IEEE JSAC SPECIAL ISSUE ON 5G WIRELESS COMMUNICATION SYSTEMS 1
What Will 5G Be?
Jeffrey G. Andrews, Fellow, IEEE, Stefano Buzzi, Senior Member, IEEE, Wan Choi, Senior Member, IEEE,
Stephen Hanly, Member, IEEE, Angel Lozano, Fellow, IEEE, Anthony C. K. Soong, Fellow, IEEE,
Jianzhong C. Zhang, Senior Member, IEEE
Abstract—What will 5G be? What it will not be is an in-
cremental advance on 4G. The previous four generations of
cellular technology have each been a major paradigm shift
that has broken backwards compatibility. And indeed, 5G will
need to be a paradigm shift that includes very high carrier
frequencies with massive bandwidths, extreme base station and
device densities and unprecedented numbers of antennas. But
unlike the previous four generations, it will also be highly
integrative: tying any new 5G air interface and spectrum together
with LTE and WiFi to provide universal high-rate coverage and
a seamless user experience. To support this, the core network
will also have to reach unprecedented levels of flexibility and
intelligence, spectrum regulation will need to be rethought and
improved, and energy and cost efficiencies will become even more
critical considerations. This paper discusses all of these topics,
identifying key challenges for future research and preliminary
5G standardization activities, while providing a comprehensive
overview of the current literature, and in particular of the papers
appearing in this special issue.
I. INTRODUCTION
A. The Road to 5G
In just the past year, preliminary interest and discussions
about a possible 5G standard have evolved into a full-fledged
conversation that has captured the attention and imagination
of researchers and engineers around the world. As the long-
term evolution (LTE) system embodying 4G has now been
deployed and is reaching maturity, where only incremental
improvements and small amounts of new spectrum can be
expected, it is natural for researchers to ponder “what’s next?”
[1]. However, this is not a mere intellectual exercise. Thanks
largely to the annual visual network index (VNI) reports
released by Cisco, we have quantitative evidence that the
wireless data explosion is real and will continue. Driven largely
by smartphones, tablets, and video streaming, the most recent
(Feb. 2014) VNI report [2] and forecast makes plain that
an incremental approach will not come close to meeting the
demands that networks will face by 2020.
In just a decade, the amount of IP data handled by wireless
networks will have increased by well over a factor of 100:
from under 3 exabytes in 2010 to over 190 exabytes by 2018,
on pace to exceed 500 exabytes by 2020. This deluge of data
J. G. Andrews (jandrews@ece.utexas.edu) is with the University of Texas
at Austin, USA.
S. Buzzi (buzzi@unicas.it) is with University of Cassino and Southern
Lazio, Italy, and with CNIT, Italy.
W. Choi (wchoi@kaist.edu) is with Korea Advanced Institute of Science
and Technology (KAIST), Daejeon, South Korea.
S. Hanly (stephen.hanly@mq.edu.au) is with Macquarie University, Sydney,
Australia.
A. Lozano (angel.lozano@upf.edu) is with Universitat Pompeu Fabra
(UPF), Barcelona, Spain.
A. C. K. Soong (anthony.soong@huawei.com) is with Huawei Technologies,
Plano, Texas, USA.
J. C. Zhang (jianzhong.z@samsung.com) is with Samsung Electronics,
Richardson, Texas, USA.
Article last revised: June 9, 2014
has been driven chiefly by video thus far, but new unforeseen
applications can reasonably be expected to materialize by
2020. In addition to the sheer volume of data, the number of
devices and the data rates will continue to grow exponentially.
The number of devices could reach the tens or even hundreds
of billions by the time 5G comes to fruition, due to many new
applications beyond personal communications [3]–[5]. It is our
duty as engineers to meet these intense demands via innovative
new technologies that are smart and efficient yet grounded in
reality. Academia is engaging in large collaborative projects
such as METIS [6] and 5GNOW [7], while industry is driving
preliminary 5G standardization activities (cf. Sec. IV-B). To
further strengthen these activities, the public-private partner-
ship for 5G infrastructure recently constituted in Europe will
funnel massive amounts of funds into related research [8].
