This paper introduces an optimum amplify-and-forward (AF) distributed beamforming (DBF) in the presence of cochannel interference (CCI) when only local channel-state information (CSI) is available at each relay. It is shown that the proposed DBF closely achieves the performance obtained with global CSI when interference power toward relays is small or there are a large number of interferers but greatly reduces the complexity and overhead. The proposed DBF provides significant improvements over the conventional DBF designed without considering CCI at the cost of slightly increased complexity and overhead and achieves the capacity scaling of 1/2logK through K relays, where 1/2logK corresponds to the maximal capacity scaling when there is no CCI.
1. Introduction
Cooperative relaying has attracted a great deal of attention because of its appealing properties for both performance and various applications. Among various schemes, cooperative beamforming is being widely considered because it achieves optimal diversity-order performance and capacity scaling by maximizing the received signal-to-noise ratio (SNR). An upper bound on capacity scaling of dual-hop relay networks was provided in [1], in which the capacity scaling was achieved using the consequence of receive and transmit matched filtering at each relay in distributed way. However, an optimum design for beamforming weight was not taken into account in [1]. An optimal distributed beamforming (DBF) to maximize the received SNR was proposed in [2], which showed that the optimal performance is achieved only with local channel-state information (CSI) obtained at each relay. The results of [2] were extended to two-way relaying in [3]; near optimum joint DBF was introduced with which the maximal capacity scaling and full diversity order were achieved.
The above-mentioned works do not consider the impact of cochannel interference (CCI) that is one of the major limiting factors on the performance of wireless communication systems. Recently, [4] introduced optimal beamforming that maximizes the received signal-to-interference-plus-noise ratio (SINR) when N sources perform DBF toward a relay and the destination is corrupted by CCI. However, the impact of CCI was considered only at the destination. Although there is an abundance of research on cooperative beamforming with a variety of scenarios, the distributed approach based on local CSI considering CCI has not yet been thoroughly investigated.
This paper investigates the optimum DBF based on local CSI when the relays and the destination are affected by CCI. The proposed DBF has very small complexity and overhead compared to the cooperative beamforming obtained with global CSI. More details provided in this paper are summarized as follows:
An optimal amplify-and-forward (AF) DBF weight is proposed in the presence of CCI at both the relays and the destination when only local CSI is available at each relay.
The proposed DBF is shown to achieve nearly the performance obtained with global CSI when there are a large number of interferers or interference power toward relays is small.
The DBF has a capacity scaling of 1/2logK through K relays, where 1/2logK corresponds to the maximal capacity scaling when there is no CCI.
Numerical results verify that the proposed DBF represents significant improvements over the conventional DBF designed without considering CCI at the cost of slightly increased overhead and complexity.
This paper is organized as follows: Section 2 introduces the system model for DBF protocol. Section 3 presents the optimum DBF weight, and its capacity scaling law is derived. Finally, the numerical results are presented in Section 4, and concluding remarks are given in Section 5.
Notations. diag[x1,…,xK] denotes the diagonal square matrix with x1,…,xK on its main diagonal, (·)∗ the complex conjugate, and (·)H the Hermitian, respectively. a is the Euclidean norm of the vector a, and IK denotes the K×K identity matrix. E[X] and VAR[X] mean the expectation and the variance of a random variable (r.v.). X. →w.p.1 denotes convergence with probability one. For two functions f(x) and g(x), f(x)~g(x) means that limx→∞f(x)/g(x)=1, or equivalently lim1/x→0f(1/x)/g(1/x)=1.
2. System Model
Figure 1 depicts a wireless network that consists of a source, a destination, and K relays. Let Sr={1,2,…,K} be a set of the relays. Each node has a single antenna and the relays operate in half-duplex mode with AF strategy. All the relays and the destination are affected by I interferers. Hereafter, subscripts s, k, and d denote the source, the kth relay, and the destination, respectively, and i is the index of interferers. Because of the long distance between s and d, there is no direct link between them. It is assumed that the activities of interferers change slowly, and, therefore, each node is affected by the same interferers during two phases.
