See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/280881469

On the Design of 5G Transport Networks

Article  in  Photonic Network Communications · August 2015

DOI: 10.1007/s11107-015-0553-8

CITATIONS READS

32 2,705

7 authors, including:

Matteo Fiorani Bjorn Skubic

Ericsson Ericsson
61 PUBLICATIONS   511 CITATIONS    65 PUBLICATIONS   929 CITATIONS   

SEE PROFILE SEE PROFILE

Jonas Mårtensson Piero Castoldi

RISE Research Institutes of Sweden Scuola Superiore Sant'Anna
70 PUBLICATIONS   688 CITATIONS    498 PUBLICATIONS   4,394 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Physical-layer survivability and security View project

FP7 COMBO View project

All content following this page was uploaded by Matteo Fiorani on 17 August 2015.

The user has requested enhancement of the downloaded file.

Noname manuscript No.
(will be inserted by the editor)

On the Design of 5G Transport Networks

Matteo Fiorani · Björn Skubic · Jonas

Mårtensson · Luca Valcarenghi · Piero
Castoldi · Lena Wosinska · Paolo Monti

Received: date / Accepted: date

Abstract Future 5G systems will pave the way to a complete new societal paradigm
where access to information will be available anywhere, anytime, and to anyone,
anything. Most of the ongoing research and debate around 5G systems are focusing
on the radio network segment (e.g., how to offer high peak-rates per subscriber,
and how to handle a very large number of simultaneously connected devices with-
out compromising on coverage, outage probability, and latency). On the other
hand, understanding the impact that 5G systems will have on the transport net-
work (i.e., the segment in charge of the backhaul of radio base stations and/or the
fronthaul of remote radio units) is also very important. This paper provides an
analysis of the key architectural challenges for the design of a flexible 5G transport
infrastructure able to adapt in a cost efficient way to the plethora of requirements
coming from the large number of envisioned future 5G services.
Keywords 5G transport · Backhaul · Fronthaul · Network Sharing · Network
Function Virtualizion (NFV) · Flexible transport

1 Introduction

The 5th generation of mobile networks (5G) is seen as one of the enablers of
what is referred to as the Networked Society [1], i.e., a future with user- and

The work described in this paper was carried out with the support of the Kista 5G Transport
Lab (K5) project funded by VINNOVA and Ericsson.

M. Fiorani · Lena Wosinska · P. Monti

KTH Royal Institute of Technology, Kista, Sweden
E-mail: [email protected]
B. Skubic
Ericsson Research, Kista, Sweden
E-mail: [email protected]
J. Mårtensson
Acreo Swedish ICT, Kista, Sweden
E-mail: [email protected]
L. Valcarenghi · P. Castoldi
Scuola Superiore Sant’Anna (SSSUP), Pisa, Italy
E-mail: [email protected]
2 Matteo Fiorani et al.

machine-centric communications where access to information and sharing of data

is available anywhere and anytime to anyone and anything.
Even though the details of 5G are still the subject of ongoing research and
debate, it is already clear that a future 5G system will need to offer not only high
peak-rates per subscriber, but it will also be required to handle a large number
of simultaneously connected devices, to provide better coverage, to offer highly
reliable communications and low latency. Finally, the above objectives must be
met at similar cost and energy consumption levels as today’s networks [2][3]. As
a result it becomes evident that the 5G vision requires the definition of a new
generation of mobile access systems that goes beyond the mere upgrade of the
existing ones, i.e., 3G and 4G.
Most of the ongoing research efforts on 5G mobile networks target the radio
technologies that will be part of 5G [4][5]. For example, 5G will include new and
flexible spectrum resources (e.g., sub-6.5GHz, mm-wave Radio Access Technologies
RAT), multi-antenna technologies with beam forming and massive multiple-input
multiple-output (MIMO), as well as inter site coordination for higher spectral
efficiency. Increased capacity and coverage will also be achieved through continued
radio site densification (e.g., through small cells).
Considering that a 5G system may also include different deployment architec-
tures (e.g., integrated radio base station, centralized baseband, radio-over-copper
solutions) it becomes also important to assess how 5G will impact the transport
infrastructure. However, as of today not much is known about these implications.
On the other hand, it is expected that with site densification as well as with larger
numbers and variegated services to be provisioned, the role of the transport net-
work, i.e., backhaul of radio base stations or fronthaul of remote radio units, will
become more and more crucial [6].
The aim of this paper is to discuss the key challenges in designing a transport
network for the future 5G mobile systems. First the paper provides an in-depth
overview of the set of services (and respective requirements) that a 5G transport
network is expected to accommodate. This overview is then the basis to present a
number of candidate transport architectures categorized according to the type of
transport segment they belong to (e.g., small cell transport vs. metro/aggregation
networks). The paper continues discussing a few transport design options, focusing
mainly on how to accommodate high traffic volumes over specific geographical
areas while providing transport resources in a flexible manner. Finally two case
studies are presented. Their purpose is to show how some of the key design concepts
presented in the paper can be applied to deploy a transport infrastructure able to
guarantee the required level of quality of experience of the various services to be
provisioned at the lowest possible both deployment and operational cost.

2 Related Works

Initiatives related to 5G are already undergoing both in research projects and

5G-related consortia, mainly focused on the definition of new radio technologies.
However, the transport part of the 5G infrastructure will play a key role as well.
With continued site densification and with larger numbers and variegated services
to be provisioned, the role of the transport network, i.e., backhaul of radio base
stations or fronthaul of remote radio units, will become more and more crucial.
The COnvergence of fixed and Mobile BrOadband access/aggregation networks
(COMBO) project [7] proposes and investigates new integrated approaches for
Fixed-Mobile Converged (FMC) broadband access/aggregation networks for differ-
ent scenarios (dense urban, urban, rural). COMBO architectures are based on joint
On the Design of 5G Transport Networks 3

