1

Millimeter Wave Communications for Future

Mobile Networks
Ming Xiao, Senior Member, IEEE, Shahid Mumtaz, Senior Member, IEEE,
Yongming Huang, Senior Member, IEEE, Linglong Dai, Senior Member, IEEE, Yonghui Li, Senior Member, IEEE,
Michail Matthaiou, Senior Member, IEEE, George K. Karagiannidis, Fellow, IEEE, Emil Björnson, Member, IEEE,
Kai Yang, Senior Member, IEEE, Chih-Lin I., Senior Member, IEEE, Amitava Ghosh, Fellow, IEEE
arXiv:1705.06072v1 [cs.IT] 17 May 2017

Abstract—Millimeter wave (mmWave) communications have traffic volumes, with huge area spectral efficiency (hundreds
recently attracted large research interest, since the huge available of bit/s/Hz/km2 ) and the very high throughput per device
bandwidth can potentially lead to rates of multiple Gbps (gigabit (multiple Gbps). For instance, it is predicted that the world
per second) per user. Though mmWave can be readily used
in stationary scenarios such as indoor hotspots or backhaul, it monthly traffic of smartphones will be about 50 petabytes in
is challenging to use mmWave in mobile networks, where the 2021 [1], which is about 12 times of the traffic in 2016.
transmitting/receiving nodes may be moving, channels may have In order to meet these requirements, the research and
a complicated structure, and the coordination among multiple deployment for the future mobile networks [2]–[4] have
nodes is difficult. To fully exploit the high potential rates of
mmWave in mobile networks, lots of technical problems must
already been launched. Since 2013, the national-level 5G
be addressed. This paper presents a comprehensive survey of research organizations and projects (including European Union
mmWave communications for future mobile networks (5G and (EU) 5GPPP/METIS, China IMT-2020 (5G) Promotion Group,
beyond). We first summarize the recent channel measurement Korea 5G Forum, and Japan ARIB) have been set up one
campaigns and modeling results. Then, we discuss in detail recent after the other to achieve the 2020 technical targets. In
progresses in multiple input multiple output (MIMO) transceiver
design for mmWave communications. After that, we provide an
2015, ITU-R officially named 5G systems as IMT-2020,
overview of the solution for multiple access and backhauling, and released recommendation on its framework and overall
followed by analysis of coverage and connectivity. Finally, the objectives. Currently, Phase-1 of 5G is being standardized
progresses in the standardization and deployment of mmWave in 3GPP (http://www.3gpp.org/news-events/3gpp-news). Fig. 1
for mobile networks are discussed. illustrates potential usage scenarios and capabilities of IMT-
Index Terms—Millimeter wave communications, mobile net- 2020 [4]. Note that 5G is envisaged not only to expand and
works, channel model, MIMO beamforming, multiple access, support diverse usage scenarios and applications that will
standardization. continue beyond the current networks, but also to support a
broad variety of new application scenarios, including:
I. I NTRODUCTION AND BACKGROUND
1) Enhanced Mobile BroadBand (eMBB);

W ITH the fast development of electronic devices and

computer science, various emerging applications (e.g.,
virtual reality, augmented reality, big data analytics, artificial
2) Massive machine type communications (mMTC);
3) Ultra Reliable Low Latency Communication (URLLC).

intelligence, three-dimensional (3D) media, ultra-high defini- It is expected that IMT-2020 can provide the following
tion transmission video, etc.) have entered our society and eight key performance indicators (KPIs) [4]: greater than
created a significant growth in the data volume of wire- 10 Gbit/s peak data rate, 100 Mbit/s user-experienced data
less networks. Meanwhile, mobile networks have become rate, 3x spectrum efficiency, greater than 100 Mbps cell edge
indispensable to our society as a key service for personal- rates, 10 Mbit/s/km2 area traffic capacity, 100x network energy
computing devices. One of the main characteristics of fu- efficiency, 1 ms over-the-air latency, support for 500 km/h mo-
ture mobile networks (5G and beyond) is the unprecedented bility, and 106 km2 connection density [2]. The multiplicative
improvements are measured with respect to IMT-Advanced.
M. Xiao is with the department of information science and engineering, Recently, EU has also launched Beyond 5G research within the
School of Electrical Engineering, Royal Institute of Technology, Sweden.
Email: [email protected]; Shahid Mumtaz, is with Campus Universitario de
H2020 framework (ICT 2017-09 Call), where a key technology
Santiago, Portugal, Email: [email protected]; Yongming Huang, is with foundation is to use the millimeter wave (mmWave) bands
Southeast University, China, Email: [email protected]; Linglong Dai is from 30 GHz to 300 GHz, and also THz frequency bands.
with Tsinghua University, China, Email: [email protected]; Yonghui
Li, University of Sydney, Australia, Email: [email protected] To achieve the magnificent objectives and visions listed
Michail Matthaiou is with Queen’s University Belfast, UK, Email: above, several key enabling technologies have been identified,
[email protected]; George K. Karagiannidis is with Aristotle University such as mmWave communications, massive multiple-input
of Thessaloniki, Greece, Email: [email protected]; Emil Björnson is with
Linköping University, Sweden, Email: [email protected]; Kai Yang is with and multiple-output (MIMO), small cell deployment, full-
TongJi University, China, [email protected]; Chih-Lin I., China Mobile, duplex relaying, D2D communications, interference manage-
China, Email: [email protected]; Amitava Ghosh, Nokia, USA, Email: ment techniques, dynamic TDD with self-backhauling and
[email protected]
Ming Xiao would like to thank Mr. Zhengquan Zhang for valuable novel access technologies. Many of these technologies have
discussion and drawing some of the tables. complementary benefits and need to be combined to achieve all
 2

Enhanced Mobile Broadband

been devoted to mmWave communications research and
Peak data rate User experienced mmWave wireless local area networks (WLAN); for example,
(Gbit/s)
20
data rate (Mbit/s)
IEEE 802.11ad technology operating at 60 GHz, is already
100
IMT-2020 available. The more challenging development of mmWave
10
Area traffic capacity
1
mobile communications is ongoing and is what this paper
Spectrum
(Mbit/s/m²) 10
1 efficiency focuses on. Samsung first achieved 1 Gb/s data transmission
3x
0.1 1x
at 28 GHz in May 2013. Google also put substantial
research efforts into mmWave communications. Verizon
10x
1x has submitted applications to the Federal Communications
100x
Network energy
350400 Mobility
500 (km/h)
Commission (FCC) to obtain special temporary authorization
efficiency
IMT-Advanced (STA) to test mmWave communications technology at
10
105
Ultra-Reliable
28 GHz. T-Mobile is also expected to obtain STA at
106
Massive Machine Connectivity density
1
Latency
and Low Latency 28 GHz and 39 GHz. Nokia in collaboration with National
Type (devices/km²) (ms)
Communications
Instruments achieved a peak rate of 15 Gbps using their
Communications
proof-of-concept system at 73GHz band in April 2015.
To further promote the development of mmWave mobile
communications, Millimetre-wave evolution for backhaul and
access (MiWEBA) (http://www.miweba.eu/, a joint EU-Japan
Fig. 1. 5G usage scenarios and key capabilities of IMT-2020, as compared
to IMT-Advanced [2]. project), Beyond 2020 heterogeneous wireless networks
with mmWave small cell access and backhauling (MiWaves)
(http://www.miwaves.eu/), and mmWave based mobile radio
the key capabilities of 5G. For example, mmWave communica- access network for fifth generation integrated communications
tions [4], [5], [7], [9], [12], [13] is widely considered the most (mmMAGIC) (https://5g-mmmagic.eu/) projects have been
important technologies to achieve 10 Gbit/s peak data rates. initiated by EU. In Japan, to prepare for 2020 Tokyo
This is because there are a large amount of bandwidth available Olympic, DOCOMO and Ericsson tested at 15GHz to
in the mmWave bands, and expanding the bandwidth is an reach the rates of 4.5Gbps in outdoor environments and at
efficient approach to enhance system capacity. In particular, 70GHz to reach the rates of 2Gbps in indoor environments
the channel capacity of an additive white Gaussian noise (https://www.nttdocomo.co.jp/english/info/media_center/pr/201
channel operating over B Hz is 5/030203.html). In China, the minister of science and
! technology (MOST) has supported a few projects (through
P
C = B log2 1 + , (1) project 863) in mmWave in mobile networks and the
N0 B
RF chips of 60GHz and 42-48GHz have been produced
where P is the signal power and N0 is the noise power (http://www.most.gov.cn/kjbgz/201609/t20160923_127867.htm).
spectral density [15]. Hence, the capacity increases linearly Huawei and China mobile also demonstrated the Ka-band
with the bandwidth B, if we also let P grow proportionally (26.5–40GHz) mobile access with 20Gbps rates in the Mobile
to B. Since P is limited by regulations in practice, mmWave World Congress 2017.
communication is particularly well-suited for scenarios with The mmWave communication has a rich history. Recently
good channel conditions, such as short-range small cell access the main research interest on mmWave is shifting from local
and line-of-sight backhauling in mobile networks [16]–[19]. area networks to mobile networks. The main focus of this
Self-bcakhaul, where the same wireless spectrum is shared paper is on development of mmWave mobile communications.
between access and backhaul [20], [23], can provide flexi- We acknowledge the initial contribution in this field in the
ble and cost-efficient solutions to overcome the difficulty of 1990s and 2000s including channel measurement and system
deploying dedicated backhaul, especially in an ultra dense design, but we particularly concentrate on its new research
network (UDN) [24]–[26]. It refers to a set of solutions where contributions for future mmWave mobile networks. Note that
small base stations (BSs) without dedicated backhaul connect though there have been many research progresses in the
to other BSs that have dedicated backhaul link, by utilizing a area, the existing literature mainly focuses on addressing
similar radio access technology as the one used by the user specific technical challenges. A comprehensive survey, which
equipments (UEs) to access the network. organically combines these numerous but disjointed works and
The study of mmWave can trace back to more than 100 provides the summary of latest research progress, is basically
years ago. For instance, the experiments at wavelengths missing.
as short as 5 and 6 mm were performed by Bose and Note that references [7], [12]–[14], [16] gave excellent
Lebedew in the 1890s [27]. As for its application in radio overviews of some aspects of mmWave mobile networks from
communications, the mmWave mobile communications different aspects. Reference [7] is a pioneering overview paper
were originally invented in the 1990s and early 2000s, on the mmWave for mobile networks. Yet, as published in
including system design and channel measurements [28]– 2013, [7] did not contain the technology development of
[30]. Due to the fact that the spectral resource below very recent years. Moreover, some topics, such as, mmWave
6 GHz is becoming scaring while abundant bandwidth MIMO, multiple-access, backhual and standardization are not
is availabe at mmWave band, recently large efforts have included in [7]. Reference [12] is also an overview paper
 3

networks.
2
10
A. Main Technical Challenges
Despite the theoretical potentials for extremely high data
101 rates, there are several key technical challenges for using
Specific Attenuation (dB/km)

mmWave in mobile networks, including severe pathloss, high

penetration loss, high power consumption, blockage due to
100 shadowing, hardware impairment, etc. In what follows, we
will give a brief introduction to these topics.
Pathloss: In free-space transmission, the power of the
10-1 received signal (outside the Kirchhoff area) can be determined
by Friis transmission formula [7]:
!2
λ
10-2 Pr (d) = Pt Gt Gr d −n, (2)
4π
where Pt is the transmit power and Gt and Gr are the
antenna gains of the transmitter and receiver, respectively.
10-3
50 100 150 200 250 300Moreover, λ is wavelength, d is the transmission distance,
Frequency (GHz) and the pathloss exponent n equals to 2 in free space. The
formula in (2) can be also used to, approximately, describe
Fig. 2. Atmospheric and molecular absorption in different frequency bands the power of the received signal in non-free-space propagation
[35].
as well, by making channel measurements and then finding a
suitable value of n that approximately describes the pathloss
on mmWave 5G paper published in early 2014, which did measurements. The value of n is usually in the range from 2
not include recent progress of the topic. Moreover, topics to 6. There are also refined models, e.g., for cellular networks
of mmWave MIMO and standardization are not included in and n can in some scenarios be smaller than 2 [31].
[12]. Reference [16] gives a good brief overview of mmWave The wavelength of mmWave signals is much shorter than
for 5G. However, the introduction in [16] does not include conventional microwave communication signals, operating
technical details as a magazine paper. [14] is a recent overview at carrier frequency below 6 GHz. Hence, the pathloss of
paper on signal processing for mmWave MIMO. Yet, reference mmWave signals is much higher than that of microwave
[14] is focused in mmWave MIMO and other topics e.g., signals, if all other conditions including the antenna gains are
recent developments of channel measure/modeling, multiple the same. Although the pathloss of mmWave is generally quite
access, standardization, field test are missing. Reference [13], high, it is feasible to communicate over the distances that are
published in 2014, discussed the mmWave mobile in channel common in urban mobile networks, such as a few hundreds of
modeling, modulation, network architecture. Yet, [13] did meters [7] or even a few kilometers [32]. By using directive
not discuss the topics of mmWave MIMO, standardization, antennas, it has been demonstrated that 10 km communication
multiple access. Different from these papers, we provide such ranges are possible under clean air conditions [32]. If the air
a systematic survey on the main progress and technical content is not clean, the rain attenuation and atmospheric/molecular
of mmWave communications for mobile networks, and further absorption increase the pathloss and limit the communication
discuss some related research issues and challenges. Especially range [33], [34]. The impact of these factors varies with the
we include the recent technical development of related topics, carrier frequency; for instance, the atmospheric and molecular
including the accepted papers of this IEEE JSAC special issue. absorption are shown in Fig. 2 and the rain attenuation is
The rest of the paper is organized as follows. We first give shown in Fig. 3.
an overview of the key challenges and technical potentials of Penetration loss: The pathloss discussion assumes line-of-
mmWave communications in future mobile networks in Sec- sight (LoS) communications, but the high penetration loss is
tion II. Then, we discuss mmWave propagation characteristics compounded in non-line-of-sight (NLoS) scenarios. In indoor
and channel modeling in Section III, and then present MIMO environments, although the penetration losses for clear glass
design for mmWave communications in Section IV. Then, we and dry walls are relatively low for 28 GHz signals (compa-
discuss the multiple access technologies, in-band backhauling, rable to microwave bands), the penetration losses for brick
and the coverage performance in Section V. We present the and tinted glass are high for 28 GHz signals (about 28 dB and
progress of the standardization and deployment of mmWave 40 dB), which is much higher than at microwave bands [37],
mobile networks in Section VI. The conclusions are finally [112]. The penetration losses are typically larger at higher
given in Section VII. frequencies. Hence, it is difficult to cover inside with mmWave
nodes deployed outside and vice-versa due to high penetration
II. K EY C HALLENGES AND T ECHNICAL P OTENTIALS loss.
Next, we will describe some important challenges and po- High power consumption: In addition to the challenges
tential gains from using mmWave communications in mobile imposed by high pathloss, (1) shows that the transmit power
 4

elsewhere. This idealized radiation pattern, often referred as

102 the “flat-top model”, was used in [22], [38]–[40] for system-
level performance analysis. However, in practice, the radiation
patterns are more complicated and implementation-dependent;
the main-lobe gain is not constant and the side-lobe radiation
101
is non-zero. The effect of side-lobe radiation and the gradual
Rain Attenuation (dB/km)

reduction of main-lobe gain caused by beam misalignment

cannot be ignored. The maximum beamforming gain, which
100 can be achieved only if the main-lobes of the transmitter
0.25 mm/h
and receiver are perfectly aligned, is rare due to practical
2.5 mm/h implementation constraints. In 3GPP, a more practical two-
12.5 mm/h
dimensional directional antenna pattern [41] is adopted, where
25 mm/h
10-1
50 mm/h the antenna gain G(θ), with respect to the relative angle θ to
100 mm/h its boresight, is given by
150 mm/h


200 mm/hr

 

3 2θ 2 θm
10-2

 G m 10− 10 ω , |θ| ≤ ,
G (θ) =  2
50 100 150 200 250 300
Frequency (GHz) 


(3)
 G s,
θm
≤ |θ| ≤ π,
Fig. 3. Rain attenuation in different frequency bands [35].
 2

where ω denotes the half-power (3 dB) beamwidth and θ m is

the main-lobe beamwidth. G m and G s represent the maximum
needs to increase with the bandwidth if the signal-to-noise main-lobe gain and averaged side-lobe gain, respectively.
ratio (SNR) should remain intact [14]. Alternatively, directive Eq. (3) models how the signal changes with the beamwidth
antennas or MIMO technology can be used to direct the and receiving angles. The narrow beamwidth has a two-fold
signal power spatially, which leads to an array gain and impact. While the pros will be discussed later, for the cons,
latter also provides the flexibility of spatial multiplexing narrow beamwidth leads to higher sensitivity to misalignment
[36]. MIMO/beamforming is considered essential for mmWave between the transmitter and the receiver, especially in mobile
communications, particularly since the short wavelength at networks that should support high mobility. The reason for
mmWave frequencies makes it possible to fit many half- beam misalignment can be coarsely divided into two cate-
wavelength spaced antennas into a small area. MIMO arrays gories: 1) Imperfection of existing antenna and beamforming
are normally fully digital for sub 6 GHz systems, where techniques [42]–[44], such as the analog beamforming im-
each antenna requires a dedicated radio-frequency (RF) chain, pairments, array perturbations, oscillator locking-range based
including power amplifier (PA), low noise amplifier (LNA), phase error, and the direction-of-arrival (DoA) estimation er-
data converter (ADC/DACs), mixer, etc. However, realizing a rors. Moreover, limited feedback may also cause the transmit-
fully digital MIMO implementation at mmWave frequencies ter having only partial channel information and thus beaming
is a non-trivial task, using current circuit design technology misalignment [45]. 2) Mobility of communication UE [46],
[14]. Having hundreds (or even thousands) of antennas, each [47], which invokes tracking error and system reaction delay.
supported by a separate RF chain, requires a very compact Hardware impairments and design challenges: In addi-
circuit implementation. Moreover, due to the high bandwidth, tion to above challenges, practical transceiver hardware are
the PAs and data converters are expensive and power con- impaired by phase noise (PN), non-linear PAs, I/Q imbalance,
suming. Hence, it appears that fully digital mmWave MIMO and limited ADC resolution [48]. These effects limit the chan-
implementations are currently infeasible from a cost-efficiency nel capacity [49], particularly when high spectral efficiency
perspective. This is likely to change in the future, but, in is envisioned. On the other hand, it was proved in [50] that
the meantime, alternative low RF-complexity architectures MIMO communication links are less affected by hardware
have received much attention from the research community. impairments than single-antenna links.
In particular, hybrid analog/digital architectures are being In mmWave communication systems, mixers are applied for
considered, where the corresponding signal processing tech- signal up-conversion at the transmitter and down-conversion at
niques must be redesigned to enable channel estimation and the receiver using local oscillators to generate carrier signals
a good tradeoff between the spectral efficiency and energy operating at the desired carrier frequency. However, due to the
consumption/hardware cost [37]. random deviation of the output signal frequency around the
Narrow beamwidth and side-lobes: To increase the trans- carrier, it is infeasible that both oscillators at the transmitter
mission distance for mmWave, an array gain can be obtained and the receiver operate exactly at the same carrier frequency.
by using directional antennas, MIMO, and beamforming. Such a mismatch can be described by PN since the frequency
Consequently, the beamwidth of mmWave signals is normally offset yields a random phase difference for the time domain
narrow. When modeling the directivity, the radiation patterns samples. Due to the high carrier frequency, mmWave commu-
are usually modeled in an idealized fashion, e.g., a constant nication systems are more sensitive to PN than conventional
large antenna gain within the narrow main-lobe and zero ones.
 5