System model. A source communicates with a destination through K intermediate relays, where the relays and the destination are affected by a number of I interferers. The relays operate in a half-duplex mode with AF strategy.
Frequency-flat block-fading channels are assumed, where hi,j denotes the channel coefficient between node i and node j (i,j∈{s,Sr,d}) and gm,n is the channel coefficient between the mth interferer and receiving node n (n∈{Sr,d}). Channel reciprocity is assumed and each node has the receivers’ CSI. The channel coefficients are modelled by independent but not identically distributed (i.n.i.d.) complex Gaussian r.v.’s. That is, channel powers |hs,k|2, |hk,d|2, |gi,k|2, and |gi,d|2 are independent and exponentially distributed r.v.’s whose means are σs,k2, σk,d2, σI,i,k2, and σI,i,d2, respectively.
During the first phase, s transmits xs with power Ps. The received signal at relay k is corrupted by multiple interfering signals xi,1’s with power Pi’s:(1)yk=hs,kPsxs+∑i=1Igi,kPixi,1+nk,where nk~Nc(0,1) is complex additive white Gaussian noise (AWGN) at relay k. During the second phase, each relay simultaneously retransmits the signal:(2)xk=akwkyk,where wk denotes the beamforming weight for relay k to be optimized. When the normalized amplifying gain is considered as(3)ak=1Pshs,k2+∑i=1IPigi,k2+1,the transmission power of xk becomes |wk|2.
Aggregate transmit power over all relays is assumed to be constrained by ∑k=1K|wk|2≤Pr, where Pr is the maximum transmission power available at each relay. The assumption makes the DBF more practical at the network point of view. With the constraint, the total used power remains constant regardless of the number of relays K. It is an effective way to constrain the interference to other nodes in the network. Moreover, under the assumption, the transmission power cannot be shared among different nodes, which may not be practical. The received signal at d is given by(4)yd=∑k=1Khk,dakwkyk+∑i=1Igi,dPixi,2+nd,where nd is complex AWGN and xi,2’s are the interfering signals during the second phase with powers Pi’s. It is assumed that E[|xj|2]=1, where xj∈{xs,xi,1,xi,2}.
Using a beamforming weight vector w=[w1∗,…,wK∗], the SINR of the received signal at d is represented by(5)γd=PswaHawHwAHBAwH+wAHEnHnAwH+∑i=1IPigi,d2+1,where(6)a=a1hs,1h1,d,…,aKhs,KhK,d,A=diaga1h1,d,…,aKhK,d,B=∑i=1IPigi,12∑i=1IPigi,1∗gi,2∑i=1IPigi,1∗gi,K∑i=1IPigi,2∗gi,1∑i=1IPigi,22⋮⋮⋱∑i=1IPigi,K∗gi,1⋯∑i=1IPigi,K2,n=n1,…,nK.
3. Distributed Beamforming with CCI Based on Local CSIFact 1.
When T is positive definite Hermitian, the following modified Rayleigh-Ritz theorem holds for any row vector x [2, Proposition 1]:(7)xsHsxHxTxH≤λmax,where λmax=Tr[(TH/2)-1(sHs)(T1/2)-1] is the largest eigenvalue of (TH/2)-1(sHs)(T1/2)-1 and the equality holds when x=csT-1 for any nonzero constant c.
When there is no limit on available CSI at each relay, that is, global CSI is available, the optimal beamforming weight vector wgc that maximizes the received SINR γd in (5) is given by(8)wgc=aV-1PraV-12,where(9)V=AHBA+AHA+∑i=1IPigi,d2+1PrIK.The proof is as follows. The received SINR γd in (5) becomes(10)γd=PswaHawHwAHBA+AHA+∑i=1IPigi,d2+1/PrIKwH=PswaHawHwVwH,where E[nHn]=IK. From Fact 1, the optimal vector wgc in (8) is obtained, where the value of c=Pr/∥aV-1∥2 is chosen to meet the aggregate the power constraint Pr.
