A rchitecture for N ext G eneration W ireless N etworks :

S calability , F lexibilty , and I nteroperability

Scalable and Flexible

Massive MIMO Precoding for 5G H-CRAN
Na Chen, Bo Rong, Xinran Zhang, and Michel Kadoch

Abstract are also invested with much expectation for high

spectral and spatial resources utilization, regarded
Scalability and flexibility are widely consid- as promising solutions for 5G HetNets. With the
ered as two major design goals for 5G networks. ultra-high frequency of waveform and large scale
Aiming at these goals, this article first identifies of antennas, they are able to provide significant
a promising architecture based on the hetero- improvements in both the quality of service (QoS)
geneous cloud radio access network (H-CRAN), and efficiency of the system [4].
reviews the challenges in MIMO precoding for In recent years, Long Term Evolution (LTE) and
H-CRAN, and then proposes a scalable and flexi- LTE-Advanced (LTE-A) have been recognized as
ble massive MIMO precoding scheme by exploit- mature technologies and gradually pushed into
ing the null-space of user signals. Specifically, the commercial use all over the world. In LTE/LTE-A
system can accomplish effective radio resource networks, the main challenge for RRM is the time
management and flexible spatial coordination by and frequency domain management, such as fre-
distinguishing the intended and victim users CSI, quency reuse [5] and cooperative communica-
and avoid the interference by precoding within tion [6]. In 5G HetNets, however, the system has
the null-space for the CSI of victim users. Simu- to be capable of dealing with the overwhelming
lation results indicate that the proposed scheme densities of heterogeneous devices and network
is capable to effectively alleviate the interference infrastructure nodes, as well as supporting aston-
to victim users and support high QoS as well as ishing numbers of antennas in massive MIMO
spectral efficiency. equipped nodes.
Conventional MIMO precoding schemes usu-
Introduction ally perform in the baseband domain and utilize
Aiming at designing a mobile network architec- a fully digital precoder, modifying both the ampli-
ture that satisfies the requirements of the fifth-gen- tude and phase of transmitted signals. Although a
eration (5G) system, both academia and industry fully digital precoding scheme usually brings opti-
have made many achievements to support sys- mal performance, enormous hardware complexity
tems with higher density, mobility, diversity, and and energy consumption become serious prob-
complexity. The 5G heterogeneous network (Het- lems because of the expensive RF chain for each
Net) has become a popular choice to dominate antenna. This becomes a serious problem due
future telecommunication implementations [1]. to the huge amount of antennas at the massive
Although HetNets can provide seamless cov- MIMO transmitter side. Fortunately, the hybrid
erage and high performance, commercial appli- precoding scheme that combines digital precod-
cation is challenged by the large volume of data ing and analog precoding together has been pro-
traffic and the limitation of resources caused by posed to significantly decrease the number of
high density of small cells [2]. Meanwhile, the RF chains, which is considered to be an essential
developing scale of HetNets induces an explosive technique for realistic 5G MIMO systems [7].
growing volume of mobile data traffic which leads In this article, we focus on 5G MIMO precod-
to critical inter-tier interference. ing schemes as well as how H-CRAN benefits the
Na Chen and Xinran Zhang To provide sufficient data rates with high ener- system scalability and flexibility, as well as the perfor-
are with Beijing University gy efficiency (EE) and spectral efficiency (SE), mance such as EE and SE. We address an H-CRAN-
of Posts and Telecommuni-
cations.
many technologies are incorporated into HetNets, based network architecture that is compatible with
including cloud computing, coordinated transmis- the current Third Generation Partnership Project
Bo Rong is with the sion, radio resource management (RRM), network (3GPP) LTE-A system and also consistent with 5G
Communications Research virtualization, and self-organizing networks (SONs) concepts. In the proposed H-CRAN architecture,
Centre Canada.
[3]. The heterogeneous cloud radio access net- the baseband unit (BBU) pool takes responsibility
Michel Kadoch is with the work (H-CRAN) is especially highlighted due to its for integrating the large amount of spatial domain
Universit du Qubec. inherent advantage in integrating and processing information and performing null-space calculation
information that is of definite scalability and flex- concerning massive MIMO coordination. Then the
Digital Object Identifier: ibility. Millimeter-wave (mmWave) and massive null-space information is integrated to enable the
10.1109/MWC.2017.1600139WC multiple-input multiple-output (MIMO) techniques whole network coordination to achieve better per-