Another important hardware impairment in mmWave is

the nonlinear PAs, since it is challenging to provide linear Water
vapor
amplification to a signal with very wide bandwidth. In practice, absorption
band
each amplifier has a non-linear behavior, e.g., input signals
of large amplitude are clipped and different frequencies are
amplified differently. For the modeling of such non-linear 24.25GHz 42.5
57
64 76 86
71 81
109.5
102.2
164 200 300 GHz

characteristics, the modified Rapp model [51], is commonly Oxygen

used to describe the input-output relationship. Denoting Vi absorption bands

the input voltage level of the PA, the output voltage level
Vo is described by the amplitude modulation to amplitude
modulation (AM-AM),
rVi
Vo = (4)
27.5GHz 28.35 29.1 29.25 31 31.3 GHz 37 GHz 38.6 40 42 42.5 GHz
1
,
(1 + (|rVi |/Vs ) 2p ) 2p
Bands proposed Potential bands Other bands
Absorption bands
for mobile use for mobile use raised in NOI
where r is the small signal gain, Vs is the limiting output
amplitude, and p controls the smoothness of the transition from Fig. 4. Spectrum usage in mmWave bands.
linear operation to saturated operation. There are also other .
models, such as the Saleh model [52].
Moreover, the higher and larger frequency bands of
performance less affected by hardware impairments. The large
mmWave communication systems cause many technical chal-
unused frequency bands have therefore already attracted lots
lenges in the design of circuit components and antennas.
of interest. For instance, in October 2003, the FCC announced
In [53], authors discussed in depth the challenges against
that the 71–76 GHz, 81–86 GHz, and 92–95 GHz frequency
the design of mmWave CMOS Radios, ranging from the
bands (collectively referred as the E-band) will be available for
device-level challenges to the architecture-level challenges. In
ultra-high-speed data communication including point-to-point
particular, phase noise and IQ imbalance also cause severe
WLAN, mobile backhaul, and broadband Internet access. A
technical challenges against realizing mmWave RF circuits
total of 12.9 GHz bandwidth is available in the E-band (60-
[54]–[56]. In [57], authors have discussed and summarized
90 GHz). More recently, in July 2016, FCC dedicated large
the latest research progress on integrated circuits for mmWave
bandwidths in mmWave bands for future cut-edge wireless
communication systems. Authors have specifically summa-
communications, namely, 64-71 GHz unlicensed bands (plus
rized in details a few key technologies, including RF power
previous 57-64 GHz) and 27.5-28.35 and 37-40 GHz licensed
amplifiers , mixers, high speed analog-to digital converters, on-
bands.
chip and in-package antennas and 60 GHz voltage-controlled
Short wavelength and narrow beamwidth: Contrary to
oscillators.
signals at sub-6 GHz bands, the mmWave signal has much
shorter wavelength, which facilitates packing a large number
B. Technical Potentials of antennas into an array of compact size [6], [7]. This
Having listed the main challenges, we should not forget the greatly expands the application range of large-scale antenna
main reasons for using mmWave communications. communications in the future mobile networks [9], [10]. At
Large continuous unused bandwidth: Compared to mi- the same time, when having many antenna elements, the
crowave communications, one of the major benefits of beamwidth is narrow [11]. The positive side of this property is
mmWave communications is the availability of large band- the higher security against eavesdropping and jamming, and
width, though wider bandwidth does not always lead to a larger resilience against co-user interference. This implies
higher rates in the noise-limited region [8]. Currently, the that the spectrum can be reused frequently in space, so many
available bandwidth for mobile networks (2G, 3G, 4G and interfering point-to-point MIMO systems (or multiuser MIMO
LTE-Advanced spectrum) is globally smaller than 780 MHz systems) can be deployed in a limited spatial region.
and each major wireless provider has only a total of about
III. C HANNEL M EASUREMENTS AND M ODELING
200 MHz spectrum [7]. This bandwidth is not sufficient for
providing rates of Gbps to multiple devices, since a huge A. Millimeter Wave Measurement Campaigns
per-device spectral efficiency would be required. However, in Since the wavelength of mmWave bands is far shorter than
mmWave bands, there are large chunks of bandwidth available in microwave bands below 6 GHz, the parameters for radio
for future mobile networks. As shown in Fig. 4, in the channel models will be quite different. Thus, understanding
mmWave bands, the potentially available bandwidth can be the mmWave propagation characteristics is the first task to
more than 150 GHz [5], even excluding unfavorable bands design and develop mmWave communication systems. In
such as the 60 GHz oxygen absorption band (57–64 GHz) and general, parameters such as pathloss, delay spread, shadow-
the water vapor (H2 O) absorption band (164–200 GHz). With ing, and angular spread are used to characterize the radio
150 GHz of spectrum, a low spectral efficiency of 1 b/s/Hz is propagation, which can be obtained through analyzing the data
sufficient to deliver a rate of 150 Gbps. Having a low spectral collected by various channel measurement campaigns in differ-
efficiency simplifies the implementation and makes the end ent environments. Extensive channel measurement campaigns
 6

TABLE I
S UMMARY OF MM WAVE MEASUREMENT CAMPAIGNS

Frequency Scenario Site Parameters Ref.

(GHz)
10 Street canyon Berlin Large-scale parameters, time variance, and frequency dependence [58]
Outdoor to indoor Belfort Penetration losses, delay spread, and time variation [58]
11 Urban outdoor Ishigaki Pathloss, RMS delay spread, shadow, and power delay profile [59]
15 Street canyon Helsinki Pathloss, delay spread, and angular spread [58]
Street canyon Stockholm Pathloss and delay spread [58]
Office Pathloss, directional spread, and delay spread [58]
Airport Helsinki Pathloss, delay spread, and angular spread [58]
Outdoor to indoor Stockholm Penetration loss and RMS delay spread [58]
16 Urban outdoor Tokyo Pathloss [60]
17 Outdoor to indoor Belfort Penetration losses, delay spread, and time variation [58]
26 Urban steet and Sidewalk Tokyo Pathloss [61]
28 Street canyon Helsinki Delay spread, angular spread, and pathloss [58]
Street canyon Berlin Large-scale parameters, time variance, and frequency dependence [58]
Open square Helsinki Pathloss, delay spread, and angular spread [58]
Airport Helsinki Pathloss, delay spread, and angular spread [58]
Indoor (lab) and outdoor urban Dallas Penetration loss, pathloss, reflectivity, and AoA [62]
Dense urban New York Pathloss, RMS delay spread, shadow, PDP, AoA, and AoD [63]–[65]
Indoor and outdoor to indoor New York Penetration loss and reflection coefficients [66]
Dense urban New York Outage [67]
Outdoor to indoor Göteborg Excess loss, received power [68]
Office New York Pathloss, RMS delay spread, and shadow [69], [70]
Laboratory RMS delay spread and PDP [71]
Station and airport Seoul Pathloss and shadow [72]
Urban Daejeon Pathloss, RMS delay spread, angle spread, AoA, and AoD [73], [74]
37 Urban steet and Sidewalk Tokyo Pathloss [61]
38 Urban (Campus) Austin Pathloss, RMS delay spread, and PDP [75]
Urban (Campus) Austin Outage statistics [76]
Urban (Campus) Austin Pathloss, RMS delay spread, shadow, PDP, AOA, outage statistics [77]
40 Indoor (lab) Penetration loss, reflectivity [62]
41 Street canyon Berlin Large-scale parameters, time variance, and frequency dependence [58]
55 Urban (street) London Pathloss, fading envelope, coherence bandwidth [78]
57 Urban (street) Oslo Mean delay, delay spread, delay interval and delay window [79]
60 Street canyon Helsinki Delay spread, angular spread, and pathloss [58]
Street canyon Stockholm Pathloss and delay spread [58]
Office Pathloss, directional spread, and delay spread [58]
Shopping mall University of Bristol Direction [58]
Outdoor to indoor Stockholm Penetration loss and RMS delay spread [58]
Indoor (lab) Penetration loss [62]
Urban (Campus) Austin Pathloss, RMS delay spread, and PDP [75]
Hospital Japan Pathloss, delay spread, and PDP [80]
Outdoor courtyard and in vehicle Pathloss, RMS delay spread, PDP, and AoA [81]
Street canyon Berlin Pathloss [82]
Street canyon Berlin Pathloss and delay spread [83]
Urban Aachen Angular, RSS [84]
72 Office New York Penetration loss, delay spread, PDP [85]
73 Dense urban New York Pathloss, delay spread, shadow, PDP, AoA, and AoD [65], [86]
Dense urban New York Outage [67]
Office New York Pathloss, RMS delay spread and shadow [69], [70]
81-86 Roof-to-street Otaniemi Frequency response, impulse responses, and delay [87]
and street canyon and Kaisaniemi (Helsinki)
82 Street canyon Berlin Large-scale parameters, time variance, and frequency dependence [58]
Office CEA-Leti Delay and angular spread [58]
10 Indoor hotspot Oulu Large-scale parameters, pathloss model [88]
11, 16, 28, 38 Indoor Jinan, China Power delay, azimuth, elevation profile [89]
73 rural Virginia pathloss model [90]
26 Indoor BeiJing pathloss, shadow fading, and coherence bandwidth [91]
 7

TABLE II
S UMMARY OF MM WAVE MEASUREMENT RESULTS ON DIRECTIONAL (D) AND OMNIDIRECTIONAL (O) PATHLOSS , DELAY SPREAD AND SHADOWING

Freq. Environment Scenario Site Tx/Rx Antenna PL exp. Delay spread Shadowing Ref.
(GHz) (Height (m) etc.) (ns) (dB)
11 Macro cellular LoS/NLoS Ishigaki (O), Tx: 28, Rx: 3 2.6/3.4 4.4/6.7 [59]
11 Micro cellular LoS Ishigaki (O), Tx: 8, Rx: 3 2.7 2.5 [59]
11 Outdoor hotspot LoS Ishigaki (O), Tx: 3, Rx: 3 2.2 2.5 [59]
28 Outdoor cellular LoS/NLoS New York (D), 24.5-dBi Tx/Rx 2.55/5.76 8.66/9.02 [7], [63]
(Manhattan)
28 Dense urban LoS/NLoS New York (O), Tx:7; 17, Rx: 1.5 2.1/3.4 3.6/9.7 [65]
(Manhattan)
28 Office LoS/NLoS New York (D), V-V, 15-dBi, Tx: 2.5, Rx: 1.5 1.7/4.5 4.1/18.4 2.6/11.6 [69]
(Brooklyn)
28 Office LoS/NLoS New York (O), V-V, 15-dBi, Tx: 2.5, Rx: 1.5 1.1/2.7 1.7/9.6 [69]
(Brooklyn)
28 Office LoS/NLoS New York (D), V-H, 15-dBi, Tx: 2.5, Rx: 1.5 4.1/5.1 12.8/18.7 8.0/10.9 [69]
(Brooklyn)
28 Office LoS/NLoS New York (O), V-H, 15-dBi, Tx: 2.5, Rx: 1.5 2.5/3.6 3.0/9.4 [69]
(Brooklyn)
28 Station LoS/NLoS Seoul (D), Tx: 8, Rx: 1.5 2.15/4.06 1.19/10.67 [72]
28 Airport LoS/NLoS Seoul (D), Tx: 8, Rx: 1.5 2.17/3.55 1.33/7.61 [72]
28 Urban NLoS Daejeon (O), 24.5-dBi Tx:15 3.53 22.29 6.69 [73]
28 Urban street LoS/NLoS Daejeon (O), Tx:15, Rx: 1.6 1.90/3.15 0.63/22.09 [74]
28 Urban street LoS/NLoS New York (O), Tx:15, Rx: 1.6 1.81/3.03 2.05/17.99 [74]
(Manhattan)
28 Urban street LoS/NLoS New York (O), Tx:15, Rx: 1.6 1.87/2.97 1.74/15.92 [74]
(Manhattan)
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 13.3-dBi RX 2.13/2.54 10.1 8.14/7.74 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 13.3-dBi RX 2.16/2.52 17.3 8.78/7.82 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 13.3-dBi RX 2.03/2.40 4.8 5.31/5.27 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 13.3-dBi RX 2.74/2.97 17.5 12.46/11.16 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 25-dBi RX 2.25/3.29 13.5 6.51/11.63 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 25-dBi RX 2.38/3.20 15.1 14.12/8.97 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 25-dBi RX 2.01/2.70 5.3 6.56/5.54 [77]
38 Outdoor Cellular LoS/NLoS Austin (D), 25-dBi TX and 25-dBi RX 2.99/4.17 16.5 13.92/8.90 [77]
55 Urban (street) LoS/NLoS London (D), Tx: 10, Rx: 2 3.6/10.4 [78]
60 Hospital LoS Japan Tx (O): 1.31, Rx (D): 2.04 1.34 6.7 [80]
60 Hospital LoS Japan (O), Tx : 2.30, Rx: 2.04 1.86 6.3 [80]
60 Hospital LoS Japan (O), Tx: 0.75, Rx: 1.58/2.29 2.23 3.8 [80]
60 Street canyon LoS/NLoS Berlin (O) 2.02/3.06 [83]
72 Indoor NLoS New York 20-dBi Tx/Rx: 1.61 [85]
(Brooklyn)
73 Office LoS/NLoS New York (D), V-V, 15-dBi, Tx: 2.5, Rx: 1.5 1.7/5.3 3.3/13.3 2.1/15.6 [69]
(Brooklyn)
73 Office LoS/NLoS New York (O), V-V, 15-dBi, Tx: 2.5, Rx: 1.5 1.3/3.2 1.9/11.3 [69]
(Brooklyn)
73 Office LoS/NLoS New York (D), V-H, 15-dBi, Tx: 2.5, Rx: 1.5 4.7/6.4 21.2/10.3 9.0/15.8 [69]
(Brooklyn)
73 Office LoS/NLoS New York (O), V-H, 15-dBi, Tx: 2.5, Rx: 1.5 3.5/4.6 6.3/9.7 [69]
(Brooklyn)
73 Outdoor cellular LoS/NLoS New York (D), Tx: 7, Rx: 4 2.68/4.72 6.79/9.52 [86]
73 Outdoor cellular LoS/NLoS New York (D), Tx: 7, Rx: 2 2.77/4.64 5.33/8.32 [86]
73 Outdoor cellular LoS/NLoS New York (D), Tx: 17, Rx: 4 2.52/3.90 4.00/10.81 [86]
73 Outdoor cellular LoS/NLoS New York (D), Tx: 17, Rx: 2 2.32/3.76 2.71/8.81 [86]
73 Dense urban LoS/NLoS New York (O), Tx:7; 17, Rx: 2 2.0/3.3 5.2/7.6 [65]
(Manhattan)
73 Dense urban LoS/NLoS New York (O), Tx:7; 17, Rx: 4.06 2.0/3.5 4.2/7.9 [65]
(Manhattan)
10 Indoor LoS/NLoS Oulu Tx: 2, Rx: 1.6 1.4/3.3 17.4/30.4 1.4/2.6 [88]
11/16/28/38 Indoor JiNan Tx: 2.6, Rx: 1.45 9.1/10.2/10.6/8.5 [89]
73 rural LoS/NLoS Virginia Tx:110, Rx: 1.6-2 2.16/2.75 1.7/6.7 [90]
26 Hall LoS BeiJing Tx: 2.5, Rx: 2 16.08 0.098 [91]
 8

[58]–[87], covering potential mmWave mobile communication 2003 3GPP-SCM

bands, like 10, 28, 38, 60, and 82 GHz, have already been
performed. Table I summarizes the performed measurement 2004

campaigns, where in particular NYU WIRELESS and the EU

mmMAGIC project make contributions.1 The researchers from 2005
SCM-E
WINNER-I

NYU WIRELESS have performed many measurements at 28, 2006 COST 273
38, 60, 72 and 73 GHz since year 2011, and obtained abundant
WINNER-II
measurement results. Based on various measurement results 2007

(including those from NYU WIRELESS) and ray tracing

analysis, 3GPP [21], [99] also provides new modeling features 2008

like: i) dynamic LoS/non-LoS (NLoS) blockage 2 , spatial 2009

IMT-Advanced
IEEE 802.11ad
consistency and penetration modeling, ii) extension of power
WINNER+
delay/angle profiles, and iii) the pathloss model based on one- 2010

meter reference distance. The EU mmMAGIC project plans

2011
more than 60 single-frequency measurement campaigns, cov-
ering eight frequency bands from 6 to 100 GHz, under typical 2012
COST 2100

environments and scenarios, including urban micro-cellular

(UMi) (street canyon, open square), indoor (office, shopping 2013

mall, airport), outdoor-to-indoor (O2I) and two scenarios with QuaDRiGa

2014
very high user densities (stadium and metro station) [58].
1) Pathloss and Shadowing: Pathloss and shadowing are 2015 3GPP 3D
METIS MiWEBA

the two most important large-scale characteristics of the radio

channel, which have been reported for various environments 2016
mmMAGIC
both in LoS and NLoS cases. It is common to determine 2017
the pathloss exponent that best fits the measurements to Friis
Year
transmission formula in (2), while the shadowing parameters
are used to model the random deviations from this model. Fig. 5. Available channel models for mobile and wireless networks [58].
Table II summarizes the main mmWave measurement results
for directional and omnidirectional pathloss and shadowing.
2) Power Delay Profile and Delay Spread: The shape of Macro (SMa), Urban Macro (UMa), and UMi. The SCM-
power delay profile (PDP) in measurements is a single or su- Extended model [99] further extended the SCM, by supporting
perposition of multiple exponentially decaying spectrums. The bandwidths of up to 100 MHz.
delay spread denotes an extent of the multipath power spread 2) WINNER I/II/+ Model: As the EU flagship mobile
over the PDP, which is an important parameter that determines technological projects for 4G, the WINNER I/II/+ projects
the inter-symbol interference in single-carrier transmission developed several channel models for mobile networks. The
and the frequency-flatness of the subcarriers in multi-carrier WINNER I model [100] is an antenna-independent model, in
transmission. In Table II, we also summarize the delay spread which different antenna configurations and different element
of various measurement results. patterns can be used. In order to support more scenarios, such
as outdoor-to-indoor and indoor-to-outdoor, and elevation in
indoor scenarios, the WINNER II model [101] was devel-
B. Millimeter Wave Channel Modeling oped. It includes scenario-dependent polarization modeling,
and thus improves the accuracy for cross-polarized MIMO
Channel models are important for evaluating and analyzing
antennas. The parameter tables were reviewed and additional
the performance in system-level simulations. For mmWave,
measurements were done in order to cover the complete 1–
some recent channel modeling results are reported in [92]–
6 GHz frequency range. As a major upgrade of the WINNER
[98]. For more general mobile and wireless networks, several
II model, the WINNER+ model [102] supports 3D propagation
channel models have proposed continuously, which are illus-
effects. WINNER II and WINNER+ models are also antenna
trated in Fig. 5. To give a historical perspective, we give a
independent since they are based on double-directional channel
brief introduction as follows (even though some models are
representations.
not for mmWave).
3) 3GPP 3D Model [103], [104]: This model is defined
1) 3GPP Spatial Channel Model and SCM-Extended: The in the 2 GHz band at a relatively small bandwidth of 10 MHz.
3GPP Spatial Channel Model (SCM) [41] was proposed in It has consolidated parameters for the two most commonly
2003 and supports six delay paths with 5 MHz bandwidth in used scenarios: UMa and UMi, which are further split into
the 2 GHz frequency band under three scenarios: Suburban LoS, NLoS and O2I propagation. The core part of this model,
i.e., the small-scale fading (SSF) model, is identical to the
1 Note that there are a lot of excellent measurement campaigns and results.
WINNER+ model. Thus, the same parameters can be used
For space limits, we can only list a part of the existed literatures here. The
results, especially before 2012, are mostly omitted. and similar functionality is provided. More recently, 3GPP
2 COST 259 also include the LoS/NLoS blockage probability model. TR38.900 [98] summarizes the recent channel modeling re-
 9