This article is an attempt to summarize and overview many
of these exciting developments, including the papers in this
special issue. In addition to the highly visible demand for
ever more network capacity, there are a number of other
factors that make 5G interesting, including the potentially
disruptive move to millimeter wave (mmWave) spectrum, new
market-driven ways of allocating and re-allocating bandwidth,
a major ongoing virtualization in the core network that might
progressively spread to the edges, the possibility of an “Internet
of Things” comprised of billions of miscellaneous devices,
and the increasing integration of past and current cellular and
WiFi standards to provide a ubiquitous high-rate, low-latency
experience for network users.
This editorial commences with our view of the “big three”
5G technologies: ultra-densification, mmWave, and massive
multiple-input multiple-output (MIMO). Then, we consider
important issues concerning the basic transmission waveform,
the increasing virtualization of the network infrastructure, and
the need for greatly increased energy efficiency. Finally, we
provide a comprehensive discussion of the equally important
regulatory and standardization issues that will need to be
addressed for 5G, with a particular focus on needed innovation
in spectrum regulation.
B. Engineering Requirements for 5G
In order to more concretely understand the engineering chal-
lenges facing 5G, and to plan to meet them, it is necessary to
first identify the requirements for a 5G system. The following
items are requirements in each key dimension, but it should be
stressed that not all of these need to be satisfied simultaneously.
Different applications will place different requirements on
the performance, and peak requirements that will need to
be satisfied in certain configurations are mentioned below.
For example, very-high-rate applications such as streaming
high-definition video may have relaxed latency and reliability
requirements compared to driverless cars or public safety
applications, where latency and reliability are paramount but
lower data rates can be tolerated.
IEEE JSAC SPECIAL ISSUE ON 5G WIRELESS COMMUNICATION SYSTEMS 2
1) Data Rate: The need to support the mobile data traffic
explosion is unquestionably the main driver behind 5G. Data
rate can be measured in several different ways, and there will
be a 5G goal target for each such metric:
a) Aggregate data rate or area capacity refers to the total
amount of data the network can serve, characterized in bits/s
per unit area. The general consensus is that this quantity will
need to increase by roughly 1000x from 4G to 5G.
b) Edge rate or 5% rate is the worst data rate that a
user can reasonably expect to receive when in range of the
network, and so is an important metric and has a concrete
engineering meaning. Goals for the 5G edge rate range from
100 Mbps (easily enough to support high-definition streaming)
to as much as 1 Gbps. Meeting 100 Mbps for 95% of users will
be extraordinarily challenging, even with major technological
advances. This requires about a 100x advance since current
4G systems have a typical 5% rate of about 1 Mbps, although
the precise number varies quite widely depending on the load,
the cell size, and other factors.
c) Peak rate is the best-case data rate that a user can hope
to achieve under any conceivable network configuration. The
peak rate is a marketing number, devoid of much meaning to
engineers and likely to be in the range of tens of Gbps.
Meeting the requirements in (a)-(b), which are about 1000x
and 100x current 4G technology, respectively, are the main
focus of this paper.
2) Latency: Current 4G roundtrip latencies are on the order
of about 15 ms, and are based on the 1 ms subframe time
with necessary overheads for resource allocation and access.
Although this latency is sufficient for most current services,
anticipated 5G applications include two-way gaming, novel
cloud-based technologies such as those that may be touch-
screen activated (the “tactile Internet” [9]), and virtual and
enhanced reality (e.g., Google glass or other wearable comput-
ing devices). As a result, 5G will need to be able to support
a roundtrip latency of about 1 ms, an order of magnitude
faster than 4G. In addition to shrinking down the subframe
structure, such severe latency constraints may have important
implications on design choices at several layers of the protocol
stack and the core network (cf. Sect. III).
3) Energy and Cost: As we move to 5G, costs and energy
consumption will, ideally, decrease, but at least they should
not increase on a per-link basis. Since the per-link data rates
being offered will be increasing by about 100x, this means
that the Joules per bit and cost per bit will need to fall by at
least 100x. In this article, we do not address energy and cost
in a quantitative fashion, but we are intentionally advocating
technological solutions that promise reasonable cost and power
scaling. For example, mmWave spectrum should be 10-100x
cheaper per Hz than the 3G and 4G spectrum below 3 GHz.