optimisation of fixed and mobile access/aggregation networks around the innova-

tive concept of Next Generation Point of Presence (NG-POP). Within COMBO
a first resource orchestration across DWDM optical transport, radio access net-
works (RANs) and cloud domains based on SDN has been experimentally demon-
strated [8].
The Mobile and wireless communications Enablers for the Twenty-twenty In-
formation Society (METIS) project [2], which ended in March 2015, had as an
objective to lay the foundation of 5G. Most of the attention of the METIS project
is dedicated to define future scenario and requirements for 5G.
The 5G-Infrastructure-PPP (5G-PPP) is a 1.4 Billion Euro joint initiative
between the European ICT industry and the European Commission to rethink the
infrastructure and to create the next generation of communication networks and
services that will provide ubiquitous super-fast connectivity and seamless service
delivery in all circumstances. 5G-PPP released recently a vision document [9]
for 5G infrastructure that defines key drivers, 5G disruptive capabilities, design
principles, key enabling technologies, and timeline.
The iJOIN project [10] addresses the issue that wireless access and backhaul
are individually designed and therefore not optimised jointly. The approach is
based on the concept of RAN-as-a-Service (RANaaS), where RAN functionality
is centralised through an open IT platform based on a cloud infrastructure. The
SOftware-Defined Access using Low-Energy Subsystems (SODALES) project [11]
aimed at offering transparent transport services for fixed and mobile subscribers to
support 10 Gbps fixed access and radio technologies over a converged network ar-
chitecture. The transport solution is based on low cost 10 Gbps customer premises
equipments (CPE) for FTTH and fixed radio access.
Some recent works focus on specific aspect of the 5G transport infrastructure.
In [12], for example, an overview of the issues and requirements of next gener-
ation mobile fronthaul architectures is provided. In particular, it is stated that
the utilization of Common Packet Radio Interface (CPRI) might jeopardize the
savings expected from the C-RAN architecture. Thus, possible ways of avoiding
this are “the reduction of the digital fronthaul bandwidth requirements based on
a modified RAN architecture (dual-site processing) or in quasi-analogue optical
transmission schemes for certain use cases”.

3 Service Requirements for 5G Networks

5G networks will be required to support both the evolution of current services

as well as completely new services. These services will imply new requirements
on the RAN and consequently new requirements on the RAN transport network
(backhaul and fronthaul). We analyze the requirements on 5G networks accord-
ing to three groups of services: (i) end-user broadband services, , (ii) Machine
Type Communications (MTC), and (iii) cloud services. They are described in the
following subsections.

3.1 End-user Broadband Services

5G end-user broadband services represent an evolution of current broadband ser-

vices. They are mainly driven by the continued request for an enhanced user expe-
rience (e.g., 8K television and 3D video conferencing) and they pose a challenge in
terms of very high aggregated capacity to the transport network. End-user broad-
band services can be provided through either fixed or mobile access. 5G transport
4 Matteo Fiorani et al.

Fig. 1 Comparison among different mobile traffic forecasts. For the Ericsson Mobility Report
and Cisco VNI the area traffic demand was calculated from the average monthly generated
traffic using the model presented in [13].

networks will be required to support both, realizing the so called Fixed-Mobile

Convergence (FMC). In the following, we analyze separately mobile and fixed
access traffic requirements.

Mobile access services The traffic in mobile access networks has been increasing
exponentially over the last years, a trend that it is not expected to change in
the near future. Traffic forecasts for mobile access are usually based on modeling
end-user broadband services. According to the traffic forecast model presented in
the EU FP6 project EARTH [13], the area traffic demand is expected to increase
100 fold in the decade between 2010 and 2020 reaching 474 M bps/km2 average
at busy hours in dense urban areas. This prediction is in line with the latest
estimates provided by the Ericsson Mobility Report [14] and the Cisco Visual
Networking Index (VNI) [15]. A comparison among the different traffic forecast
models is presented in Figure 1. It is worth noting that the Ericsson Mobility and
Cisco curves were derived from the values of the average monthly traffic per mobile
terminal [14, 15] using the analytical formulation presented in [13].
More aggressive traffic scenarios have also been recently proposed and investi-
gated [16][3], e.g., the 5G test cases identified in the EU FP7 project METIS [3]. A
driver for more aggressive traffic densities in the mobile access could be a possible
replacement of the fixed access. METIS defines 5G in terms of five fundamental
scenarios which the next generation wireless access networks will have to sup-
port. Future applications may be associated with one or several of these scenarios
imposing different challenges to the network. METIS also defines twelve specific
test cases that are mapped onto the five scenarios. These test cases are used to
sample the space of future applications. There are two test cases that are par-
ticularly challenging in terms of traffic requirements. The first one, called dense
urban information society (i.e., METIS TC2), envisions a dense urban scenario
where humans interact with the cloud (i.e., any type of content remotely stored)
and the surrounding environment (e.g., sensors, video cameras) via high capacity
connections. As a result the traffic requirements for this test case are in the order
of 700 Gbps/km2 (i.e., 1000 times higher than the traffic forecasted for 2020 by
EARTH, Figure 1). The other test case, called virtual reality office (i.e., METIS
TC1), refers to the possibility to have high resolution 3D tele-presence and virtual
reality in indoor offices to remotely interact with colleagues. As a result this test
On the Design of 5G Transport Networks 5

case drives the traffic requirements up to 15 T bps/km2 [16][3], almost 100 times
higher than TC2.
Fixed access services The fixed access traffic is also increasing, even if at a lower
rate compared to its mobile counterpart. Fixed access services can be categorized
in residential and business. According to the traffic forecast model presented in
the EU FP7 project OASE [17], the residential traffic is expected to increase 10
fold in the current decade and reach average values at busy hours of 16 M bps per
household. On the other hand, peak residential traffic is expected to increase at
higher rate and reach 1 Gbps per household by the year 2020. Similar trends are
forecasted for the business traffic, i.e., a 10 fold increase in the current decade
reaching average and peak values per business of 10 Gbps and200 M bps, respec-
tively. In addition, business traffic introduces also strict availability requirements
on the transport network, i.e., 99.99% or higher.

