large-scale mimo-based

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 2

S M A RT B A C K H A U L I N G A N D F R O N T H A U L I N G F O R 5G N E T W O R K S

LARGE-SCALE MIMO-BASED WIRELESS BACKHAUL IN 5G NETWORKS ZHONGSHAN ZHANG, XIYUAN WANG, KEPING LONG, ATHANASIOS V. VASILAKOS, AND LAJOS HANZO

ABSTRACT In order to enhance the attainable transmission rates to the levels specified by future wireless communications, a paradigm shift from conventional small-scale MIMO to LS-MIMO is highly desirable. LS-MIMO technology as a “clean-slate” approach is shown to be capable of dramatically increasing the area spectral efficiency (SE, as measured in bits per second per Hertz per square kilometer) while simultaneously improving the energy efficiency as measured in bits per Joule. Furthermore, the concept of LSMIMO has established itself as a beneficial transmission/detection paradigm, thus substantially reducing the impact of interference relying on some advanced transmit precoding/beamforming/detection techniques. This article is intended to offer a state-of-the-art survey on LSMIMO research, to promote the discussion of its beneficial application areas and the research challenges associated with BF aided wireless backhaul, LS-MIMO channel modeling, signal detection, and so on. Additionally, a joint group power allocation and pre-beamforming scheme called JGPAPBF is proposed to substantially improve the performance of LS-MIMO-based wireless backhaul in heterogeneous networks. Our hope is that this article will stimulate future research efforts.

INTRODUCTION AND MOTIVATION OF LS-MIMO Zhongshan Zhang, Xiyuan Wang, and Keping Long are with the University of Science and Technology Beijing (USTB). Athanasios V. Vasilakos is with Lulea University of Technology. Lajos Hanzo is with the University of Southampton.

2

With the rapid development of wireless communication technologies and systems, the explosive growth of wireless device usage has increased the customers’ demand for wireless data rate to unforeseen levels. The gains already achieved in area spectral efficiency (SE, as measured in bits per second per Hertz per square kilometer) still fail to meet the expected growth of data demand, despite some advanced SE improvement technologies having achieved steady improvements. Meanwhile, the power consumption of information and communication systems is becoming a major concern of both academia and industry, urgently requiring substantially improved energy efficiency (EE, as measured in bits per Joule).

1536-1284/15/$25.00 © 2015 IEEE

In light of the fact that the SE of singleantenna-aided systems has been approaching the Shannon capacity, multiple-input multiple-output (MIMO) techniques have been considered as an attractive solution for increasing SE through providing extra degrees of freedom (DoF) [1]. Theoretically, the more antenna elements (AEs) devices are equipped with, the higher the DoF provided by the propagation channel, enabling the performance, quantified in terms of channel capacity and/or link reliability, to be improved. However, the existing studies show that conventional small-scale MIMO technologies are gradually approaching SE satisfaction in fourth generation (4G) wireless systems. It is generally recognized that any further improvement in SE will only be marginal. In order to meet the SE/EE requirements in the fifth generation (5G) of wireless standards, two widely recognized schemes — the large-scale MIMO (LS-MIMO) and small-cell technologies — are generally considered. •In order to further enhance the MIMO transmission rates to the levels required by future wireless communications, new ways of achieving substantial SE/EE improvements for defining 5G standards must be investigated. A paradigm shift from small-scale MIMOs to the LS-MIMOs is thus highly desirable. In [2], LSMIMO is defined and distinguished from classical MIMO schemes as a particular operating condition of cellular systems that is capable of providing a promising means of meeting the growing throughput demand, while simultaneously improving both the quality of service (QoS) and quality of experience (QoE) in next-generation wireless communication systems. Furthermore, this technology may become an integral part of future wireless cellular systems due to its capability of simultaneously serving multiple users while counteracting the inter/intra-cell interference imposed by pilot contamination. •The exponential growth of users’ demand for mobile traffic has also motivated extensive research into spectrally efficient techniques by substantially improving the geographic spectrum reusability. Small-cell deployments, in which densely installed pico/femtocell base stations (BSs) consume a power much lower than macro-

IEEE Wireless Communications • October 2015

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 3

kh au l ac sb les ire W

ul ha ck ba

IEEE Wireless Communications • October 2015

ess

The backhaul technology represents the family of potential connections spanning from the BSs to the core network. In HetNets, spatial reuse gains can be exploited by extreme network densification relying on small-cell deployments, in which small-cell BSs have to be connected to their donor BSs (i.e., the macrocell BSs) through wired or wireless backhaul, as illustrated in Fig. 1. Although the classic wired solution may provide a high-speed, high-reliability backhaul to support the small cells, the extra cost of installing a wired backhaul forces the operators to resort to more economical, easy-to-deploy wireless solutions. Unlike conventional macrocell BSs, which are typically connected through a highcapacity wired backhaul network, small-cell BSs are likely to be connected via a time-variant wireless backhaul infrastructure rather than expensive wired backhaul. As compared to the wired-backhaul solution, the wireless backhaul exhibits a significant benefit in terms of both hardware costs and deployment difficulties. However, the specific choice of wireless backhaul solutions in HetNets depends on a number of system considerations, such as the network capacity, small-cell deployment density, data rate requirements, and electromagnetic interference issues, as well as maintaining low power consumption and the availability of orthogonal backhaul spectrum. It thus imposes new requirements, including: • In HetNets comprising both macrocells and densely deployed small cells, forwarding the massive backhaul traffic into the core network becomes a critical problem. • The wireless backhaul has demanding communication requirements, such as high throughput and low delay. For instance, high throughput achieved at low delay is capable of supporting flawless broadband multimedia communications. • Considering the fact that pilot contamination may lead to performance erosion in LSMIMO systems, which may become further aggravated by additionally supporting back-

cces s

re l

REQUIREMENT OF WIRELESS BACKHAUL IN HETNETS

User a Wi

cell BSs due to the reduced path loss in the shrinking cell size, could significantly improve both the SE and EE of the cellular systems simultaneously. However, each individual technology alone is unlikely to totally satisfy the QoS/QoE and throughput requirements for 5G systems. A sophisticated solution is thus required to be proposed for enabling the coexistence and interplay of the above-mentioned two technologies. Heterogeneous networks (HetNets), in which a macrocell tier is overlaid with a dense tier of small cells (i.e., comprising pico/femtocells), have been regarded as a promising architecture for wireless access networks in terms of both spectral and power efficiencies. In particular, LS-MIMO-aided BSs can be deployed to provide macrocell coverage, while small-cell deployments may act as the main capacity driver for high-throughput user equipments (UEs) with low mobility.

