The Impact of Beamforming and Coordination on

Spectrum Pooling in mmWave Cellular Networks

(Invited Paper)
Hossein Shokri-Ghadikolaei† , Federico Boccardi∗ , Elza Erkip‡ , Carlo Fischione† , Gábor Fodor†∗∗ ,
Marios Kountouris§ , Petar Popovski¶ , and Michele Zorzik¶
† Electrical
Engineering, KTH Royal Institute of Technology, Stockholm, Sweden
∗ Ofcom, UK
‡ NYU Tandon School of Engineering, New York University, Brooklyn, NY, USA
∗∗ Ericsson Research, Stockholm, Sweden
§ Mathematical and Algorithmic Sciences Lab, France Research Center, Huawei Technologies Co. Ltd., France
¶ Department of Electronic Systems, Aalborg University, Aalborg, Denmark
k Department of Information Engineering, University of Padova, Padova, Italy

Emails: hshokri@kth.se, federico.boccardi@ieee.org, elza@nyu.edu, gaborf@kth.se, carlofi@kth.se,

marios.kountouris@huawei.com, petarp@es.aau.dk, and zorzi@dei.unipd.it

Abstract—Spectrum pooling is not typically used in current
cellular networks, because it only provides a slight performance
improvement while requiring heavy coordination among different
cellular operators. However, these problems can be potentially
overcome in millimeter-wave (mmWave) networks, thanks to the
use of beamforming both at base stations and at user equipments.
In this paper, we develop a joint beamforming and cell association
optimization problem to characterize the performance gain that
can be obtained when spectrum pooling is used, as a function
of the underlying beamforming and coordination strategies. Our
performance analysis reveals that beamforming can substantially
reduce the need for coordination and simplify the implementation
of spectrum pooling. These benefits are more prominent at higher
mmWave frequencies (for example, 73 GHz) due to the possibility
of having antenna arrays with more elements within the radome.
The results of this paper provides useful insights on the feasibility
of spectrum pooling at mmWave networks.
Index Terms—Spectrum pooling, millimeter wave networks,
coordination, beamforming.
I. I NTRODUCTION
Millimeter-wave (mmWave) communication systems are a
frontier for wireless networks to address both the scarcity of
available sub-6 GHz spectrum and the significant increase in
mobile traffic demands [1], [2]. Unfortunately, the availability
of spectrum for mobile services presents limitations even at
mmWave frequencies, particularly considering the require-
ments of other systems that may use these bands in the future,
including satellite and fixed services [3], and also the need for
licensing this band to multiple operators to ensure a healthy
competition. Therefore, it is essential to seek an optimal use
of the spectrum, with the ultimate goal of maximizing the
benefits for the citizens.
Spectrum pooling, namely spectrum sharing for the same
use — mobile services in this case —, among multiple mobile
operators is a promising solution to allow a more efficient
spectrum utilization. Preliminary studies have shown that
the specific features of mmWave frequencies, including their
unique propagation characteristics and beamforming with a
large number of antennas, are important enablers for spectrum
pooling. In [4], a mechanism that allows two different IEEE
802.11ad access points to share the same time/frequency
resources was proposed. This mechanism is based on a novel
signaling report, which is broadcast by each access point to
establish an interference database that facilitates scheduling
decisions. Reference [5] showed that, under certain conditions,
simple scheduling policies with no coordination can be as good
as the complicated ones with full coordination in mmWave
networks. Reference [6] showed that, with an idealized antenna
pattern, spectrum pooling may be beneficial even without any
coordination in the entire mmWave network. Reference [7]
extended this study and illustrated the importance of coor-
dination in non-idealized settings. Reference [8] studied the
performance of a hybrid spectrum access scheme in which
exclusive access is used at frequencies in the 20/30 GHz
range while spectrum pooling (or even unlicensed spectrum)
is used at frequencies around 70 GHz. It is clear from these
studies that beamforming, both at the transmitter and at the
receiver, and coordination, both within an operator (here-
after called intra-operator coordination) and among different
operators (hereafter called inter-operator coordination) have
major effects on the performance of spectrum pooling in
mmWave networks. However, an accurate characterization of
their effects is a largely open problem.
In [7], we discussed various architectural options to support
spectrum pooling and provided example results to illustrate
the main trends. In the present paper, we add a mathemat-
ically detailed technical framework to evaluate the gains of
beamforming and coordination for spectrum pooling schemes
in mmWave networks. Specifically, we first discuss technical
enablers for spectrum pooling including various supporting
architectures and enabling functionalities [7]. We then consider
a particular pooling architecture and mathematically formulate
a multi-objective optimization problem that finds the optimal
association and beamforming when full coordination is avail-

