Content uploaded by Zhang Peize
Author content
All content in this area was uploaded by Zhang Peize on Aug 30, 2019
Content may be subject to copyright.
China Communications • May 2019 1
well as some implications on future mmWave
cellular network designs.
Keywords: fifth generation (5G); channel
modeling; large-scale antenna array (LSAA);
millimeter wave (mmWave) communications;
radio propagation measurements; wireless
coverage
A multigigabit-per-second user-experience
data rate with lower latency and higher reli-
ability is required to support various emerging
applications for fifth-generation (5G) mobile
communication systems. This enhancement
is hardly achieved under existing network
schemes due to spectrum congestion in cur-
rent commercial microwave bands below
6 GHz [1]. For this purpose, millimeter-wave
(mmWave) mobile communications are wide-
ly considered to fulfill 5G key performance
indicators (KPIs) dened by the International
Mobile Telecommunications for 2020 (IMT-
2020) [2, 3]. The candidate mmWave bands
(e.g., 24.25-27.5 GHz, 37-40.5 GHz, 42.5-
Abstract: Millimeter-wave (mmWave) com-
munications will be used in fifth-generation
(5G) mobile communication systems, but
they experience severe path loss and have
high sensitivity to physical objects, leading
to smaller cell radii and complicated network
architectures. A coverage extension scheme
using large-scale antenna arrays (LSAAs) has
been suggested and theoretically proven to be
cost-efficient in combination with ultradense
small cell networks. To analyze and optimize
the LSAA-based network deployments, a
comprehensive survey of recent advances
in statistical mmWave channel modeling is
first presented in terms of channel parameter
estimation, large-scale path loss models, and
small-scale cluster models. Next, the measure-
ment and modeling results at two 5G candidate
mmWave bands (e.g., 28 GHz and 39 GHz)
are reviewed and compared in several outdoor
scenarios of interest, where the propagation
characteristics make crucial contributions to
wireless network designs. Finally, the cover-
age behaviors of systems employing a large
number of antenna arrays are discussed, as
Received: Sep. 24, 2018
Revised: Dec. 30, 2018
Editor: Tao Jiang
Radio Propagation and Wireless Coverage of LSAA-
Based 5G Millimeter-Wave Mobile Communication
Systems
Haiming Wang1, 2,*, Peize Zhang1, 2, Jing Li1, 2, Xiaohu You1, 3,*
1 School of Information Science and Engineering, Southeast University, Nanjing 211111, China
2 State Key Laboratory of Millimeter Waves, Southeast University, Nanjing 211111, China
3 National Mobile Communications Research Laboratory, Southeast University, Nanjing 211111, China
* The corresponding authors, email: xhyu@seu.edu.cn, hmwang@seu.edu.cn
COVER PAPERS
China Communications • May 2019
2
term observation data [2].
From the perspective of radio propagation
characteristics, an individual mmWave cellu-
lar network not only provides a smaller cell
radius but also has challenges in continuous
coverage considering user mobility. Hence,
mmWave radio access networks require ultra-
dense network deployment and a complicated
network structure to satisfy the demand from
multiple smart mobile services in hotspot ar-
eas with multitudinous users (e.g., business
centers, ofce buildings, large-scale stadiums,
and train stations) and improve the quality of
service (QoS) [3, 10]. Network densification
using small cells can provide directional con-
nections and ultrahigh throughput for users,
corresponding to a network capacity increase
and negligible intercell interference. However,
this densication not only implies a higher de-
ployment cost and is impracticable for equip-
ping every small cell with fiber connectivity
but also cannot realize ubiquitous coverage in-
dependently. Furthermore, for high-speed mo-
bile users, overcoming the effect of frequent
handover in small cells is also a challenge.
Due to severe attenuation and path loss be-
tween transmitters (TXs) and receivers (RXs),
LSAA-based mmWave mobile communi-
cation systems are necessary to extend the
coverage up to several hundred meters, with
the increase in the number of array elements
using directional transmissions [6, 11, 12].
Traditional single-directional beams must be
steered either electronically or mechanically
to detect optimized links, and communication
is broken when the transceivers are moving
or no LoS path exists. Thanks to the employ-
ment of multibeam or beam-steerable antenna
arrays (e.g., passive multibeam antenna, the
lens-based beam-switching antenna system,
and the active phased array), these arrays can
overcome the above shortcomings and provide
robust beam alignment [13]. Combined with
massive MIMO techniques, the systems using
multibeam arrays and advanced beamforming
precoding provide tremendous increases in
spectral efciency and anti-interference capa-
bility at base station (BS) sides [3, 14]. With
43.5 GHz, 45.5-47 GHz, 47.2-50.2 GHz, 50.4-
52.6 GHz, 66-76 GHz, and 81-86 GHz) have
been taken into account for the 5G mobile
service in the World Radiocommunication
Conference 2015 (WRC-15) and will finally
be determined in WRC-19 based on technical
and economic considerations. Meanwhile,
with the rapid development of radio frequency
(RF) hardware techniques [4], the implemen-
tation of several enabling technologies (e.g.,
massive multiple-input and multiple-output
(MIMO), hybrid beamforming using miniatur-
ized high-density large-scale antenna arrays
(LSAAs) and low-power complementary
metal–oxide–semiconductor (CMOS) circuits)
is possible for mmWave communications [5,
6]. However, there are still several challenges
and requirements in the deployment of LSAA-
based mmWave systems with large coverage.
In contrast to traditional wireless commu-
nication systems, the path loss of mmWave
wireless systems increases much more rapid-
ly with propagation distance according to the
Friis transmission formula [2]. Based on ex-
tensive eld mmWave channel measurement
campaigns, the path loss exponent (PLE) is
usually larger than 2 in free space [7, 8], and
only for cases in waveguide-like environ-
ments (e.g., corridors, tunnels, and street can-
yons) it can be smaller than 2 [9]. In addition,
line-of-sight (LoS) propagation is dominant
in mmWave channels due to shorter wave-
lengths, and diffracting or scattering occurs
with a large angular spread for obstacle-LoS
(OLoS) scenarios with human and foliage
blockages. In non-LoS (NLoS) scenarios,
received signals are mostly composed of
first- and second-order reflected paths from
exterior walls and the ground in outdoor en-
vironments. Moreover, mmWave signals suf-
fer from large building entry loss (BEL) and
high sensitivity to blockages, which creates
challenges in providing continuous coverage
in outdoor-to-indoor (O2I) scenarios. The
impact of atmospheric absorption by oxygen
molecules, water vapor, and rain attenuation
is nonnegligible in mmWave bands and var-
ies with the carrier frequency based on long-
Based on the ob-
servations of recent
mmWave channel
models from various
standard bodies and
measurement activi-
ties, appropriate large-
scale path loss models
and small-scale cluster
models of LSAA-based
mmWave systems
have been given, re-
spectively.
China Communications • May 2019 3
lar spread and the corresponding statistics are
presented for the optimization of LSAA-based
mmWave transmission schemes. As a credible
approach, propagation measurement-based
network design and deployment significantly
reduce the outage probability and provide
a reasonable tradeoff between the network
complexity and coverage enhancement of 5G
mmWave mobile communications.
The rest of the article is organized as fol-
lows. We start with a brief review of the recent
advances in mmWave channel modeling and
discuss the impacts of path losses, blockages,
and temporal/spatial dispersions on network
deployment using the LSAA in section II.
Then, a comparison of the eld measurement
results in three typical outdoor access sce-
narios (e.g., urban microcell (UMi), urban
macrocell (UMa), and rural macrocell (RMa))
is presented in section III. Finally, several im-
plications on the LSAA-based mmWave radio
network design, with respect to the coverage
analysis and network deployment, are derived
in section IV, and the conclusions are drawn in
section V.