3.2 Machine Type Communications

MTC applications are expected to play an important role in the design of 5G

transport networks. They introduce potentially new challenges which might go
beyond the pure connectivity. Due to the large number of possible scenarios and use
cases for MTC, it is difficult, if not impossible, to define a single set of requirements.
In the following, we present a few examples coming from a selected number of use
cases.
In MTC devices interact among themselves without human intervention [18].
MTC applications can be broadly categorized as massive and mission critical [18].
Massive MTC applications are provided through the extensive deployment of con-
nected devices (e.g., sensors and actuators), which perform a large number of dif-
ferent tasks (e.g., monitoring and surveillance). Each of these devices is expected
to generate a relatively small amount of data. The large number of connected ma-
chines will create a challenge for the wireless network, due to the large overhead
generated by the control information. On the other hand, massive MTC will most
probably not affect significantly the transport network. This is due to the fact that
the traffic generated by a large number of devices over a geographical area will be
aggregated in the transport.
Mission critical MTC applications are used to perform specific tasks in an
automated manner (e.g., industrial processes). Some mission critical MTC appli-
cations are expected to create major impact on transport networks due to the
high availability and strict latency requirements. Different applications introduce
different level of challenges. An example of mission critical MTC application is
traffic safety control. Cooperative active safety systems can warn drivers of dan-
gerous situations and intervene through automatic braking or steering if the driver
is unable to avoid an accident. The main requirement deriving from this MTC ap-
plication is related to the latency that should be as small as possible to minimize
the risk of accidents and damages to people. For instance, in case of automated
braking the maximum network reaction time should be lower than 10 ms [19] to
minimize the risk of collisions.

3.3 Cloud services

Businesses and consumers are increasingly relying on cloud-based services. This

is proven by the fact that the cloud traffic is increasing at a Compound Annual
Growth Rate (CAGR) of 32%, i.e., much faster than any other type of traffic [20].
6 Matteo Fiorani et al.

Cloud services are very heterogeneous in nature and may impose a large number of
different requirements on the transport network. As a consequence, 5G transport
networks must be adapt in a flexible way (i.e., in terms of capacity and latency,
etc.) in order to effectively support a large set of different requirements.
An example of a cloud application that has a severe impact on the transport
network is cloud gaming. In this application the entire game is hosted on a server
in a data center while the user runs only a simple client transmitting game con-
troller actions upstream to the game server. One single cloud gamer may require a
guaranteed streaming bit-rate of up to 15 M bps and round trip delays (including
processing in the data center) lower than 100 ms [21]. If we now consider that we
can have several of them playing simultaneously on the same transport network
segment, it is easy to understand how they can severely impact both the transport
network and the data center performance. An even more extreme example is High
Frequency Trading (HFT) [22]. HFT is the trading of financial assets based on
computerized strategies with brief holding time and high volume. At the heart of
such a trading algorithm would be the computation of short-term stock correla-
tion in real-time. HFT introduces very strict requirements in terms of transport
delay, since the lower the network delay the higher the potential revenues for the
investors. In order for HFT to be effective delays should be in the order of few
hundreds of µs. This translates also in requirements on the bit-rate, which should
be as high as possible in order to reduce the serialization delays (i.e., order of
several hundreds of M bps for the trader) [22].

4 RAN Transport Services

The service requirements described above trigger an evolution of both the RAN
transport architecture and services. The latter include the evolution of current
LTE fronthaul (i.e., centralized RAN (C-RAN)) and backhaul functionalities as
well as new service instances driven by the upcoming 5G radio technologies. LTE
fronthaul and backhaul services are expected to evolve to support the increased
amount of traffic generated per LTE base station (i.e., eNodeB). On the other
hand, innovative 5G radio access technologies will drive new requirements for 5G
transport networks. We identified three specific examples which are expected to
impact the transport network design: (i) ultra-densification, (ii) advanced radio
coordination, and (iii) massive multiple -input multiple -output (M-MIMO).
LTE backhaul services. LTE radio interfaces are continuously evolving to more
advanced configurations for supporting the increasing mobile traffic requirements.
As a consequence, the backhaul traffic generated by an LTE base station also
grows. In order to dimension the LTE backhaul network the following general rule
applies [23]. Given a number of sectors N (from one or more base stations), the
backhaul capacity is given by the maximum between the peak bit-rate for a single
sector and N times the busy hour average bit-rate. It is worth noting that the
peak bit-rate for a single sector is generated when a single user-equipment (UE)
is connected in close proximity of the base station, so that it experiences ideal
channel conditions and obtains the highest possible transmission rate [23]. The
peak bit-rate of a sector depends on the radio access network configuration (i.e.,
channel bandwidth, MIMO, etc.) as well as the UE category of the UE served by
the sector [24]. It is worth noting that the average bit-rate per sector at busy hours
of a small cell is usually higher than the one of a macro cell [25]. The reason is
twofold. First, small cells are usually deployed to cover a ”hot spot”, thus they are
more likely to have higher concentration of users towards the cell center and away
On the Design of 5G Transport Networks 7

Table 1 Expected backhaul capacity requirements for a LTE base station. We consider the
LTE configuration with highest possible spectral efficiency.

Base station Downlink Uplink

Macro (3 sectors) 228 M bps 126 M bps
Micro (1 sector) 90 M bps 50 M bps
Indoor (1 sector) 132 M bps 116 M bps

from the edge, which improves the signal quality distribution. Second, small cells
are likely to have better inter-cell isolation, reducing interference and improving
the signal quality distribution.
Table 1 shows the expected backhaul capacity for three LTE base station types
calculated according to the dimensioning guidelines just described. We considered
macro cells, micro cells, and indoor small cells. We assumed radio configurations
with the highest possible spectral efficiency available for LTE [24] and we consid-
ered 20 MHz of channel bandwidth. In addition, we considered that macro cells
are equipped with three sectors, while micro and indoor cells are omnidirectional
(one sector) and we excluded the transport overhead.
Centralized RAN. Centralized Radio Access Network (C-RAN) is a relatively new
architectural concept aimed at reducing the cost and at providing a better man-
agement of mobile networks [26]. Differently from conventional RANs, where radio
units and the baseband unit (BBU) are co-located at the base station site, in a
C-RAN architecture BBUs are decoupled from the base stations and centralized
in a Central Office (CO) or in a BBU Hotel. Radio units (also referred to as Re-
mote Radio Head (RRH)) are connected to BBU hotels via a fronthaul network
using Radio over Fiber (RoF) transmission technologies. RoF transmission can be
digital or analog. The main standards for Digital RoF (D-RoF) are the Common
Public Radio Interface (CPRI) [27] and the Open Base Station Architecture Ini-
tiative (OBSAI) [28], while there is an ongoing work in IEEE 1904.3 to define a
Radio over Ethernet solution [29]. On the other hand, Analog RoF (A-RoF) [30]
solutions have not been standardized yet. RoF transmission poses strict latency
and high capacity constraints on the transport network. In the following we will
use CPRI fronthaul as an example to describe these constraints.
The constraint on the latency for the transmission between a RRH and a BBU
comes from the LTE physical layer Hybrid Automated Repeat Request process [31]
which requires a maximum round trip delay of 3 ms, including both transport
and BBU processing time. As a consequence, the higher the transport delay the
lower the available time for BBU processing. A good practice is to limit the RoF
transmission delay to around 100 µs, which in turn means that the maximum
distance between a RRH and a BBU is limited to around 20 km.
The CPRI bit-rate can be calculated as a function of the radio configuration
and the analog-to-digital conversion (ADC) resolution according to the following
formula [27]:

