1

Towards the 5th Generation

of Wireless Communication Systems
Nicola Marchetti
CTVR / The Telecommunications Research Centre
Trinity College Dublin, Ireland

[email protected]

Abstract

In this article, one first introduces the general landscape of the next generation of wireless
communication systems (5G), including its driver and requirements, and the candidate technologies
that might help to achieve its intended goals. The following areas, which the author considers to be of
particular relevance for 5G, are then introduced: detection of and access to free spectrum over bands of
an heterogeneous nature, extreme densification of networks (massive base station deployments),
extreme increase in the number of antennas in transmitter arrays and their interaction with a novel
waveform, integration of both wireless and optical sides of telecom networks, and study of wireless
networks using the magnifying glass provided by complex systems science. In particular, recent
results from the author’s research team are shortly discussed for each of these research areas.

Keywords: 5G, Spectrum, Small Cells, Massive MIMO, Waveforms, Optical Wireless
Integration, Complex Systems Science.

I. INTRODUCTION

Ongoing and future societal development will lead to changes in the way communication
systems are used. On-demand information and entertainment will increasingly be delivered
over mobile and wireless communication systems. These developments will lead to a big rise
of mobile and wireless traffic volume, predicted to increase a thousand-fold over the next
years [1][2]. It is also predicted that today’s dominating scenarios of human-centric
communication will be complemented by a huge increase in the numbers of communicating
machines; there are forecasts of a total of 50 billion connected devices by 2020 [3]. The
coexistence of human-centric and machine-type applications will lead to a large diversity of
communication characteristics.
Different applications will impose very diverse requirements on mobile and wireless
communication systems that the fifth generation (5G) will have to support [4]: very stringent
latency and reliability requirements, a wide range of data rates, network scalability and
flexibility, very low complexity and very long battery lifetime. One of the main challenges is
to satisfy these requirements while at the same time addressing the growing cost pressure.
We have fresh quantitative evidence that the wireless data explosion is real and will
continue.
A recent Visual Networking Index (VNI) report [5] makes it clear that an incremental
approach to 5G will not be enough to meet the demands that networks will face in the next
years; indeed likely 5G will have to be a paradigm shift that includes (among other things)
very high carrier frequencies with large bandwidths, extreme base station and device
densities, and massive numbers of antennas. The motivation behind chasing spectrum in
high frequencies is the scarcity of Radio Frequency (RF) spectra allocated for cellular
communications. Cellular frequencies use UHF bands for cellular phones, but these
frequency spectra have been used heavily, making it difficult for operators to acquire more.
Another challenge is that the deployment of advanced wireless technologies comes with high
 2

energy consumption; it has been reported by cellular operators that the energy consumption
of Base Stations (BS) contributes to over 70% of their electricity bill [6].
A difference compared to previous generations, is that 5G will also be highly integrative,
tying any new air interface and spectrum together with Long Term Evolution (LTE) and
WiFi to provide universal high-rate coverage and a seamless user experience [7].
Academia is engaging in large collaborative projects such as METIS [4] and 5GNOW [8],
while the industry is driving preliminary 5G standardisation activities. To further
strengthen these activities, the public-private partnership for 5G infrastructure recently
constituted in Europe [7][9]. Fig. 1 shows at a glance the recent evolution of wireless
communication systems standards. It is apparent that 5G encompasses elements that are
disruptive as compared to the past.

Fig. 1: Some characteristics of recent wireless systems generations.

II. DRIVERS AND REQUIREMENTS

While 5G requirements will span every key dimension, not all of them will need to be
satisfied simultaneously, as different applications will place different demands on the
performance of the systems. For example, very-high-rate applications such as streaming
high-definition video may have relaxed latency and reliability requirements compared to e.g.
driverless cars or public safety applications, where latency and reliability are paramount but
lower data rates can be tolerated [7]. 5G aims to connect the whole planet, achieving
seamless and ubiquitous communications between anybody and anything, anywhere,
anytime, and anyhow, i.e. by whatever device, service or network [10]. 5G will provide the
foundational infrastructure for building smart cities, which will push mobile network
performance and capability requirements to their extremes [11]. Among the main 5G
drivers, we have Internet of Things (IoT), Gigabit wireless connectivity and Tactile Internet.
For IoT, the main challenge is the scalability problem, with more than 100,000 Machine-to-
Machine (M2M) nodes in a cell under the premises of low cost and long lifetime. In terms of
Gigabit connectivity, for example users might request quick downloads of 3D streaming
content with data rates on the order of 100 Mbps; such levels of connectivity are also
expected in large crowd gatherings with possibly interactively connected devices. Tactile
Internet comprises a large amount of real-time applications with extremely low latency
requirements. Motivated by the tactile sense of the human body, which can distinguish
 3

latencies on the order of 1 ms accuracy, 5G aims to be applied to steering and control

scenarios, implying a disruptive change with respect to today’s communications [12].
As we move to 5G, costs and energy consumption 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; one should therefore try
to advocate technological solutions that promise reasonable cost and power scaling. A major
cost consideration for 5G, due to the increased BS densities and bandwidth, is how to do the
backhauling from the network edges into the core [7].