46 1536-1284/17/$25.00 2017 IEEE IEEE Wireless Communications February 2017

 Internet

Sync/control/transport/baseband
Sync/Control/Transport/
Sync
/Control
/Transport
/ ...
BBU pool

RRH RRH
RRH eNB
RRH eNB

RRH RRH
Macro cell
eNB
IoT RRH
RRH Small cell: RRH 2
RRH 1 heterogeneous
Small cell: ultra-density devices
UEs

Figure 1. A typical H-CRAN system.

formance. This novel architecture implements coor- requirements make spatial domain management a
dinated RRM and eases the operation pressure of new problem to solve, which is quite suitable for
access points significantly. the application of the cloud RAN (C-RAN).
The rest of this article is organized as follows. IBM first proposed the concept of C-RAN for a
We first address the 5G H-CRAN architecture distributed wireless communication system. Aim-
supporting RRM. We present the 5G MIMO pre- ing at balancing the load between different base
coding problem. Then we propose our two-tier stations, the centralized BBU pool clustered with
interference-free hybrid precoding scheme, fol- multiple BBUs performs baseband (physical/medi-
lowed by the performance evaluation. In the end, um access control, PHY/MAC) processing. For
we conclude the article. inter-cluster communication, the X2 interface is
usually referred to as X2+.
Architecture of H-CRAN A typical C-RAN architecture mainly consists
In this section, we overview H-CRAN systems and of three different parts [2]:
introduce the massive MIMO assisted H-CRAN 1. The remote radio heads (RRHs), distributed radio
architecture. units with antennas located at a remote site
2. The BBU pool composed of high-performance
Derivation of 5G H-CRAN programmable equipment and real-time net-
In 5G systems, there will be a tremendous work virtualization and RRM technologies
increase in the requirements of traffic volume for 3. The network for transmission that links the
the following reasons [8]: RRHs and the BBU pool with high bandwidth
1. There will be large volume and high diversity of and low latency
heterogeneous devices. In 5G H-CRAN systems, the BBU pool is
2. The devices will boast high performance levels. involved in collecting and managing spatial infor-
3 More services will be accessed for high diversi- mation from a variety of access points, as shown
ty, requiring wide bands. in Fig. 1. The BBU pool takes charge of most pro-
4. Application of cloud will result in more trans- cessing and computing procedures, such as RRM.
mission between terminal devices and cloud In contrast, the RRHs only perform radio function-
devices. alities, such as digital processing and power ampli-
Along with the increasing requirements, 5G fication. To maintain backward compatibility with
is believed to greatly depend on centralized existing LTE/LTE-A systems, together with support
high-performance processors and distributed gen- for seamless coverage of the serviced area, eNBs
eralized equipment to implement the scalability perform a critical role, while RRHs are for high
and flexibility of the network. capacity in special areas. The eNB can harmonize
Considering the volume and diversity features multiple radio networks to operate well with each
of accessed devices, the HetNet has been regard- other, and signals for system control can easily be
ed as one of the most exciting solutions to the delivered throughout the network. With these out-
impending mobile data traffic crunch in the con- standing advantages, H-CRANs are able to com-
text of existing LTE-A and 5G. It is constructed of bine HetNets with C-RANs and benefit from them.
different network tiers and cell sizes by deploying
nodes of different transmission powers, such as Massive MIMO Assisted
macro, micro, pico, and femto nodes. The HetNet
becomes an inevitable trend for future develop- H-CRAN Architecture
ment of information networks since it makes full In H-CRANs equipped with massive MIMO, eNBs
use of the complementary characteristics of differ- and RRHs consisting of several hundred elements
ent network tiers. Nevertheless, different tiers net- simultaneously serve mobile terminals within
working access technologies and various service the specific time-frequency resources. Massive