BCP RI = NS · NAnt · RS · 2NRes · OCW · OLC , (1)

where NS is the number of sectors and NAnt is the number of MIMO elements
per sector. In addition, RS is the sampling rate (typically 30.72 M Hz for a single
20 M Hz carrier) and NRes is the number of bits per sample (typically 15 bits per
sample). Finally, OCW represents the overhead introduced by CPRI control words
(typically 1 control word for 15 words of payload) and OLC represents the line
8 Matteo Fiorani et al.

Table 2 CPRI transport capacity for different radio configurations using of a single 20 M Hz
carrier.
Sectors MIMO CPRI bitrate
1 1 1.228 Gbps (option 2)
1 2 2.457 Gbps (option 3)
1 4 4.915 Gbps (option 5)
1 8 9.830 Gbps (option 7)
3 4 12.165 Gbps (option 9)

coding overhead (typically 10/8 or 66/64 Byte). Table 2 shows a few values of the
transport capacity required by CPRI as a function of radio configuration using
OLC = 10/8 Byte.
Ultra-densification. Ultra-densification aims at increasing the radio network ca-
pacity via the dense deployment of low-cost and low-power small cells (indoor
and/or outdoor) [32]. The small-cells layer is added on top of the conventional
macro base station layer, which serves mainly for coverage purposes. As a result
the number of sites that need to be connected to the transport infrastructure be-
comes very high increasing substantially the complexity (mostly in terms of cost
and power consumption) of the transport network. This will drive the need for
more cost- and energy-efficient transport solutions [33][34].
Advanced radio coordination. Advanced radio coordination solutions, such as Co-
ordinated Multipoint (CoMP) and Enhanced Inter-Cell Interference Coordination
(eICIC), aim at improving the spectral efficiency in the radio network and in par-
ticular at the cell edges [35]. Different coordination levels (i.e., moderate, tight, and
very tight) pose different constraints on the transport network. Moderate coordi-
nation techniques (e.g., eICIC) have no specific requirement in terms of transport
performance. On the other hand, tight coordination techniques introduce strict
latency constraints (i.e., between 1 and 10 ms), while they do not introduce very
strict capacity constraints (i.e., lower than 20 M bps). Finally, very tight coordi-
nation techniques introduce very demanding constraints in terms of both latency
(lower than 0.5 ms) and capacity (in the range of several Gbps using CPRI).
Massive MIMO. The idea behind M-MIMO is to provide base stations with large
spatial multiplexing gains and beamforming capabilities thanks to hundreds of
antennas elements [36]. M-MIMO techniques allow for a better spectral efficiency
and for smoothed channel responses thanks to the vast spatial diversity they can
provide. As a result M-MIMO base stations are able to provide significantly higher
data rates to end-users. From the transport perspective this translates in higher
capacity that needs to be provided at each base station, i.e., M-MIMO drives
for transport networks able to support very high transmission capacities. The
required capacity per M-MIMO base station using backhaul services is in the range
of several Gbps. On the other hand, the required capacity using CPRI fronthaul
services is in the range of hundreds of Gbps or even a few T bps depending on
the specific radio configuration (see formula (1)). This raises doubts about the
feasibility of CPRI fronthaul for serving M-MIMO base stations.

5 Candidate Transport Network Architectures

This section provides an overview of candidate technologies and architectures for

5G transport networks. One aspect of 5G transport is providing higher capacity to
On the Design of 5G Transport Networks 9

Fig. 2 Reference transport network architecture.

more densely deployed macro sites. Another aspect is providing connectivity to a

large number of small cells. There are different avenues along which the transport
architecture could evolve. One scenario is the existence of a dominant/pre-existing
deployment of fixed access (DSL, PON, etc) that could be reused for connecting
small cells and possibly new macro sites. Such reuse could occur on an infrastruc-
ture level, reusing the installed base of copper/fiber or on a system level where
fixed access systems provide packet backhaul to new cells. In this case the ex-
isting fixed access infrastructure could reduce costs for connecting new cells, but
it will also restrict the RAN deployment options depending on the structure of
the existing infrastructure, e.g., limiting the possible degree of RAN centralization
and coordination. Another scenario is a dedicated infrastructure (separate from
the fixed access) to cater for the densification of macro and small cells. Such a
deployment could involve a large degree of wireless backhaul or it could be driven
by RAN transport requirements that cannot be fulfilled by the fixed access infras-
tructure. This scenario would also be relevant for more aggressive scenarios where
mobile access is the dominant access technology. In this work, we divide the trans-
port network in two segments, i.e., small cell transport and metro/aggregation
(Figure 2). The small cell transport segment aggregates the traffic to/from the
wireless small cells towards the metro/aggregation segment. The main challenge
in deploying a transport segment for small cell lies in having to reach a large num-
ber of geographically disperse cell sites. These sites could be indoor (e.g., offices
and/or residential premises) and might be difficult to be reached in an econom-
ically feasible way. For this reason cost-efficient deployments are necessary. The
metro/aggregation segment, on the other hand, connects different site types (i.e.,
macro, small cells, and/or fixed access points) among themselves and to the core
network via the service edge (i.e., the node(s) used for interconnecting different
network domains). Metro/aggregation networks are characterized by high capac-
ity requirements and thus high speed transmission and switching technologies are
required in this network segment.