978-1-5386-3954-2/16/$31.00 ©2016 IEEE 21 Asilomar 2016

able, and finally apply this optimization framework to the cases
when either only intra-operator coordination or no coordina-
tion is available. We show that beamforming, especially at
the user equipment (UE), can substantially reduce the need
for coordination and simplify the implementation of spectrum
pooling. This benefit is more prominent at higher mmWave
frequencies (for example, 73 GHz). Furthermore, while inter-
operator coordination can be neglected in the large-antenna
regime, intra-operator coordination can still bring gains by
balancing the network load. We believe that the results of this
paper provide useful insights on the feasibility of spectrum
pooling at mmWave networks.
The rest of this paper is organized as follows. Section II dis-
cusses some protocol and architectural enablers for spectrum
pooling. Sections III presents the system model considered in
this paper. Sections IV formulates optimization problems to
study the gains of coordination and beamforming in spectrum
pooling schemes in mmWave networks, followed by numer-
ical results in Section V. We provide further discussions in
Section VI and conclude the paper in Section VII.
II. P ROTOCOL AND A RCHITECTURAL E NABLERS
In this section, following [7], we discuss different types of
coordination, supporting architectures, and functional enablers
for spectrum pooling in mmWave networks.
A. Coordination
The modality of coordination directly affects the feasibility
of different spectrum pooling options. With an uncoordinated
approach, network operators make independent decisions on
how to allocate their spectrum using some predefined common
rules that ensure equitable spectrum allocation. However, pool-
ing performance can be improved by distributed coordination
when the decision of each operator may be aided by the
exchange of supporting information from other operators.
For instance, high interference levels measured in subbands
within the shared spectrum pool can trigger a request from
a participating operator to its peer operator(s) to reschedule
some of the served traffic to other resources. The pooling
performance can be further improved by allowing centralized
coordination where the actions are decided by a logical central
entity, such as a spectrum broker [9]. We should note that the
latency requirements may make some of these coordination
modalities infeasible.
B. Supporting Architectures
There are different architectures that support spectrum pool-
ing in mmWave networks, illustrated in Fig. 1. We may add a
new interface either at the radio access or in the core network
(RAN or CN) to enable coordination of different operators.
The RAN interface allows a near real-time coordination,
whereas the CN interface allows handling a large number of
cellular base stations (BSs) by exchanging only a few protocol
messages, since typically a large number of BSs are associated
with a few CN nodes.
22
Fig. 1: Architectural solutions supporting spectrum pooling between two
different network operators [7]: (a) interface at the RAN (BS); (b) interface
in the CN; (c) RAN sharing, (d) CN sharing; (e) via a spectrum broker; and
(f) uncoordinated.
Alternatively, two or more network operators may share
their BSs or CNs. As resource allocation and scheduling deci-
sions are made by a single unit, these alternatives are effective
ways to implement centralized coordination. Note that CN
sharing may not allow a real-time centralized coordination.
Spectrum broker is another alternative architecture to imple-
ment coordination. It is a central resource management entity
that grants spectrum resources on an exclusive basis during
some time window [9]. Finally, when the number of networks
in the pool is not limited, (e.g., in Wi-Fi), uncoordinated
spectrum pooling is reminiscent of a license-exempt regime.
C. Supporting Functions
Depending on the type of coordination and on the ar-
chitectural solution, spectrum pooling may require various
supporting functions, including spectrum sensing, efficient
channel estimation approaches, and distributed synchroniza-
tion mechanisms.
Spectrum sensing can help identify subbands that have the
least instantaneous traffic load, so as to reduce the coor-
dination requirements. Efficient approaches to estimate the
channel at both transmitters and receivers is an important
enabler of spectrum pooling in mmWave networks, given the
importance of accurate narrow beamforming. With the large
number of antennas available in mmWave systems, a good
channel estimate usually reduces the need for interference
coordination. As we show in [10], the benefits of inter-operator
coordination vanish in the asymptotic case of perfect channel
estimate and infinite number of antennas. Synchronization of
the participating networks is an indisputable part of spectrum
pooling that helps avoid BS-BS and UE-UE interference. Due
to their importance, these synchronization signals should be
protected from the adverse effects of spectrum pooling (such
as higher multiuser interference), e.g., by exclusive resource
allocation [10].
Table I [7] compares the main characteristics of the different
architectural solutions. In the following, we present technical
evidence that reveals under which assumptions and conditions
spectrum pooling at mmWave frequencies is beneficial. To
this end, as a first step, we consider ideal assumptions on
the channel estimation (no error) and coordination (no delay).
III. S YSTEM M ODEL
A. Network Model
We consider the downlink of a multi-operator cellular net-
work with total bandwidth W to be shared among Z operators
in the network. Let Bz be the set of BSs of operator z, and
B = B1 ∪ B1 ∪ . . . ∪ BZ be the set of all BSs in the network.
We denote by U the set of all UEs, by Uz the set of all UEs of
operator z, and by Wz the bandwidth of operator z. Without
loss of generality, we assume universal frequency reuse within
an operator, so every UE is receiving interference from all
non-serving BSs of that operator.
Let xbu be a binary variable that is equal to 1 if UE u ∈ U is
served by (or associated to) BS b ∈ B. We assume that each BS
can serve only UEs of the same operator (no national roaming),
namely xbu = 0 for all b ∈ Bz and u ∈ Uk where z 6= k. An
association remains unchanged over many coherence intervals,
For a UPA with half wavelength antenna spacing at the BS,
we have
1 h
aBS (θ, φ) = √ 1, . . . , e−jπ(m sin(θ) sin(φ)+n cos(φ)) ,
MBS NBS
i∗
. . . , e−jπ((MBS −1) sin(θ) sin(φ)+(NBS −1) cos(φ)) ,
(2)
where 1 ≤ m ≤ MBS and 1 ≤ n ≤ NBS . The parameters of
the channel model depend both on the carrier frequency and
on being in line-of-sight (LoS) or non-LoS conditions and are
given in [1, Table I].
C. Beamforming Model
We consider one RF chain at each UE and Nr RF chains
at each BS, where 1 ≤ Nr  MBS NBS . Each RF chain can
serve only one UE at a time; however, a BS can serve multiple
UEs at the same time by exploiting different RF chains. We
UE
denote by wbu the combining vector of UE u when being
served by BS b and by WbBS the precoding matrix at BS b
BS
whose u-th column wbu is the precoding vector for UE u.
We assume that the total transmission power p is divided
equally among all active RF chains, so the transmission
power P of BS b toward its individual UEs is p/Nb , where
Nb = u∈U xbu is the load of BS b (number of its active
RF chains). The received power of UE u from BS b is
UE ∗