III. CANDIDATE TECHNOLOGY COMPONENTS

A popular view is that the required increase in data rate will be achieved, for the most
part, through combined gains in three categories [7]: 1) extreme network densification to
improve the area spectral efficiency (more nodes per unit area and Hz); 2) increased
bandwidth, primarily by moving towards mm-wave spectrum and making better use of
unlicensed spectrum in the 5 GHz band (more Hz); 3) increased spectral efficiency, chiefly
through advances in Multiple Input Multiple Output (MIMO) techniques (more bits/s/Hz per
node).
As per network densification, its motivation is that a straightforward but extremely
effective way to increase the network capacity is to make the cells smaller. Networks are
now rapidly evolving to include nested small cells such as picocells (range under 100 meters)
and femtocells (WiFi-like range), as well as distributed antenna systems [13][14][15].
In the quest for bandwidth, particular challenges that need to be addressed in 5G systems
are fragmented spectrum and spectrum agility. It is unlikely that these challenges can be
met using Orthogonal Frequency Division Multiplexing (OFDM), and new waveforms that
are more flexible and robust are required. In the 5GNOW project [16] several alternative
candidate waveforms are proposed, such as Filter Banks Multi-Carrier (FBMC) [17]. Recent
studies suggest that mm-wave frequencies could be used to augment the currently saturated
radio spectrum bands for wireless communications [18]. By increasing the RF channel
bandwidth for mobile radio channels, the data capacity can be greatly increased, while the
latency for digital traffic much decreased, thus supporting much better Internet-based
access and applications that require minimal latency. mm-wave frequencies, due to the
much smaller wavelength, may exploit new spatial processing techniques such as massive
MIMO [19], which can help to achieve larger spectral efficiency.
A native inclusion of M2M communication in 5G involves satisfying three fundamentally
different requirements associated with different classes of low data rate services: support of
a massive number of low rate devices, sustaining a minimal data rate in virtually all
circumstances, and very low latency data transfer [20]. A question for the industry is
whether we should have the same network designed for both human and machine
communications, a new dedicated network for machines, or a hybrid [21].
It is unlikely that one standard and one model of network deployment will be able to fit
all use cases and scenarios in the future. Mobile networks and deployed equipment need to
be flexible in order to be optimised for individual scenarios, which may be dynamic in space
and time. This requisite for flexibility will have a significant impact on the design of new
network architectures. In [22] the authors present one way to provide this flexibility by
leveraging cloud technology and exploiting it to operate Radio Access Networks (RAN).
Radio access infrastructures based on cloud architecture will provide on-demand resource
processing, storage and network capacity wherever needed. Software-defined air interface
technologies will be seamlessly integrated into 5G wireless access network architectures,
allowing RAN sites to evolve toward a “hyper transceiver” approach to mobile access,
helping to realize the joint-layer optimization of radio resources [11].
The ultimate goal of the communication networks is to provide access to information
when, where, and in whatever format we need it in. To achieve this goal, wireless and
optical technologies play a key role. Wireless and optical access networks can be thought of
as complementary. Optical fiber does not go everywhere, but where it does go, it provides a
 4

huge amount of available bandwidth. Wireless access networks, on the other hand,
potentially go almost everywhere, but provide a highly bandwidth-constrained transmission
channel susceptible to various impairments. Future broadband access networks must
leverage on both technologies and converge them seamlessly [23].
Fig. 2 summarises some of the candidate technology components that the author believes
will play a prominent role in the building up and affirmation of 5G. The following sections of
the paper will present some highlights about the work done to date in these areas by my
group.

Fig. 2: Some 5G candidate technology components.