5.1 Small Cells Transport

The best small cells transport network technology depends on the specific de-
ployment scenario. The main technology options can be categorized as copper-
, wireless- and fiber-based [33][34]. Recent copper-based technologies (such as
G.fast [37]) are able to offer rates in the order of few Gbps over relatively short
distances (i.e., few hundreds of meters). They represent an optimal solution for
backhauling indoor LTE small cells in areas where there is an already installed
copper infrastructure that can be reused (e.g., in order to save costs). In principle,
advanced copper-based technologies could also be used for fronthaul (e.g., CPRI
or analog radio over copper [30], but only for low bit-rate options (e.g., option 2
10 Matteo Fiorani et al.

in Table 2) and given that the BBUs are located maximum few hundreds meters
away from the RRHs.
Wireless-based solutions are an attractive option in locations where deploy-
ing a wired transport infrastructure is difficult (e.g., due to geographical condi-
tions) or uneconomical. Advanced wireless-based technologies, i.e., millimeter wave
(mmw) [38] and Free Space Optics (FSO) [39], achieve very high transmission ca-
pacities (up to several Gbps) over short and medium distances (i.e., up to few
km). As a result, they can be used both for backhauling LTE base stations and
for fronthauling RRHs (when the maximum distance between RRHs and BBUs is
limited to few km).
Fiber-based transmission technologies are able to provide very high data-rates
(i.e., up to hundreds of Gbps) over long distances (i.e., tens of km or more). This
is the reason why fiber-based solutions are seen as a good and long term candidate
for small cells transport. In particular, next generation Passive Optical Networks
(PON) [40] technologies (such as Ultra Dense Wavelength Division Multiplex-
ing PON (UD-WDM-PON) and Time Wavelength Division Multiplexing PON
(TWDM-PON)) are particularly attractive due to their reduced energy consump-
tion and potentially lower costs compared to Active Optical Networks (AON). For
a more detailed comparison of different optical and wireless technology options for
small cells transport the reader is referred to [6].

5.2 Metro/aggregation Network

A promising candidate for the metro/aggregation segment is represented by Dense-

Wavelength-Division Multiplexing (DWDM) networks [41][42]. An example of DWDM
metro network based on optical DWDM rings is shown on the right hand side of
Figure 2. In this network several optical DWDM access rings (NR ) collect the
traffic from a large number of access points (i.e., macro base stations plus small
cells and fixed access points) and transmit it to the metro node. The metro node
aggregates the traffic from the various access rings and forwards it to the metro
ring and the service edge. The access points as well as the service edge have elec-
tronic packet aggregation functionalities to allow for the the multiplexing of low
rate traffic. On the other hand, for the metro node a number of different possible
architectures can be used, as described next.
Optical metro node. With this solution, switching at the metro node is done
completely in the optical domain thanks to active optical elements such as Wave-
length Selective Switches (WSSs) and Reconfigurable Optical Add-Drop Multi-
plexers (ROADMs). This solution is referred to as DWDM-centric[42]. An advan-
tage of this approach is that the architecture of the metro node is simple and can
potentially reduce the cost and the energy consumption levels as well as allow
for an easy maintenance. On the other hand, since the traffic in the access rings
cannot be statistically multiplexed in the direction toward the metro ring, this
solution might require several fibers in the metro ring.
Electronic metro node. In this solution, packet aggregation is employed in the
metro node to statistically multiplex the traffic toward the metro ring, e.g., by
using a high capacity Ethernet packet switch. Electronic packet aggregation allows
for electronic traffic grooming, i.e., a large number of wavelength channels in the
access rings can be multiplexed over a lower number of higher capacity channels in
the metro ring. As a consequence, this solution might reduce the number of fibers
required in the metro ring compared to the solution based on optical switching at
the metro node. However, an electronic packet switch might consume more power
in addition to having high deployment and maintenance costs.
On the Design of 5G Transport Networks 11

Micro data center. In this last approach the electronic metro node not only
performs packet aggregation, but it is also directly connected to a micro data
center that allows for advanced processing and storing functionalities inside the
transport network. For example cache servers can be placed in the micro data
center to provide video delivery services to the end-users. In this way, part of the
traffic is terminated locally at the metro node and is not transmitted toward the
service edge, i.e., offloading the metro ring. This solution may potentially reduce
the number of fibers required in the metro ring (i.e., with respect to electronic
aggregation-only solution) and it also provides better Quality of Experience (QoE)
levels for the end-users. On the other hand, the extra IT equipment installed at
the metro node increases both costs and energy consumption.

6 Flexible Transport Networks

In this section we discuss the pros and cons of a number of design strategies for
5G transport networks. There are two different approaches that can be used while
deciding where to deploy resources in a transport infrastructure. The first one is
based on the overprovisioning concept, i.e., dimensioning for the worst case traffic
scenario (i.e., peak traffic). This is the approach mobile operators have been mainly
using so far. Its main advantage is the relatively low complexity since no dynamic
control of transport resources is required. With overprovisioning the 5G transport
capacity constraints are satisfied thanks to the ubiquitous availability of transport
resources. However, this solution comes with high cost in terms of equipment and
suffers from an inefficient use of transport resources.
The second design approach allocates transport resources on-demand to sup-
port specific transport needs, that may vary over time. This can be achieved using
advanced network functionalities, such as dynamic resource sharing and network
function virtualization (NFV) [43]. Dynamic resource sharing is based on the in-
tuition that the same transport resource can be dynamically shared over time for
different transport purposes. The type of resources and how they can be shared
depends on the specific transport service, the deployment scenario, the radio de-
ployment architecture, and on the choice of transport technology. Examples of
transport resources that can be dynamically shared are: optical fibers (e.g., using
dynamic wavelength allocation), wavelength channels (e.g., using time division
multiplexing), and spectrum (e.g., using time, frequency, or code division multiple
access).
NFV provides flexibility by dynamically placing network functions in differ-
ent locations depending on the specific need of a service, e.g., close to the users
to exploit traffic locality. Dynamically pushing network functions closer to the
users enables to serve requests locally and to offload the central part of the net-
work infrastructure, thus reducing the risk of congestions. In addition, NFV allows
for reduced end-to-end service delays, fulfilling the extreme latency requirements
imposed by some critical MTC applications. The network functions that can be
dynamically virtualized can be categorized in: (i) radio network functions, (ii)
transport functions, and (iii) IT functions. Radio network functions may include,
for instance, the virtualization of the Evolved Packet Core (EPC) functionalities
for local breakout (approach referred to as Virtual EPC (vEPC [44])). The vEPC
will be an important feature in 5G transport as it enables to terminate the mobile
traffic, and to serve mobile service requests locally. Transport functions may in-
clude the virtualization of packet aggregation, while IT functions may include the
virtualization of computing and storing functionalities, e.g., network caching. On
12 Matteo Fiorani et al.