sults of 3GPP for the band above 6 GHz and up to 100 GHz. 10) mmMAGIC Model [58]: This is a statistical channel
The models in [98] are comprehensive including street canyon, model for link- and system-level simulations that is designed
open area, rooftop, indoor, backhaul, D2D/V2V and stadium for the entire frequency range from 6 to 100 GHz for a large
etc. variety of scenarios. The model uses the theoretical approach
4) COST 273/2100 Model: The COST 273 model [105] of the existing 3GPP 3D channel model. However, it extends
was an evolution of the earlier COST 259 channel model it in the following ways: i) the spatial accuracy regarding
towards mobile broadband by using MIMO technologies. The path and sub-path distributions is substantially improved, ii)
model is based on the concept of geometrically located multi- a realistic non-uniform distribution of sub-path amplitudes is
path clusters in 2D propagation environment to model the in- included, iii) sub-paths can be modeled using spherical waves;
terrelationship between Angle-of-Arrivals (AoAs) and Angle- iv) there is consistency over the frequency range (6−100 GHz),
of-Departures (AoDs). This concept is effective to keep spatial v) there is frequency-dependent antenna models; vi) providing
consistency, and can evaluate the performance of MIMO continuous variations over time, vii) mmWave-specific random
beamforming and multi-cell transmission more accurately. As blockage, clustering and scattering objects are being modeled,
an evolution of the COST 273 model, the COST 2100 model and viii) the reflection from the ground or the floor is modeled.
[106] used visibility regions (VRs) introduced in COST 259 to 11) 3GPP-like 5G Model [22], [112]: In [112], the outdoor
model the scenario-variation. These VRs make the evaluation model is established for the bands from 6 GHz to 100 GHz.
of multi-cell and heterogeneous transmissions more practical An initial 3D channel model which includes: 1) typical de-
by considering the VR of BSs from each UE. This model also ployment scenarios for urban microcells and urban macrocells,
extends the multipath clusters to 3D propagation environments. and 2) a baseline model for incorporating pathloss, shadow
5) QuaDRiGa Model [107]: As an open source imple- fading, LoS probability, penetration, and blockage models
mentation of the 3GPP-3D channel model, the QuaDRiGa for the typical scenarios. Various processing methodologies
channel model is further extended with the features of spatial such as clustering and antenna decoupling algorithms are also
consistency (to accurately evaluate the performance of mas- included in [112]. In [21], [22], the indoor model is established
sive MIMO) and multi-cell transmissions by exploiting the for office and shopping mall environments. The measurement
approach in SCM-E and COST 273. results show that the smaller wavelengths introduce an in-
creased sensitivity of the propagation models to the scale of
6) IEEE 802.11ad Model [108]: This model was developed
the environment and show some frequency dependence of the
in 2010 to support indoor short-range communications, such
pathloss as well as increased occurrence of blockage.
as in offices and homes using 60 GHz unlicensed band. The
model is quasi-deterministic (QD): specular components, such IV. MIMO D ESIGN FOR MM WAVE C OMMUNICATIONS
as the LoS path and single and double bounce reflections are While conventional microwave communications, operating
modeled deterministically in 3D propagation environments, below 6 GHz, can cover many users in wide coverage areas,
while other contributions are modeled stochastically as random mmWave communications mainly provide local-area coverage
components in the cluster. One of the important features is the and thus feature fewer users. Moreover, there is limited
support for blockage. scattering in outdoor mmWave communications, in contrast
7) MiWEBA Model [109]: The model is an extension of to the rich scattering in conventional microwave communica-
the IEEE802.11ad channel model towards outdoor access, tions [7]. Due to these differences, mmWave MIMO systems
backhaul/fronthaul, and device-to-device (D2D) scenarios. The have different constraints and requirements, requiring different
approach is QD, where specular components are modeled de- transceiver designs. In this section, we first describe hardware
terministically, while other components are modeled stochasti- architectures for mmWave MIMO systems, for the use in mo-
cally. In this way, beamforming and path-blocking models can bile networks. We discuss how to design the signal processing
be supported. Furthermore, the first 60 GHz pathloss model in techniques, including channel estimation, channel-tracking,
an UMi environment was developed in MiWEBA. The effect and precoding/combining for millimeter MIMO systems.
of the ground reflection paths traveling closely in space to the
LoS path has been found to be of high significance. A. MIMO Architectures
8) IMT-Advanced Model [110]: This model consists of The conventional MIMO system is fully digital, where all
a primary module and an extension module (not specific the signal processing techniques are performed at baseband
to mmWave), where the former is based on the WINNER as shown in Fig. 6 (a) [36]. As explained in Section II,
II channel model, while the latter enhances the support for for mmWave communication with many antennas and high
variable BS antenna heights, street widths, and city structures. bandwidth, the conventional fully digital MIMO architecture
9) METIS Model [111]: This model consists of a map- requires many energy-intensive RF chains, leading to an unaf-
based (deterministic) model, a stochastic model, and a hybrid fordable energy consumption and hardware cost. Therefore,
model as a combination of both. The stochastic model extends although the fully digital architecture most likely will be
the WINNER+ and 3GPP-3D models to support 3D shadowing available at some point in the future, alternative architectures
maps, mmWave parameters, direct sampling of the power are required for emerging mmWave mobile networks.
angular spectrum, and frequency dependent pathloss models. To reduce the implementation complexity, a fully analog
Based on extensive measurement campaigns, lists of channel architecture has been adopted in indoor mmWave communi-
parameters for <6 GHz and 50 to 70 GHz bands are available. cations, such as 60 GHz WLAN [114]. As shown in Fig. 6
 10

to the multi-stream scenario. As shown in Fig. 6 (c), the key

RF chain idea is to divide the conventional digital signal processing
(e.g., precoding and combining) of large size into two parts:
RF chain
a large-size analog signal processing (realized by analog
circuits) and a dimension-reduced digital signal processing
Digital signal
processing
(requiring a small number of RF chains). Since there is
often only a small number of effective scatterers at mmWave
frequencies, each user has a MIMO channel matrix with
low rank [7]. Hence, the optimal number of data streams is
RF chain generally much smaller than the number of antennas. Since the
number of streams determines the minimum required number
(a)
of RF chains, this number can be significantly reduced by
the hybrid architecture, leading to reduced cost and energy
consumption.
The analog circuits of the hybrid architecture can be im-
Digital signal Analog plemented by different circuit networks, leading to different
RF chain
processing circuit hardware constraints [14], [115], [116]. Next, we describe
the three typical implementation networks that are illustrated
in Fig. 7. The choice of architecture affects not only the
signal processing design but also the performance of mmWave
MIMO systems.
(b) N1) Fully-connected network with phase-shifters: In this
network, each RF chain is connected to all antennas via phase-
shifters, as shown in Fig. 7 (a) [117]. Hence, a highly directive
signal can be achieved by adjusting the phases of transmitted
RF chain signals on all antennas [117]. By employing such a network,
Dimension-
Analog all elements of the analog precoder/combiner have the same
reduced digital
signal processing
circuit fixed amplitude.3 Difficult factors in the implementation can be
RF chain the addition of the analog signals at each antenna and selection
of a collection of phase-shifts that are suitable over the entire
bandwidth.
N2) Sub-connected network with phase-shifters: In this
(c) network, each RF chain is only connected to a subarray
via phase-shifters, as shown in Fig. 7 (b) [113], [121]. Due
Fig. 6. MIMO architectures: (a) fully digital architecture; (b) fully analog
architecture; (c) hybrid architecture. to this limitation, for each RF chain, only the transmitted
signals on a subset of antennas can be adjusted. Therefore,
compared with N1, the achieved array gain and directivity
is reduced proportionally to the number of subarrays [113].
(b), only one RF chain is employed to transmit a single
However, this network might be preferred in practice, since
data stream, and the analog circuit (e.g., realized by analog
there is no need to add analog signals at the antenna inputs
phase-shifters) is utilized to partially adjust the signals (e.g.,
and the number of phase-shifters required in this network
phases of signals) to achieve an array gain. The advantage
is also significantly reduced. Moreover, it has been shown
of the fully analog architecture is that it only requires one
to achieve the performance close to that of N1 [121]. This
RF chain, leading to quite low hardware cost and energy
network imposes two hardware constraints [121]: i) the analog
consumption [114]. However, since the analog circuit can
precoder/combiner should be a block diagonal matrix; and
only partly adjust the signals, it is hard to adjust the beam
ii) All the nonzero elements of the analog precoder/combiner
to the channel conditions and this leads to a considerable
have the same fixed amplitude.
performance loss [14], particularly for mobile users. In addi-
N3) Lens antenna array: An alternative quite different from
tion, fully analog architecture can only support single-stream
the networks discussed above is to utilize a lens antenna array,
transmission, which cannot achieve the multiplexing gain to
as shown in Fig. 7 (c) [126]. The lens antenna array (a feed
improve spectral efficiency [14].
antenna array placed beneath the lens) can realize the functions
New MIMO architectures need to be designed to balance of signal emitting and phase-shifting simultaneously [126].
between the benefits of fully digital and fully analog ar-
chitectures. The hybrid analog-digital architecture is the key 3 Here we assume the resolution of phase-shifter is sufficiently high, so that
solution [122]–[124]; the hybrid architecture was agreed to be the phase of transmitting signal can be arbitrarily adjusted. In practice, the
deployed in future 5G systems at the 3GPP RAN1 meeting phase-shifters with finite resolution will incur phase noise and degrade the
performance of hybrid precoding and combing. In this case, how to design
in June 2016 [125], as described later. The hybrid architecture the signal processing techniques is an interesting topic of future research, and
can be considered an extension of the fully analog architecture some initial works can be found in [118]–[120].
 11

Beam Beam

(a) (b)
Fig. 8. Beam-training: (a) Wide beamwidth; (b) Narrow beamwidth.

Phase shifters
Phase shifters
use of pilot transmission is the most efficient way to acquire
(a) (b)
CSI [6], [130]. The channel estimation is more complicated
Lens antenna array in a hybrid mmWave architecture, since we cannot extract the
actual received signals on all antennas simultaneously. We will
now discuss alternative methods for acquiring CSI in a hybrid
architecture.
To exemplify the signal processing, we consider a single-
user multi-stream system. The transmitter employs NT anten-
T RF chains, the receiver employs N antennas
nas and NRF ( ) R
R
and NRF RF chains, and ND ≤ min NRF T , NR
Switches
RF parallel data
streams are transmitted from the transmitter to the receiver. In
(c) this subsection, we mainly consider narrowband systems. Note
that for broadband systems, the analog circuit is fixed for the
Fig. 7. Analog circuit with different networks: (a) Fully-connected network whole bandwidth. As a result, the analog signal processing
with phase-shifters (N1); (b) Sub-connected network with phase-shifters (N2);
(c) Lens antenna array (N3).
(e.g., precoding) cannot be adaptively adjusted according to
different frequencies. This will lead to more challenges in
signal processing design, which requires future investigation,
It can concentrate the signals from different propagation but some pioneering works can be found in [137], [140]. The
directions (beams) on different feed antennas. As the scattering narrowband system model of the hybrid architecture can be
for outdoor mmWave communications is not rich [7], the presented as [117]
number of effective propagation paths is usually limited, and
the channel power will be concentrated on only a small number y = W H HFs + W H n, (5)
of beams. Therefore, the selecting network can be used to where s and y both of size ND × 1 are the transmitted and
significantly reduce the MIMO dimension as well as the received signal vectors, respectively, H of size NR × NT is the
number of RF chains without major performance loss [126]. mmWave MIMO channel, which can be modeled as described
With careful design, the lens antenna array can excite several in Section III, n of size NR × 1 is the noise vector. F = FA FD
orthogonal beams spanning the whole space. If the directions of size NT × ND is the hybrid precoder, where FA of size
of channel paths coincide with the directions of the orthogonal NT × NRFT is the analog precoder and F of size N T × N is
D RF D
beams, an array gain similar to N1 can be achieved4 [126]. the digital precoder. Similarly, W = WA WD of size NR × ND
Moreover, compared with phase-shifter network (including is the hybrid combiner, where WA of size NR × NRF R is the
a large number of phase-shifters, power splitters/combiners, R
analog combiner and WD of size NRF × ND is the digital
and signal/control lines) in N1, the hardware cost and energy combiner.
consumption incurred by lens antenna array in N3 is relatively The precoder and combiner matrices should be selected
low [128], [251]. Essentially, the lens antenna array plays based on the current channel realization H, but estimating
the role of a discrete Fourier transform (DFT).5 Therefore, this matrix is non-trivial [141]. Firstly, due to the lack of
in this network, each column of the analog precoder/combiner antenna gain before the establishment of the transmission link,
restricts to a DFT column [126]. the SNR for channel estimation can be quite low. Secondly,
the number of RF chains in the hybrid architecture is usually
B. Channel Estimation with Hybrid Architecture much smaller than the number of antennas (i.e., NRF T  N and
T
R  N ), so we cannot simultaneously obtain the sampled
NRF
Channel state information (CSI) is essential to benefit from R
the array gain provided by multiple antennas. Acquiring CSI signals on all receive antennas. As a result, the traditional
is particularly challenging in mobile networks, where the channel estimation schemes [138], [139] requiring the sampled
channels can change rapidly. With a fully digital receiver, the signals on all antennas will involve unaffordable pilot overhead
in a hybrid architecture. To solve this problem, two dominant
4 If not, power leakage will happen, leading to some performance loss after
categories of channel estimation schemes have been proposed
the selecting network [127]. in the mmWave literature.
5 It is worth pointing out that mathematically, N3 is equivalent to the
structure based on butter matrix proposed in [129], which aims to improve the The key idea of the first category is to reduce the dimension
performance of antenna selection for microwave MIMO with high correlation. of channel estimation problem, by dividing it into two steps.
 12

In the first step, it performs the beam-training between the i) enjoys the best performance; ii) is not the tabu beam pair
transmitter and receiver to obtain the analog precoder FA and according to some criteria. After a small number of iterations,
analog combiner WA . In the second step, the effective channel it can obtain FA and WA with satisfying performance. Other
matrix WAH HFA in the analog domain is estimated by classical algorithms, such as local search algorithm [148], can be also
algorithms, such as least squares (LS) [139]. Note that the size used. They can usually significantly reduce the pilot overhead,
of WAH HFA is ND × ND , which is much smaller than that of but the robustness cannot be guaranteed, since these algorithms
the original channel matrix (i.e., ND  NT, NR ). Therefore, the are performed in a random way.
pilot overhead in the second step is relatively low, which is The second category of channel estimation schemes is to
proportional to ND . The remaining difficulty lies in how to exploit the sparsity of mmWave MIMO channels. Instead of
design the efficient beam-training scheme to find the optimal estimating the effective channel matrix WAH HFA of small size,
FA and WA . it can directly obtain the complete channel matrix H with
To achieve this goal, two primary approaches have been low pilot overhead. To explain the basic idea, we consider
proposed. The first one is to extend the traditional single-beam the simplest case where both the transmitter and receiver use
training schemes standardized in IEEE 802.11ad/802.15.3c one RF chain for channel estimation6 [141]. In time slot m,
to multi-beam training [47], [142]. For example, the single- the transmitter uses a hybrid precoder fm of size NT × 1 to
beam training scheme in IEEE 802.11ad consists of three transmit the pilot s m to the receiver. The received pilot ym at
phases [142]: i) Sector level sweep (SLS): the wide beamwidth the receiver using a hybrid combiner wm of size NR × 1 can
is considered at first, as shown in Fig. 8 (a). The transmitter be presented as
and receiver try all possible wide beam pairs. With the channel √ H
ym = ρwm Hfm s m + wmH
n
feedback to indicate the largest received SNR [143], [144], the √ T 
best beam pair (according to some criterion) can be selected = ρ fm ⊗ wm vec (H) + wm
H H
n. (6)
for the next phase; ii) Beam-refinement protocol (BRP): the Note that the mmWave MIMO channel matrix H can be well
narrow beamwidth is considered as shown in Fig. 8 (b), and approximated by the extended virtual channel model with
the beams are trained in a similar way within the previous sufficiently quantized AoAs/AoDs [141]. Therefore, vec (H)
selected wide beam pair; iii) Beam tracking: a periodic beam- can be rewritten as vec (H) = AD hb where hb = vec H̃b of
refinement is performed for the time-varying channels. Re-
size G2 × 1 is a sparse channel vector, and G is number of
peating such procedure above for each RF chain to select its
quantized AoAs/AoDs. The position of each nonzero element
corresponding beams, we can obtain the optimal FA and WA .
in hb indicates the AoA and AoD of one channel path, and the
This hierarchical search could significantly reduce the training
value of the nonzero element is the corresponding complex
overhead, but its performance heavily depends on the designed
gain of this path. AD of size NT NR × G2 is the dictionary
training beam codebook. For the fully analog architecture,
matrix of quantified AoAs/AoDs. After receiving M pilots,
forming a wide beam usually requires the deactivation of some
we have
antennas and thus reduces the total transmit power. While for
 fT ⊗ w H 
the hybrid architecture, it is possible to design wide beam  1T 1 
√  f2 ⊗ w2 
H
without reducing the transmit power [145], [146]. Another √
y = ρ  ..  AD hb + neff = ρΨAD hb + neff, (7)
scheme can be found in [164], where a novel codebook-based  
 fT ⊗ w H 
.
beam-training together with the following hybrid precoding
M M
design was proposed. Specifically, an RF codebook, denoted  
as FC B , is used to specify the possible set of RF beamform- where y = y1, y2, · · · y M T and neff is the effective noise
ing vectors by considering the practical limitation of phase- vector.
shifters. Then, the original hybrid precoding problem is trans- To estimate hb from (7), there are two primary approaches
formed into a joint codeword selection and precoding design proposed in the literature. The first one is to combine the idea
problem, which is essentially a group sparsity constrained of beam-training with sparse signal recovery. For example,
optimization problem and only requires effective channels in [141], an adaptive channel estimation scheme is proposed.
HFC B . Based on that, a beam-training procedure is proposed This scheme divides the total channel estimation problem
to obtain effective channels with less signaling feedback by into several subproblems, each of which only considers one
utilizing the beam-domain sparse property of mmWave chan- channel path. For each channel path, it first starts with a
nels, and efficient algorithms are developed for maximizing coarse AoA/AoD grid, and determines the AoA and AoD of
both the spectral efficiency and the energy efficiency. The this path belonging to which angle range by employing OMP
second approach is to employ algorithms developed from algorithm. Then, the AoA/AoD grid around the determined
machine learning to realize the beam-training, since it can be angle range is narrowed and the AoA and AoD of this path
considered as a combinatorial problem with a finite number are further refined also by the OMP algorithm. In each step,
of beams to be searched. For example, in [147], a tabu search fm and wm are designed based on the corresponding AD
(TS)-based beam-training scheme is proposed. It first selects (which varies due to different AoA/AoD grid in each step)
an initial beam pair, and defines the neighbors of this pair to make the effective sensing matrix ΨAD have a fixed gain
(e.g., only one RF chain changes its beam and the others keep 6 When multiple RF chains are used, the transmitted pilots on different RF
fixed). Then, the training procedure is only executed within chains can be designed to be orthogonal to simplify the channel estimation
the neighbors with much smaller size to find a pair which: at the receiver.
 13

in a specific angle range. This requirement is the same as that the sparse structure of mmWave MIMO channels, it utilizes
of the hierarchical beam-training introduced above. Therefore, the obtained channels in the previous time slots to predict
a good multi-resolution codebook such as [145], [146] can the prior information, i.e., the support of the channel, in the
greatly improve the performance. It has been shown that the following time slot without channel estimation. Finally, with
adaptive channel estimation scheme enjoys high accuracy with the known supports, the time-varying channels can be tracked
reduced pilot overhead. Some improved schemes following with a low pilot overhead. The related schemes can be found
the similar idea can be found in [146], [149]. The second in [160], [161]. Such schemes can perform well for the LoS
approach is to regard (7) as a classical sparse signal recovery path. For the NLoS paths caused by complicated scattering,
problem [150]–[152]. Then, it designs fm and wm to make the it is difficult to analyze the geometrical relationship. As a
sensing matrix ΨAD enjoy sufficiently low mutual coherence, result, a more promising solution in practice is to utilize the
which is crucial to achieve high signal recovery accuracy in idea of the second category to track the LoS path. Then, by
CS theory [153]. In [154], several design methods have been eliminating the influence of this path, the NLoS paths can be
proposed to achieve this goal. Finally, (7) can be solved by lots tracked following the idea of classical Kalman filters [162].
of efficient CS algorithms, such as OMP and LASSO [153], To realize mmWave mobile networks, the efficiency of
and the structural sparsity of mmWave MIMO channels (e.g., the beam-tracking schemes must be carefully tested in real
common support or partial common support) can be also environments, to understand which mobility speeds that can be
exploited to further reduce the pilot overhead and improve supported and which channel characteristics that can reliably
the recovery accuracy [155]. imposed to aid the tracking procedure.