MMWAVE CHANNEL
CM
MmWave statistical channel models mainly
concentrate on feature extraction from field
channel measurement data or ray-tracing sim-
ulation data. The concerned channel param-
eters are used to model large-scale path loss
and characterize small-scale multipath effects,
which further impact channel generation and
network deployment.
Models
As the basis of link budget and interference
calculations, distance-dependent path loss
models provide realistic insights into the
large-scale propagation characteristics of ra-
dio channels, which determine the coverage
of cellular and backhaul networks. The link
budget in mmWave channel measurements is
regard to potential beamforming architectures,
fully digital beamforming requires one ded-
icated RF chain per antenna element, which
is limited by hardware constraints, including
cost, unaffordable energy power consumption,
and integrated size [6]. Hence, several hybrid
analog-digital beamforming architectures
have been designed to reduce the number of
RF chains, resulting in fewer data streams and
reduced power consumption [15]. Even more
so, with the development of the substrate-in-
tegrated waveguide (SIW) technique, a fully
digital beamforming MIMO system is possi-
ble when using a high-performance mmWave
transceiver front-end [16]. To support multiple
users in the mmWave cellular network, multi-
ple-access techniques are necessary to exploit
the spatial resolution, spectral efciency, and
connectivity density, which require accurate
channel state information (CSI) and good
backward compatibility with other mmWave
physical layer-enabled techniques [3]. From
a wireless networking perspective, multitier
cooperative cell deployment, combined with
LSAA-based small cells, is expected to be a
key enabler to vastly boost 5G mmWave radio
network coverage and capacity with lower en-
ergy consumption and networking costs [10].
Recently, a novel low-latency heterogeneous
network (HetNet) using an adaptive low-laten-
cy strategy was developed, where the coopera-
tive network among the mmWave macrocells,
small cells, and wireless backhaul improved
the overall area coverage and system through-
put simultaneously [17, 18].
In this article, we focus on the coverage
and connectivity evaluation of LSAA-based
mmWave communication systems from radio
channel modeling and wireless networking
perspectives. The motivation is to show the
impact of large-scale and small-scale channel
characteristics on a 5G mmWave system de-
sign. Specifically, we discuss state-of-the-art
characteristics in terms of mmWave statistical
channel modeling based on eld measurement
data, including channel parameter estimation
and channel model standardization. On the
other hand, a site-specic analysis of the angu-
China Communications • May 2019
4
P P tt
r omni
=∫t
t
0
1( )d (3)
where t0 and t1 represent the rst and last times
when Pt
omni
( )
exceeds the noise floor (NF),
respectively. The NF can be calculated by the
variance in the last hundred ns of each CIR,
which represents the zero-mean-value random
measured noise following a Gaussian distribu-
tion. In practice, the detection level is devel-
oped based on a signal-to-noise ratio (SNR)
threshold relative to the NF, which helps to
determine the integration interval and estimate
effective MPCs. The SNR threshold is an
empirical value that is observed from a large
amount of mmWave channel sounding data,
which is closely related to the carrier frequen-
cy, bandwidth, and environment. An alterna-
tive method to find the detection level is to
compute the cumulative distribution function
(CDF) for the last hundred ns of each PDP and
then estimate the 95% condence interval. The
former method reects the receiver sensitivity,
which will be implemented with downlink
channel estimation in future 5G mmWave user
devices. The latter, however, is more practical
for individual TX-RX pairs when lacking suf-
ficient sounding data verification. Otherwise,
there are still some issues in the PDP range
extraction. For example, Fig. 1 shows a syn-
thesized PDP measured in the urban NLoS
scenario, where the start of the excess delay
t0 is the first time that the PDP is above the
detection level, with a 15 ns safety margin,
as opposed to the delay of the path with the
largest received power due to a lacking LoS
path. Note that random noise may exceed the
detection level with a large propagation delay,
which is not a detectable multipath, as shown
in figure 1. Therefore, we design a sliding
window with a specific duration to compute
the ratio of effective MPCs to all received
paths from the end of the PDP above the de-
tection threshold. If the ratio is larger than the
default, the current delay represents t1 in (3).
Here, we advocate the use of this credible PDP
estimation method in future mmWave channel
modeling activities.
given by
P PG L G G
rt tP r s
=+−++, (1)
here P
t represents transmitted power in the RF
output port; Gt and Gr represent the gain of the
transmit and receive antennas, respectively;
and Gs represents the system gain considering
the external RF amplifier gain, which can be
calculated through back-to-back calibration.
To accurately compute the path loss LP in ,
it is essential to obtain the received power P
r
through a power delay profile (PDP), which
reflects the decay of multipath components
(MPCs) with propagation delay and is also the
foundation of channel parameter estimation.
For an omnidirectional channel, the PDP is
synthesized with overlapping directional PDPs
when using narrow beamwidth antennas with
the directional scanning sounding method.
Here, we advocate for a standardized method
to synthesize the omnidirectional PDP by
P t ht
omni
( )
=∑∑
Θ
Φ
(
,Θ,Φ
)
2
, (2)
where ht
(
,Θ,Φ
)
represents the unique channel
impulse response (CIR) at the angle of depar-
ture (AoD) Θ
and the angle of arrival (AoA)
Φ
. Once the raw PDPs are recorded, the total
received powers for the unique TX-RX pairs
are computed using the denition in [2],
Fig. 1. Illustration of the effective PDP extraction method based on the eld mea-
surement results in the outdoor NLoS scenario.
0 500 1000 1500 2000 2500
-110
-100
-90
-80
-70
-60
-50
Random Noise: -71.49 dBm
Detection Deration: 20 ns
Diffraction from the
Building Corner
Reflection from the Metal Billboard
Outside the Opposite Building
Nosie Floor = -84.05 dBm
Detection Level = -74.05 dBm
Excess Delay [ns]
PDP [dBm]
t
0
t
1
China Communications • May 2019 5
XPD, which is a random variable defined as
the power ratio of the vertical-to-vertical (V-
V) and vertical-to-horizontal (V-H) channels
and the horizontal-to-horizontal (H-H) and
horizontal-to-vertical (H-V) channels [20].
2) Multi-Frequency Models
Standard bodies are also interested in the
utilization of multifrequency path loss models
to cover a broad range of frequencies via lim-
ited channel sounding data. The CI model can
be used for both single- and multifrequency
datasets, while it does not reect the coherence
between different frequencies. An extension of
the CI model with frequency weighting is the
CIF path loss model [21],
L fd L fd
CIF 0
++ +
(,) (, )
10 1 lognX
=
bf f
FS
(
fd
−
00
0
)
10
d
σ
CIF, (6)
where n denotes the distance dependency
of the path loss, b represents a model-fitting
parameter that captures the linear frequency
dependency of the path loss, and f0 represents
a fixed reference frequency that balances the
linear frequency dependence computed by
f0=round
∑
∑
k
K
=
k
K
1
=1
fN
N
kk
k
, (7)
where Nk represents the number of measure-
ments at a specic frequency fk, K represents
the number of measured frequencies, and the
function round{}⋅ calculates the nearest integer
of a value. In addition, an extension of the FI
model for multiple frequencies is the alpha-be-
ta-gamma (ABG) model [21],
L fd d
ABG ( , ) 10 log= +
++
αβ
10 log
γ
10 fX
10
σ
ABG, (8)
where
α
represents an optimized offset pa-
rameter, while
β
and
γ
describe the distance
and frequency dependency on the path loss,
respectively. Previous tting results at 28 GHz
and 73 GHz show unacceptable standard de-
viation values of more than 10 dB [21], which
indicates that multifrequency path loss models
are only suitable for a small frequency range
with similar propagation characteristics.
