1

5G-XHaul: A novel wireless-optical SDN transport

network to support joint 5G backhaul and fronthaul
services
D. Camps-Mur, J. Gutiérrez, E. Grass, A. Tzanakaki, P. Flegkas, K. Choumas, D. Giatsios, A. Beldachi, T.
Diallo, J. Zou, P. Legg, J. Bartelt, J. K. Chaudhary, A. Betzler, J. J. Aleixendri, R. González, and D. Simeonidou

Abstract—The increased carrier bandwidth and the number We argue that several RAN functional splits will coexist in
of antenna elements expected in 5G networks require a redesign the upcoming 4G/5G landscape, featuring D-RAN in situations
of the traditional IP-based backhaul and CPRI-based fronthaul where backhaul is limited (e.g. Small Cells), traditional CPRI-
interfaces used in 4G networks. We envision future mobile
networks to encompass these legacy interfaces together with novel based fronthaul for 4G networks, and novel eCPRI-based
5G RAN functional splits. In this scenario, a consistent transport fronthaul for future 5G RANs. A unified transport network
network architecture able to jointly support backhaul and 4G/5G architecture is needed to serve all these interfaces while
fronthaul interfaces is of paramount importance. In this paper supporting multi-tenancy, through a cohesive set of data-plane
we present 5G-XHaul, a novel transport network architecture technologies and a common control and management plane
featuring wireless and optical technologies and a multi-technology
software defined control plane, which is able to jointly support that minimizes operational costs.
backhaul and fronthaul services. We have deployed and validated This paper describes and evaluates a novel transport ar-
the 5G-XHaul architecture in a city-wide testbed in Bristol. chitecture referred to as 5G-XHaul. It features wireless and
Index Terms—Radio Access Networks, 5G, fronthaul, back- optical technologies and a control plane based on Software
haul, optical networks, SDN Defined Networking (SDN) that is able to jointly transport
fronthaul and backhaul services. Our contribution is comple-
mentary to Ericsson’s Transport Intelligent Function (TIFs)
I. I NTRODUCTION described in [4], whereby the 5G-XHaul network could be con-
To date, 4G networks have been deployed using two main sidered an SDN enabled underlay interacting with Ericsson’s
types of architectures: Distributed Radio Access Networks TIF. In addition, unlike China Mobile’s Slice Packet Network
(D-RANs), where a full base station stack is included in (SPN) [5], which uses an Ethernet-based transport, 5G-XHaul
each cell site; and Centralized RANs (C-RANs), where the advocates for a solution that provides flexible allocations
cell site only features the Remote Radio Heads (RRHs) and for a transport slice directly within the optical domain. To
the radiating elements, and the Baseband Units (BBUs) are the best of our knowledge, this is the first work where a
centralized in a remote location. The C-RAN architecture is joint backhaul/fronthaul transport network is experimentally
more energy efficient and augments network capacity through demonstrated in a realistic city-wide testbed.
inter-cell coordination, but imposes very strict requirements
on the transport network connecting the RRHs and the BBUs, II. THE 5G-XHAUL ARCHITECTURE
known as fronthaul. Instead, D-RAN only requires the trans- A. 5G-XHaul Data Plane architecture
port of IP packets between the cell site and the core network
We envision 5G RANs consisting of two different layers. A
through a transport network, commonly known as backhaul. In
first layer of macro-cells collocated with the already deployed
current deployments, fronthaul and backhaul are implemented
4G grid. 5G macro-cells are expected to operate with an
as entirely separate networks based on different technologies.
exemplary carrier bandwidth of 100 MHz in the 3.5 GHz
The C-RAN architecture is, in its current form, based on
band. This layer will be complemented with a dense layer
digitized radio samples, e.g. Common Public Radio Interface
of Small Cells deployed on lamp posts or street furniture,
(CPRI), which does not scale to 5G RANs. Hence, alternative
which may operate at millimeter wave (mmWave) frequencies,
RAN functional splits between the cell site and a centralized
thus providing additional area capacity. While various RAN
location have been proposed that trade-off centralization gains
functional splits may be considered for the macro-cell layer,
with reduced requirements to the transport network [1]. This
Small Cells will likely feature higher functional splits, thus
work has led to the eCPRI standard [2], supporting a variety
relaxing the requirements on the transport.
of functional splits. 3GPP has also embraced a 5G RAN
The 5G-XHaul data plane architecture is depicted in Figure
architecture able to support multiple functional splits, where a
1. A wireless transport segment, potentially including multiple
base station is split into a Centralized Unit (CU), a Distributed
hops, connects the Small Cells to the wired network. Macro-
Unit (DU), and a Remote Unit (RU) [3].
cells featuring Massive MIMO antenna arrays are connected
D. Camps-Mur is with i2CAT Foundation, Barcelona, Spain, e-mail: to an optical transport segment comprising two different
[email protected]). technologies: i) a passive high capacity WDM-PON network
2