IV. SPECTRUM

5G systems are expected to provide data rates on the order of Gbps, anytime and anywhere.
This can only be realised with much more spectrum than that currently available to
International Mobile Telecommunications (IMT) systems through the International
Telecommunication Union (ITU) process. All spectrum currently available to cellular mobile
systems, including IMT, is concentrated in bands below 6 GHz due to the favourable
propagation conditions in such bands. As a result, these bands have become extremely
crowded, and prospects for large chunks of new spectrum for IMT systems below 6 GHz are
not favourable [24]. As a consequence of this situation, Opportunistic Spectrum Access
(OSA) has been considered by regulators for a number of different spectrum bands. In [25]
we discuss and qualitatively evaluate techniques used in the discovery of spectrum
opportunities, also called white spaces, in the radar, TV, and cellular bands (see Fig. 3).
These techniques include spectrum sensing, cooperative spectrum sensing, geolocation
databases, and the use of beacons.
 5

Fig. 3: Which access technique for which band?

Each of the three bands mentioned above, calls for a different set of spectrum access
techniques. While TV bands are well suited to the adoption of geolocation databases, a
database-assisted spectrum sensing mechanism may represent the most efficient solution to
exploit the spectrum holes in radar bands. The unpredictability of cellular systems calls for a
more coordinated spectrum access approach, namely beacon signalling, that could be
implemented using the already established cellular infrastructure and spare bits of its
logical channels [25].
Another potential means to make large amounts of spectrum available to future IMT
systems is through Licensed Shared Access (LSA), whereby certain underutilised non-IMT
spectrum could be integrated with other IMT spectrum in a licensed pre-determined manner
following mutual agreement among the licensees [24]. In [26], we propose a cloud-RAN MD-
MIMO (Massive Distributed MIMO) platform as architecture to take advantage of the LSA
concept. This particular architecture is worth exploring in the context of LSA since similar
principles about the use of resources can be identified. LSA is a spectrum sharing approach
to a pool of virtual spectrum resources and cloud-RAN provides the way of managing the
pool of virtual network resources.
The resources available in a cloud-based RAN (antennas) and LSA (spectrum) are ideally
infinite (in practice, much larger than what a single virtual operator requires) but there is a
cost associated with their utilisation. Therefore, resource allocation needs to take into
account the (unconstrained) pool of resources and their utilisation cost. In [26], we consider
the problem of choosing the optimal set of spectrum and antenna resources that maximise
the resource efficiency, defined as the number of bits transmitted per cost unit (the resource
utilisation cost). In particular, we assume that K users demand a wireless service from a
virtual network operator. The operator rents antennas from the cloud-RAN and the LSA
spectrum for the time needed to transmit the information. Using the cloud-RAN
infrastructure (processing, backhaul, antennas, etc.) and LSA spectrum has a cost, in
currency units per second. The aim of the network operator is to choose the optimal number
of antennas M and bandwidth W such that the number of transmitted bits per currency unit
is maximised (see Fig. 4).
 6

Fig. 4: An example of cloud-based MD-MIMO RAN with Mmax = 8 and Wmax = 100 Mhz available
antennas and bandwidth, respectively. The virtual network operator uses a subset of the available
spectrum (W = 50 MHz) and antennas (M = 5) to maximise the number of bits transmitted to the K = 4
users per cost unit. The red lines indicate the antennas that are being used by the virtual network
operator.

The service provider wants to serve K users and chooses M antennas and W MHz from the
pool of resources offered by the cloud-based RAN and LSA. Using the infrastructure and the
spectrum has a cost, which we identify by the spectrum cost cw (cost of using 1 MHz of the
bandwidth from LSA for 1 second), the antenna cost cm (cost of using one of the distributed
antennas for 1 second), and the operative cost co (cost of using the cloud infrastructure, e.g.
backhaul, processing, etc. for 1 second).
The cost efficiency is the number of transmitted bits per cost unit (bits/cu). It is computed as
the ratio of the total rate to the total costs:

K
 r 
W ∑ log1 + k 
η (M , W ) =
k =1  N 0W  (1)
cm M + cwW + co

where rk is the power received by the kth user and N0 is the noise power spectral density.
Fig. 5 compares the optimal cost efficiency with the efficiency obtained by an arbitrary
strategy that either maximises the number of antennas or the bandwidth. We have assumed
that Wmax = 50 MHz and Mmax = 20. The results show that the optimal solution is able to
transmit up to an order of magnitude more information for the same cost. The figure also
shows that when the bandwidth cost is very small, maximising the bandwidth is near-
optimal. Similarly, if the bandwidth is expensive, the number of active antennas should be
maximised.
 7

Fig. 5: Cost efficiency when the optimal spectrum and number of antennas are chosen, compared with
the cost efficiency when the spectrum or the number of antennas is maximised.