IEEE Wireless Communications February 2017 47

 based on specific standardized interfaces. This
Application Multimedia Real-time mechanism extremely limits the openness of the
Management Security
layer communication system. In 5G systems, network elements are
expected to be physically generalized and apply
programmable customized function so that the
RRH intended users RRH victim users eNB users network architecture can have high scalability and
Control Mobility
management
Null-space
analysis
flexibility. Meanwhile, it further keeps the opera-
RRM
layer tors capital expenditures (CAPEX) and operating
User
grouping
SVD expenses (OPEX) reasonable.
Employing H-CRAN technology can further
enhance the scalability and flexibility of the sys-
tem since the BBU pool exploits complete knowl-
Infrastructure edge of the network. Furthermore, all forwarding
layer equipment can be managed flexibly via the con-
BBU pool trol layer. RRHs can be simple in structure, while
eNB complex computing is handled by the BBU pool
and realized by massive MIMO. To optimize
RRH1 the performance and efficiency, system parame-
Internet ters and procedures should accordingly be well
Users RRH2 determined to support the service requirements.
Additionally, new network functions can be real-
ized and deployed in programmable equipment,
Figure 2. Proposed 5G system architecture based on H-CRAN controlling mas- encouraging rapid network innovation.
sive MIMO precoding procedure.
On Massive MIMO Precoding
MIMO further enables broadband transmission, In this section, we focus on the developing
which requires sufficient SE, EE, and robustness. MIMO precoding schemes. First, we analyze the
As a large number of antennas are involved, the challenges, and then summarize some outstand-
energy consumption becomes higher. To make ing existing solutions.
better use of the limited energy, well designed
signal processing algorithms such as precoding/ Challenges and Relevant Considerations
beamforming schemes become critical to the sys- We state current challenges and our consider-
tem. Meanwhile, massive MIMO antenna arrays ations below.
engender huge amounts of spatial domain infor- The Requirements of Future Networks: Mas-
mation in real time, which significantly increas- sive MIMO is a promising technology for future
es the computational complexity and may bring 5G networks, which are expected to bring high
coordination problems. capacity, EE, security, and robustness for the wire-
To deal with the specified requirements, less broadband access network with high density
we here address a massive MIMO assisted 5G and high mobility. Relying on spatial multiplex-
H-CRAN to better manage the large volume of ing, the small and active antenna units provided
spatial information and conduct massive MIMO should be supported by a set of effective signal
coordination. In this architecture, eNBs and processing procedures with good knowledge of
RRHs collect the information of mobile termi- channel state information (CSI).
nals and report to the BBU pool. The BBU pool The Usage of System Resources: A massive
then performs the information processing such as MIMO system applies antenna arrays with tens or
null-space analysis, then sends the processed pre- even hundreds of antennas simultaneously work-
coding instructions back to eNB/RRHs to perform ing in the same time-frequency resource blocks
appropriate coordination concerning the global (RBs). Future 5G networks equipped with massive
knowledge of the system spatial information. MIMO and applying mmWave can achieve super
As Fig. 2 shows, the infrastructure layer con- wide bandwidth with frequency ranges from 26.5
sists of all the heterogeneous physical equipment to 300 GHz. With the carrier at high frequency,
of the system, including the BBU processors, mac- massive MIMO has a much narrower antenna
rocell eNBs, RRHs, and user devices. The control beam size compared to that in conventional com-
layer implements the network functions based on munication; with proper deployment, it can aim at
the physical equipment in the infrastructure layer targets more precisely and effectively.
and supports the application layer. In our scenario, Balance between Users: The two-tier structure
it takes over the processing and computing tasks, of the network provides a solution to the large vol-
and enables our null-space-based MIMO precoding ume and high density of users accessing the future
scheme. The application layer contains the support- network. However, how to reduce the interference
ed services, including multimedia, real-time com- between users is a significant problem because of
munication, and other services provided by the 5G the requirement for much higher device density
MIMO system. In this architecture, the H-CRAN and volume in future networks. Although the appli-
takes over the heavy operation from the eNB and cation of small cells has reduced the pressure of
RRHs to promote the efficiency and performance to a single eNB, it is still challenging to serve a large
implement a scalable and flexible system. number of users with sufficient QoS.