C. Channel Tracking D. Hybrid Precoding and Combining

Since the mmWave channel varies over time in mo- After the CSI has been acquired, we can design the pre-
bile networks, the conventional real-time channel estimation coding and combining in the hybrid architecture to achieve
should be executed frequently. This has two important conse- the multiplexing gain and array gain offered by MIMO.
quences [156]: i) In mmWave bands, the user mobility leads However, the precoder/combiner design is considerably dif-
to huge Doppler effects and very limited channel coherence ferent from the precoding/combining optimizations in [163]
time. This means that the mmWave MIMO channels will for the conventional fully digital MIMO systems, due to the
vary quickly, even if we consider the short symbol duration special hardware constraints in hybrid architecture of mmWave
associated with the wide bandwidth; ii) In hybrid mmWave systems. In this subsection, we discuss how to design the
implementations, there is no enough time to continuously redo hybrid precoder F = FA FD and combiner W = WA WD for N1-
the beam-training from scratch. Hence, channel-tracking ex- N3 architectures discussed in Section III-A with perfect CSI,
ploiting the temporal correlation of the time-varying channels which helps understand the fundamental limits of the mmWave
is preferable [156]. In this subsection, we describe two main MIMO with a hybrid architecture. The practically more rele-
categories of channel-tracking schemes. vant case of imperfect CSI is largely open and deserves much
The first channel-tracking approach is an improved version attention in the upcoming years; in particular, because it is
of beam-training. The key idea is to select several candidate well-known from conventional MIMO communications that
beam pairs instead of only the optimal beam pair, during each precoding/combining schemes that work well under perfect
beam-training procedure [142], [157]. For example, when the CSI can be widely different from the schemes that work well
beam pair with the highest SNR is selected, the beam pairs in practice.
achieving the second and third highest SNRs are also retained. Let us for simplicity consider a non-fading channel H.
When the channel varies, only the candidate beam pairs are Specifically, we focus on the optimization problem of hybrid
tested and switched on to keep the SNR above a certain precoding and combining with the aim of maximizing the
threshold. If all the candidate beam pairs fail, the complete achievable rate, given by [117]
beam-training will be executed again. This idea involves quite  
Fopt, Wopt = arg max log2 I+ ρR−1n W HFF H W ,
H H H
low complexity, which makes it easy to implement. Thus, it F,W
has been applied in current commercial mmWave communi- s.t. FA ∈ F , WA ∈ W, kFk 2F ≤ ND, (8)
cations, such as WLAN (IEEE 802.11ad) [114]. However, this
idea is only efficient for single-stream data transmission. For where Rn is the noise/interference covariance matrix, while
multi-stream data transmission, the beam training itself will F and W are the sets with all possible analog precoders
incur high pilot overhead, not to mention the search among and combiners satisfying the hardware constraints (which
all candidate beams for all data streams. are different for N1-N3 as discussed in Section IV-A), re-
An alternative solution to track the mmWave MIMO channel spectively. Generally, obtaining the optimal solution to (8)
is to utilize the geometric relationship between the transmitter is a non-trivial task, since F and W are much different
and receiver to track the LoS path of the channel. Specifically, from conventional fully digital communication. To address
in [156], a priori aided channel-tracking scheme was proposed. (8), several hybrid precoding/combining schemes have been
By considering a motion model, this scheme first excavates proposed in the literature to achieve feasible solutions.
a temporal variation law of the AoA and AoD of the LoS We first discuss the hybrid precoding and combining
path. After that, by combining the temporal variation law with schemes for the architecture N1. One effective approach is
 14

to decompose the original optimization problem into several Re ( ×)

subproblems, and each subproblem is approximated as a RF chain Im ( ×)
1-bit ADC

convex one and then solved by standard convex optimization

Re ( ×)
algorithms. Particularly, a spatially sparse precoding scheme RF chain 1-bit ADC
Im ( ×)
based on the orthogonal pursuit matching (OMP) has been Digital signal
proposed in [117], which can fully exploit the sparsity of processing
mmWave MIMO channel and achieve the near-optimal perfor-
mance. Using a similar idea, there are some other advanced
Re ( ×)
schemes proposed for N1 [164], [166], [167]. RF chain 1-bit ADC
Im ( ×)
For the subconnected architecture N2, a few hybrid pre-
coding solutions have been developed for maximizing the Fig. 9. 1-bit ADC based architecture at the receiver.
achievable rate or the spectral efficiency [121], [168]. More
recently, the energy efficiency optimization was studied and
a solution was proposed in [165]. Since the subconnected E. Low-Resolution ADC Based Architecture
architecture adopts a subarray structure, it is natural to solve Besides the promising hybrid analog-digital architecture
the complicated hybrid precoding problem by decomposing it discussed above, some other advanced architectures have also
into several subproblems and optimizing them in an alternating been proposed to reduce the RF complexity of mmWave
manner. In particular, a successive interference cancelation MIMO systems. The low-resolution (e.g., 1-bit) ADC based
(SIC)-based precoding scheme was proposed in [121], which architecture7 is one typical example as shown in Fig. 9 [131].
enjoys higher energy efficiency than the spatially sparse pre- Different from hybrid analog-digital architecture which aims
coding scheme. The hybrid precoding problem with energy to reduce the number of RF chains, the key idea of the low-
efficiency as the objective is more complicated. In [165], resolution ADC based architecture is to replace the high-
this optimization problem is solved by jointly exploiting the resolution (e.g., 15-bit) ADCs by low-resolution ADCs to
interference alignment and fractional programming. First, the reduce the energy consumption and hardware cost, while the
analog precoder and combiner are optimized via the alternating total number of RF chains or ADCs is still the same as that
direction optimization method. Then, the digital precoder and in the fully digital architecture [131].
combiner without hardware constraints are obtained based on
The main advantage of low-resolution ADC based architec-
the effective channel matrix WAH HFA .
ture is that it can significantly reduce the energy consumption
Finally, for the hybrid precoding/combining schemes for and hardware cost since it only requires a small number of
N3, the main difficulty lies in the designs of FA and WA , compactor in ADC [132]. Moreover, this architecture can
as FD and WD can be easily obtained based on WAH HFA . also simplify other circuit modules such as the automatic
Therefore, it is essential to design an appropriate selecting gain control (AGC) [132]. It has been proved that low-
matrix to select DFT columns (beams) to form FA and WA . resolution ADC based architecture can achieve the capacity-
Following this idea, a magnitude maximization (MM) beam- approaching performance in the low and medium SNR re-
selection scheme was proposed in [169], where several beams gion [133]. However, due to the severe quantization effect
with large power are selected. Alternatively, an interference- of low-resolution ADCs, the capacity of this architecture is
aware beam-selection scheme was proposed in [127]. The key usually limited in the high SNR region [133] In addition, the
idea is to classify all users into two user groups according nonlinearity of the quantization also imposes new challenges
to the potential inter-beam interference. For users with low on the signal processing designs. Take the 1-bit ADC for
inter-beam interference, it directly selects the beams with large example, the sampled signal in the digital domain can only
power, while for users with severe inter-beam interference, be one of the two discrete values instead of one continuous
a low-complexity incremental algorithm was proposed. More value. To solve this problem, some promising solutions have
beam-selection schemes for N3 can be found in [170], [171]. been proposed. For example, in [134], a channel estimation
It worth pointing out that although we have only discussed method using expectation-maximization (EM) algorithm is
the single-user multi-stream scenario, the schemes designed proposed for low-resolution ADC based architecture to find the
for N3 can be extended to multi-user multi-stream scenario. maximum a posteriori probability estimate. In [135], a sum-
In principle, we can replace the digital SVD precoder by the product-algorithm (SPA) based signal detector is designed by
classical zero-forcing (ZF) precoder to suppress multi-user utilizing the concept of clustered factor graph. In [136], the
interference. However, such an extension is not straightforward authors develop an efficient algorithm for optimal Bayesian
for the schemes designed for N1 and N2. To tackle this data detection in the mmWave OFDM system with low-
problem, in [172], a two-stage precoding scheme was proposed resolution ADCs. A power allocation (PA) scheme is also
for architecture N1. In the first stage, FA is searched from a proposed to minimize the average symbol error rate in [136].
predefined codebook to maximize the desired signal power Nevertheless, besides these works above, there are still
of each user. In the second stage, a digital precoder similar
7 Note that the DACs at the transmitter usually consume less power than
to ZF precoder is designed to cancel multi-user interference.
ADCs for mmWave MIMO systems. Therefore, employing low-resolution
The multi-user hybrid precoding scheme for N2, however, is ADCs instead of DACs is more promising to reduce the energy consumption
still an open problem requiring further research efforts. and hardware cost [131].
 15

some open issues for the low-resolution ADC based architec- V. ACCESS , BACKHAULING , AND C OVERAGE
ture [131], e.g., how to design signal processing for broadband
A. Multiple-access Technologies
mmWave MIMO channels, which require future investigation.
Multiple-access technologies are necessary for supporting
multiple users in mobile networks, and they have been widely
F. Recent Progress investigated in the lower frequency bands. Different multiple
In the current special issue of IEEE JSAC, the signal access technologies have been utilized in practical systems,
processing schemes on mmWave MIMO are investigated including frequency division multiple access (FDMA), time
in [174]–[189]. Specifically, in [174], a receive antenna selec- division multiple access (TDMA), code division multiple
tion (RAS)-aid spatial modulation scheme is proposed to re- access (CDMA), and orthogonal frequency division multiple
duce the RF complexity, and an iterative algorithm is proposed access (OFDMA). These multiple-access technologies are also
to search the antenna index with low complexity. In [175], applicable to mmWave, but in different flavors than in lower
a channel estimation scheme is proposed by utilizing the frequency bands due to the increased complexity caused by
structured sparsity of mmWave MIMO channel, and a training the greatly increased bandwidth, and the different channel
sequence (TS) is designed by the genetic algorithm. In [176], characteristics in mmWave bands, e.g., highly directional
a low complexity non-iterative interference alignment (IA) transmissions. More importantly, as described in Section IV,
scheme for multi-cell mmWave MIMO systems is proposed. the shorter wavelengths at mmWave frequencies make MIMO
In [177], a CANDECOMP/PARAFAC decomposition-based technology suitable for mmWave, since it is possible to use
channel estimation scheme is proposed for wideband mmWave more antennas in the same physical space. One of the key
MIMO by utilizing the tensor theory. The Cramer-Rao bounds advantages of MIMO is the spatial resolution and beamform-
of the estimated channel parameters are also derived to show ing capability that it provides. In this subsection, major non-
the advantages of the proposed scheme. In [178], a new orthogonal multiple-access technologies that exploit the spatial
antipodal curvedly tapered slot antenna is designed to generate or power domains will be discussed in the context of mmWave.
circularly polarized field. It can achieve high gains with low This includes spatial division multiple access (SDMA), non-
hardware complexity in E-band and W-band. [179] proposes a orthogonal multiple access (NOMA), and finally random ac-
single-user multi-beam transmission scheme in the beamspace, cess.
and the corresponding multi-beam selection, cooperative beam 1) SDMA: In addition to the orders-of-magnitude larger
tracking, multi-beam power allocation, and synchronization bandwidths, as compared to conventional systems operating
are investigated. [180] develops a low-complexity channel esti- below 6 GHz, a multiplexing gain can be achieved by mul-
mation for hybrid mmWave MIMO systems, and investigates tiplexing of users in the spatial domain. This technology is
the achievable sum-rate of zero forcing with the estimated known as SDMA was introduced in 1990s [192], and Massive
channel under the consideration of system imperfection. [181] MIMO is the latest branch of this tree [36]. 802.11ac was
investigates the hybrid precoding for wideband mmWave the first major wireless standard that integrated SDMA [193].
MIMO, and proposes a unified heuristic design for two dif- SDMA can ideally increase the sum rate proportionally to
ferent hybrid precoding architectures, i.e., the fully-connected the number of multiplexed users, provided that the BS is
and the partially-connected architectures, to maximize the equipped with at least as many RF transceiver chains as
overall spectral efficiency. [182] proposes a novel unified all the users have in total. Due to the highly directional
hybrid precoding design for fully- and partially-connected transmissions in mmWave communication systems, users from
hybrid architectures from the view of energy efficiency instead different directions may be well separated using different
of spectral efficiency. [183] characterizes the gains of pilot spatial beams, which is also known as favorable propagation
precoding and combining in terms of channel estimation qual- [191]. The design of transmit precoding/beamforming and
ity and achievable data rate. [184] considers the sub-28 GHz receive combining for SDMA is generally a mature topic
communications that do not exhibit enough directivity and [194], [195], but the special channel estimation and channel
selectively, and tackles the sum-rate maximization problem characteristics of hybrid systems require some further work on
based on the concept of difference of log and trace (DLT) this topic [196], [197].
bound. [185] analyzes the achievable rate and energy efficiency One key difference between systems operating in mmWave
of hybrid precoding receivers with low resolution ADC, and band and at lower frequencies is the coverage area, which
shows it robustness to small automatic gain control imperfec- is substantially smaller in mmWave. Hence, there is generally
tions. [186] investigates the hybrid precoding design with an- fewer users to multiplex by SDMA in mmWave. Nevertheless,
tenna selection, and decomposes the whole problem into three one of the critical challenges of SDMA is how to serve
sub-problems which are solved via an alternating optimization multiple users when the number of users is larger than the
method. [187] proposes a hybrid precoding scheme based on number of antennas. It is necessary to group users so that users
Kronecker decomposition for multi-cell multi-user mmWave from different groups may access the BS at the same time,
MIMO systems. [188] investigates beamforming training for while not causing significant interference to each other. After
partially-connected hybrid architectures, and proposes two realizing the user grouping, user scheduling should also be
multi-resolution time-delay codebooks. In [188], fundamental considered to select the users from different groups to access
limits of beam alignments are studied under different search the BS at the same time. Users in the same group can be
approaches (exhaustive or hierarchical). served orthogonally (or semi-orthogonal) in time, frequency,
 16

or code domain. A few effective scheduling algorithms have [213]. An overview of random access in mmWave mobile
been proposed in [202]–[210] to enhance throughput. networks has been presented in [211], where the important
2) NOMA: An alternative to the non-orthogonal spatial issues for the design of a random access channel with respect
multiplexing provided by SDMA is to perform non-orthogonal to initial access, handover, uplink-downlink configuration, and
multiple access (NOMA) in the power domain. This approach scheduling have been discussed in detail.
is considered as one of the candidates for improving the More recently, there are some recent works emphasize on
spectral efficiency and connectivity density in 5G [223]– lower frequencies in ad hoc wireless network scenarios or,
[227]. At the BS, different signals intended for different more recently, on the 60 GHz IEEE 802.11ad WLAN and
users are superimposed on each other after classic channel WPAN scenarios [214]. While the authors in [215] consid-
coding and modulation. Multiple users share the same time- ered an exhaustive method which sequentially scans the 360
frequency resources, and are then detected at the receivers degree angular space, the authors in [216] proposed the use
by successive interference cancellation (SIC). In this way, of directional cell discovery procedure where the angular
the spectral efficiency can be enhanced at the cost of an space is scanned in time-varying random directions using
increased receiver complexity compared to conventional or- synchronization signals periodically transmitted by the base
thogonal multiple access (OMA), where the potential inter- stations. In [217], different scanning and signaling methods
ference is treated as noise. NOMA has been considered in were compared, with respect to the initial access design
mmWave communications in recent literatures. Specifically, options, in order to assess the respective access delays and
the performance of NOMA in mmWave communications was system overheads. The analysis shows that low-resolution fully
evaluated in [228] and [229], and simulation results have digital architectures have significant benefits when compared
shown that NOMA can achieve better channel capacity than with single-stream analog beamforming. In order to reduce
OMA in both uplink and downlink mmWave communication the delay due to exhaustive search procedures, the authors
systems. Furthermore, the capacity performance of NOMA- in [211], [217], [218] implemented a faster user discovery
mmWave-massive-MIMO systems was investigated in [230], method which employs a two-stage hierarchical procedure,
and simulation results indicated that enormous capacity im- while the use of context information about user and/or BS
provement can be achieved compared to the existing LTE positions, provided by a separate control plane, was considered
systems. In addition, sum rate and outage probabilities of in [214], [219]. In a refinement to the method in [219], the
mmWave-NOMA systems with random beamforming were procedure to capture the effects of position inaccuracy and
analyzed in [231], where two users can be served by NOMA obstacles was implemented in [220]. In addition, the use
in each beam. Furthermore, a transmission scheme that uses of booster cells which operate at mm-Waves and under the
NOMA in mmWave beamspace MIMO has been proposed coverage of a microwave-based anchor cell was proposed by
in [232]. By using intra-beam superposition coding and SIC the authors in [221]. In this arrangement, the booster BS gets
under the framework of NOMA in the proposed beamspace information about user locations from the anchor BS, enabling
MIMO-NOMA system, more than one user can be simul- it to directly steer the transmit beams towards the user position.
taneously supported in each beam, which is different from Finally, the authors in [222] showed that the performance of
conventional SDMA, where only the signal intended for one analog beamforming degrades in presence of errors in the
user is transmitted in each beam. Consequently, the number of available context information provided during the initial cell
served users can be significantly increased for a given number search procedures.
of beams and RF chains, and the system achievable sum
rate can be also improved. Besides, beam division multiple
access (BDMA) with per-beam synchronization (PBS) in time B. Backhauling
and frequency for wideband massive MIMO transmission To meet the aggressive 5G KPIs discussed in Section I,
over mmWave/Terahertz (THz) bands has been proposed in an UDN deployment is considered as a promising system
[233], beam scheduling for both UL and DL BDMA and a architecture to enable Gbps user experience and seamless
greedy beam scheduling algorithm has also been developed. coverage in mobile networks [235], [236]. In other words,
Additionally, the design challenges of NOMA in mmWave many small-cell BSs are densely deployed as hotspots (e.g., in
due to beamforming was investigated in [234]. Note that office buildings, shopping malls, residential apartments) that
with the enhanced pathloss in mmWave communications, the would greatly offload the macro cells. Hence, the backhaul
interference experienced by the users in NOMA can be signif- between the macro-cell BS and the associated small-cell BSs
icantly reduced [228]. Considering this harmony between the should provide large bandwidth with reliable link transmission.
mmWave channel characteristics and the principle of NOMA, Besides, energy efficiency and deployment cost are also key
the use of NOMA for mmWave communications is a research considerations for operators. It has been demonstrated that
direction deserves further investigation. backhaul links with 1–10 Gbps is required to effectively
3) Random access: This is primarily used for initial access support UDN [235]. Conventional optical fiber supports large
and handover, which is very important in system design. data rates and reliable link transmission, but its application
Since random access cannot fully benefit from beamforming to UDN as backhaul may not be an economical choice for
due to the lack of information on the best transmit-receive operators due to the restriction of deployment and installation
beam pair, the design of the random access channel becomes [237]. Hence, wireless backhaul is more attractive to overcome
more challenging in mmWave communication systems [211]– the geographical constraints. Microwave backhaul in sub-6
 17

