588 IEEE COMMUNICATIONS LETTERS, VOL. 21, NO. 3, MARCH 2017 Antenna Port Selection in a Coordinated Cloud Radio Access Network Gurhan Bulu, Talha Ahmad, Ramy H. Gohary, Cenk Toker, and Halim Yanikomeroglu Abstract—We investigate the optimization of antenna port selection in the downlink of a cloud radio access network in which a user terminal can be served by multiple ports. The goal is to maximize the minimum weighted signal-to-interference-plus- noise ratios observed by the users while satisfying their quality-of- service requirements. This optimization is formulated as a mixed integer programming problem and solved using semidefinite relaxation and Gaussian randomization. It is shown that our technique outperforms baseline schemes by 6–8 dB. Index Terms— Distributed antennas, C-RAN, antenna selection. I. I NTRODUCTION D ISTRIBUTED antenna systems (DAS) offer an effective means for improving the spectral efficiency of wireless communication networks, including those with heterogeneous traffic. In DAS, antennas are not collocated at a base station (BS), but dispersed over the coverage area. A key DAS objective is to improve coverage and capacity of indoor and cellular wireless communication systems [1]–[3]. However, realizing this potential requires proper selection of the antenna ports [4], [5]. Approaches for optimizing this selection include those based on maximizing the sum-rate [6] and maximizing the minimum signal-to- interference-plus-noise ratio (SINR) [7], [8]. To elaborate, in [7] the transmission parameters are optimized in a DAS in which the user terminals (UTs) are restricted to be served by the ports of their respective BSs. We advance the antenna port selection problem to an emerging area of “cloud” radio access networks. In particular, we consider a DAS in which UTs are served by arbitrary ports, not necessarily those of their original BSs. Such a broader con- figuration enables cell-edge UTs to be served by neighbouring BSs, thereby creating UT-centric virtual cells with boundaries that flex freely to accommodate mobility and time-varying demand. The creation of such virtual cells enriches the design with valuable degrees of freedom and offers the potential of achieving a significantly better performance. To realize this potential, we develop an optimization-based framework that uses the semidefinite relaxation (SDR) and Gaussian randomization technique [9]. This technique was originally devised in [10] to provide efficiently computable solutions for a class of NP-hard problems, including the max-cut and the Manuscript received September 1, 2016; revised November 1, 2016; accepted November 9, 2016. Date of publication November 14, 2016; date of current version March 8, 2017. The associate editor coordinating the review of this letter and approving it for publication was J. Ben Othman. G. Bulu and C. Toker are with the Department of Electrical and Elec- tronics Engineering, Hacettepe University, Ankara 06800, Turkey (e-mail: bulu@ee.hacettepe.edu.tr). T. Ahmad is with Ericsson Canada, Ottawa, ON K2K 2V6, Canada. R. H. Gohary and H. Yanikomeroglu are with the Department of Systems and Computer Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada. Digital Object Identifier 10.1109/LCOMM.2016.2628380 satisfiability problems. It was shown in [10] that the solution yielded by this approach is at least 87.5% of the optimal solution, but in practice, this algorithm is known to yield solutions that are typically closer to the optimal solution [9]. Hence, it can be seen that SDR with Gaussian approximation offers performance guarantees that are usually not available to other polynomial-complexity algorithms. Simulation results demonstrate that although the proposed technique does not necessarily yield the optimal solution, it captures the flexibility offered by virtual cells for and yields superior performance for both homogeneous and heterogeneous traffic models [11]. The main contributions of this work are as follows: 1) We propose a DAS cloud configuration in which ports can transmit to any UT using binary or ternary states. 2) We apply the SDR technique to DAS clouds with homogeneous and heterogeneous UT distributions. II. SYSTEM MODEL We consider a DAS cloud with L ports and K UTs. Transmissions in this cloud are coordinated by a central entity, which can be an elect BS with computational capabilities sufficient to coordinate transmissions from ports to UTs. To perform this task, the central entity is assumed to know the channel gains between the ports and the UTs. Each UT can be served by any port in the cloud. However, a port can transmit to at most one UT at any time instant. Let Ŵ ∈{0, 1} L ×K be a matrix whose ℓk -th entry, γ ℓk = 1 if the ℓ-th port is used to serve the k -th UT and γ ℓk = 0 otherwise. Restricting the ℓ-th port to transmit to at most one UT yields ∑ K k=1 [Ŵ] ℓ,k ≤ 1, ℓ = 1,..., L . Let P ℓ be the transmit power of this port. Then the M R k × 1 received signal of the k -th UT, k = 1,..., K , is y k = L  ℓ=1 γ ℓk  P ℓ H ℓk x k + K  i =1,i  =k L  ℓ=1 γ ℓi  P ℓ H ℓk x i + η k , where, H ℓk ∈ C M R k ×M T ℓ is the channel matrix between the ℓ-th port and the k -th UT, M T ℓ is the number of transmit antennas of the ℓ-th port, M R k is the number of receive antennas of the k -th UT, x k is the normalized data vector and η k ∼ CN (0,σ 2 I M R k ) is the additive noise. Letting γ = vec(Ŵ), the SINR of the k -th UT is SINR k (γ ) = β k    ∑ L ℓ=1 γ ℓk √ P ℓ H ℓk    2 ∑ K i =1,i  =k    ∑ L ℓ=1 γ ℓi √ P ℓ H ℓk    2 + σ 2 M R k , = β k γ T A k γ γ T B k γ + σ 2 M R k , (1) where β k > 0 is a weight used to prioritize the SINR of the k -th UT [8], A k ∈ R LK ×LK and B k ∈ R LK ×LK are 1558-2558 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.