BS 2
p| wbu Hbu wbu | /Nb . The RF precoding and combining
vectors try to maximize the received power. Formally, they are
the solution of the following optimization problem:
 2
whereas beamforming should be recomputed every coherence  UE
∗ BS 
maximize  wbu Hbu wbu  , (3a)
time. For short-term scheduling, we ensure that each BS can wbUE BS
u , wb u
UE
serve all its UEs simultaneously (by assuming a sufficiently subject to wbu ∈ W UE , (3b)
large number of RF chains at each BS). BS BS
wbu ∈W , (3c)
where W BS and W UE are the precoding and combining
B. Antenna and Channel Models codebooks of the BSs and UEs, respectively.

We consider a half wavelength uniform planar array (UPA) D. SINR and Rate Models
of MBS × NBS antenna elements for all BSs and a UPA of The received power at each UE u ∈ Uz when the serving
NUE × MUE antennas for all UEs, albeit our mathematical BS is b ∈ Bz is the summation of five components: desired
framework can be easily extended to other antenna models. power P , intra-cell interference I1 , inter-cell interference
We consider a narrowband cluster channel model [1], with I2 , inter-operator interference I3 , and noise power spectral
Nbu paths between BS b and UE u. Let gbun be a zero-mean 2
density σ
. The received power of UE u from BS b is P =
UE ∗ BS 2
complex Normal random variable that denotes the channel gain p| wbu Hbu wbu | /Nb . Interference term I1 corresponds to
of the n-th path, and Lbu be the path loss [1]. The channel the signals transmitted to other UEs by the same BS. It is
matrix between BS b and UE u is given by straightforward to show that
Nb u x  2
 UE
∗
r
P bj
X
MBS NBS NUE MUE X UE I1 = p  wbu Hbu wbj BS 
 , (4)
, φUE


