Centralised versus Distributed Radio Access

Networks: Wireless integration into Long Reach

Passive Optical Networks

Christian Raack Julio Montalvo Garcia Roland Wessäly

atesio GmbH Telefónica I+D atesio GmbH
Berlin, Germany Madrid, Spain Berlin, Germany
[email protected] [email protected] [email protected]

Abstract—In this paper we evaluate the cost of wireless inte-

gration into an architecture based on long reach passive optical
networks (LR-PON). We will prove that including backhaul
or fronthaul signals from antenna sites into existing LR-PON
structures deployed for wired customers requires only marginal
additional resources. We will also show that scenarios with
remote radio heads and centralised base band units outperform
distributed radio access networks mainly because of operational
cost savings.

I. L ONG R EACH PON S AND M ETRO C ORE N ODES

The FP7 project D ISCUS is investigating an optimised end-
to-end optical network architecture considering Long-Reach Fig. 1. The D ISCUS LR-PON architecture: A multiple splitting hierarchy
(LR) passive optical networks (PONs) and a flat optical core between customer and LE and dual-homing between LE and MC. The local
[1], [2]. The LR-PONs technology approach in D ISCUS con- exchange houses the first stage 4x4 split and also acts as an amplifier node.
sists of a hybrid Time and Wavelength Division Multiplexing The MC node acts as the access optical line termination as well as a core
router with L1/2/3 switching and routing functionality.
(WDM) of symmetrical 10 Gbps optical channels to commu-
nicate the remote devices in a PON (ONUs) with the central
office equipment (OLTs). While the most advanced PON including backhauling or fronthauling signals stemming from
standard is currently designed to offer up to 8 of these 10 Gbps antenna sites into the PON structures originally deployed for
symmetrical channels [3], D ISCUS LR-PONs are designed to wired customers requires only marginal additional resources.
support up to 40 of these channels [4]. Moreover, D ISCUS is We will also show that scenarios with remote radio heads and
investigating and developing the technologies required in order centralised base band units outperform distributed radio access
to support simultaneously a 100km or even 125km reach and networks because of operational cost savings.
512 split ratio in LR-PONs. With these values, D ISCUS allows
a very high consolidation of PONs into a minimum number
II. C ENTRALISED VERSUS D ISTRIBUTED R ADIO ACCESS
of Points of Presence, where the access, the metro/aggregation
N ETWORKS
and photonic transmission layers are comprised in the same
network node, namely an MC (metro-core) node. The centralisation of the base-band processing of radio
networks appeared as an option to traditional radio access
In this respect, D ISCUS revisits the classical divide between
networks (RAN) because it may reduce capital and operational
access, metro, and core networks for a new future proof end-to-
cost, as well as facilitate the implementation of advanced radio
end network architecture for next-generation communication
features such as Coordinated Multi-Point (CoMP) transmission
networks. Central to the DISCUS LR-PON is a dual-homing
and reception. In centralised radio access networks (C RAN),
architecture, see Fig. 1. The optical signal is supposed to
the antenna sites are simplified with Remote Radio Heads
pass a three-stage splitting hierarchy from the MC node to
(RRH) with no base-band processing, in opposite to the
the customer. The first stage splitter already located at a local
distributed RAN (D RAN). This allows C RAN to achieve a more
exchange site, is used to provide a redundant connection to a
easy installation of small RRH, also allowing to reduce the
second MC node. In consequence, every LR-PON by-passes an
overall base-band processing resources due to the statistical
LE site and is connected with two different MC nodes. Clearly,
gain of centralisation. Nevertheless, it is well known that the
due to the quadratic scaling of connections in the full mesh of
demand of transport resources to deliver the signals from the
the optical island a small number of MC-nodes is indispensable
RRH to the Base-Band Unit (BBU), where the fronthauling
for this architecture to be cost-effective and scalable.
signals are processed, increase drastically. This may reduce
In this paper, we study the impact of wireless integration the opportunities for C RAN to be a cost-effective scalable
into an architecture based on LR-PONs. We will prove that solution when considering 4G and 5G scenarios with ultra-
high speed capacity (1-10 Gbps) offered to the mobile user
equipment in the radio layer. Due to the high split ratio
and scalable capacity of D ISCUS LR-PONs, the support of
D RAN services by the fixed fibre network seems a feasible
and cost-effective approach for fixed-wireless convergence.
Nevertheless, D ISCUS LR-PONs open also a new possibility
for a cost-effective fixed-wireless convergence with C RAN,
because a high number of high speed common public radio
interface (CPRI) signals can be transported into the same LR-
PON using dedicated wavelengths in a Point to Point fashion,
instead of requiring dedicated fibre links. We refer to Fig. 2 (a) (b)
for integrating the D RAN backhaul and C RAN fronthaul into
the LR-PON D ISCUS architecture.