GHz and 20 GHz bands have been successfully deployed A visible LoS region of location X is defined as the set of
in current mobile networks. However, the available licensed locations which can be connected to X with a LoS link. If
spectrum in these bands are limited, and insufficient to meet we consider the case where the blockages are impenetrable,
the demand of 5G backhauling, and thus mmWave with wider the average size of a visible region in a cellular network can
bandwidth is preferred. be shown to be 2πe−p / β 2 [242] and the average number
Using mmWave bands for backhauling in UDN is desirable of BSs which has a LoS link with a user is 2π µe−p / β 2 ,
due to the following reasons [237]. where µ is the density of the BS PPP. Therefore, in order
• Large bandwidth: The large amount of underutilized to achieve acceptable coverage, we can either increase the
mmWave including unlicensed V-band (57–67 GHz) and density of the BS deployment or reduce the physical length of
lightly licensed E-band (71–76 GHz and 81–86 GHz) (the the communication links between two nodes in the network by
specific regulation may vary from country to country) deploying more intermediate relays. The paper [173] system-
can provide potential GHz transmission bandwidth. For atically studies the blockage problem of mmWave networks.
example, more than 1 Gbps backhaul capacity can be The impact of reflection and relaying are also investigated. It is
supported over a 250 MHz channel in the E-band [238]. shown that if the LoS signals are blocked, the reflection signals
• Reduced interference: The coverage distance for E-band may be useful in regions, while for some regions relaying
is up to several km due to rain attenuation, while that for is preferred. Reference [173] also studies the optimal routing
V-band is about 50–700 m due to both the rain and oxygen problem in multiple relay mmWave networks. Recently, an
attenuation. Due to high pathloss, mmWave is suitable optimal opportunistic strategy for access point deployment is
for UDN, where improved frequency reuse and reduced proposed in [247] to well balance the overhead and capacity
inter-cell interference are expected. It should be pointed under random blockage.
out that rain attenuation is not a big issue for mmWave [243] takes a stochastic geometry approach to the con-
used for backhaul in UDN. For example, if we consider nectivity of mmWave networks with multi-hop relaying. It is
very heavy rainfall of 25 mm/h, the rain attenuation is shown that multi-hop relaying can greatly improve the con-
only around 2 dB in the E-band for a backhaul link of nectivity compared to the single-hop mmWave transmission.
200 m [238]. The results also show that to obtain near-optimal connectivity
• In-band backhauling: The backhaul and user access the relaying route window should be about the size of the
links are conventionally carried out in different frequency obstacle buildings. The coverage probability is an important
bands. Multiplexing backhaul and access on the same performance metric in a mmWave cellular network. It is
mmWave frequency band, also named in-band backhaul defined as the probability that the destination is able to receive
[235], has obvious cost benefits from the hardware and a signal with a certain threshold SNR T:
frequency reuse perspective. However, as in any in- Pc (T ) = Pr(SNR > T ). (10)
band backhauling scenario, the interference between the
backhaul and the user access links must be controlled; To further improve the coverage, [244] studies the possibility
for example, by beamforming and signal processing. of BS cooperation in the downlink of mmWave networks in a
stochastic geometry framework. It is shown that cooperation
among randomly located BSs can effectively increase the
C. Coverage and Connectivity
coverage probability.
Since mmWave signals in general have high penetration By using tools from stochastic geometry, a general and
loss, they are very sensitive to the blockage by walls and other tractable framework for coverage analysis with arbitrary distri-
objects. In [242], a mathematical framework for modeling the butions for interference power and arbitrary antenna patterns
random blockages that occur when users are moving around was developed and applied to mmWave ad hoc networks
was developed by leveraging concepts from random shape exploiting directional antenna arrays in [245]. It is shown
theory. In this model, both BSs and users are assumed to form that the coverage probabilities of mmWave ad hoc networks
a homogeneous Poisson point process (PPP) on the plane. increase as a non-decreasing concave function with the antenna
Random buildings are modeled as rectangles with random array size. Numerical results also show that largescale antenna
sizes and orientations whose centers form a PPP on the plane. arrays are required for satisfactory coverage in mm-wave
Let K denote the total number of blockages crossing a link. As networks. To further enhance the connectivity and session
shown in [242], K follows a Poisson distribution with the mean continuity, multi-connectivity strategies were developed to
of βR + p, where β = 2λ(E[W ] + E[L])/π, p = λE[L]E[W ], leverage multiple simultaneous small cell connections in ul-
E[L] and E[W ] represent the average length and width of a tradense urban deployments of mmWave networks in [246].
random blockage building, respectively. Then, the probability The benefits of multi-connectivity strategies were investigated
that link of length R is free from blockage has an exponential by taking into account: (i) the intricacies of mmWave radio
distribution, given by propagation in realistic urban environments; (ii) the dynamic
P(K = 0) = e−(βR+p) . (9) mmWave link blockage due to human mobility; and (iii)
the multi-connectivity network behavior to preserve session
It should be noted that in practical mmWave systems the continuity. Results show that even simpler multi-connectivity
parameters β and p in (9) need to be obtained through schemes bring notable improvements to session-level mmWave
experimental fitting. operation in realistic environments.
 18

To achieve a robust coverage and connectivity of mmWave services where high-speed data transmission is needed. This is
networks, a heterogeneous mm-wave network architecture particularly important in small cells and dense urban scenarios.
consisting of small-cell BSs operating at mmWave bands and Other applications are backhaul for last-mile fiber replacement,
macro-cell BSs, operating at microwave frequencies, should be wireless fronthauls, etc. However, the research has not been
considered in practice. The macro-cells will be used as a sig- very clear on how mmWave affects the latency and reliability
naling network for cell discovery and signaling transmission to aspects.
guarantee full coverage and provide reliable control channels. 2) mmWave and massive MIMO: The 5G vision is pushing
The small-cell BSs form a data subnetwork that provides high for a fundamental change in the mobile networks, starting with
data rates using mmWave bands for the users that they cover. the lowest physical layer (PHY). The two core PHY technolo-
gies, that will set 5G NR apart from previous radio access
VI. S TANDARDIZATION AND D EPLOYMENT technologies (RATs), are mmWave and massive MIMO. Both
Despite of the huge potential and great number of benefits put forth different paradigms that make a break with many
in using mmWave in mobile networks, there are still a lot current understandings in the wireless propagation, signal
of skepticisms, particularly from investors, as for whether processing, device manufacturing and eventual network design
the technology is suitable for cellular coverage and mobility [113]. Moreover, there is a clear trend that mmWave and
scenarios. Due to the great potential of mmWave communi- massive MIMO will work in cooperation to ensure a successful
cations as described in previous sections, 3GPP is working 5G operation. The great SDMA capability of massive MIMO
towards standardizing mmWave for in the 5G New Radio (NR) is ideal in the frequency bands below 6 GHz, where a large
interface. More specifically, 3GPP is working on Release 14 area is covered and many users with NLoS channels need
(will be finalized around June 2017), which will include the to be served simultaneously. The high capacity of mmWave
channel modeling for radio above 6GHz. From the second communications is ideal for hotspots that need not guarantee
half of 2017, 3GPP will work on release 15, which will coverage or support mobility, but can provide great service to
deliver the first set of 5G standards. Commercial use of the the LoS users that are static or moving at pedestrian speed. In
unlicensed mmWave band is not a new territory, but mmWave 3GPP, the maximum number of the RF chains for the NR BS
bands have been used since the 1980s. However, the existing and UE is determined as 32 and 8 respectively. The maximum
standards and applications have only been for static ultra- number of antenna elements can go up to 1024 at the BS and
high-definition multimedia data transfer. Specifically, IEEE 64 at the UE (for 70 GHz), respectively.
802.11ad/aj for wireless local area network (WLAN), IEEE 3) Hybrid beamforming Architecture: At the PHY layer,
802.15.3c and ECMA-387 for wireless personal area network the hybrid analog-digital beamforming technique was agreed
(WPAN), WirelessHD for video area networking (VAN) are to be used in 5G systems at the 3GPP RAN1 meeting in 2016
standards at unlicensed 60GHz and 45GHz for providing [125]. As discussed in Section IV, the hybrid architecture may
short-range point-to-point (P2P) communications. be used by both the BSs and UEs. It is worth noting that
In this section, we will introduce the key features of the the hybrid beamforming would not only be used to boost the
aforementioned existing and in-development mmWave stan- data rate for data transmission, but also be used at the control
dards, followed by an overview of the plan for commercial channel to enhance cell coverage. Thereafter, it is deemed nec-
cellular deployment. essary to have Physical Uplink Control Channel (PUCCH) to
include CSI that is related to analog beamforming information.
As for the Physical Downlink Control Channel (PDCCH), it is
A. 3GPP’s New Radio at mmWave Band desirable to have a common design for high and low frequency
1) Vision and use cases: 5G is envisioned to provide bands. Both uniform antenna arrays (i.e., antenna elements
orders-of-magnitude improvements in the peak data rate, with the same polarization from multiple panels are uniformly
network capacity, latency, availability, and reliability over distributed in horizontal and vertical dimensions, respectively)
legacy networks. Meanwhile, the deployment should cooperate and non-uniform antenna arrays [250] (i.e., antenna elements
seamlessly with legacy networks and provide fundamental with same polarization from multiple panels are not uniformly
shifts in cost and energy efficiency [120]. As discussed in distributed in horizontal or vertical dimension) should be
Section I, 3GPP has defined three use cases for 5G NR since supported in 5G systems. This means that a flexible design
2016; namely, eMBB, mMTC and URLLC. EMBB is targeted of the hybrid beamforming is needed, which is not limited to
for mobile broadband services that require extraordinary data calibrated arrays of a particular geometry. Spatial beams can be
rates; mMTC is the basis for connectivity in Internet of Things generated from separate antenna panels (or different sections
(IoT); and URLLC is needed for applications which have of the antenna array) to serve different users. Meanwhile, it
stringent latency and reliability requirements. 3GPP has also is possible to allocate different transmission tasks (i.e., data
identified its interest in the frequency bands above 6 GHz channel and control channel) to each subarray. Nevertheless,
and up to 100 GHz for NR. The channel model for above fully digital beamforming is not disregarded by industry, and
6 GHz is defined in 3GPP Specification 38.900/38.901 [21], low-resolution beamforming could work at the lower end of
[98]. Network operators around the world are exploring the the mmWave spectrum with manageable complexity.
possibility of using parts of this spectrum for licensed mo- 4) PHY layer design: Some common understandings on
bile communications to meet diversified services from their the way forward for the standardization of the PHY layer
customers. mmWave is naturally envisioned to provide eMBB for mmWave communications have been agreed upon. More
 19

TABLE III
S UMMARY OF MM WAVE PERFORMANCE MEASUREMENT CAMPAIGNS

Huawei QUALCOMM Ericsson Samsung Nokia

Freq band 28 GHz [252] 28 GHz [254] 28 GHz [257] 28 GHz [256] 28 GHz
73 GHz [253] 73 GHz [255]
Architecture Digital (@28 G) Hybrid Hybrid Hybrid Hybrid(@28GHz)
Analog (@73 G) RF(@73GHz)
Peak Throughput (Gbps) 35(@73 G) 10 14 7.5 11

detailed discussion and specifications are still open and are 2) NLoS transmission: metal and concrete walls pose high
expected to be finalized in the 3GPP Release 15 in 2018. penetration loss; heavy foliage poses severe shadowing
Research should focus on, but not be limited to, the follow- loss; reasonable reflection loss but high diffraction loss.
ing areas: i) Beam-management, which includes the beam- 3) Coverage: indoor ≤ 100 m; urban outdoor: ≤ 350 m;
sweeping procedure, beam-selection based on CSI feedback or outdoor-to-indoor: ≤ 20 m.
beam-reciprocity assumptions, beam-tracking and recovery for 4) Beam-tracking: It can be done for a single user at
mobility support, etc.; ii) the corresponding Reference Signal pedestrian (15 km/h) or low speed (40 km/h); multi-user
(RS) design for beam-management; and iii) Control channel tracking (spatially) is lacking.
design. The next phase of testing will focus on the feasibility in
Beam-sweeping is a main method for the analog part of more difficult use cases for a single channel; for instance, out-
the hybrid beamforming, and its procedure design has a door BS serving indoor UEs, maximum outdoor coverage test,
big impact on the system implementation. To name a few, high-mobility/vehicular scenarios, beam-tracking for multiple
how to design an efficient and reliable sweeping procedure UEs, dense urban services, multi-cell networking and core
during the initial access stage, how to perform beam-tacking network support. This will require the consumer equipments
for a UE as the propagation channel changes (due to fast to be developed and ready for such tests, so that end-to-
fading or mobility), and how to maintain a connection (i.e., end performance evaluation can be possible. It is most likely
beam-recovery) in the case of link failure and/or blockage. that mass deployment of mmWave technology for mobile
A well designed beam-selection mechanism can ensure good networks will occur after 2020, since the progress still largely
signal strength to reap the benefits of data transmission at the hinges on the standardization process and the resolving of the
mmWave frequencies with a wide bandwidth. The accuracy of aforementioned challenges.
the selection is dependent on the analog beam-feedback, the
digital precoding matrix feedback and the Channel Quality
Index (CQI) feedback. The common understanding is that the VII. C ONCLUSIONS
UE will report measurement results on different BS transmit Despite of high potentials of providing multiple Gbps rates,
beams to aid beam-selection at the BS. many technical challenges have to be solved for mmWave
communications to become a mainstream technology in mo-
bile networks. In recent years, large efforts have made to
B. Prototypes and Deployment plan
tackle the various challenges and many excellent results have
There has been tremendous efforts from various orga- been reported. This tutorial paper has summarized the recent
nizations (vendors, operators, universities, etc.) in building technical progresses in mmWave communications for mobile
hardware platforms for mmWave channel measurement (as networks, including channel measure/modeling, MIMO de-
described in Section II) and for proof-of-concept prototyping. sign, multiple access, performance analysis, standardization,
This is the necessary step to test for different frequency and deployment. Many directions for future research work
bands, different use cases, and their respective KPIs before have also been identified. From our point of view, finding
the mass deployment. In the past year of 2016, there have effective solutions for applying mmWave technology in high
been many feasibility tests and functionality tests for potential mobility environments, enabling enhanced transmission dis-
key techniques proposed for mmWave. A large quantity of tance, combating hardware impairments, and achieving high
measurement data has been accumulated, which is vital for energy-efficiency are very important and interesting challenges
clarification on the system architecture and the PHY design, as to tackle. From a broader perspective, the RF implementation
well as the subsequent standardization and commercialization of mmWave technology is very important and a long-term goal
for mmWave communication. Table III lists some known should be to obtain a cost-efficient fully digital implementa-
operational prototypes and their key system parameters. tion.
Current measurements are largely focused on four areas of
interest.
R EFERENCES
1) Basic throughput performance: the maximum number
of reported spatial streams is two (not counting dual- [1] Ericsson AB, “Traffic exploration tool, interactive online tool”, Available
at: http://www.ericsson.com/TET/trafficView/loadBasicEditor.ericsson8.
polarization of antenna elements) and the maximum spec- [2] “IMT vision-framework and overall objectives of the future development
tral efficiency is less than 20 bps/Hz. of IMT for 2020 and beyond,” ITU-R M. 2083-0, Sep. 2015.
 20