Hbu = gbun aUE θbun
Nbu bun
j∈Uz \{u} m∈Uz xbm
n=1
× a∗BS θbun
BS
, φBS where p/Nb is the transmit power of BS b toward UE u. I2


bun , (1)
BS corresponds to the interference from the signals transmitted by
where θbun and φBSbun are the azimuth and elevation angles
UE other BSs of the same network operator and can be calculated
of arrival (AoAs) of the n-th path, θbun and φUE
bun are the as
azimuth and elevation angles of departure (AoDs) of the n-th x  2
X X
P ij  UE
∗ BS 
path, aBS ∈ CMBS ×NBS and aUE ∈ CMUE ×NBS are the vector I2 = p  wbu Hiu wij  . (5)
m∈Uz xim
response functions of the BSs’ and UEs’ antenna arrays to i∈Bz \{b} j∈Uz
the corresponding AoAs and AoDs, and (·)∗ is the conjugate I3 corresponds to the interference from the signals transmitted
transpose operator. by all BSs of other operators B \ Bz toward their own UEs.

23
TABLE I: Main characteristics of different architectural enablers of spectrum pooling in mmWave cellular networks [7]. “Enhanced CSI” refers to enhanced
channel state information acquisition capability.

Type of coordination Time resolution Supporting functions required Information exchange overhead
Interface at RAN Distributed Real-time Enhanced CSI and distributed synchronization High
Interface in CN Distributed Long-term Enhanced CSI Low
RAN sharing Centralized Real-time Enhanced CSI –
CN sharing Centralized Long-term Enhanced CSI –
Spectrum broker Centralized Long-term Enhanced CSI Low
Uncoordinated Uncoordinated – Enhanced CSI and spectrum sensing –

By assuming Wz = W for all 1 ≤ z ≤ Z, each UE u optimal association:

receives interference from all BSs of all operators, and the P1 : maximize [f1 (X), f2 (X), . . . , fZ (X)] , (9a)
inter-operator interference can be expressed as X
Z X X 2 subject to beamforming design (3) , (9b)
x 
 UE
∗
P ij
X
BS 
I3 = p  wbu Hiu wij  . (6)
X
xbu = 1 , ∀u ∈ Uz , 1 ≤ z ≤ Z , (9c)
k=1 i∈Bk j∈Uk j∈Uk xij
k6=z b∈Bz
X
Then, the average rate that UE u can get from BS b conditioned xbu ≤ Nr , ∀b ∈ Bz , 1 ≤ z ≤ Z (9d)
on the network topology is u∈Uz
  