V. SMALL CELLS

As mentioned in Section III, the number of base stations is expected to grow considerably;
there will be more and smaller base stations. This makes it possible to accommodate more
users within the same spectrum. However, the base station densification will need to be
supported by a widely spread backhaul network. Hence, also the number of backhaul links
will increase along with the number of base stations. Backhaul links could be either wired or
wireless. Fig. 6 depicts the network densification concept.
The emerging interest in small-cells has pushed researchers to investigate the
performance gain of cell splitting. For instance, in dense networks, when the path-loss
follows an attenuation proportional to a power of the distance, cell-splitting provides linear
Area Spectral Efficiency (ASE) gain with the density of nodes [27]. However the research in
small cells has not focused to date on how the total transmit power of the network changes
as cell-splitting occurs and what transmit power levels are needed to maintain linear gain.
To address this, in [28] we first provide the expression for the minimum transmit power that
guarantees linear ASE gain while performing cell-splitting. Then, by applying this
expression for the minimum power, we show that the total transmit power of the network
(i.e., the sum of the transmit power of all the base stations within a finite portion of the
network) needed to achieve linear ASE gain by means of cell-splitting is a decreasing
function of the node density, meaning that a significant reduction in the total transmit
power can be obtained by shrinking the cell size and increasing the node density.
 8

Fig. 6: Network and backhaul densification.

The ASE can be calculated as:

ASE = d ⋅ Ccell (2)

where d=D-2 is the cell density, with D being the scaling factor, and C cell is the average cell
capacity. The total transmit power of the network, obtained by setting the nodes power at
the minimum value that still guarantees linear ASE gain when performing network scaling,
can be expressed as:

PTX ,tot = P0 D β −2 α L2 (3)

where we consider a finite network portion bounded by a square of area L2 and containing N
N −1
1
base stations, P0 >0 is an arbitrary power, β ∈ ℜ is the path loss exponent, α =
N
∑α
k =0
k and

α k is related to the transmit power of the nth base station by PTX ,k = P0 D α k .

β

Fig. 7 shows the ASE gain and total transmit power reduction, due to the increase of base
station density. As we can see from the plot, the gain is linear with the density of base
stations. Nonetheless, the total transmit power used by the network can be decreased while
maintaining a linear ASE gain. Hence, the plots show a twofold advantage of increasing the
base station density; it allows achieving higher throughput and reducing the overall power
radiated by the base station antennas. The overall transmit power reduction achievable by
setting the transmit power as indicated in [28], may have positive implications in reducing
the aggregate interference experienced by an incumbent willing to share spectrum with a
secondary system of small cells. This may be particularly useful in future scenarios involving
LSA or Authorised Shared Access (ASA) schemes in which small cell networks exploit new
spectrum sharing opportunities.
 9

Fig. 7: ASE gain and total transmit power reduction, due to the increase of base station density.

VI. COMBINATION OF MASSIVE MIMO WITH FILTER BANK MULTI-CARRIER

In recent years massive MIMO has gained significant momentum as a potential candidate
to increase the capacity of multiuser networks, due to the fact that by increasing the number
of antennas at BS, the processing gain can be increased to become arbitrarily large. As a
consequence, in theory the network capacity can be increased unboundedly [29]. An
assumption made by [29] and followed by other researchers is that OFDM may be used to
convert the frequency selective channels between each mobile terminal and the multiple
antennas at the BS into a set of flat fading channels. Accordingly, the flat gains associated
with subcarriers constitute the spreading gain vector that is used for despreading of the
respective data stream.
In [30], we introduce the application of FBMC to massive MIMO communications. We find
that in the case of massive MIMO systems, linear combining of the signal components from
different channels smooths channel distortion; hence one may relax the requirement of
having approximately flat gain for the subcarriers, and as a result one may significantly
reduce the number of subcarriers in an FBMC system. That way we can reduce both system
complexity and the latency/delay caused by the synthesis filter bank (at the transmitter) and
the analysis filter bank (at the receiver). Also, one may adopt larger constellation sizes, and
hence further improve the system bandwidth efficiency. Moreover, increasing the subcarrier
spacing has the obvious benefit of reducing the sensitivity to Carrier Frequency Offset (CFO)
and Peak-to-Average Power Ratio (PAPR). An additional benefit of FBMC is that
carrier/spectral aggregation (i.e., using non-contiguous bands of spectrum for transmission)
becomes a simpler task, since each subcarrier band is confined to an assigned range and has
a negligible interference to other bands; this is not the case with OFDM [31].
Fig. 8 presents some results that show the effect on the Signal-to-Interference Ratio (SIR)
of increasing the number of antennas at the receiver, for different number of subcarriers in
a single-user case, exploring the system capability of achieving a flat channel response over
each subcarrier band. For the channel model used here, the total bandwidth, equivalent to
the normalised frequency one in the figure, is equal to 2.8 MHz. This, in turn, means the
subcarrier spacing in each case is equal to 2800/L kHz, where L is the number of subcarrier
bands. As an example, when L = 32, subcarrier spacing is equal to 87.5 kHz. Compared to
the subcarrier spacing in OFDM-based standards (e.g., IEEE 802.16 and LTE), this is
relatively broad; 87.5/15 ≈ 6 times larger. Reducing the number of subcarriers (equivalently,
increasing symbol rate in each subcarrier band), as noted above, reduces transmission
latency, increases bandwidth efficiency, and reduces sensitivity to CFO and PAPR.
 10