[3] Afif Osseiran et al., “Scenarios for 5G mobile and wireless communica- [30] H. Xu, V. Kukshya, T. S. Rappaport, “Spatial and temporal charac-
tions: the vision of the METIS project,” IEEE Commun. Mag., vol. 52, teristics of 60-GHz indoor channels,”IEEE Journal of Selected Area in
no. 5, pp. 26–35, May 2014. Communications, vol. 20, no. 3, pp. 620–630, Apr. 2002.
[4] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. Soong, [31] X. Zhang and J. G. Andrews, “Downlink Cellular Network Analysis
and J. C. Zhang, “What will 5G be?” IEEE J. Sel. Areas Commun., With Multi-Slope Path Loss Models,” IEEE Trans. Commun., vol. 63,
vol. 32, no. 6, pp. 1065–1082, Jun. 2014. no. 6, pp. 1881–1894, May 2015.
[5] Z. Pi and F. Khan, “An introduction to millimeter-wave mobile broadband [32] G. R. MacCartney, S. Sun, T. S. Rappaport et al., “Millimeter wave
systems,” IEEE Commun. Mag., vol. 49, no. 6, pp. 101–107, Jun. 2011. wireless communications: New Results for Rural Connectivity,” in Proc.
[6] E. Björnson, E. G. Larsson, and T. L. Marzetta, “Massive MIMO: Ten All Things Cellular 16: 5th workshop on All things cellular proceedings,
myths and one critical question,” IEEE Commun. Mag., vol. 54, no. 2, in conjunction with ACM mobiCom, Oct. 2016.
pp. 114–123, Feb. 2016. [33] C. Kourogiorgas, S. Sagkriotis, and A. D. Panagopoulos, “Coverage and
[7] T. S. Rappaport et al., “Millimeter wave mobile communications for 5G outage capacity evaluation in 5G millimeter wave cellular systems: Impact
cellular: It will work!” IEEE Access, vol. 1, pp. 335–349, May 2013. of rain attenuation,” in Proc. 9th European Conf. Antennas Propagation
[8] J. Du and R. A. Valenzuela, “How Much Spectrum is Too Much in (EuCAP), Apr. 2015, pp. 1–5.
Millimeter Wave Wireless Access” IEEE J. Selected Area on Commun., [34] Y. P. Zhang, P. Wang, and A. Goldsmith, “Rainfall effect on the
2017 (to appear). performance of millimeter-wave MIMO systems,” IEEE Trans. Wireless
[9] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, Commun., vol. 14, no. 9, pp. 4857–4866, Sep. 2015.
“Five disruptive technology directions for 5G,” IEEE Commun. Mag., vol. [35] “E-Band Technology,” E-Band Communications, available:
52, no. 2, pp. 74–80, Feb. 2014. http://www.e-band.com/index.php?id=86.
[10] F. W. Vook, E. Visotsky, T. A. Thomas and A. Ghosh, “Performance [36] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta,
Characteristics of 5G mmWave Wireless-to-the-Home,” In Proc. Asilomar O. Edfors, and F. Tufvesson, “Scaling up MIMO: Opportunities and
2016. challenges with very large arrays,” IEEE Signal Process. Mag., vol. 30,
[11] H. Huang, C. B. Papadias, S. Venkatesan,“MIMO Communication for no. 1, pp. 40–60, Jan. 2013.
Cellular Networks” Springer US, 2012. [37] A. Ghosh, “The 5G mmWave radio revolution,” Microwave Journal,
[12] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellular vol. 59, no. 9, pp. 22–36, Sep. 2016.
wireless networks: Potentials and challenges,” Proc. IEEE, vol. 102, no. [38] J. E. Wieselthier, G. D. Nguyen, and A. Ephremides, “Energy-limited
3, pp. 366–385, Mar. 2014. wireless networking with directional antennas: The case of session-
[13] A. Ghosh, et. al., “Millimeter-wave Enhanced Local Area Systems: A based multicasting,” in IEEE 21st Annual Joint Conference of the IEEE
High-Data-Rate Approach for Future Wireless Networks,” IEEE Journal Computer and Communications Societies (INFOCOM 2002), Jun. 2002,
on Selected Areas in Commun., vol. 32, no. 6, pp. 1152–1163, June 2014. pp. 190–199.
[14] R. W. Heath, N. Gonzalez-Prelcic, S. Rangan, W. Roh, and A. Sayeed, [39] I. Kang, R. Poovendran, and R. Ladner, “Power-efficient broadcast rout-
“An overview of signal processing techniques for millimeter wave MIMO ing in adhoc networks using directional antennas: Technology dependence
systems,” IEEE J. Sel. Top. Signal Process., vol. 10, no. 3, pp. 436–453, and convergence issues,” University of Washington, Washington, USA,
Apr. 2016. Tech. Rep. UWEETR-2003-0015, 2003.
[15] C. E. Shannon, “A mathematical theory of communication,” Bell Sys. [40] S. Singh, R. Mudumbai, and U. Madhow, “Interference analysis for
Tech. J., vol. 27, no. 3, pp. 379–423, Jul. 1948. highly directional 60-Ghz mesh networks: The case for rethinking
[16] C. Dehos, J. L. Gonzalez, A. D. Domenico, D. Ktenas, and L. Dussopt, medium access control,” IEEE/ACM Transa. Net, vol. 19, no. 5, pp. 1513–
“Millimeter-wave access and backhauling: The solution to the exponential 1527, May 2011.
data traffic increase in 5G mobile communications systems?” IEEE [41] UMTS: Spatial channel model for Multiple Input Multiple Output
Commun. Mag., vol. 52, no. 9, pp. 88–95, Sep. 2014. (MIMO) simulations. ETSI 3rd Generation Partnership Project (3GPP).
[17] X. Ge, H. Cheng, M. Guizani, and T. Han, “5G wireless backhaul Sophia Antipolis Cedex, France. 3GPP TR 25.996, V12.0.0.
networks: Challenges and research advances,” IEEE Netw., vol. 28, no. [42] J. Yu, Y.-D. Yao, A. F. Molisch, and J. Zhang, “Performance evaluation
6, pp. 6–11, Jun. 2014. of CDMA reverse links with imperfect beamforming in a multicell
[18] Z. Gao, L. Dai, D. Mi, Z. Wang, M. A. Imran, and M. Z. Shakir, environment using a simplified beamforming model,” IEEE Trans Veh.
“MmWave massive-MIMO-based wireless backhaul for the 5G ultra- Tech., vol. 55, no. 3, pp. 1019–1031, Mar. 2006.
dense network,” IEEE Wireless Commun., vol. 22, no. 5, pp. 13–21, Oct. [43] J. Shen and L. W. Pearson, “The phase error and beam-pointing error in
2015. coupled oscillator beam-steering arrays,” IEEE Trans. Antennas Propag.,
[19] Z. Pi, J. Choi, and R. Heath, “Millimeter-wave gigabit broadband vol. 53, no. 1, pp. 386–393, Jan. 2005.
evolution toward 5G: Fixed access and backhaul,” IEEE Commun. Mag., [44] H. Li, Y.-D. Yao, and J. Yu, “Outage probabilities of wireless systems
vol. 54, no. 4, pp. 138–144, Apr. 2016. with imperfect beamforming,” IEEE Trans Veh. Tech., vol. 55, no. 5, pp.
[20] S. Singh, M. N. Kulkarni, A. Ghosh, and J. G. Andrews, “Tractable 1503–1515, 2006.
model for rate in self-backhauled millimeter wave cellular networks,” [45] T. Kim, B. Clerckx, D. J. Love and S. J. Kim, “Limited Feedback
IEEE J. Sel. Areas Commun., vol. 33, no. 10, pp. 2196–2211, Oct. 2015. Beamforming Systems for Dual-Polarized MIMO Channel,” IEEE Trans.
[21] 3GPP TR 38.901, http://www.3gpp.org/DynaReport/38901.htm. Wireless Commun., vol. 9, no. 11, pp. 3425–3439, Nov. 2010.
[22] K. Haneda, et. al. “Indoor 5G 3GPP-like Channel Models for Office [46] A. W. Doff, K. Chandra, and R. V. Prasad, “Sensor assisted movement
and Shopping Mall Environments,” in in Proc. 2016 IEEE International identification and prediction for beamformed 60 GHz links,” in 12th
Conference on Communications Workshops (ICCW), May 2016. Annual IEEE Consumer Communications and Networking Conference
[23] R.-A. Pitaval, O. Tirkkonen, R. Wichman, K. Pajukoski, E. Lahetkangas, (CCNC), 2015, pp. 648–653.
and E. Tiirola, “Full-duplex self-backhauling for small-cell 5G networks,” [47] S. Hur, T. Kim, D. J. Love, J. V. Krogmeier, T. Thomas, A. Ghosh et al.,
IEEE Wireless Commun., vol. 22, no. 5, pp. 83–89, Oct. 2015. “Millimeter wave beamforming for wireless backhaul and access in small
[24] R. Baldemair et al., “Ultra-dense networks in millimeter-wave frequen- cell networks,” IEEE Trans. Commun., vol. 61, no. 10, pp. 4391–4403,
cies,” IEEE Commun. Mag., vol. 53, no. 1, pp. 202–208, Jan. 2015. Oct. 2013.
[25] X. Ge, S. Tu, G. Mao, C. Wang, and T. Han, “5G ultra-dense cellular [48] C.-S. Choi, Y. Shoji, H. Harada, R. Funada, S. Kato, K. Maruhashi,
networks,” IEEE Wireless Commun., vol. 23, no. 1, pp. 72–79, Feb. 2016. I. Toyoda, and K. Takahashi, “RF impairment models for 60GHz-band
[26] S. Samarakoon, M. Bennis, W. Saad, M. Debbah, and M. Latva-aho, SYS/PHY simulation,” Tech. Rep., IEEE 802.15-06-0477-01-003c, Nov.
“Ultra dense small cell networks: Turning density into energy efficiency,” 2006.
IEEE J. Sel. Areas Commun., vol. 34, no. 5, pp. 1267–1280, May 2016. [49] E. Björnson, P. Zetterberg, M. Bengtsson, and B. Ottersten, “Capacity
[27] D. T. Emerson, “The work of Jagadis Chandra Bose: 100 years of limits and multiplexing gains of MIMO channels with transceiver impair-
millimeter-wave research,” IEEE Trans. Microwave Theory and Tech., ments,” IEEE Commun. Lett., vol. 17, no. 1, pp. 91–94, Jan. 2013.
vol. 45, no. 12, pp. 2267–2273, Dec. 1997. [50] E. Björnson, M. Matthaiou, and M. Debbah, “Massive MIMO with non-
[28] R. E. Ziemer, “An overview of millimeter wave communications,”in ideal arbitrary arrays: Hardware scaling laws and circuit-aware design,”
European Micorwave Conference, 1985. IEEE Trans. Wireless Commun., vol. 14, no. 8, pp. 4353–4368, Aug.
[29] H. H. Hmimy, S. C. Gupta, “Performance of frequency-hopped NPC- 2015.
SMA for broad-band personal communication services (B-PCS) at mil- [51] C. Rapp, “Effects of HPA-nonlinearity on a 4-DPSK/OFDM-signal for
limeter waves in an urban mobile radio environment,”IEEE Trans. Vehi. a digital sound broadcasting system,” in Proc. of the Second European
Tech., vol. 49, no. 1, pp. 90–97, Jan. 1999. Conference on Satellite Communications, Liege, Belgium, Oct. 1991.
 21

[52] A. Saleh, “Frequency-independent and frequency-dependent nonlinear [75] T. S. Rappaport, E. Ben-Dor, J. N. Murdock, and Y. Qiao, “38 GHz and
models of TWT amplifiers,” IEEE Trans. Commun., vol. 29, no. 11, pp. 60 GHz angle-dependent propagation for cellular & peer-to-peer wireless
1715–1720, Nov. 1981. communications,” in Proc. IEEE Int. Commun. Conf. (ICC), 2012, pp.
[53] B. Razavi, “Design of Millimeter-Wave CMOS Radios: A Tutorial,” 4568–4573.
IEEE Transactions on Circuits and Systems, vol. 56, no. 1, Jan. 2009. [76] J. N. Murdock, E. Ben-Dor, Y. Qiao, J. I. Tamir, and T. S. Rappaport, “A
[54] S. K. Yong, P. Xia, and A. Valdes-Garcia, “60GHz Technology for Gbps 38 GHz cellular outage study for an urban outdoor campus environment,”
WLAN and WPAN, John Wiley & Sons, 2011. in Rroc. IEEE Wireless Commun. Netw. Conf. (WCNC), 2012, pp. 3085–
[55] A. M. Niknejad and H. Hashemi, “Mm-Wave Silicon Technology, 60 3090.
GHz and Beyond, Springer, 2008. [77] T. S. Rappaport, F. Gutierrez, E. Ben-Dor, J. N. Murdock, Y. Qiao,
[56] B. Razavi, “Design considerations for direct-conversion receivers,” IEEE and J. I. Tamir, “Broadband millimeter-wave propagation measurements
Transactions on Circuits and Systems II, vol. 44, no. 6, pp. 428-435, June and models using adaptive-beam antennas for outdoor urban cellular
1997. communications,” IEEE Trans. Antenna. Propag., vol. 61, no. 4, pp.
[57] T. S. Rappaport, J. N. Murdock and F. Gutierrez, “State of the art in 1850–1859, Apr. 2013.
60-GHz integrated circuits and systems for wireless communications,” [78] H. J. Thomas, R. S. Cole, and G. L. Siqueira, “An experimental study
Proceedings of the IEEE, vol. 99, no. 8, pp. 1390–1436, Aug. 2011. of the propagation of 55 GHz millimeter waves in an urban mobile radio
[58] mmMAGIC Deliverable D2.1, “Measurement Campaigns and Initial environment,” IEEE Trans. Veh. Technol., vol. 43, no. 1, pp. 140–146,
Channel Models for Preferred Suitable Frequency Ranges,” Mar. 2016 Jan. 1994.
[Online]. Available: https://5g-mmmagic.eu/results/#deliverables. [79] G. Lovnes, J. J. Reis, and R. H. Raekken, “Channel sounding measure-
[59] M. Kim, J. Takada, Y. Chang, J. Shen, and Y. Oda, “Large scale ments at 59 GHz in city streets,” in Proc. IEEE 5th Int. Symp. Personal,
characteristics of urban cellular wideband channels at 11 GHz,” in Proc. Indoor, and Mobile Radio Commun. (PIMRC), 1994, pp. 496–500.
9th European Conf. Ant. Prop (EuCAP), 2015, pp. 1–4. [80] M. Kyro, K. Haneda, J. Simola, K. Nakai, K. i. Takizawa, H. Hagiwara,
and P. Vainikainen, “Measurement based path loss and delay spread
[60] H. Masui, M. Ishii, K. Sakawa, H. Shimizu, T. Kobayashi, and M.
modeling in hospital environments at 60 GHz,” IEEE Trans. Wireless
Akaike, “Microwave path-loss characteristics in urban LOS and NLOS
Commun., vol. 10, no. 8, pp. 2423–2427, Aug. 2011.
environments,” in Proc. IEEE Vehi. Technol. Conf. (VTC Spring), 2001,
pp. 395–398. [81] E. Ben-Dor, T. S. Rappaport, Y. Qiao, and S. J. Lauffenburger,
“Millimeter-wave 60 GHz outdoor and vehicle AOA propagation mea-
[61] M. Sasaki, W. Yamada, T. Sugiyama, M. Mizoguchi, and T. Imai, “Path
surements using a broadband channel sounder,” in Proc. IEEE Global
loss characteristics at 800 MHz to 37 GHz in urban street microcell
Telecommun. Conf. (GLOBECOM), 2011, pp. 1–6.
environment,” in Proc. 9th European Conf. Ant. Prop. (EuCAP), 2015,
[82] W. Keusgen, R. J. Weiler, M. Peter, M. Wisotzki, and B. Göktepe,
pp. 1–4.
“Propagation measurements and simulations for millimeter-wave mobile
[62] S. Rajagopal, S. Abu-Surra, and M. Malmirchegini, “Channel feasibility
access in a busy urban environment,” in Proc. 39th Int. Conf. Infrared,
for outdoor non-line-of-sight mmWave mobile communication,” in Proc.
Millimeter, and Terahertz waves (IRMMW-THz), 2014, pp. 1–3.
IEEE Veh. Technol. Conf. (VTC Fall), 2012, pp. 1–6.
[83] R. J. Weiler, M. Peter, T. Kĺźhne, M. Wisotzki, and W. Keusgen,
[63] Y. Azar et al., “28 GHz propagation measurements for outdoor cellular “Simultaneous millimeter-wave multi-band channel sounding in an urban
communications using steerable beam antennas in New York city,” in access scenario,” in Proc. 9th European Conf. Ant. Prop. (EuCAP), 2015,
Proc. IEEE Int. Conf. Commun. (ICC), 2013, pp. 5143–5147. pp. 1–5.
[64] M. Samimi et al., “28 GHz angle of arrival and angle of departure analy- [84] L. Simic, N. Perpinias, and M. Petrova, “60 GHz outdoor urban
sis for outdoor cellular communications using steerable beam antennas in measurement study of the feasibility of multi-Gbps mm-wave cellular
New York city,” in Proc. IEEE Veh. Technol. Conf. (VTC Spring), 2013, networks,” http://arxiv.org/abs/1603.02584.
pp. 1–6. [85] S. Nie, G. R. MacCartney, S. Sun, and T. S. Rappaport, “72 GHz
[65] G. R. MacCartney, M. K. Samimi, and T. S. Rappaport, “Omnidirec- millimeter wave indoor measurements for wireless and backhaul com-
tional path loss models in New York City at 28 GHz and 73 GHz,” munications,” in Proc. IEEE 24th Int. Symp. Personal, Indoor, Mobile
in Proc. IEEE Int. Symp. Personal, Indoor and Mobile Radio Commun. Radio Commun. (PIMRC), 2013, pp. 2429–2433.
(PIMRC), 2014, pp. 227–231. [86] G. R. MacCartney and T. S. Rappaport, “73 GHz millimeter wave
[66] H. Zhao et al., “28 GHz millimeter wave cellular communication propagation measurements foroutdoor urban mobile and backhaul com-
measurements for reflection and penetration loss in and around buildings munications in New York city,” in Proc. IEEE Int. Conf. Commun. (ICC),
in New York city,” in Proc. IEEE Int. Conf. Commun. (ICC), 2013, pp. 2014, pp. 4862–4867.
5163–5167. [87] M. Kyrö, S. Ranvier, V.-M. Kolmonen, K. Haneda, and P. Vainikainen,
[67] S. Nie, G. R. MacCartney, S. Sun, and T. S. Rappaport, “28 GHz and “Long range wideband channel measurements at 81-86 GHz frequency
73 GHz signal outage study for millimeter wave cellular and backhaul range,” in Proc. 4th European Conf. Ant. Prop., 2010, pp. 1–5.
communications,” in Proc. IEEE Int. Conf. Commun. (ICC), 2014, pp. [88] A. Roivainen, C. Ferreira Dias, N. Tervo, V. Hovinen, M. Sonkki, and
4856–4861. M. Latva-aho, “Geometry-based stochastic channel model for two-story
[68] C. Larsson, F. Harrysson, B.-. Olsson, and J.-E. Berg, “An outdoor-to- lobby environment at 10 GHz,” in IEEE Trans. on Antennas Propagation,
indoor propagation scenario at 28 GHz,” in Proc. 8th European Conf. vol. 64, no. 9, pp 3990-4003, Sept. 2016.
Ant. Prop (EuCAP), 2014, pp. 3301–3304. [89] J. Huang, C. X. Wang, R. Feng, J. Sun, W. Zhang and Y. Yang,
[69] S. Deng, M. K. Samimi, and T. S. Rappaport, “28 GHz and 73 GHz “Multi-Frequency MmWave Massive MIMO Channel Measurements and
millimeter-wave indoor propagation measurements and path loss models,” Characterization for 5G Wireless Communication Systems,” in IEEE
in Proc. IEEE Int. Conf. Commun. Workshop (ICCW), 2015, pp. 1244– Journals on Selected Areas in Communications, June 2017.
1250. [90] G. R. MacCartney and T. S. Rappaport, “Rural Macrocell Path Loss
[70] G. R. Maccartney, T. S. Rappaport, S. Sun, and S. Deng, “Indoor Models for Millimeter Wave Wireless Communications,” in IEEE Jour-
office wideband millimeter-wave propagation measurements and channel nals on Selected Areas in Communications, June 2017.
models at 28 and 73 GHz for ultra-dense 5G wireless networks,” IEEE [91] B. Ai, K. Guan, R. He, J. Li, G. Li, D. He, Z. Zhong, and K. M. Huq,
Access, vol. 3, pp. 2388–2424, Oct. 2015. “On Indoor Millimeter Wave Massive MIMO Channels: Measurement
[71] X. Wu, Y. Zhang, C. Wang, G. Goussetis, E-H. M. Aggoune, and M. and Simulation,” in IEEE Journals on Selected Areas in Communications,
M. Alwakeel, “28 GHz indoor channel measurements and modelling in June 2017.
laboratory environment using directional antennas,” in Proc. 9th European [92] A. I. Sulyman, A. T. Nassar, M. K. Samimi, G. R. Maccartney, T. S.
Conf. Ant. Prop (EuCAP), 2015, pp. 1–5. Rappaport, and A. Alsanie, “Radio propagation path loss models for 5G
[72] M. Kim, J. Liang, H. Kwon, and J. Lee, “Path loss measurement at cellular networks in the 28 GHz and 38 GHz millimeter-wave bands,”
indoor commercial areas using 28GHz channel sounding system,” in Proc. IEEE Commun. Mag., vol. 52, no. 9, pp. 78–86, Sept. 2014.
17th Int. Conf. Advanced Commun. Technol. (ICACT), 2015, pp. 535–538. [93] M. R. Akdeniz, Y. Liu, M. K. Samimi, S. Sun, S. Rangan, T. S.
[73] S. Hur, Y. Cho, T. Kim, J. Park, A. F. Molisch, K. Haneda, and M. Peter, Rappaport, and E. Erkip, “Millimeter wave channel modeling and cellular
“Wideband spatial channel model in an urban cellular environments at 28 capacity evaluation,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp.
GHz,” in Proc. 9th European Conf. Ant. Prop (EuCAP), 2015, pp. 1–5. 1164–1179, Jun. 2014.
[74] S. Hur et al., “Proposal on millimeter-wave channel modeling for 5G [94] G. R. Maccartney, T. S. Rappaport, M. K. Samimi, and S. Sun,
cellular system,” IEEE J. Sel. Topics Signal Process., vol. 10, no. 3, pp. “Millimeter-wave omnidirectional path loss data for small cell 5G channel
454–469, Mar. 2016. modeling,” IEEE Access, vol. 3, pp. 1573–1580, Aug. 2015.
 22