mmWave FH/BH (xH) link

Sub-6 BH link

Remote Node
storage
mmW xH compute OLT
VNF
mmW xH mmW xH TSON Edge
mmW xH ONU Edge Cloud
Macro TSON Core
Cell Site
Small Cells mmW xH mmWave xH Node
Sub-6 Ac/BH Sub-6 Access/BH Node

Sub-6 Ac/BH 5GX-CO 5GX-CO

Sub-6 Ac/BH Sub-6 Ac/BH Sub-6 Ac/BH

Macro ONU
Cell Site ONU Small
Cells TSON
ONU
WDM-PON
Data Centers
BBU
vBBU (vBBU/BBU Pool)
Core Network
Wireless Access VNF
VM P-S GW/GSN
EPC
5GX-CO
Internet
PDN-GW

Fig. 1: 5G-XHaul data plane architecture

in the access, connecting the macro-cells to the central offices nologies. Thus, in 5G-XHaul, high capacity 60 GHz links are
(5GX-CO in Figure 1); and ii) an active Time Shared Optical combined with lower capacity but Non-Line of Sight (NLoS)
Network (TSON) in the metro segment, connecting 5GX-COs capable Sub-6 radios to form a heterogeneous wireless mesh
together, and with the core network. The 5GX-CO is seen as a network.
virtualized environment where backhaul services and (virtual) 3) Optical Transport: The optical segment of the 5G-
BBUs are hosted. 5G-XHaul may be serving RAN networks XHaul data plane architecture involves WDM-PON in the
from different tenants featuring different functional splits, e.g. access segment and TSON in the metro segment.
a CPRI-based interface and a traditional IP-based backhaul The 5G-XHaul WDM-PON solution encompasses an Op-
interface. Thus, the 5GX-CO needs to be able to host the func- tical Networking Unit (ONU), typically located in a macro-
tions required to serve both types of interfaces. Nevertheless, cell site; and an Optical Line Termination (OLT) located at
the architecture also allows to concentrate backhaul services the 5GX-CO. Up to forty different ONUs can be multiplexed
and BBUs in only a subset of the 5GX-COs, achieving then over a single trunk fiber by means of a passive dense wave-
higher centralization gains [6]. To enable centralization, both length division multiplexing (DWDM) filter in the field. Each
the optical access and metro segments support the transmission wavelength channel is capable of a symmetric data rate of
of joint backhaul (Ethernet) and fronthaul (CPRI) connectivity 10 Gbps or 25 Gbps, depending on the reach and the optical
services. The interested reader is referred to [7] for a blueprint modulation format used. An additional feature of this solution
deployment of the 5G-XHaul architecture in a typical Euro- is the use of an inexpensive vertical-cavity surface-emitting
pean city. (VCSEL) tunable laser in the ONU. The VCSEL is controlled
The 5G-XHaul transport network considers the following by the OLT to autonomously tune to the correct wavelength,
transmission technologies: using an out-of-band communication channel between ONU
1) Scalable Massive MIMO: Large Massive MIMO arrays and OLT. This solution has been standardized in the newly-
used by 5G macro-cells pose significant challenges in a CPRI- consented ITU-T G.698.4 standard [9], lowering significantly
based C-RAN architecture. 5G-XHaul features a novel RAN the operational costs associated to WDM-PON. WDM-PON
functional split where antenna processing is offloaded to the offers to 5G-XHaul a transparent interface that can deliver
Massive MIMO array [8]. This allows the transport network both Ethernet (backhaul) and CPRI (fronthaul) services on
to scale with the number of spatial streams, rather than with different wavelengths, using a transponder that receives the
the number of antenna elements as in traditional CPRI. Ethernet/CPRI streams in colorless grey wavelength channels
2) Wireless Transport: The wireless transport segment in and converts them to DWDM wavelengths.
Figure 1 connects the Small Cells to the optical segment, and TSON is an active optical technology that provides sub-
is dimensioned to support Ethernet backhaul services. This wavelength granularity [10], supporting natively the trans-
segment consists of two types of technologies: i) unlicensed port of both Ethernet and CPRI services forwarded through
mmWave technologies operating at V-Band (60 GHz), and ii) different wavelengths. TSON features TSON edge nodes,
technologies below 6 GHz (Sub-6), operating in the unlicensed which collect/deliver Ethernet or CPRI traffic through a set of
5 GHz band. We expect these devices to be based on IEEE input/output optical interfaces; and TSON core nodes, which
802.11ad and IEEE 802.11ac/ax radios respectively, benefiting forward optical bursts multiplexed in time on each wavelength.
from the economies of scale associated to IEEE 802.11 tech- Packets from each input interface are received at the TSON
3