However, using wgc is not realistic for DBF. To calculate wgc in a distributed way, V should be delivered to each relay, but it requires a significant burden because (1) acquiring V causes very high complexity since all the individual channel coefficients of interference channel gm,n’s must be estimated and (2) sharing V causes large overhead. Therefore, using wgc in DBF is impractical, especially when K or I is large. To mitigate this problem, the following theorem introduces a simple DBF when only local CSI is available at each relay.
Theorem 1.
When only local CSI is available at each relay, the optimal beamforming weight vector wlc that maximizes the received SINR γd is given by(11)wlc=uPru2,where u is(12)u=hs,1h1,d/a1Prh1,d2∑i=1IPigi,12+1+∑i=1IPigi,d2+1/a12,…,hs,KhK,d/aKPrhK,d2∑i=1IPigi,K2+1+∑i=1IPigi,d2+1/aK2.
Proof.
To calculate the weight coefficient wk at each relay with only local CSI, V in (9) must be a diagonal matrix, and relay k needs to be able to estimate [V]k,k without communication between relays. Therefore, B must be replaced by(13)B~=diag∑i=1IPigi,12,…,∑i=1IPigi,K2.From Fact 1, wlc is obtained by(14)wlc=aV~-1PraV~-12,where(15)V~=AHB~A+AHA+∑i=1IPigi,d2+1PrIK.Because V~ is a diagonal matrix, its inverse is easily obtained from [V~-1]k,k=1/[V~]k,k, and closed-form wlc is obtained as in (11) and (12).
Each relay calculates wlc,k in a distributed way with only local CSI hs,k, hk,d, and ∑i=1IPi|gi,k|2 when (∑i=1IPi|gi,d|2+1) and u2 are delivered from the destination (to calculate u2 with very small overhead, several methods are available such as training-sequence-based channel estimation [5–7]). In this sense, wlc is called a DBF vector with local CSI. Therefore, wlc induces very small overhead. Moreover, calculating wlc causes low complexity, because each relay estimates not gm,n’s but corresponding aggregate interference plus noise power (∑i=1IPi|gi,k|2+1), which is much easier to estimate [8, 9]. Nevertheless, wlc still shows excellent performance as follows: (1) wlc achieves nearly the optimum performance of wgc when I is large enough or interference power toward relays is small and (2) wlc achieves the capacity scaling of 1/2logK, which corresponds to the maximal capacity scaling of cooperative relaying without CCI.
Corollary 2.
When the number of interferers I is sufficiently large, it becomes V~~V, and, therefore, wlc achieves the optimum performance of wgc.
Proof.
Let μk≜(1/I)∑i=1IE[Pi|gi,k|2]. When Pi is limited and I→∞, aman~(1/μmμn)(1/I) and 1/I∑i=1IPigi,m∗gi,n~{0,form≠n;μm,form=n}. Therefore,(16)AHB~Am,n=amanhm,d∗hn,d∑i=1IPigi,m∗gi,n~0,form≠nhm,d2,form=n,and V~~V.
When interference power toward relays is small, it is obvious that V~≈V, and wlc closely achieves the performance of wgc.
Theorem 3.
When K→∞ with any finite Ps, Pr, and Pi, the ergodic capacity with wlc, C¯(wlc)≜E{C(wlc)} converges to 1/2logK.
Proof.