Fig. 3 Transport network architecture for case study 1.

Table 3 Traffic requirements for case study 1 assuming an area of 2 km2 . AP: Access Point.
MN: Metro Node.
Mobile Tot. x AP Traffic x AP Traffic x MN
Macro 60 1 0.23 Gbps 13.7 Gbps
Micro 600 10 0.90 Gbps 54.0 Gbps
Indoor 6000 100 13.20 Gbps 792.0 Gbps
Fixed Tot. x AP Traffic x AP Traffic x MN
Household 20000 - - 320 Gbps
Business 4000 - - 808 Gbps

the other hand, NFV requires the availability of an IT infrastructure, e.g., micro
data centers, located within the transport network. One important aspect to con-
sider for NFV is also the business model in place in a specific transport network
instance. If the transport- and the wireless-service providers are two separate en-
tities, then the type of network functions that can be virtualized depends on the
specific agreements between these two entities.
The use of dynamic resource sharing and NFV means also moving some of the
complexity from the data plane to the control plane. There are several options for
realizing a control plane able to support dynamic resource sharing and NFV. On
the other hand, the best performance can be achieved by a control architecture
able to jointly orchestrate transport, radio, and IT resources. The implementation
of such an advanced control plane is an extremely challenging task. One possible
option for realizing this integrated multi-domain and multi-technology control ar-
chitecture is Software-Defined Networking (SDN). In [8] a proof of concept of an
SDN-based control plane for joint orchestration of radio and transport resources
has been presented, while further demonstrations are currently a major research
focus.

7 Case studies

This section presents two case studies aimed at evaluating the impact on the
transport performance of some of the concepts introduced in the paper so far. In
the first case study we analyze the cost and energy consumption levels of different
architectural options for the metro node (i.e., optical metro node, electronic metro
node or micro data center) in a DWDM transport network. In the second case
study, we evaluate the potential benefits of using advanced network functionalities
against a pure overprovisioning-based design approach.
On the Design of 5G Transport Networks 13

Fig. 4 Power consumption, equipment cost and number of wavelength channels required in
case study 1 for different metro architectural options.

7.1 Case Study 1: impact of metro node design

In this case study we consider a dense urban scenario with mobile and fixed users.
The mobile traffic requirements are the ones defined in METIS TC2 [3] (see Fig-
ure 1). We assume that mobile traffic is served using an ultra-dense heterogeneous
radio network composed of indoor small, micro and macro base stations. The traffic
generated per base station type is defined in Table 1. The area under considera-
tion is 2 km2 . The traffic from the wireless base stations is collected at the Access
Points (APs) of the DWDM network (see Figure 3). The macro base stations are
directly connected to the APs, while the small base stations (i.e., indoor small and
micro base stations) are connected to the APs via dedicated small cells transport
networks. The specific architecture of the small cells transport networks is outside
the scope of this study. In the metro network, four access rings with 15 APs each
(the area in total has 60 APs) collect the traffic from all the wireless base stations
and sends toward the Metro Node (MN). Moreover, the fixed users are served by
residential and business access networks. The average traffic generated per house-
hold and per business location is derived from [17] (see Section 3). The traffic from
the fixed access networks is aggregated directly at the MN of the DWDM network
(see Figure 3), so that it bypasses the access ring. The total number of wireless
base stations and fixed access points as well as their traffic requirements for the
considered scenario are reported in Table 3.
Three possible architectures can be considered for the metro node. The first
architecture is based on an optical metro node (i.e., with Wavelength Selective
Switches (WSS)) with wavelength channels operating at 10 Gbps in both the ac-
cess and in the metro rings. The second architecture is based on an electronic
metro node (i.e., an Ethernet packet switch) and wavelength channels operating
at 10 Gbps in the access ring and at 100 Gbps in the metro ring. In this way, the
traffic from the fixed access and the wireless base stations can be aggregated to
a lower number of higher rate channels. The third architecture is similar to the
second one, but allows for the addition of a micro data center in the MN that
is used for video content delivery. The micro data center is equipped with cache
servers that provide local delivery of content-based traffic, i.e., in this specific case
YouTube and Netflix video clips.
Figure 4 shows the power consumption, equipment cost and number of wave-
lengths in the metro ring for the three different metro architectures. In order to
14 Matteo Fiorani et al.

Fig. 5 Transport network architecture for case study 2.

calculate the power consumption and equipment cost we employed the models and
the input values presented in [45]. The equipment cost is expressed in Cost Unit
(CU) where each CU corresponds to half the cost of a 10 Gbps Ethernet interface.
In order to estimate the amount of traffic that can be served locally using the
Youtube and Netflix cache servers, we used the same offloading traffic ratios as
presented in [45]. It can be observed in Figure 4 that having an optical MN is by
far the most energy- and cost-efficient option (i.e., both energy consumption and
equipment costs are 2 to 3 times lower than in the other architectures). This is due
to the simplified architecture of the MN, which comprises only WSSs and requires
substantially less power-consuming and costly network components compared to
an electronic-based MN. The relatively high cost of the reconfigurable WSSs is
compensated by avoiding the large number of expensive optical transponders and
electronic packet switching elements required in the electronic MN. However, the
architecture with optical MN requires a very large number of 10 Gbps wavelength
channels in the metro ring (i.e., as high as 233). This high number of channels
might not be supported using a single fiber, so that more fibers might be needed
in the metro ring. As a result, in a fiber-scarce scenario a solution based on elec-
tronic switching is potentially a better option. In fact, using statistical multiplexing
and 100 Gbps transponders in the metro ring it is possible to reduce the number
of wavelength channels required by a factor of 10. Among the solutions based on
electronic packet switching, the one with micro data center provides both slightly
lower equipment cost and potentially higher user performance. The lower costs are
due to the fact that, by allowing traffic locality, this solution requires a lower num-
ber of 100 Gbps channels in the metro ring (i.e., 16 vs. 20). Given the high costs of
100 Gbps interfaces, this reduction is sufficient to compensate for the extra cost of
the cache servers. In addition, the local delivery of video content can potentially
allow for a better QoE for the end-users. Unfortunately, due to the high power
consumption of the cache servers, the solution based on micro data center is the
most energy consuming.