5G-XHaul MAC in MAC encapsulation

ETH.SA=EN ETH.SA
NBI Path-ID L2SID PAYLOAD
ETH.DA=ETN ETH.DA
L2-Ctrlr
L2-Local Ifc
COP
COP
COP
L1-Ctrlr Top-Local Ifc COP
COP
COP
Area-5
L0-Ctrlr
Area-3
Area-2
Local Ctrlr
Local Ctrlr
Area-4
OF/
NETCONF
Area-1

Edge Transport Node (ETN)

Transport Node (TN)
Inter-Area Transport Node (IATN)

VNF from tenant 1 (slice 1)

VNF from tenant 2 (slice 2)

Fig. 2: 5G-XHaul control plane architecture

edge, and are aggregated into the different slots forming the mains, where an active processing of the Ethernet header
TSON frame. The reverse process is implemented at the is performed in each hop. Native CPRI is supported only
receiver edge. TSON is currently prototyped on an FPGA in the optical domain, being the CPRI interfaces mapped
platform providing programmability, not only in terms of the transparently between TSON and WDM-PON.
various parameters that define the TSON frame, but also in the
logic mapping input ports to slots in the TSON frame. CPRI The 5G-XHaul control plane architecture is illustrated in
is natively supported in TSON edge nodes through a Xilinx Figure 2. Two main components can be distinguished: i) a
GTH Transceiver IP core supporting CPRI protocol option 5. data plane abstraction composed of three different transport
functions, and ii) a hierarchy of SDN controllers.
The three transport functions are known as the Transport
B. 5G-XHaul Control Plane architecture
Nodes (TNs), the Edge Transport Nodes (ETNs), and the
5G-XHaul features a multi-technology control plane built Inter-Area Transport Nodes (IATNs). TNs are grouped in
on the following design principles: control plane areas, controlled by a common Level-0 (L0)
• SDN architecture. A logically centralized control plane SDN controller (green diamond in Figure 2). L0 controllers
is considered, offering higher level Application Program- proactively install in each TN a set of unidirectional label
ming Interfaces (APIs), enabling automation and easing switched paths, which provide connectivity between all ETNs
the integration with other software systems, e.g. an NFV and IATNs within the control plane area. ETNs implement
ETSI MANO system. the binding between the per-tenant P/VNFs connected to the
• Hierarchical Control. 5G-XHaul considers technology 5G-XHaul transport network, and the label switched paths
specific controllers for different transport network seg- available in each control plane area. Additionally, IATNs sit
ments. For example, a wireless specific controller can between areas and perform the required stitching between the
allocate paths while considering inter-link interference, corresponding label switched paths in each area. In 5G-XHaul,
whereas a TSON controller derives the TSON TDM user plane traffic will flow mostly towards the 5GX-COs,
schedule depending on the allocated paths. In addition, to whereas control traffic between nearby base stations may be
provide end-to-end connectivity services, a higher, tech- local to a control plane area. To isolate TNs from per-tenant
nology agnostic, control layer is considered that provides state, ETNs implement a MAC-in-MAC encapsulation (Figure
forwarding across technology segments. 2). The outer VLAN tag included in the encapsulated frame
• Scalable Virtualization and Multi-tenancy. Multiple ten- is used by the TNs as forwarding label (Path-ID), whereas the
ants may connect their physical or virtual network inner VLAN tag (Layer 2 Segment ID L2SID) in the customer
functions (PNFs/VNFs) through the 5G-XHaul transport Ethernet frame is used to disambiguate different slices at the
network. To allow scalability, a layer 2 based overlay receiving ETN. Thus, although 5G-XHaul leverages MAC in
mechanism is used to enable network virtualization. MAC encapsulation, it does not use MAC learning, but instead
• Two native forwarding abstractions: Ethernet and CPRI. it proactively establishes label switched paths. A slice in 5G-
Ethernet is supported across the wireless and optical do- XHaul is the virtual layer 2 service connecting distributed
4