3) Dual-Slope Models
The single- and multifrequency path loss
Path loss models are normally modeled as a
function of the 3D propagation distance, while
those with fewer parameters, which indicate
physical meanings and simple expressions,
are preferable for application in future 5G
mmWave channel models. Here, we summa-
rize the three types of distance-dependent path
loss models in operation.
1) Single-Frequency Models
The close-in (CI) free space reference dis-
tance path loss model is a fundamental model
suiting all kinds of environments,
L d L fd n X
CI ( ) ( , ) 10 log=++
FS 100d
d
0
σ
CI, (4)
where
L fd d f
FS 0 10 0 10
(
, 20log 20log 32.44
)
= ++
rep-
resents the free space path loss (FSPL) as a
function of frequency (GHz) at the reference
distance d0, n represents the PLE using the
minimum mean square error (MMSE) t, and
X
σ
CI represents a zero mean Gaussian random
variable with standard deviation a
σ
CI in dB.
The CI model is developed from the FSPL
model; thus, the distance d should satisfy
the far-field condition. The benefits of using
a physically based reference anchor point
d0=1 have been described in detail [8], where
a standardized value has advantages in the
comparison of different measurement results.
An alternative to the CI model is the oating
intercept (FI) model based on the least-squares
(LS) method,
Ld d X
FI ( ) 10 log=++
αβ
10
σ
FI, (5)
where
α
represents the FI in dB,
β
represents
the slope, and X
σ
FI represents the shadow fad-
ing random variable. Compared with the CI
model, the FI model has no physical basis,
although it provides a simple best tting line
to the measured data with smaller standard
deviations. To study the impact of polarization
on mmWave propagation, an extension model
for a cross-polarized channel using a constant
attenuation factor (dB) (i.e., the so-called
cross-polarization discrimination (XPD) fac-
tor) is developed by sharing the same PLE in
a copolarized channel. Note that the cross-po-
larization ratio (XPR) is different from the
China Communications • May 2019
6
tion, and scattering. Thus, there is a tradeoff
between complexity and usability in modeling
mmWave path loss.
The distance-dependent path loss models
describe the received signal decay with
propagation distance, while the presence of
blockages (e.g., buildings, trees, and human
bodies) leads to additional power attenuation.
For O2I scenarios, the BEL provides a statis-
tical expression of the additional loss due to a
terminal being inside the building. Note that
the log-normally distributed shadow fading
associated with buildings is different from the
BEL, which can be considered as a random
variation in received power at the opposite
face of the building. Otherwise, penetration
loss is dened as the transmission loss through
a building material, which is calculated by
the power difference between the transmitted
signal outside the illuminated face and the
received signal outside the opposite face. This
loss can be used for assessing the impacts of
building material properties and structures on
radio wave propagation, which is a function
of operating frequency and incident direc-
tion. For the OLoS path, the received signals
blocked by humans and foliage consist of
many copies of the carrying signal generated
by diffraction and scattering. In addition, the
characterization of vehicle penetration loss
at the mmWave bands has recently received
much interest because of the development of
intelligent transportation systems. Hence, the
overall large-scale path loss models need to
consider the additional loss added to the basic
distance-dependent models.
1) BEL Model
The BEL determines the ability of the
mmWave radio system to provide continuous
coverage. A proposed BEL model in 3GPP TR
38.900 is given by [20]
LL p
BE =−⋅
+ 0,
npi 10
N
(
10log 10
σ
p
2
)
N∑
material
i=1
i
Lmater
−10
iali
(11)
where Lnpi represents the correction term
models are single-slope models, depending on
the propagation distance, plus log-normally
distributed shadowing attenuation. Based on
the observations of outdoor mmWave large-
scale channel characteristics, the LoS proba-
bility reduces with the increase in propagation
distance, corresponding to a larger path loss.
However, existing bias models do not give the
best fit with the link distance. A dual-slope
path loss model with a break-point distance of
dbp, is therefore expressed by
Ld
DS ()=
αβ
αβ β
1 1 bp
1 1 bp 2
+ +≤
++
10 log ,
10 log 10 log
+>X dd
σ
2,
10
10 10
dX dd
d
bp
σ
1
d
d
bp
,
(9)
which consists of two single-frequency mod-
els for different distance ranges. For conve-
nience, a default constant or a joint frequency
and height dependent value can be used as the
break-point distance [22]. Furthermore, sever-
al weight tting methods have been developed
to compute the optimal threshold distance,
where the propagation characteristics of dif-
ferent coverage ranges are taken into account
[23].
In [9], the omnidirectional channel models
above 6 GHz in the UMi, UMa, and RMa
scenarios from different standard bodies are
listed for comparison. Different from the three
basic path loss models, some extended mod-
els use 2D propagation distances to replace
the traditional 3D distances and add a height-
weight correction factor, which can not only
solve the problem in obtaining the 3D separa-
tion distance but also study the impact of the
BS height on path loss. However, for future
mmWave systems, directional transmission
plays a role in cellular and backhaul networks,
leading to the inaccuracy of omnidirection-
al path loss models unless antenna patterns
have been de-embedded [24]. Meanwhile,
compared with site-specic path loss models,
simplied distance-dependent models exhibit
a rather poor fit because building structures
have a signicant impact on mmWave propa-
gation, including reection, refraction, diffrac-
China Communications • May 2019 7
analytical models, are compared in terms of
frequency, foliage depth, and tree characteris-
tics. In [30], an improved coherent wave prop-
agation model considering multiple scattering
occurring among the needles of extremely
dense leaf clusters is proposed and validated
both theoretically and experimentally. For
vehicle penetration loss, a large-scale vehicle
exterior channel measurement, with the TX
inside the car at 45 GHz, was conducted in an
anechoic chamber, which can eliminate the
impact of physical objects around the car [31].
The mmWave propagation is sensitive to
surrounding objects, where the paths are suf-
ficiently separated in time, space, and polar-
ization. The interaction among the copies of
the transmitted signal traveling along different
paths, which are combined at the RX antenna
in amplitude and phase, causes rapid fluctu-
ation in the signal strength over a very short
propagation distance or duration. Multipath
fading provides natural advantages for extend-
ing the coverage of mmWave systems in NLoS
scenarios with the development of beamform-
ing and beam combining techniques. In rich
scattering environments, the larger rank of the
channel matrix indicates the increase in the
channel capacity. Thus, it is critical to model
small-scale spatiotemporal channel character-
istics, including the Rician K factor, Doppler
shifts, time dispersion, angular dispersion, and
their correlations. Measurement-based analy-
ses reveal that mmWave signals are received
in nite clusters, which are dened as sets of
MPCs with approximately the same directions
and delays from specific surrounding objects
(also called scatterers). Recent wideband
mmWave channel modeling activities have
mainly focused on the investigation of LoS
probability, path powers, and the statistics of
composite and cluster-level spatiotemporal
characteristics in different scenarios.
1) LoS Probability Model
LoS propagation can offer a credible con-
nection in mmWave communications, which
accounting for the impact of the incidence
angle on the BEL; pi represents the propor-
tion of ith materials and satisfies
N∑
material
i=1
pi=1;
L a bf
material material material
i ii
= + c represents the
penetration loss of the ith material at fc (GHz);
σ
p represents the standard deviation in the
penetration loss; Nmaterial represents the num-
ber of materials; and N
(
µσ
,2
)
represents the
normal distribution, with a mean of
µ
and a
variance of
σ
2. Meanwhile, a log-frequency
single-slope model is used to describe the
penetration loss in mmWave, with the form of
Lf
material material material 10 c
i ii
= +
αγ
10log , which has
a natural advantage when combined with the
ABG path loss model [19]. Although many
previous measurement results indicate that it
is difcult to maintain coverage indoors with
a mmWave outdoor BS [25], cooperative
communication using advanced cell selection
and transmitted power allocation algorithms
appears to solve this problem.