Benefits of Scalability and Flexibility Analysis of Existing Solutions

For existing LTE/LTE-A systems, the network To deal with the above-mentioned problem, many
elements are specialized, and the interaction previous works have made significant efforts to
between different network elements has to be meet the requirements of future mobile commu-

48 IEEE Wireless Communications February 2017

nication systems. We summarize in Table 1 the Conventional precoding Our work addresses the
existing outstanding solutions. With appropriate Hybrid precoding
processing schemes, 10 or more times capacity Linear Nonlinear
interference problem
can be achieved by massive MIMO compared to and designs a null-space
conventional MIMO with fewer antennas. Well Soft antenna selection [11]
MMSE
designed precoding schemes can reduce the
MRT
THP [10] WSMSE considered beamforming [12] based hybrid precoding
energy consumed by a large number of antennas, MMSE THP Hybrid BF with reference signal
the wavefronts of which are focused with extreme
ZF
SO THP design [13]
scheme for 5G H-CRAN
SO
sharpness to the target terminal. This precise JSDM [14] systems. The proposed
beamforming especially relies on the accurate CSI
Table 1. Existing outstanding precoding schemes. scheme considers a
acquisition of all the spatial multiplexing channels.
In practice, most of the systems tend to apply hybrid structure that
time-division duplex (TDD) calibration to simplify [14] proposed the joint spatial division and multi-
channel measurements and save uplink resourc- plexing (JSDM) approach to exploit a scheme that relies on the null-space
es. However, in certain cases, some backhaul allows for a large number of antennas at the base
systems may perform frequency-division duplex station requiring reduced-dimensional CSI at the
of CSI of the victim
(FDD) operation for their schemes. transmitter (CSIT). (unintended) users.
For conventional MIMO digital precoding With appropriate precoding schemes, the
schemes, there has been continuous work, espe- wavefronts are directed to the intended users.
cially on the classic linear precoding schemes, However, interference may occur to unintend-
such as the minimum mean square error (MMSE), ed users due to the random distribution of users,
maximum ratio transmission (MRT), and zero forc- complex environments, and unpredictable user
ing (ZF) algorithms. On the other hand, multiple motions. Our work addresses the interference
nonlinear precoding schemes are proposed to fur- problem and designs a null-space-based hybrid
ther enhance the system capacity. The nonlinear precoding scheme for 5G H-CRAN systems. The
Tomlinson-Harashima precoding (THP) algorithm proposed scheme considers a hybrid structure
can achieve better performance than MMSE at that relies on the null-space of CSI of the victim
high signal-to-noise ratio (SNR), and the authors (unintended) users, and finally results in a perfor-
of [10] further combined the linear successive mance that is comparable to the optimal digital
optimization (SO) with THP that provides better precoding schemes.
capacity than MMSE THP precoding and other
similar precoding schemes.
Conventional analog schemes generally focus
Proposed Two-Tier Interference-Free
on phase shifters, which is far less complex than Hybrid Precoding Scheme
digital ones. However, they usually lead to poor In this section, we discuss our proposed scheme
performance since the magnitude of the ele- to implement effective interference-free commu-
ments is out of control. To further improve the nication in 5G H-CRAN. Here we develop a null-
system performance and adapt to the extreme- space-based hybrid precoding scheme to mitigate
ly large amount of antennas, hybrid precoding inter-tier interference.
schemes with active antennas are suggested [9], We consider a downlink 5G H-CRAN system
which generally controls the signal phase on each with one macrocell, including 1 eNB, J RRHs, and
antenna via a network of analog phase shifters. K single-antenna users. The eNB and RRHs are all
Since then, a hybrid precoding scheme has been equipped with massive MIMO antenna elements,
developed that performs especially well in the containing NeNB and NRRH uniform transmit anten-
conditions of large-scale antenna arrays, more RF nas, respectively. Let K j denote the number of
chains than the number of data streams, and cor- users of the jth RRH. Define C j as the number
related channel matrices [11]. of users interfered by the jth RRH and served by
A hybrid precoding structure is composed other nodes. Kj and Cj should meet the constraint
of analog and digital domains. First, user data Kj + Cj NRRH. Thus, Hj C(Kj+Cj)NRRH denotes
streams are precoded by the baseband digital the stationary channel between a certain RRH j
processing. Then the streams are transferred by and the users it affects. Specifically, H 0 denotes
the corresponding RF chains and mapped into the stationary channel between the macro eNB
each antenna element by the analog phase pro- and the users it serves and interferes, and H0,k
cessing (RF processing). Notably, baseband digital denotes the channel between the macro eNB and
processing can make both amplitude and phase the kth user.
modifications, but analog phase processing only Based on the proposed system architecture
gets the phase changes with variable phase shift- mentioned above, Fig. 3 illustrates our processing
ers and combiners [6]. In the hybrid precoding diagram emphasizing the acquisition of CSI for
process, the number of RF chains is lower than RRHs and the generation of a null-space-based
that of antennas, which reduces the complexity hybrid precoding matrix. We also provide the
and power consumption in massive MIMO sys- details in the following.
tems.
The authors of [11] focused on maximizing the CSI Acquisition Method
SE as the hybrid structure was used at both the In the conventional network, a base station is only
transmitter and the receiver sides. Reference [12] able to collect the local CSI through the backhaul
considered the weighted sum mean square error of its own served users. Thus, it cannot access the
(WSMSE) in their design by applying compres- CSI of the nearby victim users. Fortunately, the
sive sensing (CS) theory. By investigating optimal hub-spoke structure of the H-CRAN enables the
designs of hybrid beamforming structures, [13] CSI of the users connected to all RRHs to be col-
improved the SE and EE of the system. Reference lected and disseminated through the backhaul link.