UE group

Macrocell Wired backhaul Picocell

Picocell group Picocell

Figure 1. The illustration of wireless backhaul in HetNet systems.

haul services, a high-performance wireless backhaul relying on high-performance interference mitigation capability is required. For example, we may allocate different spectral resources for the wireless access and the backhaul, thus mitigating the above-mentioned interference. • The EE of the wireless backhaul becomes of critical concern, as the small-cell BS density increases. Energy-efficient techniques such as beamforming can be employed to substantially reduce the power consumption of the wireless backhaul.

LS-MIMO-BASED BEAMFORMING FOR WIRELESS BACKHAUL IN HETNETS LS-MIMO technology has been advocated by virtue of its ability to significantly improve the channel capacity relying on antenna arrays (AAs) with an order of magnitude more AEs (i.e., 100 elements or more) than in systems being rolled out at the time of writing. It was claimed in [3] that LS-MIMO techniques are capable of increasing the SE by one or two orders of magnitude by exploiting a higher DoF, while simultaneously improving the EE by three orders of magnitude. All these above-mentioned properties render LS-MIMOs a promising technology for 5G wireless communication systems. Several promising characteristics of LS-MIMO technology are described in Fig. 2. Apart from the capability of SE/EE improvement, LS-MIMO technology may also provide a promising means of facilitating high-performance wireless backhaul in HetNets. Relying on beamforming (BF) technology, large-scale antenna arrays can focus their radiated power toward the intended receivers (Rxs), substantially reducing the intra-/inter-cell interference.1 Since the same spectrum can be reused among wireless backhaul and cellular BSs/UEs, the BSs in small cells can be essentially regarded as special UEs communicating with LS-MIMO-based donor BSs using wireless backhaul. Therefore, the wireless links associated with donor BSs may suffer from either quasi-static or time-varying fading.

1 By packing a large number of AEs into a limited area, LS-MIMO transceivers enable BF processing to significantly improve the signal strength in the intended direction and suppress interference simultaneously, thus leading to much higher cell throughput and cell-edge SE than traditional omnidirectional antennas.

3

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 4

Large-array gain LS-MIMO capacity t MRC: (1 – )Nr T ⋅log2 (1+ R= ZF: (1 –

(Nt – 1)P2 ) (Nr – 1)P2 + + NrP + 1

Improved SE/data rate under low SNR conditions: A large number of UEs can be accommodated by using LS-MIMO at the BS to remarkably improve the system’s performance in terms of either its sum data rate or link reliability. LS-MIMO techniques equip the system with the capability of always recovering from low SNR conditions by adding a sufficiently high number of AEs.

t )N T r

⋅log2 (1+

(Nt – Nr)P2 ) (+ Nr)P + 1

Better SER performance: Better SER performance can be attained by fitting an increased number of AEs at the LS-MIMO Tx for a fixed physical device size and for a given set of Rxs.

Reduced complexity in signal processing: At the limit, when we have an infinite number of BS-AEs, the performance of low-complexity linear processing in the form of maximum ratio combining (MRC) aided UL reception and maximum ratio transmission (MRT)-assisted DL BF becomes optimal.

Linear array Spherical array

Cylindrical array Rectangular array

Distributed array

More efficient matrix operations than in conventional MIMO: When the matrix dimensions become large, the resultant matrices tend to be “well-conditioned,” facilitating computationally efficient matrix operations. The asymptotic benefits of random matrix theory can thus be exploited, with the effects of both fast fading and uncorrelated noise vanishing in the limit for an infinite number of AEs.

Improved EE: LS-MIMO systems are capable of enormously improving the EE by offering the unique prospect of reducing the transmit power by an order of magnitude (or more), while averaging out the effects of smallscale fading (SSF). In addition, they also have the potential to reduce the power dissipated by the signal processing at the Tx relying on lowcomplexity schemes for eliminating the multiuser interference. At the limit, the transmit power per UE can by made arbitrarily low as the number of AEs tends to infinite, while the simplest form of detection relying on the classic MF algorithm becomes optimal.

Figure 2. The main benefits brought by employing LS-MIMO technology.

MOTIVATION OF THIS ARTICLE The main driving force behind the advances in LS-MIMO technology is the promise of offering substantial SE/EE improvements. It is possible to enhance the channel capacity and simultaneously reduce the energy consumption provided that increased hardware/software complexity is tolerable. One of the fundamental motivations of this article is to survey the hitherto discovered LS-MIMO techniques and propose a high-performance wireless backhaul solution. We review the state-of-the-art progress in LS-MIMO technology, and then propose a new joint group power allocation and pre-beamforming (JGPAPBF) scheme for LS-MIMO based wireless backhaul. The remainder of this article is organized as follows. The existing studies of LSMIMO techniques are listed in the following section. Then a JGPAPBF scheme conceived for joint spatial division and multiplexing in LSMIMO is proposed for performing LS-MIMObased wireless backhaul. Following that, future research directions in LS-MIMOs are discussed. Finally, our conclusions are provided in the last section.