(c) (d)

Fig. 3. (a) The Spanish network (b) The UK network, (c) The Italian network
(d) Wales: A clipping from the UK network.

a correct notion of distance. They reflect dense and sparse

structures, towns and rural areas, mountains, valleys, and river
crossings.
We constructed fibre reference networks for each of the
three countries, also see [8]. The dimension of the individual
Fig. 2. The D ISCUS LR-PON architecture with integrated antennas. a) D RAN networks is given in Table I, also see Fig. 3.
integration, 32 sites may share the same wavelength. b) C RAN integration, each
sector requires its own wavelength in the WDM PON. TABLE I. R EFERENCE NETWORKS

Instance nodes links LEs households

III. R EFERENCE N ETWORKS AND C ORE NODE Spain 18,819 26,479 8,272 20.6 Mio
UK 15,609 23,025 5,578 29.4 Mio
DISTRIBUTIONS Italy 23,689 32,700 10,620 24,9 Mio
In order to provide realistic studies of wireless integration
into LR-PONs including fibre routing and concrete positioning Based on these networks, it has been investigated how
of antenna sites and local exchanges we use reference networks many aggregation MC nodes are required within these coun-
reflecting nation-wide fibre topologies. This approach allows to tries to allow for consolidating all current local exchanges
properly incorporate the influence of topological connectivity (LEs) within a given distance and also with different assump-
(for resilience issues) and technological restrictions such as tions on the resiliency level and maximum MC node size,
maximal distances. For this study we developed reference see [6], [5]. Potential MC nodes are all LEs with a certain
networks for the UK, for Italy, and for Spain using data level of connectivity. Table II reports on solutions we obtained
from B RITHISH T ELECOM, T ELECOM I TALIA, and T ELEFON - assigning local exchanges to MC nodes, also see Fig. 4. In all
ICA . The provided data sets for the UK and Italy included these solutions each customer gets connected to two different
anonymized (shifted) coordinates of local exchanges (central MC nodes while the feeder fibres take two disjoint routes
offices) together with the number of connected customers. For (dual-homing between LE and MC) in the cable network [1],
Spain, we used detailed population statistics and high-level see Fig. 1 and 2. We allowed a maximum customer-to-MC
characteristics of the local exchange distribution in Spain to distance of 125 km. Table II also states the maximum number
estimate their positions and customer assignment.
Already in [5] and [6] we showed how to combine this op- TABLE II. MC NODE DISTRIBUTIONS .
erator specific information with public available data, namely
Instance # MCs max MC Dist max MC Size
with data from open street maps [7], in order to come up with
realistic fibre topologies. Such geo-referenced data from street Spain 110 125 km 1.0 Mio
UK 73 125 km 1.0 Mio
networks is a reasonable choice in this case as laying fibres Italy 116 125 km 2.6 Mio
is typically done along streets and street networks provide
 From the mathematical perspective, given a certain cellular
layout, covering a country with cells can be seen as a tessella-
tion of the plane. Hexagonal layouts as shown in Fig. 5 have
been studied extensively, see for instance [9], [10], [11] and
the references therein. Such simplified hexagonal tessellations
and the resulting layout characteristics such as the positioning
of antennas to each other and the number of sector can be
used to study crucial aspects in cellular deployment such as
interference between neighboured cells.
Based on a given geo-type, our deployment of antenna sites
follows a 3-sector hexagonal clover-leaf layout, see Fig. 5
([10]). Each sector antenna spans a hexagon as shown in
Fig. 5 such that the antenna site sits on the edge of the three
hexagons. In the following, if we speak of a (three-sector)
Fig. 4. MC node distribution in Italy: MC nodes with more than 1 Mio
customers in red, those with less than 50 K in white.
macro cell, we refer to these three hexagons. The radius of a
macro-cell is given by the reach of the sector antennas, which
results in a site-to-site distance of 1.5 times the radius. In the
of (primary plus secondary) customers (max MC Size) at the resulting
√ grid of antenna sites two rows have a distance of
same MC. The Spanish (mainland) and UK solutions have 3 3/4 ∼ 1.3 times the macro cell radius.
been optimised to have at most 1 Mio customers at each Notice that we will use the terms macro cell, antenna site,
MC. In these solutions we did not respect regional boundaries and base station as synonyms throughout this paper, that is,
(except for Northern-Ireland). For the Italian solution we did the antenna site houses one base station and spans one macro
not use a maximum customer constraint but respected regional cell (of one technology).
boundaries (20 administrative regions), that is, an LE in Veneto
does not connect to an MC in Lombardia, etc.. TABLE III. C ELL MODEL BASED ON POPULATION DENSITY

