251, June 2006.

[20] S. Zhang, “Quadratic maximization and semidefinite relaxation,”Math. Program., ser. A, vol. 87, pp. 453–465, 2000.

[21] P. Tseng, “Further results on approximating nonconvex quadratic op-timization by semidefinite programming relaxation,” SIAM J. Optim.,vol. 14, pp. 268–283, July 2003.

[22] M. Grant and S. Boyd, “Cvx: Matlab software for disciplined convexprogramming (web page and software),” [Online]. Available: http://cvxr.com/cvx Apr. 2010

[23] W. Ai, Y. Huang, and S. Zhang, “New results on Hermitian matrixrank-one decomposition,” Math. Program., ser. A, vol. 128, pp.253–283, Jun. 2011.

[24] Y. Rong, “Robust design for linear non-regenerativeMIMO relayswithimperfect channel state information,” IEEE Trans. Signal Process., vol.59, pp. 2455–2460, May 2011.

[25] D. P. Palomar, M. Lagunas, and J. Cioffi, “Optimum linear jointtransmit-receive processing for MIMO channels with QoS con-straints,” IEEE Trans. Signal Process., vol. 52, pp. 1179–1197, May2004.

[26] K. Phan, T. Le-Ngoc, S. Vorobyov, and C. Tellambura, “Power allo-cation in wireless multi-user relay networks,” IEEE Trans. WirelessCommun., vol. 8, pp. 2535–2545, May 2009.

[27] G. Lebrun, M. Faulkner, M. Shafi, and P. J. Smith, “MIMO Riceanchannel capacity: An asymptotic analysis,” IEEE Trans. WirelessCommun, vol. 5, pp. 1343–1350, June 2006.

Muhammad R. A. Khandaker (S’10) receivedthe B.S. (Honours) degree in computer scienceand engineering from Jahangirnagar University,Dhaka, Bangladesh, in 2006 and the M.S. degreein telecommunications engineering from East WestUniversity, Dhaka, in 2007.He is currently working toward the Ph.D. degree

at the Department of Electrical and ComputerEngineering, Curtin University, Australia. He startedhis career as a Jr. Hardware Design Engineer inVisual Magic Corporation Limited (VMCL-BD)

in November 2005. He joined the Department of Computer Science andEngineering, IBAIS University, Dhaka, in January 2006, as a Lecturer. Hethen joined the Department of Information and Communication Technology,Mawlana Bhasani Science and Technology University, Bangladesh, as aLecturer, in November 2007. Finally, he joined the Institute of InformationTechnology, Jahangirnagar University, Dhaka, as a Lecturer, in January 2008.Mr. Khandaker was awarded the Curtin International Postgraduate Research

Scholarship (CIPRS) for his Ph.D. study in 2009. He also received the BestPaper Award at the 16th Asia-Pacific Conference on Communications, Auck-land, New Zealand, 2010.


Yue Rong (S’03–M’06–SM’11) received theB.E. degree from Shanghai Jiao Tong University,Shanghai, China, the M.Sc. degree from the Univer-sity of Duisburg-Essen, Duisburg, Germany, and thePh.D. degree (summa cum laude) from DarmstadtUniversity of Technology, Darmstadt, Germany, allin electrical engineering, in 1999, 2002, and 2005,respectively.From February 2006 to November 2007, he was

a Postdoctoral Researcher with the Department ofElectrical Engineering, University of California,

Riverside. Since December 2007, he has been with the Department of Electricaland Computer Engineering, Curtin University of Technology, Australia, wherehe is now a Senior Lecturer. His research interests include signal processing forcommunications, wireless communications, wireless networks, applications of

linear algebra and optimization methods, and statistical and array signal pro-cessing. He has coauthored more than 80 referred IEEE journal and conferencepapers.Dr. Rong received the Best Paper Award at the Third International Conference

on Wireless Communications and Signal Processing, Nanjing, China, 2011; theBest Paper Award at the 16th Asia-Pacific Conference on Communications,Auckland, New Zealand, 2010; the 2010 Young Researcher of the Year Awardof the Faculty of Science and Engineering at Curtin University. He is an Editorof the IEEE WIRELESS COMMUNICATIONS LETTERS, a Guest Editor of the IEEEJOURNAL ON SELECTED AREAS IN COMMUNICATIONS Special Issue on Theoriesand Methods for Advanced Wireless Relays, a Guest Editor of the EURASIPJASP Special Issue on Signal Processing Methods for Diversity and Its Appli-cations, and has served as a TPC member for IEEE ICC, WCSP, IWCMC, andChinaCom.

Date post:	13-Mar-2020
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 60, NO. 12...

Documents