7.2 Case Study 2: benefits of deploying flexible transport networks

A virtual reality (VR) office [3] is a top-modern (virtual) office space where the
work involves the interaction with high resolution 3D scenes. The 3D video traffic
drives the average traffic density per VR office to 0.1 Gbps/m2 (assuming an aver-
age office size of 20 m2 the traffic volume is 20 Gbps per VR office). On the other
hand, the peak traffic volume density per VR office can be as high as 0.5 Gbps/m2
(i.e., 100 Gbps per VR office). In terms of wireless deployment the project METIS
envisions an ultra dense small cell network (UDN) solution [3]. The overall impact
of this test case on the metro transport depends on how many VR offices an office
building has. If we assume up to 50 VR offices then we can expect an average ag-
gregated backhaul traffic of 1 T bps and a peak of 5 T bps per building respectively.
On the Design of 5G Transport Networks 15

In this scenario it is reasonable to assume that each building is directly connected

to an access point of the DWDM metro network (see Fig. 5)), which in this case
study is supposed to be equipped with wavelength channels operating at 100 Gbps
(both in the access and metro rings). Using a pure overprovisioning-based design
approach, the metro network should be dimensioned for the peak traffic scenario,
so that 50 wavelengths per building are required. As a consequence, in order to
serve a metropolitan area with 100 office buildings a total of 5000 wavelength
channels (i.e., 10000 transponders operating at 100 Gbps) are required, leading to
an enormous costs for the network operator. On the other hand, using a SDN con-
trol plane and an efficient dynamic wavelength allocation policy it is possible to
dimension the metro network for the average traffic scenario, so that the area can
be served using 1000 wavelength channels. This reduces the number of required
transponders to 6000 (it is worth noting that the number of transponders inside
the buildings must always be calculated according to the worst case scenario). As
a consequence, using a flexible transport network design instead of overprovision-
ing the operator can reduce its equipment cost by 40%. Additional benefits can
also be achieved by placing IT resources servers at the access points or at internal
nodes of the transport network (Fig. 5) and by exploiting NFV. The servers can
be controlled by an SDN-based control plane, so that some network functions,
such as caching of data files or processing of 3D video traffic, can be dynamically
virtualized and pushed closer to the VR offices in order to save capacity inside
the metro network. A possible policy could be to monitor the traffic generated
in each office building and to activate the local servers when the traffic exceeds
a certain threshold. In this way the amount of traffic that needs to flow through
the metro network can be reduced. In terms of the dedicated small cell transport
network, optical solutions are the most suitable given the high capacity needed
between the wireless UDN and the metro aggregation point (Fig. 5). Assuming
that VR offices will be deployed in modern buildings equipped with an advanced
fiber infrastructure, the use of Grey Point-to-Point (PtP) optical links might be
an effective technology option due to low energy consumption and its relatively
high maturity [6].

8 Conclusions

This paper discusses the efficient design of 5G transport networks. Firstly, the
5G transport services are identified and categorized in terms of broadband, RAN
transport, MTC and cloud services, respectively. For each category the main re-
quirements imposed on the transport network are defined. Then, several candidate
technologies for 5G transport networks as well as three architectural options for
optical DWDM metro networks (optical metro node, electronic metro node and,
electronic metro node with micro data center) are presented. In addition, two
approaches for transport networks design are described. One is based on the over-
provisioning of transport resources while the second design approach is based on
dynamic resource sharing and network function virtualization (NFV) in conjunc-
tion of a software defined network (SDN)-based controller (referred to as flexible
transport deployment). Finally, two case studies are presented to: (i) assess the
cost and energy benefits of an all optical metro node solution, and (ii) to present
the potential savings brought by a flexible transport network deployments vs. a
pure overprovisioning-based design approach.
16 Matteo Fiorani et al.

References

1. ”Networked Society Essential”, Ericsson Booklet, http://www.ericsson.com.