Geo-Type Homes/km2 cell radius (m)

IV. A NTENNA SITE DISTRIBUTION
Dense Urban ≥ 4000 250
Distributing antennas to achieve full coverage within a Urban 2000 - 4000 500
Sub-Urban 500 - 2000 1000
country is clearly a non-trivial task. Locally, it depends on Rural 10 - 500 2000
cellular layout, antenna parameters such as antenna height, Sparse Rural 0 - 10 10000
azimuth direction, or beamwidth, as well as regional charac-
teristics such as mountains or buildings. Globally, it is further For deciding about the population density we consider
complicated by the fact that the reach of the antennas and the five geo-types: Sparse Rural, Rural, Sub-Urban, Urban, and
corresponding cell radii typically depend on the considered Dense Urban, which are defined by the number of households
geo-type. Dense urban areas contain many base stations with per square kilometre. Depending on the local geo-type we
small reach antennas (cell radii below 500m) to serve a large deploy antenna sites with the radius given by Table III. This
amount of customers within a small area, while rural areas automatically leads to a very dense distribution of sites in
are served with only a few antennas and cell radii of a few (dense) urban areas and a very sparse distribution of sites in
kilometres. (sparse) rural regions.
In order to study D RAN versus C RAN deployment and To deploy antenna sites within a given geo-type region we
cost we start from a nation-wide distribution of antenna sites use a heuristic that starts from a regular grid G as in Fig. 6(a)
implementing assumptions of population density and cell radii assuming a regular Rural tessellation with macro cells of radius
shown in Table III. In order to determine a nation-wide geo- 2 km. The heuristic then adds or removes sites if necessary,
type distribution we use the information about connected that is, if the geo-type actually differs. In fact, we make use of
customers at the LE coordinates, see below. the fact that most of the country area is actually Rural and that
the geo-type radii are multiples of each other. We further use
a description of the given country as a (typically non-convex
and simple) polygon R (or several polygons if necessary). In
this context it is important to be able to decide whether a given
coordinate point is contained in R or not, which can be done
very fast also for large polygons with many edges using the
famous crossing number algorithm [12].
The heuristic then consists of three essential steps:
1) Compute the actual geo-type geo(g) of each of the grid
points g ∈ G by counting the number of households
(connected to LEs) in a box around g. Let r(g) be the
Fig. 5. Tessellation with clover leaf layout: Sector and cell structure. The
site-to-site distance is 1.5 times the site radius. This radius corresponds to the macro cell radius from Table III corresponding to geo(g).
diameter of the shown hexagons, which in turn corresponds to the reach of 2) Iterate the grid points g ∈ G. If r(g) < 2km, that is,
the individual sector antennas. if geo(g) is either Sub-Urban, Urban, or Dense Urban,
 TABLE IV. M ACRO CELL DEPLOYMENT: G EO - TYPE AREAS AND
RESULTING MACRO CELL DEPLOYMENT.
the UK. Note that the number of sites in Table IV is larger
but in the same order of magnitude as those deployed today
Country Total S.Rural Rural S.Urban Urban D.Urban in the considered countries.
Spain Area 43.4% 53.5% 2.8% 0.2% 0.1%
Macro cells 34,689 5.1% 61.8% 20.5% 6.8% 5.8% V. PARAMETERS
Italy Area 11.1% 84.1% 4.4% 0.3% 0.1%
Macro cells 41,276 1.0% 71.0% 17.9% 4.9% 5.2% In order to perform a comparison between D RAN and
UK Area 23.4% 67.1% 7.7% 1.5% 0.3% C RAN using LR-PONs, we focus on a future possible RAN
Macro cells 44,309 1.2% 42.4% 24.1% 18.7% 13.7% scenario. This includes the offer of a peak access of 1 Gbps
at the radio layer per cell. For D RAN, a statistical aggregation
efficiency of 3.2 per antenna site is assumed, thus up to 32
antenna sites could share the same 10 Gbps channel in a single
LR-PON to perform the backhaul of the IP traffic. In order
to achieve the same throughput in the C RAN approach, the
required fronthauling capacity for each sector is around 25
Gbps, thus we assume a 1/3 compression ratio [13] in order
to transport the digital RoF (Radio over Fibre) signal from the
RRH to the BBU using a 10 Gbps Point to Point LR-PON
channel. Due to latency restrictions required to guarantee the
throughput performance, for C RAN we also consider a limited
maximum distance from the RRH to the BBU of 40km [14].
(a) (b)
Translating these assumptions to side-constraints for opti-
Fig. 6. Cell deployment in the Barcelona region. The colours correspond to mising the LR-PONs we obtain the following
the geo-types; white: Sparse Rural, yellow: Rural, light-orange: Sub-Urban,