2) Human, Foliage, and Car Blockages
Several double knife-edge diffraction
(DKED)-based models have been developed
to describe the characteristics of human body
shadowing, such as the modified DKED
METIS model, which assumes that a human
blocker is represented as a screen with four
sides or an innitely vertical screen with two
sides [26]; a multiple-edge diffraction model,
with the more general assumption of a geo-
metric model of the human body [27]; and
Vogler’s multiple knife-edge model, which
has more than one person moving along the
LoS path between the TX and RX [28]. The
simulation results show good agreement with
the measurement results in different human
blockage scenarios, including human blockers
moving along the LoS link and frontally or lat-
erally crossing the LoS link. In rich scattering
environments, foliage or vegetation obstruc-
tions cause very large foliage attenuation and
greater angular spread. In [29], three kinds of
existing foliage attenuation models, such as
empirical models, semiempirical models, and
China Communications • May 2019
8
Fig. 2. Cluster-level representation of the mmWave wideband channel in the time
and angular domains
have similar expressions but different break-
point distances. Therefore, it is necessary to
collect a sufficient amount of field channel
measurement data at random RX positions or
ray-tracing simulation data for future cellular
network designs.
2) Time-Spatial Propagation Character-
istics and Model
Multipath effects result in fast fading in
mmWave channels, where received power
not in the boresight path occurs through dif-
fraction over rooftops and around building
corners, reflection from walls and furniture,
and scattering from physical objects. Although
several clusters can be found, most of them
are very faint and even negligible in the sense
of contributed average power. Figure 2 gives
a generalized description of the time-spatial
channel characteristics, where effective clus-
ters are sparse in time and space. The simpli-
ed combined CIR is given by
ht e t
omni LoS
+ −−
⋅ Θ−Θ −Θ ⋅ Φ−Φ −Φ
δδ
∑∑
kl
K
= =
(
(
11
,, ,,
L
ΘΦ = ΘΦ
k
α δ ττ
kl k kl
k kl k kl
,,
)
et
j
ϕ
αδ
kl,
,,
(
)
j
ϕ
LoS
(
(
)
)
)
, (13)
where Θ,
=
(
θφ
AoD AoD
)
and Φ,
=
(
θφ
AoA AoA
)
represent the vectors of the azimuth and el-
evation AoD and AoA, respectively;
τ
k, Θ
k,
and Φ
k denote the delay, AoD, and AoA of the
kth cluster centroid relative to the LoS path,
respectively;
α
kl,,
τ
kl,, Θ
kl,, and Φ
kl, denote the
amplitude, delay, and two angles of the lth
ray relative to the kth cluster center, respec-
tively; and
τ
kl, denotes the random phase. The
existing statistical mmWave channel models
require the statistics of these modeling param-
eters to be obtained. The root mean square
(RMS) delay spread is defined as the square
root of the second central moment, which
increases with the number of received paths
characterized by the relative propagation de-
lay. The RMS angular spread has a similar
definition to the RMS delay spread, where it
rst needs to calculate the received 2D power
angle profile (PAP) by synthesizing the 3D
PAP in terms of azimuth or elevation. To opti-
can be used for backhaul links among small
cells and the connection between microwave
and mmWave BSs in HetNet combined with
directional transmission. However, in NLoS
scenarios, LoS propagation exhibits different
propagation mechanisms, where reflections
and diffractions cause difficulties in absolute
propagation delay measurements and various
angular dispersion characteristics. For users in
LoS scenarios with shorter TX-RX separation
distances, it is reasonable to employ wide-
beam antennas, while multibeam high-gain
array antennas are required to enhance and
track multipaths in NLoS scenarios. A basic
LoS probability model above 6 GHz follows
an exponential-like distribution as a function
of the 2D distance d2D, i.e.,
= − <≤
Pd
LoS 2 D
1,
γ
exp ,
(
exp ,
)
−<
d2D 1
d
β
2D 2
−
1
β
α
−
2
α
dd d
dd
dd
1 2D 2
2D 1
2 2D
≤
, (12)
where modeling parameters are layout-relat-
ed and frequency-independent. However, the
model requires modication in densely built-
up urban environments. In [9], recent advanc-
es in mmWave LoS probability models for
UMi and UMa scenarios are reviewed, which
k
the kth cluster
,kl
t
,
,
kl
j
kl
e
LoS
j
e
LoS Path
LoS path
the kth cluster
centroid
,,ht
,
kk
,,
,
kl kl
LoS LoS
,
the lth subpath
in kth cluster
China Communications • May 2019 9
PERFORMANCE E
ANALYSES
Understanding propagation characteristics is
fundamental to mmWave system deployments.
This section reviews several recent outdoor
mmWave channel measurement and modeling
results in various outdoor scenarios, high-
lighting the associated network deployment
considerations to enable coverage. Examples
include antenna congurations, transceiver sit-
tings, and measurement environments.
Three typical outdoor scenarios are taken into
consideration. In UMi, BSs are mounted be-
low rooftop levels of surrounding buildings
with smaller widths, indicating that seamless
coverage can be realized under the O2I condi-
tion. In UMa, larger inter-site distances, with
BS heights above 25 m, are expected due to
the rich scattering environments. In RMa, di-
rectional transmission can compensate for the
large attenuation caused by vegetation block-
mize the conguration of the massive MIMO
and beamforming at the BS, double-direc-
tional channel models consider both AoA and
AoD information. A channel exhibiting a large
angular spread is normally called a rich-scat-
tered channel. From a graph-theoretical per-
spective, the benets of rich scattering lead to
the full-rank channel matrix, corresponding
to the linear scaling up of the MIMO channel
capacity [32]. In contrast, a smaller angular
spread indicates that fewer clusters can be
detected in the absence of scattering. Thus,
the beamforming scheme is more efcient for
poor-scattered channels, where the designs of
the beamwidth and lter for each RF chain are
determined by intracluster parameters. In [8],
the correlations among large-scale dispersion
parameters were ignored, and time-spatial
statistics were computed through decoupled
PDPs and PAPs, which can include, for exam-
ple, sub-paths in a time cluster travel similar
propagation delay that may arrive from many
spatial lobes. Therefore, it is necessary to use
the joint estimation of MPCs for intracluster
delay and angle parameter calculation.
Table I. Large-scale channel parameters in path loss and shadow fading models in UMi and UMa scenarios.
Measurement conditions CI model
Ref.
Area Scen. f
[GHz]
hT
[m]
hR
[m]
Range
[m]
TX
Ant.
RX
Ant. PLE σCI
[dB]
UMi Campus
LoS 28/38 5 1.7 14-113 horn horn 2.7/3.1 4.2/4.0 [33]
NLoS 28/38 5 1.7 14-113 horn horn 3.7/3.7 6.4/5.6
NLoS 28 15 1.5 98-270 sector Omni 2.7 7.2 [35]
LoS 38 8/23/36 1.5 >200 horn horn 1.9 4.6 [8]
NLoS 38 8/23/36 1.5 >200 horn horn 3.3 12.3
UMi
Street Canyon
LoS 28 7/17 1.5 31-54 horn Omni 2.1 3.5 [34]
NLoS 28 7/17 1.5 61-186 horn Omni 3.4 9.7
LoS 29 6.5 1.5 35-256 Omni Omni 2.2 4.4 [36]
NLoS 29 6.5 1.5 35-256 Omni Omni 3.1 8.2
UMi Downtown NLoS 28 40 1.5 110-135 sector Omni 4.1 7.1 [8]
UMa
Downtown LoS 28 8/17 NA 55-207 sector Omni 2.1 3.36 [37]
UMa
Open Area
LoS 29 NA NA 35-256 Omni Omni 2.7 5.7 [36]
NLoS 29 NA NA 35-256 Omni Omni 3.4 8.0
UMa
Parking Structures
LoS 29 NA NA 35-256 Omni Omni 2.9 21.0 [36]
NLoS 29 NA NA 35-256 Omni Omni 3.44 10.5
* hT and hR represent the heights of the TX and RX antennas, respectively.