Similarly, small cells should be 10-100x cheaper and more
power efficient than macrocells. A major cost consideration
for 5G, even more so than in 4G due to the new BS densities
and increased bandwidth, is the backhaul from the network
edges into the core. We address backhaul and other economic
considerations in Sect. IV-C. As for energy efficiency, we
address this more substantially in Sect. III-C.
C. Device Types and Quantities.
5G will need to be able to efficiently support a much larger
and more diverse set of devices. With the expected rise of
machine-to-machine communication, a single macrocell may
need to support 10,000 or more low-rate devices along with its
traditional high-rate mobile users. This will require wholesale
changes to the control plane and network management relative
to 4G, whose overhead channels and state machines are not
designed for such a diverse and large subscriber base.
II. KEY TECHNOLOGIES TO GET TO 1000X DATA RATE
Of the requirements outlined in Sect. I-B, certainly the one
that gets the most attention is the need for radically higher data
rates across the board. Our view is that the required 1000x will,
for the most part, be achieved through combined gains in three
categories:
a) Extreme densification and offloading to improve the area
spectral efficiency. Put differently, more active nodes per
unit area and Hz.
b) Increased bandwidth, primarily by moving towards and
into mmWave spectrum but also by making better use
of WiFi’s unlicensed spectrum in the 5-GHz band.
Altogether, more Hz.
c) Increased spectral efficiency, primarily through advances
in MIMO, to support more bits/s/Hz per node.
The combination of more nodes per unit area and Hz, more
Hz, and more bits/s/Hz per node, will compound into many
more bits/s per unit area. Other ideas not in the above cate-
gories, e.g., interference management through BS cooperation
[10]–[23] may also contribute improvements, but the lion’s
share of the surge in capacity should come from ideas in the
above categories. In the remainder of this section, these are
distilled in some detail.
A. Extreme Densification and Offloading
A straightforward but extremely effective way to increase
the network capacity is to make the cells smaller. This ap-
proach has been demonstrated over several cellular generations
[24], [25]. The first such generation, in the early 1980s, had
cell sizes on the order of hundreds of square kms. Since then,
those sizes have been progressively shrinking and by now they
are often fractions of a square km in urban areas. In Japan,
for instance, the spacing between BSs can be as small as two
hundred meters, giving a coverage area well under a tenth of a
square km. Networks are now rapidly evolving [26] to include
nested small cells such as picocells (range under 100 meters)
and femtocells (WiFi-like range) [27], as well as distributed
antenna systems [28] that are functionally similar to picocells
from a capacity and coverage standpoint but have all their
baseband processing at a central site and share cell IDs.
Cell shrinking has numerous benefits, the most important
being the reuse of spectrum across a geographic area and
the ensuing reduction in the number of users competing for
resources at each BS. Contrary to widespread belief, as long
as power-law pathloss models hold the signal-to-interference
ratio (SIR) is preserved as the network densifies [29].
1
Thus,
in principle, cells can shrunk almost indefinitely without a
sacrifice in SIR, until nearly every BS serves a single user
(or is idle). This allows each BS to devote its resources, as
well as its backhaul connection, to an ever-smaller number of
users.
As the densification becomes extreme, some challenges
arise:
1
The power-law pathloss model ceases to apply in the near field, very close
to the transmitter [30].
IEEE JSAC SPECIAL ISSUE ON 5G WIRELESS COMMUNICATION SYSTEMS 3
• Preserving the expected cell-splitting gains as each BS
becomes more lightly loaded, particularly low-power
nodes.
• Determining appropriate associations between users and
BSs across multiple radio access technologies (RATs),
which is crucial for optimizing the edge rate.
• Supporting mobility through such a highly heteroge-
neous network.
• Affording the rising costs of installation, maintenance
and backhaul.
We next briefly discuss these challenges, particularly in view
of the other technologies raised in this article.