functions of a given tenant. This slice can be isolated either 120 and 220 meters. The wireless segment terminates at
logically and/or in terms of performance via: different prior- the WeTheCurious science museum, where dark fiber was
ities in the wireless segment, wavelengths in WDM-PON, or made available connecting to the Bristol University building,
wavelength and TDM slots in TSON. The interested reader where the optical segment of the 5G-XHaul architecture is
is referred to [11] for a detailed description of the 5G-XHaul deployed. The optical segment consists of: i) WDM-PON
approach to network virtualization. ONUs connecting to WeTheCurious, ii) the corresponding
Figure 2 depicts a first layer of technology specific (wire- WDM-PON OLT, and iii) two back-to-back TSON nodes.
less and optical) SDN controllers (L0 controllers) that use Fiber spools were used between the TSON nodes and between
OpenFlow and NETCONF to control and manage the network the OLT and the ONUs to emulate alternative deployment
devices. They interface with Level-1 (L1) controllers by means distances.
of the Control Orchestration Protocol (COP) [12], where The previous infrastructure was controlled using two L0
the service-call model has been extended to accommodate SDN controllers, one per segment (wireless and optical). L1
unidirectional VLAN-based label switched paths, and the and L2 controllers were used to coordinate the two domains,
service-topology model has been extended to accommodate the being hosted on Virtual Machines (VMs) instantiated on an
5G-XHaul ETN and IATN functions [13]. The L1 controller OpenStack cluster (not shown in Figure 3). The control plane
receives an end-to-end connectivity service request to connect provisions the label switched paths that support backhaul
two ETNs from the Level 2 (L2) controller, again through a services.
COP-based interface, and decides how to instantiate an end- Our goal was to benchmark the performance of the 5G-
to-end connectivity service spanning multiple control plane XHaul architecture and to demonstrate the joint provision of
areas. The L2 Controller programs ETNs and IATNs via the backhaul and fronthaul services in a multi-tenant fashion. For
L2-Local interface. the former we instantiate two backhaul slices, each consisting
Finally, it is worth noting that the aforementioned control of: i) a virtual Wi-Fi AP instantiated over the physical AP,
functions (L0, L1, and L2 controllers) can be virtualized and ii) two unidirectional end-to-end label switched paths, and iii)
deployed as VNFs using a MANO service platform. a VM delivering an HD video service. To provide isolation,
each backhaul slice is forwarded through a different path in
III. EXPERIMENTAL EVALUATION OF THE the wireless mesh (see section III.B for detailed information).
5G-XHAUL ARCHITECTURE To provision a fronthaul service, a Massive MIMO antenna
The 5G-XHaul architecture was evaluated in a city-wide array and a BBU were also deployed at the Bristol University
testbed deployed in Bristol, UK, whose physical topology is site. The Massive MIMO antenna array was connected to a
illustrated in Figure 3. The deployed infrastructure features WDM-PON ONU, the BBU to a TSON node, and CPRI was
all novel 5G-XHaul wireless (Sub-6 and 60 GHz) and optical used to fronthaul the time domain radio samples between the
(WDM-PON and TSON) technologies. BBU and the antenna array.
The topology consists of a high-speed Wi-Fi Access Point
(AP) backhauled through a multi-hop hybrid wireless mesh A. Benchmarking 5G-XHaul wireless and optical segments
network spanning several sites in the Bristol waterfront. The
multi-hop network comprises four 60 GHz links, operating Figure 4a depicts the latency and throughput performance
in LoS conditions, and one Sub-6 link operating in NLoS of the wireless segment and of the end-to-end backhaul slices
conditions. Link distances spanned approximately between respectively. The upper subplot of Figure 4a shows all wireless
links introducing a worst case round trip delay below 10
milliseconds. The end-to-end round trip delay experienced by
,WE each slice is below 15 milliseconds for 80 % of the packets, but
>ĂďŽƌĂƚŽƌLJ
shows a long-tail behavior of almost 100 milliseconds. This
is due to the varying performance delivered by the Wi-Fi AP
used in the access network, which is subject to interference.
The lower subplot in Figure 4a shows how the 60 GHz
links provide a stable performance between 600 Mbps and
800 Mbps, which is determined by the Modulation and
Coding Scheme (MCS) allowed by the existing propagation
conditions. The Sub-6 link, featuring an IEEE 802.11ac radio
tĞdŚĞƵƌŝŽƵƐ
operating with a bandwidth of 80 MHz, delivers a throughput
^ƵďͲϲ
between 100 Mbps and 250 Mbps, with variations again
ϲϬ',nj introduced by the use of a dynamic MCS. The end-to-end
^EƐǁŝƚĐŚ
^ƵďͲϲŶŽĚĞ performance of each backhaul slice is, however, significantly
tŝ&ŝW ŵŵtĂǀĞŶŽĚĞ
;Ɛϭ͕ƐϮͿ ^EŵŵtĂǀĞŶŽĚĞ smaller (20 Mbps), and is limited by the wireless access
ƌŝƐƚŽůĨŝďĞƌƌŝŶŐ technology in use (IEEE 802.11g Wi-Fi).
To assess the performance of the optical segment, we first
Fig. 3: 5G-XHaul BiO testbed overview. Wireless SDN and validate the error-free provision of Ethernet services up to 10
Optical Network at HPN laboratory. Gbps and CPRI line rate 5 (4.9 Gbps), using the integrated
5