Fig. 8: Comparison of the SIR of matched filter linear combining technique, for different number of
receive antennas N and for the cases of 32 subcarriers.

VII. OPTICAL-WIRELESS INTEGRATION

The main reason why it has taken many years for Fibre To The Home (FTTH) access
systems to be deployed is one of financial viability. Ultimately, the main technology that
enabled the wide spread of fibre access was the Passive Optical Network (PON), since it
allowed sharing the costs of installing fibre and deploying electronic terminations among
multiple users [32]. As mobile network operators start looking into the deployment of large
numbers of small cells, to offer a higher capacity per user, the idea of a shared, low-cost,
fibre backhaul network based on existing PON systems becomes especially attractive. PON
systems seem thus an ideal technology for mobile backhaul of small (and large) cells, due to
the potentially ubiquitous presence of access points that they can offer. Indeed, the latest
ITU-T PON standard, 10-gigabit-capable Passive Optical Network (XG-PON), already
considers Fibre To The Cell (FTTCell) scenarios [33]. In Fig. 9, we can see an illustration of
an FTTCell scenario where an XG-PON network is used to backhaul an LTE system, by
connecting the base stations to the core components of LTE. In this architecture, the core
components are connected to the root of the PON, also called Optical Line Terminal (OLT)
while the base stations are connected to the Optical Network Units (ONU), at the leaves of
the tree shaped optical distribution network.
Typically, PONs and their Dynamic Bandwidth Assignment (DBA) algorithms, are
designed for ONUs that are independent from each other (often representing individually
billed entities). However, scenarios can be envisaged, FTTCell being an example, where
wireless operators may require more than one ONU per PON to provide service in different
locations. These entities may desire to have a single service level agreement for their group
of ONUs and share the contracted capacity among the entire group of ONUs. In [34] we
discuss hierarchical DBA algorithms that allow scheduling bandwidth to a group of base
stations rather than individually. By assuring bandwidth to a group of base stations, a
mobile operator can ensure that whenever one base station is not using its assured
bandwidth, it can be assigned first to a base station of the same operator, rather than
anyone else in the PON.
 11

Fig. 9: Fibre To The Cell architecture.

By doing this, mobile operators can make use of the properties of statistical multiplexing,
taking advantage of heterogeneity of the traffic from the base stations. With careful
dimensioning, it is possible for base stations to transmit at their peak rate, without assuring
the peak rate to each base station. While we could argue that the same effect could be
achieved with best-effort bandwidth, without assuring bandwidth to the group, the
performance of the backhaul would in that case depend on other users on the PON, possibly
competing mobile operators.
In [34] we propose group-GIANT (gGIANT), an algorithm developed to enable the
assignment of group assured bandwidth, and evaluate its performance through simulations
in homogeneous and heterogeneous traffic scenarios. In particular, the heterogeneous
simulation consists of changing the load of only one ONU, and keeping constant the load of
all the other ONUs. The results for the average delay can be seen in Figure 10, where N
indicates the size of the group, i.e. the number of ONUs belonging to it.
We can see that by adding ONUs to the group, extra capacity is available to the ONUs
that need it. Comparing the more homogeneous and the more heterogeneous scenarios, we
indeed saw in [34] that the results support the idea that gGIANT algorithm provides a
larger performance increase when the traffic load is unbalanced. This supports the idea that
the more heterogeneous and bursty the traffic is, the bigger the gains from group assured
bandwidth that can be obtained.