EXISTING STUDIES OF LS-MIMOS The LS-MIMO technology, which constitutes a promising approach of improving the attainable SE by employing a large number of AEs to serve a lower number of UEs, has drawn considerable interest from both academia and industry. Apart from its SE, the LS-MIMO technology is also capable of significantly improving the EE by orders of magnitude compared to a single-anten-

4

na-aided system. Furthermore, the other performance indices such as link reliability, the capability of combatting fast fading and uncorrelated noise, and the complexity reduction in transmit precoding (TPC) may also be significantly improved when the number of AEs grows large. However, as distinguished from a classical MIMO scheme, the issues in LS-MIMO systems associated with multiuser interference, channel estimation errors, and uncorrelated noise would have a negligible impact compared to pilot contamination observed in multicellular environments. LS-MIMO technology has to be designed for realistic assumptions in order to provide a unified performance analysis in practical scenarios, because most of the previous studies are based on several idealized simplifying assumptions about the propagation conditions, hardware implementation, and number of AEs. In this section, the state-of-the-art studies of LS-MIMO techniques are surveyed and compared, followed by quantifying the performance of LS-MIMO techniques in terms of capacity, SE/EE, channel estimation, and signal detection.

LS-MIMO CAPACITY ANALYSIS The channel capacity achieved by considering an asymptotically large number of AEs has been studied by Marzetta [4], showing that the LSMIMO throughput in the limit becomes saturated and depends only on the specific topology of the network. Some random parameters, such as the singular values of the channel matrix in small-scale MIMO, become deterministic in LSMIMO, substantially enabling capacity optimization in a cost-effective manner. Furthermore, the

IEEE Wireless Communications • October 2015

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 5

LS-MIMO channel

3D LS-MIMO channel measurement at 2.6 GHz

2D linear array

Finite AEs

Infinite AEs

Cylindrical array with 128 AEs

Scalable array up to 112 AEs

128-port cylindrical patch array

50 MHz Cylindrical array having 16 dualpolarized directional patch AEs in each circle

20 MHz Outdoor channel measurement

50 MHz 32 transmit AEs Cylindrical patch array having 16 dualpolarized AEs in each circle 4 circles 29.4 cm diameter 28.3 cm height

4 circles /2-spacing 7.3 cm width

A vertical array of 8 custom-built dualpolarized AEs with /2-spacing A large cylindrical AA comprising 7 AEs rotated on a 1 m radius circle

Figure 3. The LS-MIMO channel measurement with 2D and 3D array configurations.

capacity or throughput of LS-MIMO systems is observed to be predominantly impacted by pilot contamination rather than other impairments such as uncorrelated noise and channel estimation errors. Capacity of the LS-MIMO Downlink: Under favorable propagation conditions [4] and provided that we have sufficient AEs, the columns of the channel matrix become nearly orthogonal. As the number of downlink BS-AEs grows without limit, considering the channel impulse response (CIR) feedback to the BS, which uses a low-complexity linear TPC based on the matched filter (MF), the total capacity can be represented in a simple form as [4] Cfsum =

  2 βlkl  B   Tslot − Tpilot  Tu .  + log 1 2  T Tslot  s ∑ j ≠l β2jkl  k =1 (1) Nr

∑  α  

Capacity of the LS-MIMO Uplink: When the number of BS-AEs exceeds that of the UEs, asymptotically orthogonal column vectors for the LS-MIMO channel matrix under a favorable propagation scenario can be attained. By considering maximal ratio combining (MRC) relying on perfect CIR knowledge at the BS, the achievable uplink ergodic rate per user has been found to satisfy [3] Rk ,mrc, estimated CSI ≥

  τP 2 ( N t − 1) τ  .  1 −  log2  1 + T P(τP + 1)( N r − 1) + (τ + 1)P + 1  (2)

IEEE Wireless Communications • October 2015

In a cellular system comprising LS-MIMO aided BSs and some single-antenna aided UEs, the uplink-downlink performance can be completely determined by the shadow-fading coefficients and the locations of mobile/BSs, thus making a strong case for a rigorous study of these performance metrics using stochastic geometric models.

LS-MIMO CHANNEL MEASUREMENT AND MODELLING Theoretically, as the number of BS-AEs tends to infinity, one LS-MIMO-aided BS can serve an arbitrarily large number of UEs, provided that an accurate CIR is available at the BS. However, in order to sufficiently exploit the potential of LS-MIMO technology, LS-MIMO setups with realistic propagation environments must be capable of providing enough decorrelation between UEs’ channels, preventing studies from only relying on favorable propagation conditions. LS-MIMO channel measurement in practical propagation environments can be used to corroborate the theoretical models. For instance, in a cellular system comprising LS-MIMO-aided BSs and some single-antenna-aided UEs, the uplink-downlink performance can be completely determined by the shadow-fading coefficients and the locations of mobile/BSs, thus making a strong case for a rigorous study of these performance metrics using stochastic geometric models. Some preliminary works associated with LS-MIMO channel measurement and modeling are shown in Fig. 3. Although several open issues with respect to the behavior in realistic channels are still needed to be further resolved, the overall LS-MIMO system’s performance has already exhibited very promising advantages over classical small-scale MIMO with the aid of practical channel measurement and modeling.