100

1
80

Round Trip Latency (us)

F(x<X)
60
e2e slice 1
0.5 e2e slice 2
40
mmWave
Sub-6
20

0
0
0 20 40 60 80 100 CPRI ETH-1 ETH-2

Round-trip Latency (ms)

1 (a) CPRI and Ethernet services (b) 16 QAM demodulated sig-

e2e slice 1
over 8 km fiber nal over GFDM
e2e slice 2
F(x<X)

0.5
mmWave Fig. 5: Evaluation of joint fronthaul and backhaul services.
Sub-6

0 ns. Notice that fibre introduces a propagation delay of 5 µs

0 100 200 300 400 500 600 700 800 900
Throughput (Mbps)
per km. In the case of CPRI the following round trip delays
are measured: i) 0.99 µs in the B2B case, ii) 84.19 µs for
(a) Benchmark of the 5G-XHaul architecture (La- the 8 km fiber, and iii) 166.86 µs for the 16 km fiber. These
tency/Throughput). results validate that the 5G-XHaul optical segment is able to
12 cope with 4G and 5G fronthaul requirements.
10 Before break
Throughput (Mbps)

After break
8 B. Evaluation of SDN features for Backhaul services
6 To evaluate the impact of the SDN control plane on the
4
reliability provided to backhaul services, each backhaul slice
2
runs through a different path in the wireless segment. Figure 3
depicts how slice 1 traverses first a 60 GHz hop, then a Sub-
0
0 10 20 30 40 50 60 70 6 hop, and two more 60 GHz hops before connecting to the
Time (seconds)
optical segment; whereas slice 2 traverses two 60 GHz hops
(b) Evaluation of SDN backhaul services and connects to the optical segment (cf. labels s1 and s2 in
Fig. 4: Benchmarking and evaluation of the wireless part. Figure 3). For each slice we run an H.264 HD video between
the VM and a tablet device attached to the slice’s virtual AP
through the optical and wireless segments. Each HD video
TSON and WDM-PON technologies. To deliver the required requires approximately 3-4 Mbps to run smoothly.
services, one of the TSON edge nodes features three SFP+ To provide reliability in the wireless backhaul, we use an
interfaces connected to the WDM-PON OLT. Two OLT ports SDN feature called Fast Local Link Reroute (FLRR) [14],
deliver the Ethernet traffic flowing through each backhaul end- which involves: i) the L0 controller proactively installing a
to-end slice, and the other port delivers the CPRI samples to main and a backup label switched path for each slice, and ii)
the Massive MIMO antenna array. The other TSON edge node having a fast recovery agent installed in the wireless devices
features two SFP+ interfaces, one connected to an Ethernet that detects the link failure and diverts affected packets towards
switch that serves backhaul traffic, and another one connected the backup path. The backup label switched path for slice 1 is
to the BBU. configured through the same links used by slice 2. To evaluate
The next performance measure is the round trip delay this feature, we start playing the HD videos, and then break
through the 5G-XHaul optical segment, for both the Ethernet the Sub-6 link traversed by slice 1.
and CPRI services. The overall delay includes TSON and Figure 4b depicts a traffic trace captured while streaming
WDM-PON propagation delays, as well as the delay stemming the HD video for slice 1, including the break of the Sub-6
from the various fiber deployment lengths. For WDM-PON, link. The trace shows the video traffic going through the Sub-
the delay is introduced by the OLT and ONU transponder 6 link before the break takes place (in blue), and the same
cards and by the WDM filters. For TSON, being an active video traffic flowing through one of the backup 60 GHz links
technology, processing delays are introduced by a PHY IP core after the break (in orange). The video traffic is streamed using
in the case of CPRI, and by the PHY+MAC IP core, plus the TCP, and the Sub-6 link is broken at around second 43. We
additional processing required to analyze label switched paths, observe at the beginning of the trace and right after the break
in the case of Ethernet. that TCP enters Slow Start, but traffic keeps flowing at all
In the case of Ethernet we observe: i) A round trip of times. Indeed, no glitch was appreciated in the HD video of
3.241 µs for the Back-to-Back (B2B) scenario, ii) 86.144 µs slice 1 while rerouting the traffic. The HD video for slice 2
for the 8 km fiber, and iii) 169.175 µs for the 16 km fiber. also played flawlessly throughout the experiment.
To understand the contribution of the WDM-PON and TSON
segments to the round trip latency, we measure for the B2B C. Evaluation of joint Fronthaul and Backhaul services
case a TSON PHY+MAC IP Core delay of 2.974 µs, a TSON We now analyze the ability of the 5G-XHaul optical seg-
processing delay of 0.167 µs, and a WDM-PON delay of 100 ment to jointly deliver backhaul and fronthaul services. To
6