1) Base Station Densification Gains: We define the BS
densification gain ρ(λ
1
, λ
2
) as the effective increase in data
rate relative to the increase in network density, which is a
proxy here for cost. Specifically, if we achieve a data rate R
1
(could be any measure thereof, e.g., edge rate or aggregate)
when the BS density is λ
1
> 0 BSs/km
2
and then we consider
a higher BS density λ
2
> λ
1
, with corresponding rate R
2
, then
the densification gain is the slope of the rate increase over that
density range:
ρ(λ
1
, λ
2
) =
(R
2
− R
1
)/R
1
(λ
2
− λ
1
)/λ
1
. (1)
For example, if the network density is doubled, and the edge
data rate increased by 50% (for example since some of the
added base stations were lightly loaded), then the densification
gain is ρ = 0.5. In some applications with channel access
protocols like CSMA that are inefficient in high density, it is
possible for ρ < 0, which is colloquially referred to as “the
tragedy of the commons”, but for cellular networks with a
centralized MAC we can safely assume ρ > 0.
In an interference-limited network with full buffers, the
signal-to-interference-plus-noise ratio (SINR) is essentially
equal to the SIR and, because the SIR distribution remains
approximately constant as the network densifies, the best case
scenario is ρ ≈ 1. In reality, buffers are not always full, and
small cells tend to become more lightly loaded than macrocells
as the network densifies. Altogether, the SINR usually in-
creases with density: in noise-limited networks because of the
increase in received signal power, and in interference-limited
networks because the lightly loaded small cells generate less
interference (while still providing an option for connectivity)
[31]. Nevertheless, at microwave frequencies the gain in SINR
is not enough to keep up with the decrease in small-cell
utilization and thus ρ < 1. In an extreme case, consider λ
1
and R
1
held fixed with λ
2
→ ∞. In this asymptotic setting,
the small cells compete for a finite pool of users, becoming
ever more lightly loaded, and thus ρ → 0.
Empirically and theoretically, we observe that ρ improves
and can approach 1 with macro-BS muting (termed eICIC in
3GPP) vs. the macrocells transmitting all the time and thus
interfering with the small cells all the time.
An intriguing aspect of mmWave frequencies is that den-
sification gains ρ 1 may be possible. This is because, as
discussed in Sect. II-B, at these frequencies communication is
largely noise-limited and increasing the density not only splits
the cell resources and lightens the load, but it may increase
the SINR dramatically. As a striking example of this, it was
recently showed in [32] that, under a plausible urban grid-
based deployment, increasing the BS count in a given area
from 36 to 96—which decreased the inter-BS distance from
170 meters down to 85 meters—increased the 5% cell-edge
Fig. 1: User association in a multi-RAT network over many frequency
bands is complex. In this simplified scenario, a mobile user in turn
associates with different BSs based on a tradeoff between the gain to
that BS and the traffic load (congestion) that it is experiencing.
rate from 24.5 Mbps up to 1396 Mbps, giving ρ = 9.9. While
conceding that this massive densification gain corresponds to
a particular setup and model, it is nevertheless remarkable.
In general, quantifying and optimizing the densification
gains in a wide variety of deployment scenarios and network
models is a key area for continued small-cell research.
2) Multi-RAT Association: Networks will continue to be-
come increasingly heterogeneous as we move towards 5G.
A key feature therein will be increased integration between
different RATs, with a typical 5G-enabled device having radios
capable of supporting not only a potentially new 5G standard
(e.g., at mmWave frequencies), but also 3G, numerous releases
of 4G LTE including possibly LTE-Unlicensed [33], several
types of WiFi, and perhaps direct device-to-device (D2D)
communication, all across a great many spectral bands. Hence,
determining which standard(s) and spectrum to utilize and
which BS(s) or users to associate with will be a truly complex
task for the network [34].