With wlc, the received SINR at d becomes(17)γd=PswlcaHawlcHwlcVwlcH=PswlcaHawlcHwlcV~wlcH·wlcV~wlcHwlcVwlcH~aPswlcaHawlcHwlcV~wlcH=∑k=1KNk,where(18)Nk=Pshs,k2/∑i=1IPigi,k2+1Prhk,d2/∑i=1IPigi,d2+1Pshs,k2/∑i=1IPigi,k2+1+Prhk,d2/∑i=1IPigi,d2+1+1,and (a) follows from the fact that wlcV~wlcH/wlcVwlcH~1 for sufficiently large I. The ergodic capacity with wlc is given by [10]: (19)C¯wlc~E12log21+∑k=1KNk,where the factor 1/2 denotes the rate loss because of the half-duplex constraint of relays. Because Nk satisfies the Kolmogorov conditions as shown in the Appendix, the following theorem can be applied [11, Theorem 1.8.D]:(20)∑k=1KNkK-∑k=1KENkK→w.p.10.Therefore, ∑k=1KNK→w.p.1∑k=1KE[Nk], and C¯(wlc)~1/2logK.
4. Numerical Results
In this section, C¯(wlc) is compared with C¯(wcv), where wcv is the weight vector of a conventional DBF that maximizes the received SNR when there is no CCI [2]:(21)wcv=vPrv2,where(22)v=hs,1h1,da11+a12Prh1,d2,…,hs,KhK,daK1+aK2PrhK,d2.Comparing with wcv, wlc requires only a slight increase in overhead and complexity in order to estimate ∑i=1IPi|gi,k|2+1 at the corresponding relay and to feed back ∑i=1IPi|gi,d|2+1 from the destination. It is assumed that the relays are located in the middle of the source and the destination, and, therefore, σs,k2=σk,d2=1, for all k. For comparison purposes, simulation results for C¯(wgc) are also plotted. According to the location of interferers, three cases are considered as follows.
Case 1.
The distances between relays-interferers and destination-interferers are the same, and, therefore, the relays and the destination are affected by the same average interfering power with σI,i,k2=1 and σI,i,d2=1.
Case 2.
The interferers are closely located to the destination with σI,i,k2=0.5 and σI,i,d2=3.
Case 3.
The interferers are closely located to the relays with σI,i,k2=3 and σI,i,d2=0.5.
Figure 2 plots the ergodic capacity for Cases 1 and 2, and Figure 3 for Case 3, with parameter values of I=5 and Ps=Pr=Pi=10dB. For all cases, the figures show that C¯(wgc)>C¯(wlc)>C¯(wcv) and wlc achieves remarkable performance gains over wcv; when K=40, 21%, 20%, and 29% gains are obtained for Cases 1, 2, and 3, respectively. Moreover, wlc closely achieves C¯(wgc) for Case 2 because interference power toward relays is small, but wgc is superior to wlc for Cases 1 and 3 at the cost of greatly increased overhead and complexity.
Comparison of ergodic capacity: I=5, Ps=Pr=Pi=10dB, and σs,k2=σk,d2=1, for all k: Case 1 with σI,i,k2=1 and σI,i,d2=1 and Case 2 with σI,i,k2=0.5 and σI,i,d2=3.
Comparison of ergodic capacity: I=5, Ps=Pr=Pi=10dB, and σs,k2=σk,d2=1, for all k: Case 3 with σI,i,k2=3 and σI,i,d2=0.5.
As I increases, however, wlc closely achieves C¯(wgc) for all cases as shown in Figures 4 and 5, in which the ergodic capacity is plotted for I=30, Ps=Pr=10dB, and Pi=5dB. The figures shows that wlc achieves nearly C¯(wgc) for all cases and also represents remarkable performance gains over wcv, greater than 21% for all cases when K=40.
Comparison of ergodic capacity: I=30, Ps=Pr=10dB, Pi=5dB, and σs,k2=σk,d2=1, for all k: Case 1 with σI,i,k2=1 and σI,i,d2=1, and Case 2 with σI,i,k2=0.5 and σI,i,d2=3.
Comparison of ergodic capacity: I=30, Ps=Pr=10dB, Pi=5dB, and σs,k2=σk,d2=1, for all k: Case 3 with σI,i,k2=3 and σI,i,d2=0.5.