VIII. A COMPLEX SYSTEMS SCIENCE VIEW OF 5G

Modern ICT systems are becoming increasingly larger by englobing more and more
components, while at the same time there is an evergrowing flow of information in the
systems. As the technological and social trend of communications is shifting from systems
based on closed hierarchical or semi-hierarchical structures to open and distributed,
networked organisations, new paradigms are needed to model, design, monitor, and control
this new kind of systems. Communication engineers are defied by the task of designing
networks capable of self-organisation, self-adaptation and self-optimisation of their
interactions, and that can satisfy user demand without disruptions. In this regard, help
comes from recent studies of complex systems in nature, society and engineering, which
suggest ideas on how to design and control modern communication systems [35].
 12

Fig. 10: Average delay of upstream transmission when increasing the load of only one ONU.

In [36] we move a step towards a comprehensive and rigorous study of communication

systems, by using understanding and tools from complex systems science. As a specific
application of complex system science to wireless systems, we tackle the problem of
self-organising frequency allocation, using local information and adaptations to achieve
global network-wide behaviour. We support the claim that the system we are considering is
a complex one, both in terms of entropy and complexity metrics, and show that simple
agents such as cellular automata cells, can actually achieve a nontrivial interference-free
frequency allocation.
It is known from complex systems science literature that complexity and entropy are two
distinct quantities. For studies on the relation between complexity and information/entropy,
readers can refer to [37] [38]. In [36] we use excess entropy to measure complexity. Excess
entropy can be expressed in different ways; the form we utilise is the convergence excess
entropy EC, which is obtained by considering how the entropy density estimates converge to
their asymptotic value h. In two dimensions the entropy density h can be expressed as [39]:

h = lim h( M ) (4)
M →∞

where h(M) is the entropy of a target cell X conditioned on the cells in a certain
neighbourhood of X. Then the excess entropy EC is defined as:

∞
EC = ∑ (h( M ) − h )
M =1
(5)

We then study the global network-wide behaviour with respect to the complexity of the
channel allocation matrix. Fig. 11(a) shows the channel allocation matrix resulting from the
regularly spaced assignment of N = 6 channels. This is a typical example of the frequency
allocation resulting from a centralised frequency planning. Fig. 11(b) shows an example of
the channel allocation matrix resulting from the self-organising algorithm described in [36].
Finally, a random allocation of N = 6 channels is shown in Fig. 11(c).
 13

Fig. 11: (a) Regular channel assignment. (b) Channel assignment resulting from the self-organising
algorithm described in [36]. (c) Random channel assignment.

We estimate EC and h for the three types of channel assignments in Figure 11 by using
104 × 104 matrices. For the channel assignment in Fig. 11(a), the entropy estimates are
h( M ) = 0 , ∀M = {1,2,  ,6}. Hence, EC = 0 and h = 0. This is consistent with the crystal-like
completely ordered structure of the channel allocation matrix. For the random channel
assignment matrix, the entropy estimates are h ( M ) = 2.58 , ∀M = {1,2,  ,6}. Hence, EC = 0
and h =2.58. As the channel assignment matrix is completely disordered, the entropy is the
maximum possible for N = 6 channels, i.e. log 2 (6) . Finally, in the case of the channel
assignment matrix in Fig. 11(b), the entropy estimates are h ( M ) = 1.29 , ∀M = {4,5,6}.
Hence, h = 1.29 and the resulting EC is 2.04. Therefore the channel assignment emerging
from self-organisation exhibits a high amount of structure which neither the centralised nor
the random channel allocation can reach.
For networks that allocate frequencies in a self-organised manner, we then showed that it
makes sense to talk in terms of complex systems. Instead, in the case networks manage
frequencies in a centralised fashion, as the resulting allocation would be very much like a
regular crystal-like configuration, there is no point in studying them using complex systems
science. We consider the work of [36] as a step towards a comprehensive and rigorous study
of complex communication systems, adopting in communication networks design and
analysis, philosophy and results from the emerging multi-disciplinary field of complex
systems science.