Determining the optimal user association is, for general
utility functions, a massive combinatorial optimization prob-
lem that depends on the SINR from every user to every
BS, the instantaneous load at each BS, the choices of other
users in the network, and possibly other constraints such as
the requirement to utilize the same BS and standard in both
uplink and downlink (to facilitate functioning control channels
for resource allocation and feedback) [35], [36]. Therefore,
simplified procedures must be adopted [37], an example of
which appears in this special issue [38]. The key such sim-
plified procedures are “biasing” and macrocell “blanking”
or ”muting”. Biasing refers to associating with a small cell
even if it provides a lower SINR than the macrocell, and is
useful for pushing users off of the heavily loaded macrocell
and onto the lightly loaded small cell. Everyone wins: the
remaining macrocell users get more resources while the biased
users have a lower SINR/spectral efficiency but can utilize a
large number of resource blocks on the small cell, ultimately
attaining a higher data rate. Blanking refers to shutting off
the macrocell transmissions for some fraction of the time,
preferably while the biased small cell users are being served.
This raises all the small-cell SINRs considerably—enough to
justify actually shutting down even congested macrocell BSs—
while also providing a mechanism for the biased users to hear
common control channels that would otherwise be swamped
by the macrocells.
Even a simple, seemingly highly suboptimal association
approach based on aggressive but static biasing (about 10-
20 dB, depending on various factors) towards small cells
IEEE JSAC SPECIAL ISSUE ON 5G WIRELESS COMMUNICATION SYSTEMS 4
and blanking about half of the macrocell transmissions has
been shown to increase edge rates by as much as 500%
[39], [40]. The joint problem of user association and resource
allocation in two-tier heterogeneous networks (HetNets), with
adaptive tuning of the biasing and blanking in each cell, is
considered in [36], [41]–[46]. An interesting model of hotspot
traffic is considered in [42]–[44] where it is shown that, under
various network utility metrics, the optimal cell association is
determined by rate ratio bias, rather than power (or SINR)
bias.
It will be interesting to extend these models to more
general scenarios. A dynamic model of cell range expansion is
considered in [47], where traffic arrives as a Poisson process
in time and the feasible arrival rates, for which a stabilizing
scheduling policy exists, are characterized. User association
and load balancing in a HetNet, with massive MIMO at the
BSs, is considered in [48]. The problem of determining the
optimal associations when there are multiple RATS, operating
at different frequencies and using different protocols, has not
yet received much attention. However, an interesting game the-
oretic approach is taken in [49] to the RAT-selection problem,
where convergence to Nash equilibria and the Pareto-efficiency
of these equilibria are studied. A related paper in this special
issue [50] explores the interaction between cellular operators
and WiFi network owners.
Adding mmWave into the picture adds significant additional
complexity, since even the notion of a cell boundary is blurry
at mmWave frequencies given the strong impact of blockages,
which often result in nearby BSs being bypassed in favor of
farther ones that are unblocked (cf. Fig. 2). On the positive
side, interference is much less important in mmWave (cf.
Sect. II-B) and thus the need for blanking is reduced.
In summary, there is a great deal of scope for modeling,
analyzing and optimizing BS-user associations in 5G.
3) Mobility Support: Clearly, the continued network densi-
fication and increased heterogeneity poses challenges for the
support of mobility. Although a hefty share of data is served
to stationary indoor users, the support of mobility and always-
on connectivity is arguably the single most important feature
of cellular networks relative to WiFi. Because modeling and
analyzing the effect of mobility on network performance is
difficult, we expect to see somewhat ad hoc solutions such as in
LTE Rel-11 [51] where user-specific virtual cells are defined to
distinguish the physical cell from a broader area where the user
can roam without the need for handoff, communicating with
any BS or subset of BSs in that area. Or in mmWave, restricting
highly mobile users to macrocells and microwave frequencies,
thereby forcing them to tolerate lower rates. Handoffs will be
particularly challenging at mmWave frequencies since transmit
and receive beams must be aligned to communicate. Indeed,
the entire paradigm of a handoff initiated and managed at
layer 3 by the core network will likely not exist in 5G;
instead, handoffs may be opportunistic, based on mmWave
beam alignments, or indistinguishable from PHY/MAC inter-
ference management techniques whereby users communicate
with multiple coordinated BSs, as exemplified by [52] in this
special issue.