orange: Urban, red: Dense Urban; (a) Initial grid G (b) Resulting antenna site • There is no constraint on the maximum fibre distance
deployment
between the antenna site and the metro core node for
D RAN. That is, we may use the full LR-PON distance
then a single antenna at g does not suffice to cover the of at most 125-km.
neighbourhood of g. If A(Rural) is the area covered by • There is a hard distance constraint for C RAN deployment.
a Rural antenna site and A(geo(g)) the area covered by The base band unit can only be centralised at the MC
a site of geo-type geo(g), then dA(Rural)/A(geo(g))e node if the (primary) fibre route from the antenna site to
gives the number of required antenna sites. We use the the MC node has at most 40 km.
clover-leaf layout (Fig. 5) to distribute these sites in the • At most 32 D RAN cells share the same colour in the same
considered area around g. For each new site we check PON, each sites requires one port. In contrast C RAN cells
whether it is contained in the countries polygon R. require 3 colours, one per sector.
3) Randomly chose one of the grid points g of type Sparse • The WDM LR-PON supports up to 40 colours.
Rural. We place a Sparse Rural site at g. This yields a
coverage area of A(SparseRural) larger than A(Rural).
VI. W IRELESS INTEGRATION
Remove all grid points and antenna sites in this area from
G and iterate. We will study the integration of wireless backhaul and
fronthaul signals starting from an existing deployment for
Notice that the tessellation with Sparse Rural antenna wired customers (households). We will of course ignore PON
sites in Step 3) does not necessarily follow a regular clover- specific detailed aspects such as man-holes, splicing or floating
leaf structure. Further, as mentioned above, we are not given of cables, micro-ducts, etc., but concentrate on the main
geographical coordinates of households but of local exchanges. structure of the LR-PONs in terms of cabling, splitting, and
That is, in Step 1) we can only estimate the geo-type of a grid port fill. We will also not consider all cost factors that are
point. Given a grid point we start with a box of 3 km width crucial when considering PON deployments but concentrate
having the grid point in the centre to estimate the geo-type. We on those resources that are effected or differ in a D RAN versus
then iteratively increase the radius to 10 km. We only change C RAN deployment. For instance, we will see that we can safely
the geo-type in this iteration if the change is towards a more ignore duct resources as the required fibre/cable resources do
urban geo-type (higher population density). not increase significantly enough neither in a D RAN nor in
the C RAN approach compared to the LR-PON deployment
Fig. 6(b) shows an exemplary deployment using our heuris-
based on wired clients only. In the following, if we speak
tic in the Barcelona area. Table IV provides detailed numbers
of customers, we refer to any wired client of the PON, so
of the cell deployment for the countries Spain, Italy, and UK.
residential households or business customers. This is opposed
It can be seen that most of the countries area is Sparse Rural
to the wireless clients, which are the antenna sites.
and Rural. Clearly, among the eventual macro cells there are
only few Sparse Rural because of the large radius. In contrast, No coordinates for residential households are available for
the few Dense Urban regions create a significant number of the reference countries. That is, we can only estimate the
cells because of the small cell radius. The deployment depends required LR-PON resources in the optical distribution network
strongly on the considered country. While Spain is relatively (ODN), which refers the PON segment below the LE towards
Rural there is a significant portion of (Dense) Urban cells in the clients. However, we are able to determine the (lower
 solutions we can reach more than 50% of the customers with a
centralised radio access architecture. The share of C RAN cells
increases with the population density. For (dense) urban areas
we achieve a coverage of typically more than 70% with values
up to over 90% depending on the MC node density. Notice that
the average MC node density is larger for Italy as we respect
regional boundaries in the 116 node solution.
Notice that we deploy D RAN in areas where C RAN
is not feasible due to longer distances providing a mixed
D RAN/C RAN deployment. Whenever stating a C RAN solution
we provide the percentage C RAN coverage in brackets.
Fig. 7. Trees among site location in order to evaluate cell-to-LE distances.
TABLE V. P ERCENTAGE OF CELLS AND CUSTOMERS THAT CAN BE
Orange dots indicate site locations (Sub-Urban and Urban). Blue dots are the
REACHED WITHIN 40 KM FROM THE MC FOR DIFFERENT MC- NODE
local exchanges.
DISTRIBUTIONS . T HESE SITES ARE C RAN FEASIBLE .