IEEE Wireless Communications February 2017 49

In the conventional
network, a base station BBU CSI
eNB users
is only able to collect Null-space RRH 1
the local CSI through Users
Victim users Victim users
the backhaul of its own for RRH1 for RRH1

served users. Thus it can Victim users RRH 2

for RRH2
not access to the CSI of Users
Victim users
the nearby victim users Precoding ... for RRH2
Channel
interfered. Fortunately, info info ...
RX
the hub-spoke structure eNB users

of the H-CRAN enables TX eNB RRH1

Precoding Users
the CSI of the users Source
RF chain Data

...
bits DL
Victim users
Channel
connected to all RRHs Null-space for RRH1
CSI Backhaul
to be collected and dis- RRH2
RRH 1 RRH 2
Users
seminated through the
...

... ... Victim users

backhaul link. for RRH2

...

Figure 3. Block diagram of the proposed scheme.

In our proposed CSI acquisition method,

A a,b = 1 N eNB .
the eNB and RRHs collect and report the CSI
to the BBU pool. More exactly, the eNB sends To be concise, we mainly discuss the RRH in the
the BBU pool the compressed CSI for both following parts; the process is similar for the eNB.
serviced users and victim users of each RRH. We perform the RF precoding according to
In this way, the BBU pool can gather all the the phase of the element of RF precoding matrix
CSI and do the proposed precoding operation. and the conjugate transpose of the stationary
The BBU pool controller performs the block-di- channel.
agonal (BD) algorithm [15] and generates the Consider a user k i that is served by RRH i.
null-space vector of the victim users CSI; that Then the RRH analyzes the received signal of the
is, the vector that is always orthogonal to the user ki as a sum of the transmitted signal vector
CSI matrix so that the interference can be from the jth RRH, surrounding interference sourc-
avoided. Then it computes precoding matrix es (nearby RRHs or the eNB) and the zero-mean
based on the null-space vector by the hybrid additive white Gaussian noise.
precoding algorithm proposed in the next sub- To avoid interference to victim users, the
section. Finally, the BBU pool sends the results hybrid precoding matrix should be orthogonal
(i.e., the precoding information) to the eNB with the channel matrix of victim users, that is,
and RRHs. In this way, the eNB and RRHs can within the null-space of the channel matrix of
get rid of most computing tasks, and imple- victim users. Thus, we have H ~ j,vA jD j = 0, where
ment the precoding process with the received ~ j,v CCjNRRH denotes the submatrix of Hj that
H
information and their RF module. includes the channel vectors of the victim users
This framework is more helpful when the RRHs of RRH j. By performing singular value decompo-
are too simplified to handle the computational sition (SVD), we can decompose the victim users
work. If the eNB is powerful enough, it can help channel matrix by H ~ j,v = U ~ j,vV
~ j,v ~ j,vH and get its
itself and take over some processing procedure to null-space V ~ (0)
j,v , the C j + 1 to N RRH columns of Vj,v.
~
reduce the transmission pressure. Then we define the projection matrix Mj based on
the null-space of victim users by Mj = V ~ (0) ~ (0) H
j,v = (V j,v) ,
Hybrid Precoding Based on Null-Space and the analog precoding matrix after projection
In the proposed scheme, the hybrid precoding is can be expressed as Fj = MjAj; with the transmit
divided among baseband and RF processing, with power defined, we further normalize Fj to satisfy
the precoding matrix of RRH j denoted by Dj the total transmit power constraint.
C KjKj and A j C NRRHKj, with each entry of A j The signal-to-interference-plus-noise ratio
normalized to satisfy (SINR) of the kth user served by RRH j is given as
Pj 2
A a,b = 1 N RRH . H Fd
K j j,k j j,k
SINR j,k = Kj
Here |Aa,b| denotes any given (a, b)th element of Pj 2 (1)
Aj. Specifically, D0 and A0 stand for the baseband 2j,k + H j,k Fj d j,r
r=1 K j
and RF precoding matrix of eNB, and A0 CNeN-
rk
BK0 with normalized entries

50 IEEE Wireless Communications February 2017

 25 20
Hybrid precoding 18 Hybrid precoding
ZF precoding ZF precoding
MRT precoding MRT precoding
20 16
Spectral efficiency (b/s/Hz)

Spectral efficiency (b/s/Hz)

14
15 12
10
10 8
6
5 4
2
0 0
-20 -15 -10 -5 0 5 10 15 20 25 30 -20 -15 -10 -5 0 5 10 15 20 25 30
SNR (dB) SNR (dB)

Figure 4. Performance of the proposed scheme with different numbers of users: a) SE/16 users; b) SE/32 users.