operate at the full capacity provided by the optical segment, we 1

loopback the fibers between WeTheCurious and the laboratory 0.8

L1 service-call
Top service-call
at Bristol University, thus disconnecting the wireless segment.

F(x < X)
0.6
An 8 km fiber spool is used in this experiment introducing a
base delay of 80 µs. We evaluate Ethernet services using an 0.4

Ethernet analyzer that generates two 4.9 Gbps streams with 0.2

the VLAN tag corresponding to the label switched paths for 0

150 200 250 300 350 400
slice 1 and slice 2. Concurrently to the Ethernet traffic, we delay (ms)
generate a CPRI line rate 5 stream (4.9 Gbps) between the
BBU and the Massive MIMO antenna array. Fig. 6: Control plane latency to set-up multi-domain connec-
We verify in the Ethernet and CPRI analyzers that the tivity service.
correct 4.9 Gbps are delivered for each Ethernet slice and for
the CPRI service while operating concurrently. Additionally,
in the overall connection provisioning time, the dashed blue
Figure 5a depicts the round trip delay experienced by each
line in Figure 6 represents the CDF of the time between the
service. We observe that the CPRI service and the two Ethernet
moment the L2 controller issues the service-call request to the
slices experience respectively a round trip delay of 85 µs,
L1 controller (step i.), until the time when the label switched
92 µs, and 87 µs. The processing introduced by TSON
paths in each domain are provisioned and the L2 controller
explains the slightly higher delay in the case of Ethernet.
receives the response from the L1 controller (step iv.). The
Figure 5b depicts a 16-QAM constellation diagram decoded by
measured provisioning time as a result of these two steps
a software defined radio (SDR) tapping into one of the antenna
is 250 ms. These service provisioning times are orders of
elements of the Massive MIMO array. Generalized Frequency
magnitude smaller than the times required in current networks
Division Multiplexing (GFDM) [15] was used to modulate
to set up multi-domain connectivity services, and can be
the RF signal between the BBU and the SDR. The correct
considered a stepping stone towards fully automating control
reception of the 16-QAM constellation, while concurrently
and management functions in future transport networks.
transporting error-free Ethernet traffic, validates the ability of
the 5G-XHaul architecture to jointly transport backhaul and
IV. C ONCLUSIONS
fronthaul services.
The upcoming deployment of 5G RANs leads to an in-
creased variety of the transport interfaces required to connect
D. Evaluation of 5G-XHaul hierarchical Control Plane base stations to the mobile core network. It is therefore of key
We conclude our evaluation studying the time required by importance to design transport networks able to serve all these
the 5G-XHaul hierarchical control plane to instantiate an end- interfaces in a cohesive way. In this paper we presented 5G-
to-end connectivity service. The following process is required: XHaul, a novel wireless-optical transport network architecture
i. The L1 controller receives a COP service-call request that supports concurrent backhaul and fronthaul services,
through its northbound interface issued by the L2 con- under a unified SDN control plane. We have used a city-
troller. wide testbed in Bristol to characterize the performance of the
ii. The L1 controller derives the L0 controllers that need to wireless and optical components of the 5G-XHaul architecture,
be involved in this particular service-call, and issues the and to demonstrate the ability to concurrently support backhaul
corresponding COP service-calls. and fronthaul services. We claim that the features delivered by
iii. L0 controllers return the details of the (pre-)provisioned 5G-XHaul are an important step to support the deployment of
label switched paths to the L1 controller in their COP next generation mobile networks.
response.
iv. The L1 controller builds the COP response to the L2 con- V. ACKNOWLEDGEMENTS
troller, including the information of the IATN functions The research leading to these results has received funding
involved in the end-to-end service. from the European Unions Horizon 2020 research and inno-
v. The L2 controller contacts the involved ETN and IATNs vation programme under grant agreement No 671551 (5G-
to install the necessary bindings to perform the MAC- XHaul). The authors thank Bristol is Open (BiO) for their
in-MAC encapsulation in the ETNs, and the stitching of support in the experimentation.
label switched paths between domains in the IATNs.
To increase statistical confidence we instantiate 100 dif- R EFERENCES
ferent connections and measure the Cumulative Distribution [1] J. Bartelt, N. Vucic, D. Camps-Mur, E. Garcia-Villegas, I. Demirkol,
Function (CDF) of the aggregated results. The solid red line A. Fehske, M. Grieger, A. Tzanakaki, J. Gutiérrez, E. Grass,
in the right part of Figure 6 represents the overall connection G. Lyberopoulos, and G. Fettweis, “5g transport network requirements
for the next generation fronthaul interface,” EURASIP Journal on
provisioning time when the 5G-XHaul L2 controller estab- Wireless Communications and Networking, vol. 2017, no. 1, p. 89, May
lishes a connection between two ETN functions located in 2017. [Online]. Available: https://doi.org/10.1186/s13638-017-0874-7
two different control plane domains. We can see that in the [2] “Common Public Radio Interface.” [Online]. Available:
http://www.cpri.info/
worst case, the overall service provisioning time is well below [3] “3GPP TS 23.501, Technical Specification Group Services and Systems
500 ms. To understand the contribution of the L1 controller Aspects; System Architecture for the 5G system,” Stage 2, March, 2018.
7