Reached Reached Sites

bound) of required PONs below a given local exchange as Country MCs Customers S.Rural Rural S.Urban Urban D.Urban
both the number of customers at the LE as well as the number UK 73 65.0% 23.0% 53.0% 68.0% 70.0% 70.0%
of antenna sites is given to us. We simply assume that each Spain 110 52.0% 26.0% 40.0% 60.0% 63.0% 64.0%
Italy 116 70.0% 27.0% 50.0% 75.0% 93.0% 98.0%
macro site is assigned to the closest LE. Similarly, the total
number of required ONUs and OLTs does not depend so much
on distances but on total numbers of households, antenna sites, a) Filling the PONs: When filling the PONs we ignore
and resulting PONs. exact positioning of both antennas and wired customers. In-
stead we use the fact that each PON has two capacities: at most
In contrast, cabling and splitting in the ODN depends
512 ports and at most 40 colours can be used simultaneously.
on the location of all PON clients. Nevertheless, with the
To cope with topological characteristic when filling all ports of
coordinates of antenna sites from the cellular deployment and
the same PON, we assume a filling factor of 80%, which means
with some additional assumptions on average distances for
that at most 410 ports can actually be used, c. f. Fig. 8(c).
households as well as assumptions on the splitter hierarchy
There are three types of clients that use resources of the PON
we will at least estimate fibre distances in the ODN in order
in a different way. Recall that each colour of the PON has
to approximate required fibre resources. For the feeder section
a capacity of 10 Gbps, which may be shared among the 410
(between LE and MC) we have relatively accurate numbers
ports: (i) A wired customer uses one port of the PON. At
as the cabling here only depends on the number of PONs
most 410 such customer clients may use the same colour. (ii)
bypassing the LE and the disjoint fibre routes towards the MC
A D RAN macro cell uses one port of the PON. At most 32
nodes, which we are given from the MC node distributions
such D RAN cell clients may use the same colour. (iii) A C RAN
presented in Section III.
macro cell has three sectors and uses three ports of the PON.
Summarising, even without exact coordinates for buildings Each sector needs its own colour. Our algorithm to fill the
we will have reasonable numbers for PON deployments in- PONS works iteratively. We simply start with the customers
cluding wired customers and including D RAN, C RAN cells. and fill ports, opening new PONs if necessary. We then iterate
The numbers for fibre kilometres and splitter resources in the first the D RAN cells and then the C RAN sectors successively
ODN can be seen as rough estimations while the numbers for filling the ports. We open new PONs and use new colours