P xbu ∈ {0, 1} , ∀b ∈ B, u ∈ U , (9e)
rbu = E W log 1 + , (7)
I1 + I2 + I3 + W σ 2 xbu = 0 , ∀b ∈ Bk , u ∈ Uz , k 6= z, 1 ≤ z, k ≤ Z.
where the expectation is over all random channel gains. The (9f)
P u will receive from all BSs of operator
long-term rate that UE Constraint (9c) guarantees association of each UE to only one
z is equal to ru = b∈Bz xbu rbu . BS. Constraint (9d) ensures that Nb ≤ Nr , so all Nb UEs that
are associated to BS b can be concurrently served. The case
Nb < Nr implies that some RF chains will be off, so the BS
IV. S PECTRUM P OOLING WITH A NALOG B EAMFORMING gives higher transmit power to the active ones. To address P1 ,
we use the weighted Tchebycheff scalarization method, which
In this section, we formulate the relevant optimization can provide the complete Pareto optimal solutions with low
problems to optimize the performance of spectrum pooling computational complexity [11]. Note that the main aim of this
for analog beamforming and for different coordination strate- paper is to understand the feasibility of spectrum pooling in
gies. The comparison of the solutions of these optimization mmWave networks, and efficient solution methods for P1 are
problems enables us to quantify the gains of beamforming left for future work.
and coordinations. To optimally solve P1 , a central controller should know CSI
toward all UEs at all BSs of all operators, which requires full
intra- and inter-operator coordination. In addition to privacy
A. Optimal Association with Pooling and Full Coordination challenges, the complexity and cost of such level of channel
estimation and coordination grow large with the number of
We first collect all control variables xbu in matrix Xz for BSs and UEs, and is in general overwhelming for mmWave
all operators 1 ≤ z ≤ Z, and let X = diag(X1 , X2 , . . . , XZ ). networks with dense BS deployment. The main source of
For each operator z, I1 and I2 depend only on Xz . However, this complexity is the need for inter-operator coordination to
the inter-operator interference I3 depends on the associations calculate I3 . In the following, we relax this problem.
of other operators {Xi }Z i=1,i6=z . Therefore, ru depends on the
association solutions of all operators X.
In this paper, we consider proportional fairness as the B. Optimal Association with Pooling and No Inter-operator
network utility of each operator to ensure both high net- Coordination
work throughput and fairness among individual UEs. In
P for each operator z, UE u ∈ Uz achieves rate
particular,
To alleviate the cost and complexity of P1 , we enforce the
ru = b∈Bz xbu rbu , and the objective function of the opera-
following design constraints. First, each operator maximizes
tor is
only its own benefit without taking the objectives of other
!
X X operators into account. Second, the inter-cell interference I3
fz (X) = log xbu rbu . (8) is approximated by a quantity Ib3 that depends only on Xz
u∈Uz b∈Bz for each operator z. In the following, we assume Ib3 = 0.
Consequently, fz can be calculated without any inter-operator
The following multi-objective optimization problem gives the coordination. Each operator z can then independently solve