[4] S. Dahlfort, A. De Gregorio, G. Fiaschi, S. Khan, J. Rosenberg, and Dimitris Giatsios is with the Network Implementation Testbed Laboratory
T. Thyni, “Enabling intelligent transport in 5g networks,” Ericsson since 2009. Since 2012, he is pursuing a PhD in University of Thessaly. He
Technology Review, March 2018. has participated in several research projects, including OpenLab, FLEX and
[5] “Proposal of architecture of slicing packet network (spn) for 5g trans- 5G-XHaul.
port,” ITU-T SG 15 Contribution 678, January, 2018.
[6] A. Tzanakaki, M. Anastasopoulos, I. Berberana, D. Syrivelis, P. Flegkas,
T. Korakis, D. C. Mur, I. Demirkol, J. Gutiérrez, E. Grass, Q. Wei, E. Pa-
teromichelakis, N. Vucic, A. Fehske, M. Grieger, M. Eiselt, J. Bartelt, Arash Farhadi Beldachi is FPGA team leader and a Senior Research
G. Fettweis, G. Lyberopoulos, E. Theodoropoulou, and D. Simeonidou, Associate in High-speed FPGA and embedded systems design at High
“Wireless-optical network convergence: Enabling the 5g architecture Performance Networks Group, University of Bristol. He holds a PhD in
to support operational and end-user services,” IEEE Communications Dynamically reconfigurable network-on-chip from the University of Bristol.
Magazine, vol. 55, no. 10, pp. 184–192, OCTOBER 2017.
[7] I. Demirkol, D. Camps-Mur, J. Bartelt, and J. Zou, “5g transport
network blueprint and dimensioning for a dense urban scenario,” in 2017
European Conference on Networks and Communications (EuCNC), June Thierno Diallo got his master degree in high frequencies communication
2017, pp. 1–6. systems in 2012 at the Paris-Est University. He worked in Orange Labs in
[8] “5G-XHaul Deliverable D5.2, Evaluation of wireless-optical converged Lannion where it got his PhD in 2016. He joined High Performance Network
functionalities at UNIVBRIS testbed,” July,, 2017. [Online]. Available: group of university of Bristol in 2017 where he is working on 5G technologies.
https://www.5g-xhaul-project.eu/
[9] Multichannel bi-directional DWDM applications with port agnostic
single-channel optical interfaces, ITU Std. G698.4, 2018.
Jim Zou is a senior engineer in the Advanced Technology department at
[10] G. S. Zervas, J. Triay, N. Amaya, Y. Qin, C. Cervelló-Pastor, and
ADVA Optical Networking SE. He has been participating in various EU FP7
D. Simeonidou, “Time shared optical network (tson): A novel metro
and Horizon 2020 research.He received the PhD degree from Electro-Optical
architecture for flexible multi-granular services,” in 2011 37th European
Communication group, Eindhoven University of Technology, The Netherlands,
Conference and Exhibition on Optical Communication, Sept 2011, pp.
in 2015.
1–3.
[11] D. Giatsios, K. Choumas, P. Flegkas, T. Korakis, J. Aleixendri, and
D. Camps-Mur, “Design and evaluation of a hierarchical SDN control
plane for 5G transport networks,” in 2019 IEEE International Confer- Peter Legg works in the CTO office as a System Architect for SME Blu
ence on Communications (ICC), June 2019, pp. 1–6. Wireless Technology based in Bristol, UK. After a MA in Physics and a PhD
[12] R. Muñoz, A. Mayoral, R. Vilalta, R. Casellas, R. Martı́nez, and in Electronic Engineering, he has worked in mobile communications R&D at
V. López, “The need for a transport api in 5g networks: The control Motorola, IPWireless and Huawei.
orchestration protocol,” in 2016 Optical Fiber Communications Confer-
ence and Exhibition (OFC), March 2016, pp. 1–3.
[13] “5G-XHaul Deliverable D3.3, 5G-xhaul algorithms and services Design
and Evaluation,” July,, 2018. [Online]. Available: https://www.5g-xhaul- Jens Bartelt received his M.S. and PhD degrees in Electrical Engineering
project.eu/ from Technische Universitt Dresden, Germany, in 2012 and 2017, respectively.
[14] “5G-XHaul Deliverable D3.2, Design and evaluation of scalable control From 2013 to 2017, he was a Resource Associate at the Vodafone Chair
plane, and of mobility aware capabilities and spatio-temporal demand Mobile Communication Systems at TU Dresden. Since 2018, he is a System
prediction models,” July,, 2017. [Online]. Available: https://www.5g- Architect at Airrays GmbH.
xhaul-project.eu/
[15] N. Michailow, M. Matth, I. S. Gaspar, A. N. Caldevilla, L. L. Mendes,
A. Festag, and G. Fettweis, “Generalized frequency division multi-
plexing for 5th generation cellular networks,” IEEE Transactions on Jay Kant Chaudhary received his M.Sc. in Electrical Engineering from
Communications, vol. 62, no. 9, pp. 3045–3061, Sept 2014. Tampere University of Technology (TUT), Finland in Dec. 2014. In Nov
2015, he joined the Vodafone Chair Mobile Communications Systems at TU
Dresden, Germany.