CHANNEL ESTIMATION IN THE PRESENCE OF PILOT CONTAMINATION In a cellular system, the multi-antenna-aided BS must acquire CIR via channel estimation before performing downlink TPC and uplink signal detection. Note that CIR can be obtained through (limited) feedback in frequency-division duplexing (FDD) systems. Unlike FDD, timedivision duplexing (TDD) systems enable CIR

5

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 6

Detector

Linear

Optimal linear Rx

UL training is used to estimate the UL channel An optimal linear receiver is developed by maximixing the received SINR

Nonlinear

Iterative linear filtering scheme

Random step methods

Soft information is propagated between different iterations of the hard detector The operations of matrix inversion are employed

Choosing the neighboring vector with the smallest MSE

The computational complexity increases with the Nt

Restart the process until the number of repeat times reaches a predefined value

Tree-based algorithms

Practical

Approximate matrix inversion

Enumerating all combinations of the first scalar symbols in the received vector The remaining symbols are detected using ZF detector with decision feedback

A small number of Neumann-series terms are used to obtain the near-optimal performance The computational complexity is reduced

Low-complexity Low-complexity SSG detector MCMC algorithm Considering many AEs at the BS to serve a similar number of MSs The stalling problem at high SNRs can be effectively alleviated

Built on a group detection strategy Substantial complexity reduction The performance is close to the full dimensional joint detector

Figure 4. The LS-MIMO detectors (including linear detectors, nonlinear detectors and some practical detectors) and their main features. acquisition by exploiting the feature of reciprocity. Therefore, TDD rather than FDD offers a suitable way to acquire timely CIR in LS-MIMO systems due to the reduced training burden. However, as the number of served users increases, the pilot contamination effect predetermines the ultimate performance limit of the LS-MIMO system, because the channel estimation error imposed by pilot contamination would limit the attainable throughput in the presence of sophisticated precoding/decoding schemes. Several approaches have been proposed for reducing the impact of pilot contamination. In multicell LS-MIMO systems, a new pilot design criterion, which can be described as minimizing the inner product of the correlation vectors of the pilots transmitted by difference cells, has been employed for exploiting the orthogonality between channel vectors of different users so as to mitigate the impact of pilot contamination. Following the above-mentioned criterion, pilots based on Chu sequences [5] can be employed to substantially reduce the inter-cell interference level. In particular, if the sequence’s duration is not less than the number of simultaneously activated co-cell UEs, orthogonal pilot allocation inside a cell can be guaranteed, making each cell only interfered by some rather than all of the other cells.

EFFICIENT SIGNAL DETECTION In LS-MIMO systems, multiple UEs may transmit simultaneously to the BS, requiring sophisticated signal processing to be performed at the BS for extracting the signal of each UE from the aggregate received signal. As the number of BSAEs grows large, the random channel vectors between the BS and the UEs become pair-wise orthogonal, enabling the linear Rxs with simple signal processing to provide a cost-efficient way toward low-complexity signal detection at BSs. However, linear detectors are no longer optimal in multicellular systems, because the conventional linear detectors do not take into account the correlation between the channel estimate and

6

inter-cell interference. In order to mitigate the above-mentioned disadvantages in linear detectors, several nonlinear detectors such as random step (RS) and treebased (TB) algorithms [6] have been proposed to effectively exploit the correlation between the channel estimate and inter-cell interference. Furthermore, since practical detection algorithms scale favorably to high-dimensional problems, some cost-efficient algorithms such as soft-input soft-output successive group (SSG) [7] have been developed for combating the increased computational complexity in LS-MIMO’s signal detection. The existing LS-MIMO detectors are surveyed and compared in Fig. 4.

THE DEVELOPED LS-MIMO PROTOTYPES In the rapid development of LS-MIMO techniques, basic prototyping work has also been widely carried out, including the testbed jointly developed by Linköping University and Lund University [8], the Argos testbed developed at Rice University in cooperation with AlcatelLucent [9], and the Ngara testbed developed in Australia [10]. A substantial performance improvement over single-antenna-based systems can be achieved in the developed LS-MIMO prototypes. In the Argos testbed, for example, LS-MIMO technology can achieve 12-fold capacity gain and/or 64-fold EE gain over singleantenna-based systems by employing 64 BS-AEs.

BEAMFORMING TECHNOLOGY FOR LS-MIMOBASED WIRELESS BACKHAUL IN HETNETS In multi-tier HetNets, the system is required to provide backhaul to all the small cells and simultaneously find efficient methods to leverage higher frequency bands for UE access and backhaul [11]. However, a large directional gain in wireless backhaul waveform propagation has to be maintained in order to combat the path loss and mitigate the inter-cell interference simultaneously. BF technology, which is capable of exploiting the DoF of both the spatial and tem-

IEEE Wireless Communications • October 2015

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 7

TPC/beamforming

Classical

Practical

Linear TPC

Nonlinear precoding

BF under per-node power constraint

The iterative BF algorithm applies nonlinear precoding and signal detection

The channel can be modeled as a multi-banded matrix

Each of the multiple data streams is multiplied by k complex coefficients

The obtained communication rate approximates to the optimal waterfilling under sumpower constraint

The complexity of ZFBF can be reduced by keeping the most significant entries of both channel and TPC metrics

The proper k-regularity beamformer implements functionalities from antenna selection to optimal eigen beamforming

ZF-based TPC

Matched filter

MMSE-based TPC

Converting the channel by means of pseudo-inverse

Matrix inversion for LS-MIMO becomes unnecessary when the array is scaled up

The impact of pilot contamination is mitigated

Nonlinear techniques apply to the Nt Nr regime

TPC in multicellular environment can be derived by maximizing the minimum of the data rates

Nonlinear precoding such as vector perturbation shows an evident advantage when Nt is not much larger than Nr

It is the conceptually simplest approach

The SINR can be made as high as desired by scaling up the array

Low-complexity ZFBF

BF design based on k-regularity

Figure 5. Classification of TPC/beamforming techniques in LS-MIMO systems.

poral domains to substantially decrease the inter-symbol interference and improve EE/SE simultaneously, may provide a promising solution for relieving the above-mentioned challenges.