24
the following optimization problem:
P2 : maximize fz (Xz ) , (10a)
Xz
subject to beamforming design (3) , (10b)
X 25
xbu = 1 , ∀u ∈ Uz , (10c)
b∈Bz
X −25
xbu ≤ Nr , ∀b ∈ Bz , (10d)
u∈Uz
xbu ∈ {0, 1} , ∀u ∈ Uz , b ∈ Bz , (10e)
Due to the parallelism, this approach is much more com-
putationally efficient than P1 . Optimization problem P2 re-
quires no inter-operator CSI knowledge nor inter-operator
coordination. However, it still needs full CSI knowledge and
coordination within each operator.
C. Optimal Association with No Pooling
In the no spectrum pooling scenario, we assume that the
bandwidth W is equally divided among all Z operators.
Therefore, the bandwidth available to each BS is Wz = W/Z,
the inter-operator interference is I3 = 0, and consequently
fz depends only on Xz for each operator z. Now, we can
formulate a new optimization problem P3 that optimizes the
association of the no spectrum pooling scenario for every
operator z independent of the other operators.
V. N UMERICAL R ESULTS
To numerically investigate the feasibility of spectrum pool-
ing in mmWave networks, we assume that BSs and UEs are
randomly distributed on the plane according to independent
Poisson point processes. We consider an ideal centralized
coordination scheme. Moreover, we assume that all BSs are
synchronized, and that there is no delay in the interface
between the BSs and the central controller. We also consider
Nr = 6 RF chains with 25 dBm total transmission power at
each BS (at the input of the transmit array) and only one RF
chain at each UE. Without loss of generality, we consider four
operators and a total bandwidth of 1.2 GHz, pooled according
to the following three scenarios:
• Exclusive: each operator uses a 300 MHz exclusive
bandwidth;
• Partial bandwidth pooling: operators 1 and 2 share the
first 600 GHz, and operators 3 and 4 share the second
600 GHz; and
• Full bandwidth pooling: all four operators share the whole
1.2 GHz bandwidth.
For the spectrum pooling scenarios with no inter-operator
coordination (P2 ), we consider Ib3 = 0 and find the corre-
sponding optimal association. However, note that the actual
value of I3 is considered when computing the performance of
that association. We simulate 100 random topologies and find
the optimal association and beamforming for every case.
Fig. 2 illustrates the gain of partial and full pooling under
the assumption of no inter-operator coordination (P2 ). The
baseline is an exclusive spectrum allocation (P3 ). We show
the 5th, 50th, and 95th percentiles of the UE downlink rates
25
100
UE rate enhancement (%)
Partial BW pooling
75
Full BW pooling
50
0
−50
5% 50% 95%
(a) Omnidirectional antenna at the UE.
300
UE rate enhancement (%)
250
200
150
100
50
0
5% 50% 95%
(b) 4x4 UPA at the UE.
Fig. 2: Pooling performance gains at 32 GHz, under the assumption of no
inter-operator coordination (P2 ). The baseline is exclusive spectrum allocation
(P3 ).
at 32 GHz, assuming a 32x32 uniform planar array (UPA) at
each BS, a BS density of 100 BSs/km2 , and a user density of
800 UEs/km2 . We observe in Fig. 2(a) that most UEs suffer
from spectrum pooling. Full bandwidth pooling leads to the
worst performance for the 5th and 50th percentile UEs, due to
the high inter-operator interference I3 .1 In Fig. 2(b), we repeat
the previous comparison under the assumption of a sufficiently
large number of antennas at the UEs. In this case, both partial
and full pooling enhance the 5th, 50th, and 95th percentiles
compared to the baseline (i.e., exclusive), although only the
best users approach the performance upper bound in the two
cases.
Fig. 3 shows the impact of the operating frequency and
the BS density on the full pooling performance. We keep the
physical size of the antenna array constant as a function of
the frequency, i.e., at 73 GHz we consider twice as many
antenna elements in each dimension as at 32 GHz, at both BS
and UE. We note that, without inter-operator coordination, the
benefits of spectrum pooling decrease with the BS density of
individual operators, due to higher inter-operator interference.
This effect is less pronounced at 73 GHz, due to large antenna
arrays and consequently higher directionality of the beams.
Fig. 3 indicates the importance of inter-operator coordination
for the weakest UEs (5th percentile UEs) and in ultra dense
mmWave networks (200 BS/km2 ). Also, the observation that
the effectiveness and the performance of pooling depends on
1 To put the results in perspective, we note that in case of partial pooling
each operator has access to twice the bandwidth compared to the baseline
exclusive scheme, and therefore the achievable gain cannot exceed 100%.
Similarly, in the full pooling case, the maximum gain is 300%. These gains
are upper bounds, and can be approached only when effective countermeasures
are taken against interference, otherwise the performance will be significantly
lower, e.g., as shown in Fig. 2(a).
 300
UE rate enhancement (%)
250 N-I-C. 50 BS/km2
N-I-C. 200 BS/km2
200
W-I-C. 50 BS/km2
150 W-I-C. 200 BS/km2
100
50
0 beamforming can substantially reduce the need for coordina-
5%
(a) 32 GHz, 4x4 UPA at the UE, 32x32 UPA at the BS.
300
UE rate enhancement (%)
250
200
150
100
50
0 for Strategic Research under Grant SM13-0008. The work of
5%
(b) 73 GHz, 8x8 UPA at the UE, 64x64 UPA at the BS.