2. ”The METIS 2020 Project: Laying the Fundation of 5G”, http://www.metis2020.com.
3. ”Scenarios, Requirements and KPIs for 5G Mobile and Wireless System”, EU FP7 Project
METIS, Deliverable D1.1, April 2013.
4. ”The 5G Infrastructure Public Private Partnership”, http://5g-ppp.eu/projects/.
5. ”User Equipment (UE) Radio Access Capabilities (Release 12)”, 3GPP TS 136.306, June
2011.
6. M. Fiorani, P. Monti, B. Skubic, J. Mårtensson, L. Valcarenghi, P. Castoldi, L. Wosinska,
”Challenges for 5G Transport Networks,” in Proc. of IEEE ANTS, December 2014.
7. http://www.ict-combo.eu/
8. A. Rostami, K. Wang, Z. Ghebretensae, P. Öhlen, B. Skubic, ”First Experimental Demon-
stration of Orchestration of Optical Transport, Ran and Cloud Based on SDN”, in Proc. of
OFC, March 2015.
9. http://5g-ppp.eu/wp-content/uploads/2015/02/5G-Vision-Brochure-v1.pdf
10. http://www.ict-ijoin.eu
11. http://www.fp7-sodales.eu
12. T. Pfeiffer, ”Next Generation Mobile Fronthaul Architectures”, in Proc. of OFC, March
2015.
13. ”Energy Efficiency Analysis of the Reference Systems, Areas of Improvements and Target
Breakdown”, EU FP6 Project EARTH Deliverable D2.3, January 2012.
14. ”Ericsson Mobility Report Mobile World Congress Edition”, Ericsson White Paper, Febru-
ary 2015.
15. ”Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2014-2019”,
Cisco White Paper, February 2015.
16. ”5G White Paper”, Next Generation Mobile Networks (NGMN) Alliance White Paper,
February 2015.
17. ”Consolidated requirements for European next-generation optical access networks”, EU
FP7 Project OASE, Deliverable D2.2.2, November 2012.
18. D. Astely, E. Dahlman, G. Fodor, S. Parkvall, J. Sachs, ”LTE Release 12 and Beyond
[Accepted From Open Call]”, IEEE Communications Magazine, vol.51, no.7, pp.154-160,
July 2013.
19. S. Strin, M. Kang, J. Jin, S. Kim, H. Kim, S. Moh, ”Vehicle-to-Vehicle emergency message
dissemination through the WiBro network,” in Proc. of INC, pp.1-6, May 2010.
20. ”Cisco Global Cloud Index: Forecast and Methodology, 2013-2018”, Cisco White Paper,
June 2014.
21. M. Hemmati, A. Javadtalab, A. Shirehjini, S. Shirmohammadi, T. Arici, ”Game as Video:
Bit Rate Reduction Through Adaptive Object Encoding”, in Proc. of the ACM NOSSDAV,
pp.7-12, February 2013.
22. B. Satzger, W. Hummer, P. Leitner, S. Dustdar, ”Esc: Towards an Elastic Stream Com-
puting Platform for the Cloud,”in Proc. of IEEE CLOUD, pp.348-355, July 2011.
23. ”Guidelines for LTE Backhaul Traffic Estimation”, Next Generation Mobile Networks
(NGMN) Alliance White Paper, July 2011.
24. ”Feasibility study for Further Advancements for E-UTRA (LTE-Advanced)”, 3GPP TR
36.912, April 2014.
25. ”Small cell backhaul requirements”, Next Generation Mobile Networks (NGMN) Alliance
White Paper, April 2012.
26. I. Chih-Lin, C. Rowell, Shuangfeng Han, Zhikun Xu, Gang Li, Zhengang Pan, ”Toward
Green and Soft: A 5G Perspective”, IEEE Communications Magazine, vol.52, no.2, pp.66-73,
February 2014.
27. http://www.cpri.info
28. http://www.obsai.com
29. ”Standard for Radio Over Ethernet Encapsulations and Mappings”, IEEE 1904.3 Task
Force.
30. A. Haddad, M. Gagnaire, ”Radio-over-Fiber (RoF) for mobile backhauling: A technical
and economic comparison between analog and digitized RoF”, in Proc. of ONDM, pp.132-
137, May 2014.
31. ”C-RAN The Road Towards Green RAN v2.6”, China Mobile White Paper, June 2014.
32. J.G. Andrews, S. Buzzi, Wan Choi, S.V. Hanly, A. Lozano, A.C.K. Soong, J.C. Zhang,
”What Will 5G Be?”, IEEE Journal on Selected Areas in Communications, vol.32, no.6,
pp.1065-1082, June 2014.
33. S. Tombaz, P. Monti, F. Farias, M. Fiorani, L. Wosinska, J. Zander, ”Is backhaul becoming
a bottleneck for green wireless access networks?”, in Proc. of IEEE ICC, pp. 4029-4035, June
2014.
 On the Design of 5G Transport Networks 17

34. M. Fiorani, S. Tombaz, P. Monti, M. Casoni, L. Wosinska, ”Green Backhauling for Rural
Areas”, in Proc. of ONDM, May 2014.
35. M. Hogan, ”Mobile Backhaul and Synchronization for Heterogeneous Networks”, Interne-
tional Telecom Sync Formum (ITSF), November 2012.
36. F. Rusek, D. Persson, L. Buon Kiong, E.G. Larsson, T.L. Marzetta, T.L., O. Edfors, F.
Tufvesson, F., ”Scaling Up MIMO: Opportunities and Challenges with Very Large Arrays”,
IEEE Signal Processing Magazine, vol.30, no.1, pp. 40-60, January 2013.
37. M. Timmers, M. Guenach, C. Nuzman, J. Maes, ”G.Fast: Evolving the Copper Access
Network,” IEEE Communications Magazine, vol.51, no.8, pp.74-79, 2013.
38. R. Weiler, R.J. Weiler, M. Peter, W. Keusgen, E. Calvanese-Strinati, A. De Domenico,
I. Filippini, A. Capone, I. Siaud, A. Ulmer-Moll, A. Maltsev, T. Haustein, K. Sakaguchi,
”Enabling 5G Backhaul and Access with Millimeter-waves”, in Proc. of EuCNC, pp.1-5,
June 2014.
39. Li Yuan Li, N. Pappas, V. Angelakis, M. Pioro, Di Yuan, ”Resilient Topology Design for
Free Sspace Optical Cellular Backhaul Networking,” in Proc. of IEEE Globecom pp.487-492,
December 2014
40. B. Skubic, I. Pappa, ”Energy Consumption Analysis of Converged Networks: Node Con-
solidation vs Metro Simplification,” in Proc. of OFC, pp.1-3, March 2013.
41. Shuqiang Zhang, Ming Xia, S. Dahlfort, ”Fiber Routing, Wavelength Assignment and Mul-
tiplexing for DWDM-centric Converged Metro/aggregation Networks,” in Proc. of ECOC,
pp. 1-3, Sept. 2013.
42. P. Öhlen, B. Skubic, Z. Ghebretensae, W. John, M. Shirazipour, ”Software-defined Net-
working in a Multi-purpose DWDM-centric Metro/aggregation Network,” in Proc. of IEEE
Globecom 2013.
43. ”Network Functions Virtualisation”, ETSI White Paper, October 2012.
44. A. Basta, W. Kellerer, M. Hoffmann, K. Hoffmann, E. Schmidt, ”A Virtual SDN-Enabled
LTE EPC Architecture: A Case Study for S-/P-Gateways Functions”, in Proc. of IEEE
SDN4FNS, pp.1-7, November 2013.
45. M. R. Raza, M. Fiorani, P. Monti, B. Skubic, J. Mårtensson, L. Wosinska, ”Power and
Cost Modeling for 5G Transport Networks,” to appear in Proc. of IEEE ICTON, July 2015.

View publication stats