August Betzler is a research engineer at i2CAT in Barcelona, Spain. He

Daniel Camps-Mur leads the Mobile and Wireless Internet group at I2CAT contributes to the standardization of new communication protocols for IoT as
in Barcelona. Previously, he was a senior researcher at NEC Network a member of the IETF. In 2010 he received his Diplom degree in computer
Laboratories in Heidelberg, Germany. Daniel was the technical coordinator science from the Technical University of Hamburg and in 2015 his Ph.D. from
of the H2020 5G-XHaul project. the Polytechnic University of Catalonia.

Jess Gutiérrez received the B.S. degree and the Ph.D. in Telecommunication
Engineering from the University of Cantabria, Spain, in 2008 and 2013,
respectively. Since 2013, he is with IHP. Joan Josep Aleixendri is a research engineer at i2CAT in Barcelona, Spain.
In 2016 he received his Bachelor degree in computer science from the
Polytechnic University of Catalonia.
Eckhard Grass is Team Leader of the Wireless Broadband Communications
Group at IHP, and Professor at the Department of Computer Science at
Humboldt-University Berlin. Previously he was Senior Lecturer in Micro-
electronics at the University of Westminster, London, U.K. Ricardo González is a Software Engineer at i2CAT in Barcelona. His primary
interests include Software Design and Architecture.
Anna Tzanakaki is a Research Fellow at the University of Bristol, UK, and
an Assistant Professor at the University of Athens, Greece. Dr. Tzanakaki also
serves as an Associate Editor of the JOCN (IEEE/OSA) and a TPC member
of several international conferences. Dimitra Simeonidou is a Full Professor of High Performance Networks at the
University of Bristol and the Director of the Smart Internet Lab. Dimitra is in
the editorial team of leading Journals and she chairs committees, conferences,
Paris Flegkas received a Diploma in electrical and computer engineering from standardisation groups and fora in the relevant bodies. She is the author
Aristotle University, Greece, and an M.Sc. and a Ph.D. from the University of and coauthor of over 400 papers in peer reviewed journals and international
Surrey, UK. He is an adjunct lecturer and a senior researcher at the University conferences, book chapters, several standardisation documents and patents.
of Thessaly.

Kostas Choumas received the Diploma, M.S. and Ph.D. degrees in electrical
and computer engineering from the University of Thessaly, Volos, Greece, in
2007, 2008 and 2015 respectively. From 2015 until now, he is Postdoctoral
Associate with the University of Thessaly.