EXISTING BEAMFORMING TECHNIQUES FOR LS-MIMO SYSTEMS BF/TPC techniques have been widely studied in LS-MIMO systems. Among them, the typical linear techniques include zero-forcing (ZF), block diagonalization (BD), and MF-based TPC. Furthermore, some nonlinear approaches such as vector perturbation have also been proposed for improving the downlink transmission, exhibiting an evident technological advantage when the number of BS-AEs is not much larger than that of UEs. In Fig. 5, the existing TPC/BF techniques are classified and compared, showing both the advantages and disadvantages of each algorithm.

PRE-BEAMFORMING FOR BACKHAUL AGGREGATION IN HETNETS In HetNets, the small cells may rely on diverse wireless backhaul networks constituted by a variety of architectures, such as point-to-point MIMO links, in order to connect to a mesh network [11]. Pre-beamforming can be effectively utilized in HetNets for creating a high-performance wireless backhaul, while simultaneously mitigating the impact of interference. For instance, a per-group processing (PGP) approach such as the so-called joint spatial division and multiplexing (JSDM) scheme can be employed to both improve the efficiency of FDD-based LS-MIMOs and substantially reduce the training burden imposed. Pre-beamforming can thus be applied to the low-dimensional channel matrix of

IEEE Wireless Communications • October 2015

each group/aggregation, thus allowing the conventional ZF or regularized ZF (RZF) based BF to be utilized within each group. Consequently, in the JSDM-based LS-MIMO system comprising N user groups, the received signal vector of the users in the nth group can be written as N

y n = Hn Bn Fns n + H n ∑ Bl Fl sl + v n , l ≠n

(3)

where sn having unit-power elements denotes the transmitted signal of the users in the nth group (n = 1, … , N), H n represents the channel matrix, Bn and Fn describe the pre-beamformer and the user beamformer of the nth group, respectively. Finally, v n stands for the additive noise. Let ^ H n = H nB n be the effective channel matrix; we define pn = tr(BnFnFnHBnH) to denote the power allocated to group n, where Fn can be given by ^ H + a I)–1, RZF beamformer, HnH(^ HnH Fn = xn^ n n (4a) ^ H)–1, Fn = xn^ HnH(^ HnH n

ZF beamformer (4b)

a n is the regularization constant, and the variable x n is the coefficient that makes F n satisfy the group power constraint p n . Typically, the number of columns in Bn is much lower than the number of BS antennas, thus resulting in a low dimension at user beamformer Fn. In the original JSDM scheme, the pre-beamforming matrices {Bn} were derived by invoking the approximated block diagonalization (ABD) method, subject to the semi-unitary constraints, where the column vectors of Bn are normalized and mutually orthogonal.

7

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 8

10 9 8 7 6 5 4

Nt = 128 OPT: N = 3 JGPAPBF: N = 3; Theory JGPAPBF: N = 3; Simulation ABD−OGPA: N = 3; Theory ABD−OGPA: N = 3; Simulation ABD−UGPA: N = 3; Theory ABD−UGPA: N = 3; Simulation OPT: N = 9 JGPAPBF: N = 9; Theory JGPAPBF: N = 9; Simulation ABD−OGPA: N = 9; Theory ABD−OGPA: N = 9; Simulation ABD−UGPA: N = 9; Theory ABD−UGPA: N = 9; Simulation OPT: N = 15 JGPAPBF: N = 15; Theory JGPAPBF: N = 15; Simulation ABD−OGPA: N = 15; Theory ABD−OGPA: N = 15; Simulation ABD−UGPA: N = 15; Theory ABD−UGPA: N = 15; Simulation

3 2 1 0 0

5

10

15

20

25

P T (dB)

Figure 6. Minimum spectral efficiency vs. PT in the proposed JGPAPBF algorithm with Nt = 128.

PRE-BEAMFORMING WITH GROUP POWER ALLOCATION FOR LS-MIMO-BASED BACKHAUL IN HETNETS Since the ABD-based pre-beamforming method is heuristic and imposes semi-unitary constraints on the pre-beamforming matrices, it cannot totally cancel the inter-group interference (IGI). In order to overcome the above-mentioned drawbacks, an optimization framework in which the group power allocation and the pre-beamformer are jointly optimized [12] can be proposed for designing the JSDM pre-beamformer. The above-mentioned joint optimization can be achieved by employing the per-user ergodic rate balancing criterion of max min Rnk ,

p,{B n } nk , n

s.t.

N

∑ pn ≤ PT ,

n =1

(5)

where R nk = E[log 2(1 + g nk)] denotes the peruser ergodic rate, g n k represents the instantaneous signal-to-interference-plus-noise ratio (SINR) of user k in the nth group, p = [p1, … , pN]T is the power allocation vector, and PT stands for the total BS power dissipation. Since the exact ergodic rate is hard to derive due to the requirement of taking the expectation of the cost function, the approximate solution proposed by Wagner et al. can be employed [13], in which the instantaneous SINR g n k in (5) is replaced by its deterministic approximation. In this way, the original rate-balancing problem can be reduced to an approximate SINR-balancing problem, which is much easier to solve. Further-