IX. CONCLUSIONS AND OUTLOOK

It is a time of unprecedented change, where traffic on telecommunication networks is

growing exponentially, and many new services and applications are continuously emerging.
In this scenario we cannot predict exactly what lies ahead, and therefore the best we can do
is to extrapolate some trends and make educated guesses. A possible way out is to design
networks with change in mind, so that they will be more robust to disruption caused by
growing demands and changing user patterns and yet-unimagined applications, and that the
risks associated with investment in these kinds of networks will be lower, as they will be
more durable and scalable. Networks that are designed with change in mind will also make
effective use of resources (e.g., spectrum, bandwidth, power, processing capabilities,
backhaul, etc.) and ensure a sustainable future.
The next generation of wireless communication systems (5G) is part of the above picture.
In this article, one introduced the general landscape of 5G systems, including their likely
requirements and the candidate technologies promising to achieve such goals. A few relevant
5G areas have been discussed and recent research results in these domains from the
author’s research team have been shortly presented. In the last part of the article, one
focused on the complex systems science’s view of future communication networks. One can
 14

point out that one of the most widely accepted definitions of complex system, is that of ”a
system in which large networks of components with no central control and simple rules of
operation give rise to complex collective behaviour, sophisticated information processing,
and adaptation via learning or evolution” [40]. This view resonates with the author’s
research team understanding of future wireless networks; indeed networks are becoming
increasingly distributed, formed by an ever growing amount of nodes that must take local
decisions (due to limits in signalling exchange capacity) reacting to the surrounding
environment, and yet have to achieve a global level of good user experience and network
performance in general. There is therefore ground to believe that telecommunication
systems are evolving from being simple monolithic structures to complex ones, and that
complex systems science might prove beneficial in their analysis and design.

ACKNOWLEDGMENT

The author acknowledges support from the Irish CTVR CSET grant 10/CE/I1853.

REFERENCES

[1] Cisco, “Global mobile data traffic forecast update,” 2010-2015 White Paper, 2011.
[2] Nokia Siemens Networks, “2020: Beyond 4g radio evolution for the gigabit experience,” White
Paper, 2011.
[3] Ericsson, “More than 50 billion connected devices,” White Paper, 2011.
[4] “FP7 European Project 317669 METIS (Mobile and wireless communications Enablers for the
Twenty-twenty Information Society),” https://www.metis2020.com/, 2012.
[5] Cisco, “Visual networking index,” white paper at Cisco.com, Feb. 2014.
[6] C. Han et al., “Green Radio: Radio Techniques to Enable Energy Efficient Wireless Networks,”
IEEE Communications Magazine, vol. 49, no. 6, pp. 46–54, Jun. 2011.
[7] J.G. Andrews et al., “What Will 5G Be?,”, IEEE Journal on Selected Areas in Communications, vol.
32, no. 6, pp. 1065 – 1082, Jun. 2014.
[8] “FP7 European Project 318555 5G NOW (5th Generation Non-Orthogonal Waveforms for
asynchronous signalling),” http://www.5gnow.eu/, 2012.
[9] “5G-Infrastructure Public-Private Partnership,” http://5g-ppp.eu/, 2013.
[10] C.X. Wang et al., “Cellular Architecture and Key Technologies for 5G Wireless Communication
Networks,” IEEE Communications Magazine, vol. 52, no. 2, pp. 97 – 105, Feb. 2014.
[11] Huawei, “5G: A Technology Vision,” White Paper, 2013.
[12] G.P. Fettweis, “A 5G Wireless Communications Vision,” Microwave Journal, Dec. 2012.
[13] J.G. Andrews, “Seven ways that HetNets are a cellular paradigm shift,” IEEE Communications
Magazine, vol. 51, no. 3, pp. 136 – 144, Mar. 2013.
[14] J.G. Andrews et al., “Femtocells: Past, present, and future,” IEEE Journal on Selected Areas in
Communications, vol. 30, no. 3, p. 497 508, Apr. 2012.
[15] R.W. Heath et al., “A current perspective on distributed antenna systems for the downlink of
cellular systems,” IEEE Communications Magazine, vol. 51, no. 4, pp. 161 – 167, Apr. 2013.
[16] G. Wunder et al., “5GNOW: non-orthogonal, asynchronous waveforms for future mobile
applications,” IEEE Communications Magazine, vol. 52, no. 2, pp. 97 – 105, Feb. 2014.
[17] ——, “System-Level Interfaces and Performance Evaluation Methodology for 5G Physical Layer
Based on Non-orthogonal Waveforms,” in Asilomar Conference on Signals, Systems and Computers,
Nov. 2013, pp. 1659 – 1663.
[18] Z. Pi and F. Khan, “An introduction to millimeter-wave mobile broadband systems,” IEEE
Communications Magazine, vol. 49, no. 6, pp. 101 – 107, Jun. 2011.
[19] F. Rusek et al., “Scaling up MIMO: Opportunities and challenges with very large arrays,” IEEE
Signal Processing Magazine, vol. 30, no. 1, pp. 40 – 60, Jan. 2013.
[20] F. Boccardi et al., “Five Disruptive Technology Directions for 5G,” IEEE Communications
Magazine, vol. 52, no. 2, pp. 74 – 80, Feb. 2014.
[21] G.P. Fettweis and S. Alamouti, “5G: Personal mobile internet beyond what cellular did to
telephony,” IEEE Communications Magazine, vol. 52, no. 2, pp. 140 – 145, Feb. 2014.
[22] P. Rost et al., “Cloud technologies for flexible 5G radio access networks,” IEEE Communications
Magazine, vol. 52, no. 5, pp. 68–76, May 2014.
[23] N. Ghazisaidi et al., “Fiber-wireless (FiWi) access networks: A survey,” IEEE Communications
Magazine, vol. 47, no. 2, pp. 160 – 167, Feb. 2009.
 15