China Communications • May 2019
10
Fig. 3. Channel measurements in the RMa villa district and the measured path loss, which varies with the RX location.
Fig. 4. Channel measurements in the UMi street canyon and an example of the joint azimuth AoA-AoD power spectra in a LoS link.
frequency under the same measurement con-
dition, ii) considerably lower PLEs and shad-
owing factors for LoS links over NLoS links,
and iii) higher BS positions corresponding to
coverage scaling. In addition, the cases with
PLEs smaller than 2 only appear for direction-
al propagation when using narrow beamwidth
horn antennas scanned in 3D space to detect
the strongest path. Based on a more detailed
study of the beam space representation of the
channel parameters, it is cost-efcient to em-
ploy large arrays with wider beams for user
equipment (UE) close to the BS and multiple
narrow beams for cell-edge users. For RMa,
age, although this scenario has not been thor-
oughly investigated in the 3GPP TR 38.900
model [20].
Omnidirectional and directional CI path
loss models in UMi and UMa are presented,
as described by the parameters in Table 1,
where we mainly focus on the two most po-
tential 5G mmWave bands (e.g., the 28 GHz
and 38 GHz bands). For convenience, the
measurement configurations of the TX and
RX heights, measurement areas, and antenna
scanning ranges are also given in Table 1. The
main conclusions from these studies are as fol-
lows: i) a general increase in the PLE versus
1
2
3
465
7
8
9
10
12 11
13
15
14
16
23
20
21
22
18
19
17
29
27
28
30
25
24
26
31
LoSLocation
NLoSLocation
TX
30°
Path Loss [dB]
Aroundthe
farthestvilla
Withthestepof10m
LoSpath
Themax.distance200m
Aroundthenearestvilla
(b) Measured path loss(a) Layout of the TX and RX locations
TX1
60°
LoS Location
NLoS Location
(a) Layout of the TX and RX locations
(b) 25.5 GHz @ RX 9
(c) 39.5 GHz @ RX 9
China Communications • May 2019 11
which is problematic in studying the impacts
of measurement setups, surroundings and
frequencies on coverage. To solve these prob-
lems, more than 254 effective TX-RX combi-
nations, with thousands of directional PDPs
for long-range outdoor links, such as the UMi
street canyon, UMa modern business center,
UMa residential area, and RMa villa district,
are investigated at 25.5, 28, 39, and 39.5
GHz, with a 300 MHz bandwidth. Measure-
ments in the UMi street canyon, as shown in
gure 4, conrm that the received signals are
mainly concentrated in the boresight direction
in poorly scattered environments. Thus, it is
necessary to simulate the procedure of beam
scanning and detect the optimal path. Figure
5 depicts the measured PAPs in the UMa
modern business center using wide-beam
antenna with two different pointing angles
for sector coverage. The cooperation between
adjacent sectors reduces the outage proba-
bility of the dense mmWave networks. In the
UMa residential area, the height of the BS is
above 50 m, which is much higher than that
of the surrounding buildings, indicating that
the diffraction over the rooftops and corners
of buildings increases the distance of NLoS
links. As shown in Figs. 6(b) and (c), reec-
tion from the ground is negligible, while the
paths from the opposite building are reason-
able for eliminating the blindness area (e.g.,
RX 1 and RX 5 to RX 36 in figure 6(a)).
there were very few mmWave channel mea-
surement campaigns due to the limitation of
the sounding dynamic range, except for a 73
GHz RMa channel measurement conducted
by New York University (NYU) WIRELESS,
which can provide directional transmission
with a distance over 10 km [38]. Hence, it is
essential to promote additional channel sound-
ing activities in RMa scenarios at the two
abovementioned 5G mmWave bands. Recent-
ly, a channel sounding campaign at 28 GHz
and 39 GHz in the RMa villa district using a
TX antenna height of 15 m for sector coverage
has been promoted. The TX-RX combinations
are almost blocked by vegetation and villas, as
shown in gure 3 (a). Figure 3 (b) shows the
eld-measured path loss versus the RX posi-
tions, where the cell radii of the 39 GHz chan-
nel are smaller than those of the 28 GHz chan-
nel, with an 8 dB SNR detection threshold.
The path loss for the LoS path exhibits similar
behavior to the 10 dB path loss increase com-
pared with the previous results, while there is
a slight difference in the OLoS link via foliage
blockage. Large-scale path loss measurement
results in UMi, UMa, and RMa reveal that
mmWave outdoor cellular networks can pro-
vide reliable connections, with cell radii up to
200 m.
The statistics of composite or cluster-level
small-scale channel parameters have been es-
timated under different conditions [8, 35-37],
Table II. Statistics of the composite delay and angular spread for the UMi and UMa scenarios.
Measurement condition Delay spread [log10([s])] Angular Spread [log10([˚])]
Area Scen. f [GHz] Min Max Mean Std Min Max Mean Std
UMi
Street canyon
LoS 25.5 -7.82 -7.08 -7.44 0.23 0.63 1.43 0.98 0.25
LoS 39.5 -7.73 -7.03 -7.22 0.23 0.63 1.20 0.92 0.22
NLoS 25.5 -7.41 -7.07 -7.19 0.13 1.04 1.80 1.47 0.26
UMa
Modern business
center
LoS 28 -7.42 -6.36 -6.73 0.26 1.45 2.04 1.84 0.20
LoS 39 -7.42 -6.32 -6.81 0.33 0.73 2.05 1.77 0.30
NLoS 28 -7.53 -6.77 -7.00 0.24 1.14 2.08 1.76 0.30
NLoS 39 -7.63 -6.77 -7.28 0.23 0.69 2.12 1.71 0.34
UMa
Residential area
Both 28 -7.67 -6.35 -6.87 0.29 1.06 2.21 1.84 0.27
Both 39 -7.55 -6.35 -6.96 0.34 0.73 2.21 1.78 0.43
RMa
Villas district
Both 28 -7.98 -6.77 -7.32 0.28 0.45 2.20 1.89 0.41
Both 39 -7.79 -6.69 -7.38 0.27 1.20 2.22 1.97 0.29
China Communications • May 2019
12
Fig. 5. Channel measurements in a UMa modern business center and PAPs with different sector coverages in the LoS link.
Fig. 6. Channel measurement in a UMa residential area and time-spatial propagation analyses in coverage blind spots.
ed due to the near-far effect. The mean values
of the angular spread in the RMa villa district
are larger than those in other scenarios due
to foliage blockage. These findings suggest
that propagation environments play a role in
enabling coverage via the adaptive selection
of LSAA-based transmission technology.