5. Conclusions
This paper has proposed the optimal AF DBF wlc in the presence of CCI when only local CSI is available at each relay. With slight increased overhead and complexity, wlc efficiently reduces the impact of CCI and yields significant improvements over wcv. Using wlc is more attractive when interference power toward relays is small or there are a large number of interferers where wlc achieves nearly the same performance as wgc.
AppendixLemma A.1.
For any finite Ps, Pr, and Pi with large K, Nk in (18) satisfies the Kolmogorov conditions:(A.1)∑k=1KVARNkk2<∞,μ≜1K∑k=1KENk<∞.
Proof.
Let Mk≜Ps|hs,k|2Pr|hk,d|2/(Ps|hs,k|2+Pr|hk,d|2+1). Then, Mk<Ps|hs,k|2Pr|hk,d|2/(Ps|hs,k|2+Pr|hk,d|2)<min{Ps|hs,k|2,Pr|hk,d|2}<Ps|hs,k|2. Ps|hs,k|2’s are exponentially distributed r.v.’s which mean Psσs,k2 and variance (Psσs,k2)2 are bounded. Therefore, the Kolmogorov conditions ∑k=1KVAR[Ps|hs,k|2]/k2<∞ and (1/K)∑k=1KE[Ps|hs,k|2]<∞ are satisfied. Since Nk<Mk<Ps|hs,k|2, Nk also satisfies the Kolmogorov conditions.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This work was partly supported by National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP) (NRF-2015R1A2A2A01008218), Institute for Information & Communications Technology Promotion (IITP) grant funded by the Korea government (MSIP) (no. B0101-15-0557, Resilient Cyber-Physical Systems Research), and the Robot Industry Fusion Core Technology Development Project of the Ministry of Trade, Industry & Energy of Korea (10052980).
BölcskeiH.NabarR. U.OymanÖ.PaulrajA. J.Capacity scaling laws in MIMO relay networks2006561433144410.1109/TWC.2006.16386642-s2.0-33745066114YiZ.KimI.-M.Joint optimization of relay-precoders and decoders with partial channel side information in cooperative networks200725244745810.1109/jsac.2007.0702192-s2.0-33847766668KimJ.-B.LeeI.-H.KimD.LimJ.Asymptotic performances of near-optimal selective distributed beamforming for two-way relaying with partial channel state information201115551251410.1109/LCOMM.2011.030911.1023992-s2.0-79956211971YangL.AlouiniM.-S.Outage probability of distributed beamforming with co-channel interference201216333433710.1109/LCOMM.2011.122211.1120962-s2.0-84858339641GaoF.CuiT.NallanathanA.On channel estimation and optimal training design for amplify and forward relay networks2008751907191610.1109/TWC.2008.0701182-s2.0-47149084050AhmedN.KhojastepourM. A.SabharwalA.AazhangB.Outage minimization with limited feedback for the fading relay channel200654465966910.1109/TCOMM.2006.8730742-s2.0-33645996365KimJ.-B.ChoiJ.-W.CioffiJ. M.Cooperative distributed beamforming with outdated CSI and channel estimation errors201462124269428010.1109/tcomm.2014.23648342-s2.0-84919651259DighamF. F.AlouiniM.-S.SimonM. K.On the energy detection of unknown signals over fading channels2007551212410.1109/TCOMM.2006.8874832-s2.0-33847674846KimJ.-B.KimD.Outage probability and achievable diversity order of opportunistic relaying in cognitive secondary radio networks20126092456246610.1109/TCOMM.2012.072412.1007182-s2.0-84866737063LanemanJ. N.WornellG. W.Distributed space-time coded protocols for exploiting cooperative diversity in wireless networks200349102415242510.1109/tit.2003.817829MR20279952-s2.0-0242302539SerflingR. J.1980New York, NY, USAJohn Wiley & SonsMR595165