4) Cost: Evolving to ever-smaller cells requires ever-
smaller, lower-power and cheaper BSs, and there is no fun-
damental reason a BS needs to be more expensive than a
user device or a WiFi node [26]. Nevertheless, obtaining
permits, ensuring fast and reliable backhaul connections, and
paying large monthly site rental fees for operator-controlled
small-cell placements have proven a major hindrance to the
growth of picocell, distributed antennas, and other enterprise-
quality small cell deployments. Of these, only the backhaul
is primarily a technical challenge. Regulatory reforms and
infrastructure sharing (cf. Sect. IV-C) may help address the
other challenges.
Turning to end-user-deployed femtocells and WiFi access
points, these are certainly much more cost-effective both from
a capital and operating expense perspective [24]. However,
major concerns exist here too. These include the coordination
and management of the network to provide enterprise-grade
service, which given the scale of the deployments requires
automated self-organization [53]. A further challenge is that
these end-user deployments utilize the end-user’s backhaul
connection and access point, both of which the end-user has
a vested interest in not sharing, and in some countries a legal
requirement not to. Anecdotally, all readers of this article
are familiar with the scenario where a dozen WiFi access
points are within range, but all are secured and inaccessible.
From an engineering perspective, this closed-access status
quo is highly inefficient and the cost for 5G would be
greatly reduced in an open-access paradigm for small cells.
One preliminary but successful example is Fon, which as
of press time boasts over 13 million shared WiFi access points.
5G and all networks beyond it will be extremely dense
and heterogeneous, which introduces many new challenges for
network modeling, analysis, design and optimization. We fur-
ther discuss some of the nonobvious intersections of extreme
densification with mmWave and massive MIMO, respectively,
in the next two sections. Before proceeding, however, we
briefly mention that besides cell shrinking a second approach
to densification exists in the form of D2D communication. This
allows users in close proximity to establish direct communica-
tion, replacing two relatively long radio hops via the BS with a
single and shorter direct hop. Provided there is sufficient spatial
locality in the wireless traffic, this can bring about reduced
power consumption and/or higher data rates, and a diminished
latency [54]–[56]. Reference [57] in this special issue proposes
a novel way of scheduling concurrent D2D transmissions so
as to densify while offering interference protection guarantees.
B. Millimeter Wave
Terrestrial wireless systems have largely restricted their
operation to the relatively slim range of microwave frequencies
that extends from several hundred MHz to a few GHz and
corresponds to wavelengths in the range of a few centimeters
up to about a meter. By now though, this spectral band—
often called “beachfront spectrum”—has become nearly fully
occupied, in particular at peak times and in peak markets.
Regardless of the efficacy of densification and offloading,
much more bandwidth is needed [59], [60].
Although beachfront bandwidth allocations can be made
significantly more efficient by modernizing regulatory and
allocation procedures, as discussed in Sect. IV-A, to put large
amounts of new bandwidth into play there is only one way
to go: up in frequency. Fortunately, vast amounts of relatively
idle spectrum do exist in the mmWave range of 30–300 GHz,
where wavelengths are 1–10 mm. There are also several GHz
of plausible spectrum in the 20–30 GHz range.
The main reason that mmWave spectrum lies idle is that,
until recently, it had been deemed unsuitable for mobile com-
munications because of rather hostile propagation qualities,
including strong pathloss, atmospheric and rain absorption, low
IEEE JSAC SPECIAL ISSUE ON 5G WIRELESS COMMUNICATION SYSTEMS 5
Fig. 2: Calculated mmWave BS associations with real building
locations [58]. The shaded regions correspond to association with the
BS centered at that shade. Blocking, LOS vs. non-LOS propagation,
and beam directionality render our usual notion of cell boundaries
obsolete.
diffraction around obstacles and penetration through objects,
and, further, because of strong phase noise and exorbitant
equipment costs. The dominant perception had therefore been
that such frequencies, and in particular the large unlicensed
band around 60 GHz [61], were suitable mainly for very-
short-range transmission [62]–[64]. Thus, the focus had been
on WiFi (with the WiGiG standard in the 60-GHz band)
and on fixed wireless in the 28, 38, 71–76 and 81–86 GHz.