The values of the BEL and the corresponding
statistics can be used to plan the indoor recep-
Meanwhile, it can be observed that over 7
dB, additional loss exists relative to a 2.9 dB
free space loss increase in L fd
FS 0
(
,
)
between
the 39 GHz and 28 GHz channels. This is due
to greater reflection and diffraction loss at
higher frequency bands. The statistics of the
RMS delay and angular spread for the four
concerned scenarios are provided in Table 2,
showing that UMa is a rich-scattered scenario
with a larger delay and angular spread, and in
the UMi scenario, fewer MPCs can be detect-
(a) Layout of the TX and RX locations
20°
40°
CBD-TX
Sector1
Sector2
1
2
8
3
5
4
6
7
9
10
11
17
16
15
14
13
12 18 19 20
37
29
28
27
26 25
24
23
22
21
32
30
36
35
34
33
31
38
39
45
44
43
42
41
40
46
LoS Location
NLoS Location
(b) 28 GHz PAP @ RX 18 (c) 39 GHz PAP @ RX 18
20°
40°
DRA-TX
Sector1
Sector2
1
2
4
3
5
7
8
6
9
10
11
12
13
14
15
16
17
19
18
21
20
29 28 27 26 25 24 23 22
32
30
31
33
34
35
36
LoS Location
NLoS Location
(b) 28 GHz PDAP @ RX1
(c) 39 GHz PDAP @ RX1(a) Layout of the TX and RX locations
China Communications • May 2019 13
MMWAVE
WN
MmWave propagation characteristics are
unique compared with conventional micro-
wave bands below 6 GHz. Based on the dis-
cussion of the field measurement results in
several outdoor environments across the most
potential 5G bands, the combined implications
on the design of mmWave cellular networks
using LSAAs are provided, with a focus on
coverage extension and system optimizations.
The significant path, penetration, and block-
age losses in the mmWave bands suggest that
large-scale antenna systems are necessary to
compensate for large propagation attenuation
and enable multiple parallel data streams, cor-
responding to the increases in the cell radius
and channel capacity, respectively. In general,
the antenna array gain in the boresight direc-
tion mainly depends on the number of antenna
elements N and the radiation pattern of the
element. The total gain of the antenna array on
the dBi scale is given by [42]
GG N
T/R element 10
=++
+10log10
10log 10 log
η
10
ξ
, (13)
where Gelement denotes the element gain in dBi,
10log10N represents additional antenna gain
from the array,
ξ
represents the array effi-
ciency, and
η
reflects the capability of beam
tracing. Under the assumption of an antenna
element gain by the TX and RX of 5 dBi, ap-
proximately 44 elements are required to obtain
a total antenna gain of 20 dBi, with
ξ
=0.9
and
η
=0.8. Hence, tens of antenna elements
are necessary to achieve the same cell radius
as those of previous channel sounding cong-
urations.
To enable the desired coverage and data
rate, the link budget is a basic factor for the
mmWave ultradense cellular network design.
From a radio propagation perspective, the
basic element of a downlink budget for an
LSAA-based mmWave system is given by
tion of outdoor cellular services when users
move from the outdoors to indoors with the
requirement of continuous coverage. The mea-
surement results show that an external wall or
coated glass can attenuate mmWave signals by
at least 20 dB, while only 3-5 dB can attenuate
mmWave signals for a standard single-layer
glass window. It is concluded that the UE in
indoor environments is not isolated from out-
door cells when buildings are equipped with
low-loss glasses and have a large window-to-
wall ratio. A cost-efcient deployment of the
mmWave cooperative networks to enable O2I
coverage is to employ high-gain multibeam
antennas and beamforming techniques in out-
door BSs and wide-beam antennas for indoor
access points (APs). Hence, further BEL and
penetration loss measurements with outdoor
and indoor access are necessary for system de-
sign and performance evaluation.
There are new constraints on the ultradense
cellular network deployment arising from
smaller cell radii, larger power consumption,
and higher costs. Backhaul technology in
mmWave bands provides a high-performance
solution to forward massive backhaul traffic
in the core network. The performances of the
outdoor point-to-point (P2P) fixed backhaul
links are, therefore, investigated at 60 GHz
[39] and 73 GHz [40]. By utilizing mechani-
cally steerable horn antennas, the large-scale
characteristics of the directional channel are
analyzed, where the PLEs and
σ
CI for the
outdoor LoS scenario in these two bands are
2.1/3.4 dB and 2.52/4.00 dB, respectively, and
those for the NLoS scenario are 3.7/4.8 dB
and 5.7/13.5 dB, respectively. Path loss mea-
surement results have proved that P2P connec-
tivity at a higher frequency band is viable, but
it requires dual-band deployment combined
with a 28 GHz or 38 GHz access link in a
small cell. Hence, the performance evaluation
of the backhaul links using the 28 GHz and 38
GHz channels should be carried out from the
perspective of outage probability, intersymbol
interference (ISI), and energy consumption in
the near future.
China Communications • May 2019
14
uplink and downlink in the UMa using 128
and 4 antenna elements at the BS and UE is
shown in Table 3. To successfully establish the
link at the desired throughput, sufcient SNR
is required to detect the received multipath
signals, which can be calculated by Shannon’s
channel capacity. The descriptions of the other
system default values can be found in [42].
Based on the basic CI path loss models in (4)
and the modeling results in Table 1, to achieve
a downlink data rate of 1.0 Gbps in a 500
MHz bandwidth for edge users, the maximum
cell radii are 358.7 m and 182.0 m for LoS
and NLoS scenarios, respectively, while much
smaller coverages are evident for uplinks
with a 0.5 Gbps data rate. For the 39 GHz
channel, larger antenna arrays are necessary
to provide sufficient link margins due to the
different channel characteristics and hardware
constraints. The basic link budget designed
here indicates that single-cell mmWave cellu-
lar networks using a large number of antenna
arrays can provide credible connections and
desired data rates for cell-edge users, with a
range up to approximately 200 m.
L fd P G L f L
MA ( ,) ( )
c T T add c mar
=+− −
+−G Sf
R in ()
BW
, (14)
where L fd
MA c
(
,
)
epresents the maximum al-
lowed path loss at fc GHz, with d m of the 3D
separation distance, which can further consider
the BS height-dependent factor; P
T represents
the maximum transmitted power per traffic
channel in dBm; GT and GR represent trans-
mitter and receiver antenna array gains in dBi,
which are calculated using (13), respectively;
and Lf
add c
( )
represents the additional penetra-
tion loss caused by buildings, human bodies,
foliage, and vehicles, which is a function of
the carrier frequency; Lmar represents the fading
margin, which determines the performance of
the links, and Sin represents the receiver sensi-
tivity related to the RF bandwidth fBW, which
can be calculated by the information rate, re-
quired SNR, receiver noise gure and density.
On the other hand, the uplink budget shares
similar procedures with different transceiver
congurations. The site-specic default values
of the propagation channel in can be extracted
from the mmWave channel measurement re-
sults. Here, a basic link budget for the 28 GHz
Table III. Link budget for an mmWave communication system in the UMa scenario.
Variable Unit Uplink Downlink Uplink Downlink
System
Parameter
Carrier Frequency GHz 28 39
Data Rate Gbps 0.5 1.0 0.5 1.0
Bandwidth GHz 0.5 0.5 0.5 0.5
Transmitter
Max TX Power per Channel dBm 15 20 15 20
Number of TX Antennas - 4 128 4 512
TX Antenna Element Gain dBi 5 5 5 5
EIRP dBm 24.6 44.7 24.6 50.7
Receiver
Number of RX Antennas - 128 4 512 4
RX Antenna Element Gain dBi 5 5 5 5
Noise Figure dB 7 7 8 8
Thermal noise dBm -80.0 -80.0 -79.0 -79.0
SNR dB 0.0 4.8 0.0 4.8
Receiver Sensitivity dBm -80.0 -75.2 -79.0 -74.2
Loss Ladd dB 6 3 7 4
Lmar dB 3 3 3 3
Coverage
MAPL dB 120.3 123.5 124.3 127.5
Radius (LoS) m 260.4 358.7 183.7 249.7
Radius (NLoS) m 137.1 182.0 103.0 135.2
China Communications • May 2019 15
sion, which has advantages in reducing the
space-time processing complexity and enhanc-
ing the coverage in NLoS scenarios and blind
zones through reection and diffraction paths.