[95] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, “Wide- [122] X. Zhang, A. F. Molisch, and S.-Y. Kung, “Variable-phase-shift-based
band millimeter-wave propagation measurements and channel models for RF-baseband codesign for MIMO antenna selection," IEEE Trans. Signal
future wireless communication system design,” IEEE Trans. Commun., Process., vol. 53, no. 11, pp. 4091-4103, Nov. 2005.
vol. 63, no. 9, pp. 3029–3056, Sept. 2015. [123] P. Sudarshan, N. B. Mehta, A. F. Molisch, and J. Zhang, “Channel
[96] M. K. Samimi, T. S. Rappaport, and G. R. MacCartney, “Probabilistic statistics-based RF pre-processing with antenna selection," IEEE Trans.
omnidirectional path loss models for millimeter-wave outdoor communi- Wireless Commun., vol. 5, no. 12, pp. 3501-3511, Dec. 2006.
cations,” IEEE Wireless Commun. Lett., vol. 4, no. 4, pp. 357–360, Aug. [124] V. Venkateswaran and A. Veen, “Analog beamforming in MIMO com-
2015. munications with phase shift networks and online channel estimation,"
[97] S. Sun et al., “Investigation of prediction accuracy and parameter IEEE Trans. Signal Process., vol. 58, no. 8, pp. 4131-4143, Aug. 2010.
stability of large-scale propagation path loss models for 5G wireless [125] 3GPP, “Final report of 3GPP TSG RAN WG1 #85,” 2016, available
communications,” IEEE Trans. Veh. Technol., vol. 65, no. 5, pp. 2843– at: http://www.3gpp.org.
2860, May 2016. [126] J. Brady, N. Behdad, and A. Sayeed, “Beamspace MIMO for
[98] 3GPP TR38.900: http://www.3gpp.org/ftp/Specs/archive/38_series/38.900 millimeter-wave communications: System architecture, modeling, anal-
/38900-100.zip ysis, and measurements,” IEEE Trans. Ant. and Propag., vol. 61, no. 7,
[99] D. S. Baum, J. Hansen, and J. Salo, “An interim channel model for pp. 3814–3827, Jul. 2013.
beyond-3G systems: extending the 3GPP spatial channel model (SCM),” [127] X. Gao, L. Dai, Z. Chen, Z. Wang, and Z. Zhang, “Near-optimal
in Proc. IEEE Veh. Technol. Conf. (VTC Spring), 2005, pp. 3132–3136. beam selection for beamspace mmWave massive MIMO systems,” IEEE
[100] “Final Report on Link Level and System Level Channel Models,” Commun. Lett., vol. 20, no. 5, pp. 1054–1057, May 2016.
WINNER Deliverable 5.4, IST-2003-507581, Nov. 2005.
[128] N. Behdad and A. Sayeed, “Continuous aperture phased MIMO: Basic
[101] “WINNER II Channel Models,” WINNER II, Deliverable 1.1.2, IST-
theory and applications," in Proc. Allerton Conference, Sep. 2010, pp.
4-027756, Sep. 2007.
1196-1203.
[102] “WINNER+ Final Channel Models,” CP5-026 WINNER+, Deliverable
5.3, Jun. 2010. [129] A. F. Molisch and X. Zhang, “FFT-based hybrid antenna selection
[103] Study on 3D channel model for LTE, 3GPP TR36.873, June 2015. schemes for spatially correlated MIMO channels," IEEE Commun. Lett.,
[104] A. Kammoun. H. Khanfir. Z. Altman. M. Debbah, and M. Kamoun, vol. 8, no. 1, pp. 36-38, Jan. 2004.
“Preliminary results on 3D channel modeling: From theory to standard- [130] A. Adhikary, J. Nam, J.-Y. Ahn, and G. Caire, “Joint spatial division
ization,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1219–1229, and multiplexing: The large-scale array regime," IEEE Trans. Inf. Theory,
June 2014. vol. 59, no. 10, pp. 6441-6463, Oct. 2013.
[105] L. Correia, “Mobile broadband multimedia networks,” Elsevier, 2006, [131] A. Alkhateeb, J. Mo, N. Gonzalez-Prelcic, and R. W. Heath, “MIMO
ch. 6.8: The COST 273 MIMO channel model, pp. 364–383. precoding and combining solutions for millimeter-wave systems," IEEE
[106] L. Liu et al., “The COST 2100 MIMO channel model,” IEEE Commun. Commun. Mag., vol. 52, no. 12, pp. 122-131, Dec. 2014.
Mag., vol. 19, no. 6, pp. 92–99, Dec. 2012. [132] B. Le, T. W. Rondeau, J. H. Reed, and C. W. Bostian, “Analog-to-
[107] S. Jaeckel, L. Raschkowski, K. Borner, and L. Thiele, “QuaDRiGa: digital converters," IEEE Signal Process. Mag., vol. 22, no. 6, pp. 69-77,
A 3-D multi-cell channel model with time evolution for enabling virtual Nov. 2005.
fieldtrials,” IEEE Trans. Antenna Propag., vol. 62, no. 6, pp. 3242–3256, [133] J. Mo and R. W. Heath, “Capacity analysis of one-bit quantized MIMO
Jun. 2014. systems with transmitter channel state information," IEEE Trans. Signal
[108] A. Maltsev et al., “Channel models for 60 GHz WLAN systems,” Jan. Process., vol. 63, no. 20, pp. 5498-5512, Oct. 2015.
2010. [134] A. Mezghani, F. Antreich, and J. Nossek, “Multiple parameter estima-
[109] “Channel modeling and characterization,” FP7-ICT-608637, MiWEBA, tion with quantized channel output," in Proc. ITG Workshop on Smart
Deliverable 5.1, Jun. 2014. Antennas, Feb. 2010, pp. 143-150.
[110] “Guidelines for evaluation of radio interface technologies for IMT- [135] S. Wang, Y. Li, and J. Wang, “Multiuser detection in massive spatial
Advanced,” ITU-R M.2135-1, Dec. 2009. modulation MIMO with low-resolution ADCs," IEEE Trans. Wireless
[111] “METIS Channel Models,” ICT-317669 METIS, Deliverable 1.4, Jul. Commun., vol. 14, no. 4, pp. 2156-2168, Apr. 2015.
2015. [136] H. Wang, C. K. Wen and S. Jin, “Bayesian Optimal Data Detector for
[112] K. Haneda et. al., “5G 3GPP-like Channel Models for Outdoor Urban mmWave OFDM System with Low-Resolution ADC," to appear IEEE J.
Microcellular and Macro cellular Environments,” IEEE Vehicular Tech- Selected Areas on Commun., 2017.
nical Conferences (VTC), 2016. [137] Z. Gao, C. Hu, L. Dai, and Z. Wang, “Channel estimation for
[113] S. Han, C.-L. I, Z. Xu, and C. Rowell, “Large-scale antenna systems millimeter-wave massive MIMO with hybrid precoding over frequency-
with hybrid precoding analog and digital beamforming for millimeter selective fading channels,” IEEE Commun. Lett., vol. 20, no. 6, pp. 1259–
wave 5G,” IEEE Commun. Mag., vol. 53, no. 1, pp. 186–194, Jan. 2015. 1262, Jun. 2016.
[114] J. Kim and I. Lee, “802.11 WLAN: History and new enabling MIMO [138] J. Kotecha and A. Sayeed, “Transmit signal design for optimal estima-
techniques for next generation standards,” IEEE Commun. Mag., vol. 53, tion of correlated MIMO channels,” IEEE Trans. Signal Process., vol. 52,
no. 3, pp. 134–140, Mar. 2015. no. 2, pp. 546–557, Feb. 2004.
[115] R. Mendez-Rial, C. Rusu, A. Alkhateeb, N. González-Prelcic, and
[139] M. Biguesh and A. Gershman, “Training-based MIMO channel estima-
R. W. Heath, “Channel estimation and hybrid combining for mmWave:
tion: A study of estimator tradeoffs and optimal training signals,” IEEE
Phase shifters or switches?” in Proc. ITA Workshops, Feb. 2015, pp. 90–
Trans. Signal Process., vol. 54, no. 3, pp. 884–893, Mar. 2006.
97.
[116] R. Méndez-Rial, C. Rusu, N. González-Prelcic, A. Alkhateeb, and [140] A. Alkhateeb and R. W. Heath, “Frequency selective hybrid precoding
R. W. Heath, “Hybrid MIMO architectures for millimeter wave commu- for limited feedback millimeter wave systems,” IEEE Trans. Wireless
nications: Phase shifters or switches?” IEEE Access, vol. 4, pp. 247–267, Commun., vol. 64, no. 5, pp. 1801–1818, May 2016.
Jan. 2016. [141] A. Alkhateeb, O. El Ayach, G. Leus, and R. W. Heath, “Channel
[117] O. El Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath, estimation and hybrid precoding for millimeter wave cellular systems,”
“Spatially sparse precoding in millimeter wave MIMO systems,” IEEE IEEE J. Sel. Top. Signal Process., vol. 8, no. 5, pp. 831–846, Oct. 2014.
Trans. Wireless Commun., vol. 13, no. 3, pp. 1499–1513, Mar. 2014. [142] J. Wang, Z. Lan, C.-W. Pyo, T. Baykas, C.-S. Sum, M. A. Rahman,
[118] F. Sohrabi and W. Yu, “Hybrid beamforming with finite-resolution J. Gao, R. Funada, F. Kojima, H. Harada et al., “Beam codebook based
phase shifters for large-scale MIMO systems,” in Proc. IEEE SPAWC beamforming protocol for multi-Gbps millimeter-wave WPAN systems,”
Workshops, Jul. 2015, pp. 136–140. IEEE J. Sel. Areas Commun.
[119] A. Alkhateeb, Y.-H. Nam, J. Zhang, and R. W. Heath, “Massive MIMO [143] W. Shen, L. Dai, B. Shim, S. Mumtaz, and Z. Wang, “Joint CSIT
combining with switches,” IEEE Wireless Commun. Lett., vol. 5, no. 3, acquisition based on low-rank matrix completion for FDD massive MIMO
pp. 232–235, Jun. 2016. systems,” IEEE Commun. Lett., vol. 19, no. 12, pp. 2178–2181, Dec.
[120] X. Gao, L. Dai, Y. Sun, S. Han, and C.-L. I., “Machine learning inspired 2015.
energy-efficient hybrid precoding for mmwave massive MIMO systems,” [144] W. Shen, L. Dai, Y. Shi, B. Shim, and Z. Wang, “Joint channel
in Proc. IEEE ICC’17, Paris, France, May 2017. training and feedback for FDD massive MIMO systems,” IEEE Trans.
[121] X. Gao, L. Dai, S. Han, C.-L. I, and R. W. Heath, “Energy-efficient Veh. Technol., vol. 65, no. 10, pp. 8762–8767, Oct. 2016.
hybrid analog and digital precoding for mmWave MIMO systems with [145] J. Zhang, Y. Huang, Q. Shi, J. Wang, and L. Yang, “Codebook
large antenna arrays,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp. design for beam alignment in millimeter wave communication systems,”
998–1009, Apr. 2016. submitted for publication.
 23

[146] Z. Xiao, T. He, P. Xia, and X.-G. Xia, “Hierarchical codebook design [171] J. Hogan and A. Sayeed, “Beam selection for performance-complexity
for beamforming training in millimeter-wave communication,” IEEE optimization in high-dimension MIMO systems,” in Proc. CISS, Mar.
Trans. Wireless Commun., vol. 15, no. 5, pp. 3380–3392, May 2016. 2016, pp. 337–342.
[147] X. Gao, L. Dai, C. Yuen, and Z. Wang, “Turbo-like beamforming based [172] A. Alkhateeb, G. Leus, and R. W. Heath, “Limited feedback hybrid
on tabu search algorithm for millimeter-wave massive MIMO systems,” precoding for multi-user millimeter wave systems,” IEEE Trans. Wireless
IEEE Trans. Veh. Technol., vol. 65, no. 7, pp. 5731–5737, Jul. 2016. Commun., vol. 14, no. 11, pp. 6481–6494, Nov. 2015.
[148] T. Datta, N. Srinidhi, A. Chockalingam, and B. S. Rajan, “Random- [173] G. Yang, J. Du, and M. Xiao, “Maximum throughput path selection
restart reactive tabu search algorithm for detection in large-MIMO sys- with random blockage for indoor 60 GHz relay networks,” IEEE Trans.
tems,” IEEE Commun. Lett., vol. 14, no. 12, pp. 1107–1109, Dec. 2010. Commun., vol. 63, no. 10, pp. 3511–3524, Oct. 2015.
[149] T. Kim and D. J. Love, “Virtual AoA and AoD estimation for sparse [174] P. Yang, Y. Xiao, Y. Guan, Z. Liu, S. Li, and W. Xiang, “Adaptive SM-
millimeter wave MIMO channels,” in Proc. SPAWC Workshops, Jun. MIMO for mmWave communications with reduced RF chains," IEEE J.
2015, pp. 146–150. Sel. Areas Commun., 2017 (To Appear).
[150] X. Gao, L. Dai, S. Han, C.-L. I, and X. Wang, “Reliable beamspace [175] X. Ma, F. Yang, S. Liu, J. Song, and Z. Han, “Design and optimization
channel estimation for millimeter-wave massive MIMO systems with lens on training sequence for mmWave communications: A new approach
antenna array,” to appear in IEEE Trans. Wireless Commun., 2017. for sparse channel estimation in massive MIMO," IEEE J. Sel. Areas
[151] A. Alkhateeb, G. Leus, and R. W. Heath Jr, “Compressed sensing Commun., 2017 (To Appear).
based multi-user millimeter wave systems: How many measurements are [176] C. Wang, C. Qin, Y. Yao, and Y. Li, and W. Wang, “Low complexity
needed?” in Proc. ICASSP, Apr. 2015, pp. 2909–2913. interference alignment for mmWave MIMO channels in three-cell mobile
[152] K. Venugopal, A. Alkhateeb, N. G. Prelcic, and R. W. Heath, “Channel network," IEEE J. Sel. Areas Commun., 2017 (To Appear)..
Estimation for Hybrid Architecture Based Wideband Millimeter Wave [177] Z. Zhou, J. Fang, L. Yang, H. Li, Z. Chen, and R. S. Blum, “Low-rank
Systems," IEEE J. Sel. Areas Commun., 2017, (To appear). tensor decomposition-aided channel estimation for millimeter wave MI
[153] J. A. Tropp and A. C. Gilbert, “Signal recovery from random measure- MO-OFDM systems," IEEE J. Sel. Areas Commun., 2017 (To Appear).
ments via orthogonal matching pursuit,” IEEE Trans. Inf. Theory, vol. 53, [178] Y, Yao, X. Cheng, C. Wang, J. Yu, and X. Chen, “Wideband circularly
no. 12, pp. 4655–4666, Dec. 2007. polarized antipodal curvedly tapered slot antenna array for 5G applica-
[154] W. U. Bajwa, J. Haupt, A. M. Sayeed, and R. Nowak, “Compressed tions," IEEE J. Sel. Areas Commun., 2017 (To Appear)..
channel sensing: A new approach to estimating sparse multipath chan- [179] Q. Xue, X. Fang, and C. Wang, “Beamspace SU-MIMO for future
nels,” Proc. IEEE, vol. 98, no. 6, pp. 1058–1076, Jun. 2010. millimeter wave wireless communications," IEEE J. Sel. Areas Commun.,
[155] Z. Gao, L. Dai, and Z. Wang, “Structured compressive sensing based 2017 (To Appear).
superimposed pilot design in downlink large-scale MIMO systems,” [180] L. Zhao, D. W. Kwan Ng, and J. Yuan, “Multi-user precoding and
Electron. Lett., vol. 50, no. 12, pp. 896–898, Jun. 2014. channel estimation for hybrid millimeter wave systems," IEEE J. Sel.
[156] L. Dai and X. Gao, “Priori-aided channel tracking for millimeter-wave Areas Commun., 2017 (To Appear)..
beamspace massive MIMO systems,” in Proc. IEEE RADIO, Jul. 2016, [181] F. Sohrabi and W. Yu, “Hybrid analog and digital beamforming
pp. 1493–1496. for mmWave OFDM large-scale antenna arrays," IEEE J. Sel. Areas
[157] M. Cudak, T. Kovarik, T. A. Thomas, A. Ghosh, Y. Kishiyama, and Commun., 2017 (To Appear).
T. Nakamura, “Experimental mmWave 5G cellular system,” in Proc. IEEE [182] C. G. Tsinos, S. Maleki, S. Chatzinotas, and Björn Ottersten, “On
Globecom Workshops, Dec. 2014, pp. 377–381. the energy-efficiency of hybrid analog-digital transceivers for single- and
[158] C. Zhang, D. Guo, and P. Fan, “Tracking angles of departure and arrival multi-carrier large antenna array systems," IEEE J. Sel. Areas Commun.,
in a mobile millimeter wave channel,” in Proc. IEEE ICC, May 2016, 2017 (To Appear).
pp. 1–6. [183] N. N. Moghadam, H. Shokri-Ghadikolaei, G. Fodor, M. Bengtsson,
[159] N. Kabaoğlu, “Target tracking using particle filters with support vector and C. Fischione, “Pilot precoding and combining in multiuser MIMO
regression,” IEEE Trans. Veh. Technol., vol. 58, no. 5, pp. 2569–2573, networks," IEEE J. Sel. Areas Commun., 2017, (To Appear).
Jun. 2009. [184] H. Ghauch, T. Kim, M. Bengtsson, and M. Skoglund, “Sum-rate max-
[160] Y. Zhou, P. C. Yip, and H. Leung, “Tracking the direction-of-arrival imization in sub-28 GHz millimeter-wave MIMO interfering networks,"
of multiple moving targets by passive arrays: Asymptotic performance IEEE J. Sel. Areas Commun., 2017, (To Appear).
analysis,” IEEE Trans. Signal Process., vol. 47, no. 10, pp. 2644–2654, [185] K. Roth and J. A. Nossek, “Achievable rate and energy efficiency
Oct. 1999. of hybrid and digital beamforming receivers with low resolution ADC,"
[161] X. Gao, L. Dai, T. Xie, X. Dai, and Z. Wang, “Fast channel tracking for IEEE J. Sel. Areas Commun., 2017, (To Appear).
terahertz beamspace massive MIMO systems,” to appear in IEEE Trans. [186] X. Zhai, Y. Cai, Q. Shi, M. Zhao, G. Y. Li, and B. Champagne, “Joint
Veh. Technol., 2017. transceiver design with antenna selection for large-scale MU-MIMO
[162] R. A. Iltis, “A tracking mode receiver for joint channel estimation and mmWave systems," IEEE J. Sel. Areas Commun., 2017, (To Appear).
detection of asynchronous CDMA signals,” Conference Record of the [187] G. Zhu, K. Huang, V. K. N. Lau, B. Xia, X. Li, and S. Zhang,
Thirty-Third Asilomar Conference on Signals, Systems, and Computers, “Hybrid beamforming via the kronecker decomposition for the millimeter-
vol. 2, 1999 wave massive MIMO systems," IEEE J. Sel. Areas Commun., 2017, (To
[163] T. Yoo and A. Goldsmith, “Capacity and power allocation for fading Appear).
MIMO channels with channel estimation error,” IEEE Trans. Inf. Theory, [188] C. Lin, G. Y. Li, and L. Wang, “Subarray-based coordinated beam-
vol. 52, no. 5, pp. 2203–2214, Apr. 2006. forming training for mmWave and sub-THz communications," IEEE J.
[164] S. He, J. Wang, Y. Huang, B. Ottersten, and W. Hong, “Codebook based Sel. Areas Commun., 2017, (To Appear).
hybrid precoding for millimeter wave multiuser systems,” submitted for [189] C. Liu, M. Li, S. Hanly, I. Collings and P. Whiting, “Millimeter wave
publication. beamforming alignment: large deviations analysis and design insights,"
[165] S. He, C. Qi, Y. Wu, and Y. Huang, “Energy-efficient transceiver design IEEE J. Sel. Areas Commun., 2017, (To Appear).
for hybrid sub-array architecture MIMO systems,” to appear in IEEE [190] T. Bai, A. Alkhateeb, and R. W. Heath, “Coverage and capacity of
Access, 2017. millimeter-wave cellular networks,” IEEE Commun. Mag., vol. 52, no. 9,
[166] F. Sohrabi and W. Yu, “Hybrid digital and analog beamforming design pp. 70–77, Sep. 2014.
for large-scale antenna arrays,” IEEE J. Sel. Top. Signal Process., vol. 10, [191] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Aspects of favorable
no. 3, pp. 501–513, Apr. 2016. propagation in massive MIMO,” European Signal Processing Conference
[167] X. Yu, J.-C. Shen, J. Zhang, and K. B. Letaief, “Alternating minimiza- (EUSIPCO), Sep. 2014, pp. 76–80.
tion algorithms for hybrid precoding in millimeter wave MIMO systems,” [192] R. H. Roy and B. Ottersten, “Spatial division multiple access wireless
IEEE J. Sel. Top. Signal Process., vol. 10, no. 3, pp. 485–500, Mar. 2016. communication systems,” US Patent 5515378, 1991.
[168] O. El Ayach, R. W. Heath, S. Rajagopal, and Z. Pi, “Multimode [193] IEEE Computer Society, “Wireless LAN medium access control
precoding in millimeter wave MIMO transmitters with multiple antenna (MAC) and physical layer (PHY) specifications: Enhancements for very
sub-arrays,” in Proc. IEEE GLOBECOM, Dec. 2013, pp. 3476–3480. high throughput for operation in bands below 6GHz,” IEEE P802.11ac,
[169] A. Sayeed and J. Brady, “Beamspace MIMO for high-dimensional Draft 0.1, Jan. 2011.
multiuser communication at millimeter-wave frequencies,” in Proc. IEEE [194] E. Björnson, E. Jorswieck, “Optimal resource allocation in coordinated
GLOBECOM, Dec. 2013, pp. 3679–3684. multi-cell systems,” Foundations and Trends in Communications and
[170] P. Amadori and C. Masouros, “Low RF-complexity millimeter-wave Information Theory, vol. 9, no. 2, pp. 113–381, 2013.
beamspace-MIMO systems by beam selection,” IEEE Trans. Commun., [195] T. L. Marzetta, E. G. Larsson, H. Yang, H. Q. Ngo, “Fundamentals of
vol. 63, no. 6, pp. 2212–2222, Jun. 2015. Massive MIMO,” Cambridge University Press, 2016.
 24