However, semiconductors are maturing, their costs and power
consumption rapidly falling—largely thanks to the progress of
the aforementioned short-range standard—and the other ob-
stacles related to propagation are now considered increasingly
surmountable given time and focused effort [65]–[70].
1) Propagation Issues: Concerning mmWave propagation
for 5G, the main issues under investigation are:
Pathloss. If the electrical size of the antennas (i.e., their
size measured by the wavelength λ = c/f
c
where f
c
is the
carrier frequency) is kept constant, as the frequency increases
the antennas shrink and their effective aperture scales with
λ
2
4π
; then, the free-space pathloss between a transmit and a
receive antenna grows with f
2
c
. Thus, increasing f
c
by an
order of magnitude, say from 3 to 30 GHz, adds 20 dB
of power loss regardless of the transmit-receive distance.
However, if the antenna aperture at one end of the link is
kept constant as the frequency increases, then the free-space
pathloss remains unchanged. Further, if both the transmit and
receive antenna apertures are held constant, then the free-space
pathloss actually diminishes with f
2
c
: a power gain that would
help counter the higher noise floor associated with broader
signal bandwidths.
Although preserving the electrical size of the antennas is
desirable for a number of reasons, maintaining at the same
time the aperture is possible utilizing arrays, which aggregate
the individual antenna apertures: as the antennas shrink with
frequency, progressively more of them must be added within
the original area. The main challenge becomes cophasing these
antennas so that they steer and/or collect energy productively.
This challenge becomes more pronounced when the channel
changes rapidly, for instance due to mobility (whose effect
in terms of Doppler shift increases linearly with frequency)
or due to rapid alterations in the physical orientation of the
devices.
Blocking. MmWave signals exhibit reduced diffraction and
a more specular propagation than their microwave counter-
parts, and hence they are much more susceptible to blockages.
This results in a nearly bimodal channel depending on the
presence or absence of Line-of-Sight (LoS). According to re-
cent measurements [68], [70], as the transmit-receive distance
grows the pathloss accrues close to the free-space value of 20
dB/decade under LoS propagation, but drops to 40 dB/decade
plus an additional blocking loss of 15–40 dB otherwise.
Because of the sensitivity to blockages, a given link can rapidly
transition from usable to unusable and, unlike small-scale
fading, large-scale obstructions cannot be circumvented with
standard small-scale diversity countermeasures. New channel
models capturing these effects are much needed, and in fact
currently being developed [68], [71], [72] and applied to
system-level analysis [58], [73]–[75] and simulation studies
such as [76] and [77] in this special issue.
Atmospheric and rain absorption. The absorption due to
air and rain is noticeable, especially the 15 dB/km oxygen
absorption within the 60-GHz band (which is in fact why this
band is unlicensed), but it is inconsequential for the urban
cellular deployments currently envisioned [65], [67] where
BS spacings might be on the order of 200 m. In fact, such
absorption is beneficial since it further attenuates interference
from more distant BSs, effectively increasing the isolation of
each cell.
The main conclusion is that the propagation losses for
mmWave frequencies are surmountable, but require large an-
tenna arrays to steer the beam energy and collect it coher-
ently. While physically feasible, the notion of narrow-beam
communication is new to cellular communications and poses
difficulties, which we next discuss.
2) Large Arrays, Narrow Beams: Building a wireless system
out of narrow and focused beams is highly nontrivial and
changes many traditional aspects of wireless system design.
MmWave beams are highly directional, almost like flashlights,
which completely changes the interference behavior as well as
the sensitivity to misaligned beams. The interference adopts an
on/off behavior where most beams do not interfere, but strong
interference does occur intermittently. Overall, interference is
de-emphasized and mmWave links may often be noise-limited,
which is a major reversal from 4G. Indeed, even the notion of a
“cell” is likely to be very different in a mmWave system since,
![Fig. 2: Calculated mmWave BS associations with real building locations [58]. The shaded regions correspond to association with the BS centered at that shade. Blocking, LOS vs. non-LOS propagation, and beam directionality render our usual notion of cell boundaries obsolete.](/figures/fig-2-calculated-mmwave-bs-associations-with-real-building-1vh91f33.webp)