In O2I scenarios, indoor received signals
mainly focus on the direction from windows,
leading to smaller penetration losses and angu-
lar spreads. A viable system design, thus, must
carefully consider the structure of the external
wall of a building, such as the window-to-wall
ratio and material characteristics.
C
LSAA-based 5G mmWave communications
are expected to provide users with ultrahigh
data rates, ultralow latency, and better wire-
less coverage. An overview of the emerging
challenges and requirements in mmWave
ultradense network deployments has been
presented. Based on the observations of re-
cent mmWave channel models from various
standard bodies and measurement activities,
appropriate large-scale path loss models and
small-scale cluster models of LSAA-based
mmWave systems have been given, respec-
tively. The channel measurement results in
two 5G candidate bands have shown that,
using directional transmissions and high-gain
antennas, mmWave systems can provide stable
connections and achieve decent ranges up to
hundreds of meters. The mmWave uplink and
downlink budgets have been combined with
the proposed outdoor path loss models to opti-
mize RF system congurations. Furthermore,
it is clear that there is a need to adaptively
select massive MIMO transmission schemes
based on the angular spread characteristics.
ACKNOWLEDGEMENT
The authors would like to thank the editors
and the reviewers for their detailed reviews
and constructive comments, which have
helped improve the quality of this article.
This work was supported in part by the Na-
tional Natural Science Foundation of China
under Grant No. 61671145, and the Key R&D
Thus far, a large number of antennas have
been proven reasonable to simultaneously
extend the coverage of mmWave cellular net-
works and offer the tradeoff between spectral
efciency and energy efciency for imperfect
CSI in complex propagation environments.
Based on our outdoor channel measurement
results, the differential impact of the 39 GHz
channel in terms of link margins relative to
the 28 GHz channel is dramatically worse,
especially in NLoS scenarios, showing that
single-cell LSAA-based systems operating at
28 GHz are previously deployed in outdoor
environments with larger coverage radii. For
the installation of 39 GHz radio systems, a
larger transmission power and a number of
antenna elements are required to complete the
expected QoS. Under the limitation of a nite
numbers of TX antennas, NT, and RX anten-
nas, NR, the channel capacity of traditional
MIMO systems is closely related to that of the
propagation environments, where an increase
in the rank of the channel gain matrix appears
in rich scattering environments. Meanwhile,
the distribution of the singular values of the
channel matrix approaches a deterministic
function when the MIMO arrays are large. The
very tall or very wide channel matrixes, cor-
responding to the increases in NN
TR
/, tend to
be very well conditioned and have benets for
some matrix operations.
Physically, in LoS scenarios, antenna arrays
form a beam toward the intended receiver,
with an increased eld strength in a certain di-
rection; however, in NLoS scenarios, received
signals are sparse in terms of both azimuth
and elevation, where the angles of departure
and arrival depend on scattering from diffuse
reection. In poor scattering environments, it
is more complicated to adaptively direct the
antenna array toward a certain user (i.e., beam
squint), which can result in signicant degra-
dation in performance. Hence, a key channel
dispersion factor, Δch, has been dened in [41],
which represents the number of domain beams
needed in a beam space to account for disper-
China Communications • May 2019
16
na systems for 5G wireless communications,”
Science in China Series F: Information Sciences,
vol.59, no. 8, 2016, pp. 1-18.
[15] T. Xie, et al., “Geometric mean decomposition
based hybrid precoding for millimeter-wave
massive MIMO,” China Communications, vol. 15,
no. 5, 2018, pp. 229-238.
[16] B. Yang, et al., “Digital beamforming-based
massive MIMO transceiver for 5G millime-
ter-wave communications,” IEEE Transactions on
Microwave Theory and Techniques, vol. 66, no. 7,
2018, pp. 3403-3418.
[17] K. Zheng, et al., “10 Gb/s HetNets with millime-
ter-wave communications: Access and network-
ing challenges and protocols,” IEEE Communi-
cation Magazine, vol. 53, no. 1, 2015, pp. 222-
231.
[18] G. Yang, et al., “Low-latency heterogeneous net-
works with millimeter-wave communications,”
IEEE Communication Magazine, vol. 56, no. 6,
2018, pp. 124-129.
[19] mmMAGIC (May 2017). Measurement results
and nal mmWave channel models, H2020-ICT-
671650-mmWave/D2.2 v1.0. [Online]. Available:
https://5g-mmWave.eu/ results/
[20] 3GPP, “Study on channel model for frequen-
cy spectrum above 6 GHz,” Sophia Antipolis,
France, Rep. 38.900, May 2017. [Online]. Avail-
able: http://www.3gpp.org/DynaReport/ 38900-
CRs.htm
[21] G. R. MacCartney, et al., “Indoor office wide-
band millimeter-wave propagation measure-
ments and channel models at 28 and 73 GHz
for ultra-dense 5G wireless networks,” IEEE Ac-
cess, vol. 3, 2015, pp. 2388-2424.
[22] M. Sasaki, et al., “Path loss characteristics at 800
MHz to 37 GHz in urban street microcell envi-
ronment,” Proc. EuCAP, 2015, pp. 1-4.
[23] A. Karttunen, et al., “path loss models with
distance-dependent weighted tting and esti-
mation of censored path loss data,” IET Micro-
waves, Antennas & Propagation, vol. 10, no. 14,
2016, pp. 1467-1474.
[24] R. Zhang, et al., “Antenna deembedding for
mmWave propagation modeling and eld mea-
surement validation at 73 GHz,” IEEE Transac-
tions on Microwave Theory and Techniques, vol.
65, no. 10, 2017, pp. 3648-3659.
[25] H. Zhao, et al., “28 GHz millimeter wave cellular
communication measurements for reflection
and penetration loss in and around buildings in
New York City,” Proc. IEEE ICC, 2013, pp. 5163-
5167.
[26] G. R. MacCartney, et al., “Millimeter-wave hu-
man blockage at 73 GHz with a simple double
knife-edge diraction model and extension for
directional antennas,” Proc. IEEE VTC-Fall, 2016,
pp. 1-6.
[27] X. Chen, et al., “Modelling of human body shad-
owing based on 28 GHz indoor measurement
Program of Jiangsu Province of China under
Grant BE2018121.
References
[1] J. G. Andrewset, et al., “What will 5G be?” IEEE
Journal on Selected Areas in Communications,
vol. 32, no. 6, 2014, pp. 1065-1082.
[2] T. S. Rappaport, et al., “Millimeter wave mobile
communications for 5G cellular: It will work!”
IEEE Access, vol. 1, 2013, pp. 335-349.
[3] M. Xiao, et al., “Millimeter wave communica-
tions for future mobile networks,” IEEE Journal
on Selected Areas in Communications, vol. 35,
no. 9, 2017, pp. 1909-1935.
[4] L. M. Li, et al. “MmWave communications for
5G: Implementation challenges and advances,”
Science China Information Sciences, vol. 61, no.
2, 2018: 021301.
[5] P. Zhang, et al., “A survey of testing for 5G:
Solutions, opportunities, and challenges,” China
Communications, vol. 16, no. 1, 2019, pp. 69-85.
[6] F. Sohrabi, W. Yu, “Hybrid digital and analog
beamforming design for large-scale antenna
arrays,” IEEE Journal of Selected Topics in Signal
Processing, vol. 10, no. 3, 2016, pp. 501-513.
[7] X. Zhao, et al., “Channel measurements, mod-
eling, simulation and validation at 32 GHz in
outdoor microcells for 5G radio systems,” IEEE
Access, vol. 5, 2016, pp. 1062-1072.