8

more, inspired by the SINR balancing conceived for the multiple-input single-output (MISO) downlink, the well-known uplink-downlink relationship defined in [14] can also be invoked for simplifying the proposed optimization. Additionally, the proposed JGPAPBF optimization of [12] shares a structure similar to the SINR-balancing algorithm of [14], in which both the convergence of the iteration and its global optimality can be proven. In order to evaluate the performance of the proposed JGPAPBF algorithm and the other three reference algorithms, including ABD prebeamforming with uniform group power allocation (ABD-UGPA), ABD pre-beamforming with optimal group power allocation (ABD-OGPA), and the optimal SINR balancing algorithm considering full channel state information (CSI) available at the transmitter in [14] (OPT), vs. the total-BS-power PT, the following settings have been considered: • The LS-MIMO aided BS in the proposed algorithm is equipped with a uniform circular array (UCA) having 128 AEs. • Without loss of generality, the group numbers of the small cells and/or UEs are assumed to be {3, 9, 15}. Meanwhile, the number of smallcell BSs/UEs in each group is assumed to be 3. • The channel covariance matrices (CCMs) are given by {wnRn}, where wn for group number N = {3, 9, 15} is taken from vectors [0.1, 0.5, 1, 0.1, 0.1, 0.1, 0.5, 0.5, 0.5, 1, 1, 1] and [0.1, 0.1, 0.1, 0.1, 0.1, 0.5, 0.5, 0.5, 0.5, 0.5, 1, 1, 1, 1, 1], respectively. In order to emulate the uneven channel qualities of different groups, the entries of Rn are defined by [15] [R n ]mp = 1 2∆

∫

∆+θn −∆+θn

e

2 πm  2 πp     −2 πjD cos α−  − cos  α−    N  N  

dα

(6)

where the angular spread is given by D =15°, the azimuth angle of arrival (AoA) of the nth group is qn = –p + D+2p(n – 1)/N (n = 1, … , N), and D = 1/(4sinp/M). • For the three reference algorithms, co-group users’ signals are precoded by using a ZF beamformer. Unlike the optimal SINR balancing algorithm, which is performed relying on complete instantaneous CSI, all the other algorithms are implemented by accessing both the statistical CSI of each group (i.e., the CCMs) and the perfect instantaneous CSI within each group. From Fig. 6, the effect of AoA spacing on the performance of different algorithms can be evaluated. Both the average data rates from 1000 Monte Carlo simulations and the deterministic approximation are evaluated. For the sparsegroup-angle of arrival (AoA) case (i.e., N = 3) under low-to-medium PT conditions, an almost identical performance can be obtained by both the proposed JGPAPBF algorithm and the ABD-OGPA algorithm, with the former slightly outperforming the latter. In spite of the fact that the pre-beamforming optimization might not be necessarily performed for the sparse-group-AoA case (as concluded from the above-mentioned

IEEE Wireless Communications • October 2015

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 9

observation), an obvious gain in terms of data rate over the ABD-UGPA algorithm can still be attained by both the JGPAPBF and ABD-OGPA algorithms, showing that the group-power-allocation method can still be helpful in improving the BF performance. For the cases of N = {9, 15}, on the other hand, the proposed JGPAPBF algorithm is capable of significantly outperforming the ABD-OPGA algorithm, indicating that the proposed pre-beamformer-backhaul optimization is more imperative for the dense-groupAoA case. Furthermore, an evident performance loss in terms of SE can be met in the proposed JGPAPBF algorithm compared to the optimal algorithm due to the impact of CSI-feedback reduction and the employment of a suboptimal ZF beamformer in the former.

OPEN RESEARCH ISSUES Although LS-MIMO-based BF techniques have been widely studied for achieving capacity maximization as the number of AEs grows, several challenges are still observed in practical system design. Apart from capacity optimization, the LS-MIMO technology should also take into account other issues such as low-complexity matrix inversion, hardware-impairment compensation, and efficient transmit antenna selection so as to remedy the deficiencies of the existing solutions. In the following, a number of challenges as well as potential research directions associated with the LS-MIMO BF-based wireless backhaul are discussed. Statistical CSI acquisition for the JSDN scheme: In the JSDN scheme that requires statistical CSI (or the CCM) associated with BS/UE groups, the instantaneous CSI feedback burden can be significantly reduced by considering lowdimensional channel matrices of co-group BSs/UEs. Although the CCM of wireless backhaul can be pre-calculated according to fixed spatial sectors, some form of online estimation is still favorable to adapt to the time-variant environment. Therefore, low-rank CCM has to be exploited to reduce the computational complexity. However, in JSDN-based wireless backhaul, how to exploit/acquire statistical CSI of LSMIMO channels with a tolerable training burden remains to be investigated. Scheduling for LS-MIMO-based wireless backhaul in two-tier HetNets: In HetNets, the LS-MIMO-aided BSs might play the dual role of serving the macrocell UEs (i.e., acting as macrocell BSs) and providing a wireless backhaul for small cells (i.e., acting as donor BSs). Apart from the functionalities of power control and BF, appropriate scheduling strategies (e.g., in terms of spectrum 2 and time slots as well as power) have to be implemented for the smallcell BSs and/or UEs in the spatial domain at the macrocell BSs in order to coordinate transmissions from both the small-cell BSs and the macrocell UEs. In particular, in TDD-based twotier HetNets, scheduling systems for both tiers are coupled, requiring the scheduler at the macrocell BSs to effectively balance the spectrum resource scheduling of the above-mentioned two-tiers to achieve overall throughput maximization.

IEEE Wireless Communications • October 2015

The requirements of physical realizability and cost effectiveness: Although increasing the number of AEs in the LS-MIMO transceivers usually implies a better data rate and/or link reliability performance, an additional price must be paid to compensate for the increased hardware/software complexity. Furthermore, practical LS-MIMO arrays cannot be arbitrarily large due to the systems’ physical constraints. The mismatch between the theoretical results under ideal assumptions and the practical BF would motivate a tremendous new research effort to manufacture a large number of low-cost transceiver units instead of a small number of high-cost units for facilitating a cost-effective and practically implementable LS-MIMO BF system. The requirement for low power consumption: Since LS-MIMO BF usually requires complicated hardware in digital and RF/analog domains in order to support high data rate, either the complexity or the power dissipation/comsumption would be much higher than conventional smallscale MIMO techniques. It thus requires each antenna unit in LS-MIMO systems to use extremely low power while keeping the total dissipated/transmitted power constant as the array grows large. Coordinated BF in the presence of pilot contamination: Coordinated BF may provide a promising solution for addressing the problem of multiuser interference imposed by pilot contamination. The exploitation of the covariance matrices could lead to complete removal of pilot contamination effects in the large-array regime, provided that the covariance matrices must be optimized in a multicellular setup. A suitable coordination protocol of pilot assignment is thus highly required for minimizing the channel estimation errors and maintaining the user fairness simultaneously.