[24] B. Bangerter et al., “Networks and Devices for the 5G Era,” IEEE Communications Magazine, vol.
52, no. 2, pp. 90 – 96, Feb. 2014.
[25] F. Paisana and N. Marchetti and L. DaSilva, “Radar, TV and Cellular Bands: Which Spectrum
Access Techniques for Which Bands?,” IEEE Communications Surveys & Tutorials, vol. 16, no. 3, pp.
1193 – 1120, Jul. 2014.
[26] I. Gomez-Miguelez, E. Avdic, N. Marchetti, I. Macaluso, and L. Doyle, “Cloud-RAN Platform for
LSA in 5G Networks Tradeoff within the Infrastructure,” in International Symposium on
Communications, Control, and Signal Processing, May 2014.
[27] J. Ling and D. Chizhik, “Capacity scaling of indoor pico-cellular networks via reuse,” IEEE
Communications Letters, vol. 16, no. 2, pp. 231 – 233, Feb. 2012.
[28] C. Galiotto, N. Marchetti, and L. Doyle, “The Role of the Total Transmit Power on the Linear Area
Spectral Efficiency Gain of Cell-Splitting,” IEEE Communications Letters, vol. 17, no. 12, pp. 2256 –
2259, Dec. 2013.
[29] T. Marzetta, “Noncooperative cellular wireless with unlimited numbers of base station antennas,”
IEEE Transactions on Wireless Communications, vol. 9, no. 11, pp. 3590 – 3600, Nov. 2010.
[30] A. Farhang, N. Marchetti, L. Doyle, B. Farhang-Boroujeny, “Filter Bank Multicarrier for Massive
MIMO,” in IEEE Vehicular Technology Conference, Sep. 2014.
[31] B. Farhang-Boroujeny, “OFDM Versus Filter Bank Multicarrier,” IEEE Signal Processing
Magazine, vol. 28, pp. 92 – 112, May 2011.
[32] M. Ruffini et al., “DISCUS: an end-to-end solution for ubiquitous broadband optical access,” IEEE
Communications Magazine, vol. 52, no. 2, pp. 24 – 32, Feb. 2014.
[33] ITU Std., “10-Gigabit-Capable Passive Optical Networks (XGPON) Series of Recommendations,”
Mar. 2010, G.987.x, http://www.itu.int/rec/T-REC-G/e.
[34] P. Alvarez, N. Marchetti, D. Payne, M. Ruffini, “Backhauling Mobile Systems with XG-PON
Using Grouped Assured Bandwidth,” in European Conference on Networks and Optical
Communications, Jun. 2014.
[35] P. Antoniou and A. Pitsillides, “Understanding complex systems: A communication networks
perspective,” Technical Report TR-07-01, University of Cyprus, Tech. Rep., 2007.
[36] I. Macaluso, H. Cornean, N. Marchetti, L. Doyle, “Complex Communication Systems Achieving
Interference-Free Frequency Allocation,” in IEEE International Conference on Communications, Jun.
2014.
[37] R. Lopez-Ruiz et al., “A statistical measure of complexity,” Physics Letters A, vol. 209, no. 5-6, pp.
312 – 326, Dec. 1995.
[38] H.V. Ribeiro et al., “Complexity-entropy causality plane as a complexity measure for two-
dimensional patterns,” Plos One, vol. 7, no. 8, Aug. 2012.
[39] D.P. Feldman and J.P. Crutchfield, “Structural information in twodimensional patterns: Entropy
convergence and excess entropy,” Physical Review E, vol. 67, no. 5, May 2003.
[40] M. Mitchell, “Complexity - A Guided Tour,” Oxford University Press, 2011.