where P j as the transmit power at the jth RRH, the 5G H-CRAN scenario. The results show that
and dj,r denotes the rth column of Dj. the proposed scheme achieves satisfying system
Thus, the SINR should meet each users QoS throughput and efficiency.
requirement, that is, B log (1 + SINRj,k) Rj,k (j = In Figs. 4a4d, we present the simulation results
1, 2, , J;k = 1, 2, , Kj), where Rj,k is the data rate to compare the SE of the proposed hybrid precod-
demanded by user k, and B is the total bandwidth. ing scheme with the conventional ZF and MRT pre-
Then we can solve the F j,k to get the optimum coding schemes at different SNR with NeNB = NRRH
precoding matrix that fits the users requirement. = 64 transmitting antennas and randomly distribut-
ed users. As we can see in Fig. 4a, serving 16 users
Main Improvements in the macrocell, the SE of the proposed scheme is
Here we summarize the improvements brought lower than conventional ZF precoding and is high-
to the system by the application of the proposed er than MRT precoding at higher SNR. When the
H-CRAN architecture and hybrid precoding scheme. number of users increases to 32 in Fig. 4b, the per-
Centralized CSI Acquisition: For conventional formance between the proposed hybrid precoding
HetNets, the access point only knows the CSI of and ZF becomes similar. Also, the advantage of
its own subscribed users, but the CSI is not known MRT precoding becomes more obvious. The sim-
by any other node that may be interfering with it. ulation results prove that the proposed hybrid pre-
Without specific knowledge of the CSI, it would coding scheme can provide sufficient performance
be difficult to avoid the interference; thus, the sys- for a practical 5G H-CRAN system.
tem performance will be limited. For an H-CRAN
system, it is easier to collect all the channel infor- Acknowledgment
mation and do centralized computing with a BBU This work is supported in part by the Nation-
pool. And the information collection is necessary al Natural Science Foundation of China (NSFC)
for the oncoming precoding procedure, and may under Grant No. 61471066, and a scholarship
be utilized for further applications including data from the China Scholarship Council (CSC) under
mining, network security, and so on. Note that Grant CSC No. 201506470023.
in practice the traffic between the RRHs and the
BBU pool may be of large volume. Hence, it has Conclusions
to be especially guaranteed to sufficiently transmit In this work, we present the flexible and scalable
the channel and precoding information. 5G H-CRAN architecture and focus on RRM
Null-Space-Based Hybrid Precoding: Com- and an interference mitigation massive MIMO
pared to conventional H-CRAN systems, trans- precoding scheme. Furthermore, we propose a
mission nodes in a 5G H-CRAN is expected to null-space-based hybrid precoding scheme with
be equipped with antennas on a large scale. To a flexible CSI acquisition process. By performing
satisfy the users requirements and further reduce precoding within the null-space of the neighbor-
the power consumption of RRHs, hybrid precod- ing victim users, the proposed scheme counter-
ing can satisfy both parties. Basically, the hybrid acted the interference impact that could result in
precoding scheme can reduce the number of RF degradation of performance.
chains applied in the transmitter. We particular- Performance evaluation shows that the pro-
ly consider the null-space for the victim users of posed design could acquire higher system
each RRH. With accurate knowledge of CSI, the throughput with a large amount of transmitting
proposed scheme can avoid the interference to antennas and was capable of sufficient SE com-
the victim users and achieve better QoS. pared to other outstanding precoding schemes.

Performance Evaluation References

In this section, we present the performance evalu- [1] S. Sun, Y. Ju, and Y. Yamao, Overlay Cognitive Radio
ation based on simulations applying the proposed OFDM System for 4G Cellular Networks, IEEE Wireless
null-space-based hybrid precoding scheme in Commun., vol. 20, no. 2, Apr. 2013, pp. 6873.