Apart from capacity optimization, the LS-MIMO technology should also take into account the other issues such as low-complexity matrix inversion, hardware-impairment compensation, and efficient transmit antenna selection so as to remedy the deficiencies of the existing solutions.

CONCLUSIONS In order to alleviate the severe situation in wireless channel capacity or user data rate that cannot be readily met without a considerable increase in the achievable SE, we demonstrate a number of significant benefits of LS-MIMO techniques, behind which the main driving force is the promise of substantial performance gains in terms of capacity, SE/EE, link reliability, and so on. Although the theoretical aspects of LSMIMO have shown very promising performance characteristics, the existing theoretical works are usually based on crucial assumptions about the ideal propagation conditions, which may not be sufficiently validated through practical measurements and modeling, let alone prototype design and/or practical implementations. Furthermore, several challenges, including hardware/software complexity reduction, low-power requirement in transceiver design, and performance erosion imposed by pilot contamination and so on have to be properly treated. In particular, aspects of CSI feedback reduction in BF-based wireless backhaul, BS/UE scheduling in HetNets, RFchain optimization, coordinated BF in the presence of pilot contamination, and complexity reduction in nonlinear algorithms should also be

2 For instance, in [16], an optimum decentralized spectrum allocation policy for two-tier networks is proposed in which the spectrum allocation is performed by maximizing the area SE while subject to a sensible QoS requirement.

9

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 10

Aspects of CSI-feedback reduction in BF based wireless backhaul, BS/UE scheduling in HetNets, RF-chain optimization, coordinated BF in the presence of pilot contamination, and complexity reduction in nonlinear algorithms, etc., should be specifically resolved in designing cost-effective LS-MIMO transceivers.

specifically resolved in designing cost-effective LS-MIMO transceivers.

ACKNOWLEDGMENTS This work was supported by the key project of the National Natural Science Foundation of China (No. 61431001), the 863 project No. 2014AA01A701, Program for New Century Excellent Talents in University (NECT-12-0774), the Open Research Fund of National Mobile Communications Research Laboratory Southeast University (No.2013D12), Fundamental Research Funds for the Central Universities, the Research Foundation of China Mobile, and the Foundation of Beijing Engineering and Technology Center for Convergence Networks and Ubiquitous Services.

REFERENCES [1] C. Xing, S. Ma, and Y.-C. Wu, “Robust Joint Design of Linear Relay Precoder and Destination Equalizer for Dual-Hop Amplify-and-Forward MIMO Relay Systems,” IEEE Trans. Signal Processing, vol. 58, no. 4, Apr. 2010, pp. 2273–83. [2] J. Hoydis, S. T. Brink, and M. Debbah, “Massive MIMO: How Many Antennas Do We Need?” Proc. 2011 Allerton Conf. Commun., Control, Comp., 2011, pp. 545–50. [3] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Energy and Spectral Efficiency of Very Large Multiuser MIMO Systems,” IEEE Trans. Commun., vol. 61, no. 4, Apr. 2013, pp. 1436–49. [4] T. L. Marzetta, “Noncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas,” IEEE Trans. Wireless Commun., vol. 9, no. 11, Nov. 2010, pp. 3590–3600. [5] Z. Zhang, J. Liu, and K. Long, “Low-Complexity Cell Search With Fast PSS Identification in LTE,” IEEE Trans. Vehic. Tech., vol. 61, no. 4, May 2012, pp. 1719–29. [6] E. G. Larsson, “MIMO Detection Methods: How They Work,” IEEE Signal Processing Mag., vol. 26, no. 3, May 2009, pp. 91–95. [7] J. W. Choi et al., “Low Complexity Detection and Precoding for Massive MIMO Systems,” IEEE Wireless Commun. and Networking Conf., Apr. 2013, pp. 2857–61. [8] E. Larsson et al., “Massive MIMO for Next Generation Wireless Systems,” IEEE Commun. Mag., vol. 52, no. 2, Feb. 2014, pp. 186–95. [9] C. W. Shepard, Argos: Practical Base Stations for LargeScale Beamforming, Master’s thesis, Rice Univ., Houston, TX, 2012; http://clay.rice.edu/pubs/MasterThesis.pdf [10] H. Suzuki et al., “Highly Spectrally Efficient Ngara Rural Wireless Broadband Access Demonstrator,” Proc. 2012 Int’l Symp. Commun. and Info. Technologies, Oct. 2012, pp. 914–19. [11] S. Hur et al., “Millimeter Wave Beamforming for Wireless Backhaul and Access in Small Cell Networks ,” IEEE Trans. Commun., vol. 61, no. 10, Oct. 2013, pp. 4391–4403. [12] X. Wang et al., “Joint Group Power Allocation and Prebeamforming for Joint Spatial Division and Multiplexing in Multiuser Massive MIMO Systems,” Proc. IEEE Int’l. Conf. Acoustics, Speech, and Signal Processing, Brisbane, Australia, Apr. 2015. [13] S. Wagner et al., “Large System Analysis of Linear Precoding in Correlated Miso Broadcast Channels under Limited Feedback,” IEEE Trans. Info. Theory, vol. 58, no. 7, 2012, pp. 4509–37. [14] M. Schubert and H. Boche, “Solution of the Multiuser Downlink Beamforming Problem with Individual SINR Constraints,” IEEE Trans. Vehic. Tech., vol. 53, no. 1, Jan. 2004, pp. 18–28. [15] A. Adhikary et al., “Joint Spatial Division and Multiplexing — The Large-Scale Array Regime,” IEEE Trans. Info. Theory, vol. 59, no. 10, Oct. 2013, pp. 6441–63. [16] V. Chandrasekhar and J. Andrews, “Spectrum Allocation in Tiered Cellular Networks,” IEEE Trans. Commun., vol. 57, no. 10, Oct. 2009, pp. 3059–68.