Fig. 3: Full pooling performance gains. The baseline is an exclusive spectrum
allocation. “N-I-C” and “W-I-C” stand for no inter-operator coordination (P2 )
and with inter-operator coordination (P1 ), respectively.
the frequency motivates schemes in which different pooling
strategies are used in different bands [8].
VI. F URTHER D ISCUSSIONS
The results presented above clearly indicate that spectrum
pooling provides substantial gains at the 5th, 50th and 95th
user percentiles only if proper beamforming and coordination
is employed. However, more work is required to assess the
impact of real-world effects such as pilot contamination, im-
perfect channel estimation, and mobility, where beamforming
gains may be less effective. When beamforming by itself
is not sufficient to protect the weakest users from inter-
network interference (e.g., at lower mmWave frequencies or at
ultra dense networks), pooling with full coordination or even
exclusive spectrum access is more preferable. As proposed
in [8], a hybrid spectrum access scheme is a promising option
in which exclusive access is used at lower frequencies (around
20–30 GHz) while spectrum pooling is used at higher frequen-
cies (around 70 GHz). A complementary approach especially
for dense mmWave networks, as suggested in [10], is the
use of digital beamforming that allows better interference
cancellation compared to analog beamforming.
Coordination is less effective with a non-ideal control plane
(non-zero delay and packet loss). This challenge is exacerbated
in distributed coordination schemes. Therefore, a reliable
control channel for exchanging coordinating information is
another critical enabler of spectrum pooling in mmWave
networks, which deserves a thorough investigation.
VII. C ONCLUSIONS
The typical characteristics of mmWave systems, such as
high penetration loss and directional communications, may
help overcome the traditional problems of spectrum pooling
at sub-6 GHz bands, such as high inter-operator interference.
In this paper, we have illustrated the technical enablers of
spectrum pooling in mmWave networks and proposed an
optimization framework based on a joint beamforming design
and BS association. Using this framework, we have analyzed
the gains of beamforming and coordination and showed that
50% tion and simplify the implementation of spectrum pooling. The
possibility of fitting more antenna elements within the radome
at 73 GHz boosts these benefits.
ACKNOWLEDGMENTS
The work of C. Fischione and H. Shokri-Ghadikolaei was
supported by the Swedish Research Council and the Hans
Werthén Foundation. The work of E. Erkip was partially sup-
ported by NSF grants #1302336 and #1547332. The work of
G. Fodor was partially supported by the Swedish Foundation
50%
M. Zorzi was partially supported by NYU-Wireless and the
Villum Foundation. The views expressed in this research paper
do not necessarily reflect those of the employers of F. Boccardi
or M. Kountouris.
R EFERENCES
[1] M. Akdeniz, Y. Liu, M. Samimi, S. Sun, S. Rangan, T. Rappaport,
and E. Erkip, “Millimeter wave channel modeling and cellular capacity
evaluation,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1164–1179,
Jun. 2014.
[2] H. Shokri-Ghadikolaei, C. Fischione, G. Fodor, P. Popovski, and
M. Zorzi, “Millimeter wave cellular networks: A MAC layer perspec-
tive,” IEEE Trans. Commun., vol. 63, no. 10, pp. 3437–3458, Oct. 2015.
[3] F. Guidolin, M. Nekovee, L. Badia, and M. Zorzi, “A study on the
coexistence of fixed satellite service and cellular networks in a mmWave
scenario,” in Proc. IEEE International Conference on Communications
(ICC), 2015, pp. 2444–2449.
[4] W. Feng, Y. Li, D. Jin, and L. Zeng, “Inter-network spatial sharing
with interference mitigation based on IEEE 802.11ad WLAN system,” in
Proc. IEEE Global Communications Conference (GLOBECOM) Work-
shop, Dec. 2014, pp. 725–758.
[5] H. Shokri-Ghadikolaei and C. Fischione, “The transitional behavior of
interference in millimeter wave networks and its impact on medium
access control,” IEEE Trans. Commun., vol. 62, no. 2, pp. 723–740,
Feb. 2016.
[6] A. K. Gupta, J. G. Andrews, and R. W. Heath, “On the feasibility of shar-
ing spectrum licenses in mmwave cellular systems,” arXiv:1512.01290,
2016.
[7] F. Boccardi, H. Shokri-Ghadikolaei, G. Fodor, E. Erkip, C. Fischione,
M. Kountoris, P. Popovski, and M. Zorzi, “Spectrum pooling in mmWave
networks: Opportunities, challenges, and enablers,” IEEE Commun.
Mag., vol. 54, no. 11, pp. 33–39, Nov. 2016.
[8] M. Rebato, F. Boccardi, M. Mezzavilla, S. Rangan, and M. Zorzi,
“Hybrid spectrum sharing in mmWave cellular networks,” arXiv preprint
arXiv:1610.01339, 2016.
[9] T. Irnich, J. Kronander, Y. Selen, and G. Li, “Spectrum sharing sce-
narios and resulting technical requirements for 5G systems,” in Proc.
IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC), Sept. 2013, pp. 127–132.
[10] H. Shokri-Ghadikolaei, F. Boccardi, C. Fischione, G. Fodor, and
M. Zorzi, “Spectrum sharing in mmWave cellular networks via cell asso-
ciation, coordination, and beamforming,” IEEE J. Sel. Areas Commun.,
vol. 34, no. 11, pp. 2902–2917, Nov. 2016.
[11] R. T. Marler and J. S. Arora, “Survey of multi-objective optimization
methods for engineering,” Struct. Multidiscip. O., vol. 26, no. 6, pp.
369–395, Apr. 2004.
26