IEEE Wireless Communications February 2017 51

 [2] P. Wang et al.,QoS-Aware Cell Association in 5G Heteroge-
The hybrid precoding neous Networks with Massive MIMO, IEEE Commun. Mag.,
Biographies
vol. 53 , no. 11, Nov. 2015, pp. 14247. Na Chen [M13] ([email protected]) received her B.S. degree in
scheme can reduce the [3] N. Chen et al., Self-Organizing Scheme Based on NFV and 2011 from Beijing University of Posts and Telecommunications
SDN Architecture for Future Heterogeneous Networks, (BUPT), China. She is currently a Ph.D. candidate with BUPT and a
number of RF chains Mobile Networks and Applications, vol. 20, no. 4, Aug. visiting scholar at the University of Texas at Arlington (UTA). She has
2015, pp. 46672. authored or co-authored 20 academic papers, including 4 journal
applied in the transmit- [4] E. Larsson et al., Massive MIMO for Next Generation Wire- papers and 16 conference papers, including GLOBECOM, VTC,
less Systems, IEEE Commun. Mag., vol. 52, no. 2, Feb. 2014, and others. She also edited a conference proceedings. Her research
ter. And we especially pp. 18695. interests include signal processing, wireless communications, and
[5] S. Sun, B. Rong, and Y. Qian, Artificial Frequency Selective multimedia communication. Her current work focuses on massive
consider the null-space Channel for Covert CDD-OFDM Transmission, J. Security MIMO, green heterogeneous wireless networks, and self-organizing
and Commun. Networks, Sept. 2014. networks. She has been an ACM member since 2013.
for the victim users of [6] Q. Li et al., Cooperative Communications for Wireless Net-
works: Techniques and Applications in LTE-Advanced Sys- Bo Rong [M07] ([email protected]) is a research scientist at
each RRH. With the tems, IEEE Trans. Wireless Commun., vol. 19, no. 2, Apr. Communications Research Centre Canada, Ottawa. He is also
2012, pp. 2229. an adjunct professor at Ecole de Technologie Superieure (ETS),
accurate knowledge [7] A. Alkhateeb et al., MIMO Precoding and Combining Solu- Universit du Qubec, Canada. He has authored or co-authored
tions for Millimeter-Wave Systems, IEEE Commun. Mag., over 100 technical papers in major journals and conferences, as
of CSI, the proposed vol. 52, no. 12, Dec. 2014, pp. 12231. well as one book and two book chapters in the areas of wireless
[8] A. Checko, H. L. Christiansen, and Y. Yan, Cloud RAN for networking and communications. Many of these publications
scheme can avoid the Mobile Networks A Technology Overview, IEEE Com- have significance to the research community and industry. HIs
mun. Surveys & Tutorials, vol. 17, no. 1, 1st qtr. 2015, pp. research interests include modeling, simulation, and performance