[8] T. S. Rappaport, et al., “Wideband millime-
ter-wave propagation measurements and chan-
nel models for future wireless communication
system design,” IEEE Transactions on Communi-
cations, vol. 63, no. 9, 2015, pp. 3092-3056.
[9] T. S. Rappaport, et al., “Overview of millimeter
wave communications for fth-generation (5G)
wireless networks-with a focus on propagation
models,” IEEE Transactions on Antennas and
Propagation, vol. 65, no.12, 2017, pp. 6213-
6230.
[10] S. Rangan, T. S. Rappaport, E. Erkip, “Millime-
ter-wave cellular wireless networks: Potentials
and challenges,” Proceedings of the IEEE, vol.
102, no. 3, 2014, pp. 366–385.
[11] F. Sohrabi, W. Yu, “Hybrid analog and digital
beamforming for mmWave OFDM large-scale
antenna arrays,” IEEE Journal on Selected Areas
in Communications, vol. 35, no. 7, 2017, pp.
1432-1443.
[12] X. Gao, et al., “Unifying eigen-mode MIMO
transmission,” Science in China Series F: Infor-
mation Sciences, vol.52, no. 12, 2009, pp. 2269-
2278.
[13] W. Hong, et al., “Multibeam antenna technol-
ogies for 5G wireless communications,” IEEE
Transactions on Antennas and Propagation,
vol.65, no. 12, 2017, pp. 6231-6249.
[14] D. Wang, et al., “An overview of transmission
theory and techniques of large-scale anten-
China Communications • May 2019 17
[42] I. A. Hemadeh, et al., “Millimeter-wave com-
munications: Physical channel models, design
considerations, antenna constructions, and
link-budget,” IEEE Communications Surveys &
Tutorials, vol. 20, no. 2, 2018, pp. 870-913.
Haiming Wang, received the
B.S., M.S., and Ph.D. degrees in
electronic engineering from
Southeast University, Nanjing,
China, in 1999, 2002, and 2009,
respectively. He joined the
State Key Laboratory of Milli-
meter Waves, Southeast Uni-
versity, in 2002. Now he is a Professor. His current re-
search interests include the radio propagation mea-
surement and channel modelling, millimeter-wave
antennas and arrays, and millimeter-wave wireless
communications. Dr. Wang received the first-class
Science and Technology Progress Award of Jiangsu
Province of China in 2009. He served as the vice chair
of IEEE 802.11aj task group from September 2012 to
July 2018. Email: hmwang@seu.edu.cn.
Peize Zhang, received the B.S.
degree from Beijing University
of Posts and Telecommunica-
tions, Beijing, China, in 2015,
and the M.S. degree from Chi-
na Academy of Telecommuni-
cations Technology, Beijing,
China, in 2018, both in elec-
tronic engineering. He is currently pursuing the Ph.D.
degree at State Key Laboratory of Millimeter Waves,
Southeast University, Nanjing, China. His current re-
search interests include millimeter-wave radio propa-
gation measurements and channel modeling. Email:
pzzhang@seu.edu.cn.
Jing Li, received the B.S. and
M.S. degree in electronic engi-
neering from Southeast Uni-
versity, Nanjing, China, in 2016
and 2019, respectively. Now
she is an Engineer with ZTE
Corporation, Shanghai. Her
current research interests in-
clude automatic channel sounder design and milli-
meter-wave channel modeling. Email: jing_li@seu.
edu.cn.
Xiaohu You, has been with
National Mobile Communica-
tions Research Laboratory at
Southeast University since
1990, where he held the rank
of professor, the Chair Profes-
sor of Cheung Kong Scholars
results,” Proc. IEEE VTC-Fall, 2016, pp. 1-5.
[28] X. Zhao, et al., “Attenuation by human bodies at
26- and 39.5-GHz millimeter wavebands,” IEEE
Antennas and Wireless Propagation Letter, vol.
16, 2017, pp. 1229-1232.
[29] H. M. Rahim, C. Leow, T. A. Rahman, “Millimeter
wave propagation through foliage: Comparison
of models,” Proc. IEEE ICCM, 2015, pp. 236-240.
[30] F. Wang, K. Sarabandi, “An enhanced milli-
meter-wave foliage propagation model,” IEEE
Transactions on Antennas and Propagation, vol.
53, no. 7, 2005, pp. 2138-2145.
[31] Z. Shen, et al., “Measurement-based vehicle ex-
terior channel characteristics in 45-GHz band,”
Proc. APCAP, 2017, pp. 1-3.
[32] O. Souihil, T. Ohtsuki, “Benets of rich scattering
in MIMO channels: A graph-theoretical per-
spective,” IEEE Communications Letter, vol. 17,
no. 1, 2013, pp. 23-26.
[33] M. N. Hindia, et al., “Outdoor large-scale path
loss characterization in an urban environment
at 26, 28, 36, and 38 GHz,” Physical Communi-
cation, vol. 27, 2018, pp. 150-160.
[34] S. Sun, G. R. MacCartnet, T. S. Rappaport, “Mil-
limeter-wave distance-dependent large-scale
propagation measurements and path loss mod-
els for outdoor and indoor 5G systems,” Proc.
EuCAP, 2016, pp. 1-5.
[35] J. Ko, et al., “Feasibility study and spatial-tem-
poral characteristics analysis for 28 GHz out-
door wireless channel modelling,” IET Commu-
nications, vol. 10, no. 17, 2016, pp. 2352-2362.
[36] V. Raghavan, et al., “Millimeter wave channel
measurements and implications for PHY layer
design,” IEEE Transactions on Antennas and
Propagation, vol. 65, no. 12, 2017, pp. 6521-
6533.
[37] J. Ko, et al., “28 GHz channel measurements and
modeling in a ski resort town in Pyeongchang
for 5G cellular network systems,” Proc. EuCAP,
2016, pp. 1-5.
[38] G. R. MacCartney, T. S. Rappaport, “Rural mac-
rocell path loss models for millimeter wave
wireless communications,” IEEE Journal on Se-
lected Areas in Communications, vol. 35, no. 7,
2017, pp. 1663-1677.
[39] S. I. Sulyman, et al., “Eects of solar radio emis-
sions on outdoor propagation path loss models
at 60 GHz bands for access/backhaul links and
D2D communications,” IEEE Transactions on
Antennas and Propagation, vol. 65, no. 12, 2017,
pp. 6624-6635.
[40] S. Nie, et al., “28 GHz and 73 GHz signal outage
study for millimeter wave cellular and backhaul
communications,” Proc. ICC, 2014, pp. 4856-
4861.
[41] J. Brady, A. Sayeed, “wideband communication
with high-dimensional arrays: New results and
transceiver architectures,” Proc. ICC, 2015, pp.
1042–1047.
China Communications • May 2019
18
tions R&D Activities. From 2001-2006, he was the
Principal Expert of the national 863 FuTURE Project.
He has been Section Chair of IEEE Nanjing Section
since 2010 and IEEE Fellow since 2011. He served as
the general co-chair of IEEE WCNC 2014 and the
general chair of IEEE ICC 2019. His current research
interests focus on wireless and mobile communica-
tion systems and modern digital signal processing.
Email: xhyu@seu.edu.cn
Program and served as Director of the laboratory. He
has contributed over 50 IEEE journal papers and 2
books in the areas of adaptive signal processing,
neural networks and their applications to communi-
cation systems, and holds over 80 patents. He was
the Premier Foundation Investigator of the China
National Science Foundation. From 1999 to 2002, he
was the Principal Expert of the C3G Project, responsi-
ble for organizing China’s 3G Mobile Communica-