[196] G. Kwon and H. Park, “A joint scheduling and millimeter wave networks,” IEEE Transactions on Wireless Communications, vol. 14, no.
hybrid beamforming system with partial side information,” in Proc. IEEE 12, pp. 6664-6678, Dec 2015.
International Conference on Communications (IEEE ICC’16), May 2016, [217] C. N. Barati, S. A. Hosseini, M. Mezzavilla, S. Rangan, T. Korakis, S.
pp. 1–6. S. Panwar, and M. Zorzi, “Directional initial access for millimeter wave
[197] S. Sun, T. S. Rappaport, R. W. Heath, A. Nix, and S. Rangan, cellular systems,” CoRR, vol. abs/1511.06483, 2015. [Online]. Available:
“MIMO for millimeter-wave wireless communications: Beamforming, http://arxiv.org/abs/1511.06483
spatial multiplexing, or both?,” IEEE Commun. Mag., vol. 52, no. 12, [218] V. Desai, L. Krzymien, P. Sartori, W. Xiao, A. Soong, and A.
pp. 110-121, Dec. 2016. Alkhateeb, “Initial beamforming for mmWave communications,” in 48th
[198] C. Yiu and S. Singh, “Empirical capacity of mmWave WLANs,” IEEE Asilomar Conference on Signals, Systems and Computers„ pp. 1926-
J. Sel. Areas Commun., vol. 27, no. 8, pp. 1479–1487, Oct. 2009. 1930, Nov 2014.
[199] A. Adhikary, E. Al Safadi, M. K. Samimi, R. Wang, G. Caire, T. S. [219] A. Capone, I. Filippini, and V. Sciancalepore, “Context information for
Rappaport, and A. F. Molisch, “Joint spatial division and multiplexing fast cell discovery in mm-Wave 5G networks,” in Proc. of 21th European
for mm-wave channels,” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. Wireless Conference, May 2015.
1239–1255, Jun. 2014. [220] A. Capone, I. Filippini, V. Sciancalepore, and D. Tremolada, “Obstacle
[200] C. Sun, X. Gao, S. Jin, M. Matthaiou, Z. Ding, and C. Xiao, “Beam di- avoidance cell discovery using mm-Waves directive antennas in 5G
vision multiple access transmission for massive MIMO communications,” networks,” in Proc. IEEE 26th Annual International Symposium on
IEEE Trans. Commun., vol. 63, no. 6, pp. 2170–2184, Jun. 2015. Personal, Indoor, and Mobile Radio Communications (PIMRC), pp. 2349-
[201] P. Cao and J. S. Thompson, “Practical multi-user transmission design in 2353, Aug. 2015,
millimeter wave cellular networks: Is the joint SDMA-TDMA technique
[221] Q. Li, H. Niu, G. Wu, and R. Hu, “Anchor-booster based heterogeneous
the answer?,” in Proc. IEEE International Workshop on Signal Processing
networks with mmwave capable booster cells,” in Proc. EEE Globecom
Advances in Wireless Communications (IEEE SPAWC’16), Jul. 2016, pp.
Workshops (GC Wkshps), Dec 2013, pp. 93-98.
1–5.
[222] W. B. Abbas and M. Zorzi, “Context information based initial cell
[202] C. Zhang, Y. Huang, Y. Jing, S. Jin, and L. Yang, “Sum-Rate Analysis
search for millimeter wave 5G cellular networks,” in Proc. of 25th
for Massive MIMO Downlink with Joint Statistical Beamforming and
European Conference on Networks and Communications, EuCNC, 2016.
User Scheduling,aś
˛ IEEE Transactions on Wireless Communications, vol.
16, no. 4, pp. 2181-2194, 2017. [223] K. Higuchi and A. Benjebbour, “Non-orthogonal multiple access
[203] H. Miao, M. Faerber, M. Fresia, and V. Frascolla, “Joint beam- (NOMA) with successive interference cancellation for future radio ac-
frequency multiuser scheduling for millimeter-wave downlink multi- cess,” IEICE Trans. Commun., vol. E98-B, no. 3, pp. 403-414, Mar. 2015.
plexing,” in Proc. IEEE Vehicular Technology Conference (IEEE VTC [224] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of
Spring’16), May 2016, pp. 1–5. non-orthogonal multiple access in 5G systems with randomly deployed
[204] C.-S. Sum, Z. Lan, M. A. Rahman, J. Wang, T. Baykas, R. Funada, users,” IEEE Signal Process. Lett., vol. 21, no. 12, pp. 1501-1505, Dec.
H. Harada, and S. Kato, “A multi-Gbps millimeter-wave WPAN system 2014.
based on STDMA with heuristic scheduling,” in Proc. IEEE Global [225] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, “Non-
Communications Conference (IEEE GLOBECOM’09), Dec. 2009, pp. 1- orthogonal multiple access for 5G: Solutions, challenges, opportunities,
6. and future research trends,” IEEE Commun. Mag., vol. 53, no. 9, pp.
[205] C.-S. Sum and H. Harada, “Scalable heuristic STDMA scheduling 74-81, Sep. 2015.
scheme for practical multi-Gbps millimeter-wave WPAN and WLAN [226] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
systems,” IEEE Trans. Wireless Commun., vol. 11, no. 7, pp. 2658–2669, Cambridge: Cambridge University Press, 2005.
Jul. 2012. [227] B. Wang, L. Dai, Z. Wang, N. Ge, and S. Zhou, “Spectrum and energy
[206] J. Qiao, L. X. Cai, X. Shen, and J. W. Mark, “STDMA-based schedul- efficient beamspace MIMO-NOMA for millimeter-wave communications
ing algorithm for concurrent transmissions in directional millimeter wave using lens antenna array,” submitted for publication, 2017.
networks,” in Proc. IEEE International Conference on Communications [228] S. A. R. Naqvi, and S. A. Hassan, “Combining NOMA and mmWave
(IEEE ICC’12), Jun. 2012, pp. 5221–5225. technology for cellular communication,” in Proc. IEEE Vehicular Tech-
[207] Z. Yan, B. Li, X. Zuo, and M. Yang, “A heuristic clique based STDMA nology Conference (IEEE VTC Fall’16), Sep. 2016, pp. 1-5.
scheduling algorithm for spatial concurrent transmission in mmWave [229] A. S. Marcano, and H. L. Christiansen, “Performance of non-
networks,” in Proc. IEEE Wireless Communications and Networking orthogonal multiple access (NOMA) in mmWave wireless communi-
Conference (IEEE WCNC’15), Mar. 2015, pp. 1036–1041. cations for 5G networks,” in Proc. IEEE International Conference on
[208] Q. Xue, X. Fang, M. Xiao, and Y. Li, “Multi-user millimeter wave Computing, Networking and Communications (IEEE ICNC’17), Jan.
communications with nonorthogonal beams,” to appear in IEEE Trans 2017, pp. 969-974.
Veh. Technol., 2017. [230] D. Zhang, Z. Zhou, C. Xu, Y. Zhang, J. Rodriguez, and T. Sato,
[209] G. Yan and D. Liu, “A simple adaptive STDMA scheduling scheme in “Capacity analysis of non-orthogonal multiple access with mmwave
mmWave wireless networks,” in Proc. IEEE International Conference on massive MIMO systems,” to appear in IEEE J. Sel. Areas Commun.,
Communications, Circuits and Systems (IEEE ICCCAS’13), Nov. 2016, 2017.
pp. 1-5. [231] Z. Ding, P. Fan, and H. V. Poor, “Random beamforming in millimeter-
[210] C. Li, R. Cai, and D. Liu, “A suboptimal STDMA scheduling for wave NOMA networks,” to appear in IEEE Access, 2017.
concurrent transmissions in mmWave wireless networks,” in Proc. IEEE
[232] B. Wang, L. Dai, Z. Wang, N. Ge, and S. Zhou, “Spectrum and energy
International Conference on Signal Processing, Communications and
efficient beamspace MIMO-NOMA for millimeter-wave communications
Computing (IEEE ICSPCC’14), Aug. 2014, pp. 137-141.
using lens antenna array,” submitted to IEEE J. Sel. Areas Commun.
[211] C. Jeong, J. Park, and H. Yu, “Random access in millimeter-wave
[233] L. You, X. Gao, G. Y. Li, X.-G. Xia, and N. Ma, “BDMA for
beamforming cellular networks: issues and approaches,” IEEE Commun.
millimeter-wave/terahertz massive MIMO transmission with per-beam
Mag., vol. 53, no. 1, pp. 180-185, Jan. 2015.
synchronization,” to appear in IEEE J. Sel. Areas Commun., 2017.
[212] N. Giatsoglou, K. Ntontin, E. Kartsakli, A. Antonopoulos, and C.
Verikoukis, “D2D-Aware device caching in mmWave-cellular networks,” [234] Z. Xiao, L. Dai, P. Xia, J. Choi, and X. Xia, “Millimeter-Wave
to appear in IEEE J. Sel. Areas Commun., 2017. communication with non-orthogonal multiple access for 5G,” submitted
[213] M. Polese, M. Giordani, M. Mezzavilla, S. Rangan, and M. Zorzi, to IEEE Wireless Commun. Mag.
“Improved handover through dual connectivity in 5G mmWave mobile [235] C.-L. I, C. Rowell, S. Han, Z. Xu, G. Li, and Z. Pan, “Toward green
networks,” to appear in IEEE J. Sel. Areas Commun., 2017. and soft: A 5G perspective,” IEEE Commun. Mag., vol. 52, no. 2, pp.
[214] H. Shokri-Ghadikolaei, C. Fischione, G. Fodor, P. Popovski, and M. 66–73, Feb. 2014.
Zorzi, “Millimeter wave cellular networks: A MAC layer perspective,” [236] W. Feng, Y. Wang, D. Lin, N. Ge, J. Lu and S. Li, “When mmWave
IEEE Transactions on Communications, vol. 63, no. 10, pp. 3437- 3458, Communications Meet Network Densification: A Scalable Interference
Oct 2015. Coordination Perspective," IEEE J. Sel. Areas Commun., 2017 (to appear).
[215] M. Giordani, M. Mezzavilla, S. Rangan, and M. Zorzi, “MultiConnec- [237] R. Taori and A. Sridharan, “Point-to-multipoint in-band mmWave
tivity in 5G mmwave cellular networks,” in Proc. 15th Annual Mediter- backhaul for 5G networks,” IEEE Wireless Commun., vol. 53, no. 1, pp.
ranean Ad Hoc Networking Workshop (MED-HOC-NET) (MedHoc-Net 195–201, Jan. 2015.
16), Vilanova i la Geltru, Barcelona, Spain, Jun. 2016. [238] L. Wei, R.Q. Hu, Y. Qian, and G. Wu, “Key elements to enable
[216] C. Barati, S. Hosseini, S. Rangan, P. Liu, T. Korakis, S. Panwar, and millimeter wave communications for 5G wireless systems,” IEEE Wireless
T. Rappaport, “Directional cell discovery in millimeter wave cellular Commun., vol. 21, no. 6, pp. 136–43, Dec. 2014.
 25

[239] L. Song, Y. Li and Z. Han, “Game-theoretic resource allocation for

full-duplex communications,” IEEE Commun. Mag., vol. 23, no. 3, pp.
50–56, Jun. 2016.
[240] Q. Li, G. Li, W. Lee, M. I. Lee, D. Mazzarese, B. Clerckx, and Z. Li,
“MIMO techniques in WiMAX and LTE: A feature overview,” IEEE
Commun. Mag., vol. 48, no. 5, pp. 86–92, May 2010.
[241] P. Wang, Y. Li, L. Song, and B. Vucetic, “Multi-gigabit millimeter wave
wireless communications for 5G: from fixed access to cellular networks,”
IEEE Commun. Mag., vol. 53, no. 1, pp. 168–178, Jan. 2015.
[242] T. Bai, R. Vaze, and R. W. Heath, “Analysis of blockage effects on
urban cellular networks,” IEEE Trans. Wireless Commun., vol. 13, no. 9,
pp. 5070–5083, Sep. 2014.
[243] X. Lin and J. G. Andrews, “Connectivity of millimeter wave networks
with multi-hop relaying,” IEEE Wireless Commun. Lett., vol. 4, no. 2, pp.
209–212, Apr. 2015.
[244] D. Maamari, N. Devroye, and D. Tuninetti, “Coverage in mmwave
cellular networks with base station co-operation,” IEEE Trans. Wireless
Commun., vol. 15, no. 4, pp. 2981–2994, Apr. 2016.
[245] X. Yu, J. Zhang, M. Haenggi, and K. B. Letaief, “Coverage Analysis for
Millimeter Wave Networks: The Impact of Directional Antenna Arrays,?
to appear in IEEE J. Sel. Areas Commun., 2017.
[246] V. Petrov, D. Solomitckii, A. Samuylov, M. A. Lema, M. Gapeyenko,
D. Moltchanov, S. Andreev, V. Naumov, K. Samouylov, M. Dohler, and Y.
Koucheryavy, ?Dynamic Multi-Connectivity Performance in Ultra-Dense
Urban mmWave Deployments,? to appear in IEEE J. Sel. Areas Commun.,
2017.
[247] D. Ramirez, L. Huang and B. Aazhang, ?On Opportunistic mmWave
Networks with Blockage,? to appear in IEEE J. Sel. Areas Commun.,
2017.
[248] H. Zhang, S. Huang, C. Jiang, K. Long, V. C. M. Leung, and H.
Vincent Poor, “Energy efficient user association and power allocation in
millimeter wave based ultra dense networks with energy harvesting base
stations,” to appear in IEEE J. Sel. Areas Commun., 2017.
[249] L. Wang, K.-K. Wong, R. W. Heath, and J. Yuan, “Wireless powered
dense cellular networks: How many small cells do we need?” to appear
in IEEE J. Sel. Areas Commun., 2017.
[250] Z. C. Phyo and A. Taparugssanagorn, “Hybrid analog-digital downlink
beamforming for massive MIMO system with uniform and non-uniform
linear arrays,” in Proc. 2016 13th International Conference on Electrical
Engineering/Electronics, Computer, Telecommunications and Information
Technology (ECTI-CON), Jun. 2016, pp. 1–6.
[251] J. Brady, N. Behdad, and A. Sayeed, “Beamspace MIMO for
millimeter-wave communications: System architecture, modeling, anal-
ysis, and measurements," IEEE Trans. Ant. and Propag., vol. 61, no. 7,
pp. 3814-3827, Jul. 2013.
[252] https://www.huawei.eu/blog/millimetre-wave-key-technology-5g.
[253] Huawei Technologies Co., Ltd. Huawei to Bring 73 GHz
mmWave Mu-MIMO live Demo to Deutsche Telekom. Avail-
able online: http://www.huawei.com/en/news/2016/2/73GHzmm-Wave-
Mu-MIM-livedemo (accessed on 30 May 2016).
[254] Branda, M. Qualcomm Research Demonstrates Ro-
bust mmWave Design for 5G. Available online:
https://www.qualcomm.com/news/onq/2015/11/19/qualcomm-research-
demonstrates-robust-mmwavedesign- 5g (accessed on 30 May 2016).
[255] Nokia Networks. Nokia Networks Showcases 5G speed of
10Gbps with NI at the Brooklyn 5G Summit. Available online:
http://networks.nokia.com/news-events/press-room/press-releases/nokia-
networksshowcases- 5g-speed-of-10gbps-with-ni-at-the-brooklyn-5g-
summit (accessed on 30 May 2016).
[256] Samsung Electronics Co., Ltd. Samsung Electronics and Deutsche
Telekom Demonstrate World?s First End-to-End 5G Solution at
Mobile World Congress 2016. Available online: https://news.samsung.
com/global/samsung-electronics-and-deutsche-telekom-demonstrate-
worlds-first-end-to-end-5g-solutionat- mobile-world-congress-2016
(accessed on 30 May 2016).
[257] https://www.ericsson.com/news/2076554.