interference to the 40526. analysis of next-generation wireless networks. He is a member of
the IEEE Communications Society and IEEE Broadcasting Society.
victim users and achieve [9] X. Zhang, A. F. Molisch, and S.-Y. Kung, Variable-Phase-
Shift-Based RF-Baseband Codesign for MIMO Antenna He serves as an Associate Editor for IEEE Communications Letters
Selection, IEEE Trans. Signal Proc., vol. 53, no. 11, Nov. and has served as a Guest Editor of Feature Topics and Special
better QoS. 2005, pp. 40914103. Issues of IEEE Communicaitons Magazine, IEEE Wireless Communi-
[10] V. Stankovic and M. Haardt, Successive Optimization Tom- cations, and the IEEE Internet of Things Journal.
linson-Harashima Precoding (SO THP) for Multi-User MIMO
Systems, 2005 Proc. IEEE Intl. Conf. Acoustics, Speech, and Xinran Zhang [S16] ([email protected]) received his
Signal Processing, vol. 3, Mar. 2005, pp. iii/1117iii/1120. B.S. and M.S. degrees from BUPT in 2010 and 2013, respective-
[11] F. Sohrabi and W. Yu, Hybrid Digital and Analog Beam- ly. He has been an assistant professor at the Beijing Electronic
forming Design for Large-Scale Antenna Arrays, IEEE J. Sel. Science and Technology Institute, Beijing, China, since 2013.
Topics in Signal Processing, no. 99, Apr. 2016. Since 2015 he has been working toward his Ph.D. degree at
[12] T. E. Bogale and L. B. Le, Beamforming for Multiuser BUPT. His research interests include wireless signal processing
Massive MIMO Systems: Digital versus Hybrid Analog-Dig- and network analysis for next generation wireless networks.
ital, 2014 IEEE GLOBECOM, Austin, TX, Dec. 2014, pp.
406671. Michel Kadoch ([email protected]) received his B.Eng.
[13] S. Han et al., Large-Scale Antenna Systems with Hybrid from Sir George Williams University in 1971, his M.Eng. from
Analog and Digital Beamforming for Millimeter Wave 5G, Carleton University in 1974, his M.B.A. from McGill University
IEEE Commun. Mag., vol. 53, no. 1, Jan. 2015, pp. 18694. in 1983, and his Ph.D from Concordia University in 1991. He
[14] A. Adhikary et al., Joint Spatial Division and Multiplexing: is a full professor at ETS, Universit du Qubec. He is also an
The large-Scale Array Regime, IEEE Trans. Info. Theory, vol. adjunct professor at Concordia University, Montreal, Qubec.
59, no. 10, Oct. 2013, pp. 644163. He has served as an editor for a number of journals and confer-
[15] N. Chen et al., SDN Controlled mmWave Massive MIMO ences, as well as for NSERC grants. His current research inter-
Hybrid Precoding for 5G Heterogeneous Mobile Systems, ests include cross-layer design and radio resource management
Mobile Information Systems, vol. 2016, 2016. in wireless networks.

52 IEEE Wireless Communications February 2017