BIOGRAPHIES ZHONGSHAN ZHANG received B.E. and M.S. degrees in computer science from Beijing University of Posts and Telecommunications (BUPT) in 1998 and 2001, respectively, and received his Ph.D. degree in electrical engineering in 2004

10

from BUPT. In August 2004 he joined DoCoMo Beijing Laboratories as an associate researcher, and was promoted to researcher in December 2005. In February 2006, he joined the University of Alberta, Edmonton, Canada, as a postdoctoral fellow. In April 2009, he joined the Department of Research and Innovation (R&I), Alcatel-Lucent, Shanghai, as a research scientist. From August 2010 to July 2011, he worked at NEC China Laboratories as a senior researcher. He has served or is serving as a Guest Editor and/or an editor for several technical journals, such as IEEE Communications Magazine and KSII Transactions on Internet and Information Systems. He is currently a professor of the School of Computer and Communication Engineering at the University of Science and Technology Beijing (USTB). His main research interests include statistical signal processing, self-organized networking, cognitive radio, and cooperative communications. XIYUAN WANG (S’10, M’13) received a Ph.D. degree in control sciences and engineering from the Department of Automation, Tsinghua University, Beijing, China, in 2013. He is currently a postdoctral research at USTB. His research interests include information theory, communications theory, array signal processing, and signal processing for wireless communications. K EPING L ONG [SM] received his M.S. and Ph.D. degrees at UESTC in 1995 and 1998, respectively. From September 1998 to August 2000, he worked as a postdoctoral research fellow at the National Laboratory of Switching Technology and Telecommunication Networks at BUPT. From September 2000 to June 2001, he worked as an associate professor at BUPT. From July 2001 to November 2002, he was a research fellow at the ARC Special Research Centre for Ultra Broadband Information Networks (CUBIN) at the University of Melbourne, Australia. He is now a professor and dean at th School of Computer & Communication Engineering, USTB. He is a member of the Editorial Committees of Sciences in China Series F and China Communications. He has also been a TPC and ISC member for COIN ’03/04/05/06/07/08/09/10, IEEE IWCN ’10, ICON ’04/06, APOC ’04/06/08, Co-Chair of the Organization Committee for IWCMC ’06, TPC Chair of COIN ’05/08, and TPC Co-Chair of COIN ’08/10, He was awarded by the National Science Fund for Distinguished Young Scholars of China in 2007 and selected as the Chang Jiang Scholars Program Professor of China in 2008. His research interests are optical Internet technology, new generation network technology, wireless information networks, value-added services, and secure technology of networks. He has published over 200 papers, 20 keynote speeches, and invited talks at international and local conferences. ATHANASIOS V. VASILAKOS is currently a professor at the University of Western Macedonia, Greece. He has authored or co-authored over 200 technical papers in major international journals and conferences. He is author/co-author of five books and 20 book chapters in the area of communications. He has served as General Chair/Technical Program Committee Chair for many international conferences. He has served or is serving as an Editor and/or Guest Editor for many technical journals, such as IEEE Transactions on Network and Service Management, IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions on Information Technology in Biomedicine, IEEE Transactions on Computers, ACM Transactions on Autonomous and Adaptive Systems, IEEE JSAC Special Issues in May 2009, January 2011, and March 2011, IEEE Communications Magazine, ACM/Springer Wireless Networks, and ACM/Springer Mobile Networks and Applications. He is founding Editor-in-Chief of the International Journal of Adaptive and Autonomous Communications Systems and the International Journal of Arts and Technology. He is General Chair of the Council of Computing of the European Alliances for Innovation. L AJOS H ANZO [F] (http://wwwmobile. ecs.soton.ac.uk), FREng, FIET, Fellow of EURASIP, D.Sc., received his degree in electronics in 1976 and his doctorate in 1983. In 2009 he was awarded the Doctor Honoris Causa by the Technical University of Budapest. During his 37-year career in telecommunications he has held various research and academic posts in Hungary, Germany, and the United Kingdom. Since 1986 he has been with the School of Electronics and Computer Science, University of Southampton, United Kingdom, where he holds the chair in telecommunications. He has successfully supervised 80+ PhD students, coauthored 20 John Wiley/IEEE Press books on mobile radio

IEEE Wireless Communications • October 2015

HANZO_LAYOUT_Author Layout 10/7/15 3:35 PM Page 11

communications totaling in excess of 10 000 pages, published 1400+ research entries at IEEE Xplore, acted as both TPC and General Chair of IEEE conferences, and presented keynote lectures, and has been awarded a number of distinctions. Currently he is directing a 100-strong academic research team working on a range of research projects in the field of wireless multimedia communications sponsored by industry, the Engineering and Physical Sciences Research Council (EPSRC) UK, the European Research Councils Advanced Fellow Grant, and the Royal Society’s Wolfson Research Merit Award. He is an enthusiastic supporter of industrial and academic liaison, and offers a range of industrial courses. He is also a Governor of the IEEE VTS. During 2008–2012 he was Editor-in-Chief of IEEE Press and a Chaired Professor at Tsinghua University, Beijing. His research is funded by the European Research Council’s Senior Research Fellow Grant. For further information on research in progress and associated publications please refer to http://wwwmobile. ecs.soton.ac.uk. He has 19,000+ citations.

IEEE Wireless Communications • October 2015

11