PONs, OLTs, ONUs, and required cable resources in the feeder if necessary. Customers always use the same colour. D RAN
section are very good indicators. cells use colours different from customers and C RAN sectors
use colours different from D RAN cells and different from
Given some distribution of MC nodes from Table II, a customers. See Fig. 8 for typical PON fill statistic resulting
distribution of macro sites from Table IV, and an assignment from our algorithm.
of local exchanges to a primary and secondary MC node, we
can compute the distance between an antenna site and an MC b) OLTs and ONUs: Computing the number of OLTs
node. The LE-to-MC fibre distance is taken from the fibre route required at all MC nodes to serve the PONs is simple and
to the primary MC. For the antenna-to-LE distances we build a follows directly from the number of used colours at the PONs.
spanning tree between all sites corresponding to the same LE Each colour needs a serving OLT. The number of ONUs at the
(the closest as mentioned above), see Fig. 7. The distance of client side equals the total number of used PON ports.
an antenna to the LE is defined by the unique path in that tree.
Notice that we design the weights in the minimum spanning c) Approximating the number of splitters: To approx-
tree computation so as to (heuristically) minimise distances imate the number of required splitters we fix the splitting
towards the LE. hierarchy. In case of PONs for customers or D RAN cells or
mixed PONs it is possible that all 410 available ports are
As mentioned above, a C RAN solution is feasible only actually used, that is, in these cases we assume a 512 split
if the antenna-to-MC distance is at most 40 km. However, with a 1x8 splitter at the cabinet and a 1x16 splitter at the
even with this 40 km distance restriction, LR-PONs with a DP besides the 4x4 splitter at the LE, see Fig. 1. If k is the
maximum reach of 125 km offer a good solution to cover the number of 512-PONs (customer, D RAN, or mixed PONs) then
most populated areas with C RAN. Table V shows that for all k 4x4 amplifier splitters, 4k 1x8 cabinet splitters and 32k 1x16
 TABLE VI. D RAN AND C RAN DEPLOYMENT AND REQUIRED
SUPPLEMENT RESOURCES COMPARED TO PURE RESIDENTIAL LR-PON S
(BASE ) – UK INSTANCE WITH 73 MC S , 5,578 LE S , AND 44,309
ANTENNA SITES , AND 29.4 M IO WIRED CUSTOMERS .

Supplement
Entity Base D RAN C RAN (65%)
Backhaul Fiber km 14,078,841 0.16% 0.88%
Backhaul Cable km 102,259 0.08% 0.41%
D-side Fiber km 1,737,453 0.42% 1.26%
Drop Fiber/Cable km 13,899,625 0.21% 0.64%

(a) PONs per LE PONs 74,397 0.16% 1.14%

Cabinet splitter 297,588 0.16% 0.16%
DP splitter 2,380,704 0.16% 1.14%
ONUs 29,373,914 0.15% 0.34%
OLTs 74,397 7.53% 114.80%

TABLE VII. D RAN AND C RAN DEPLOYMENT AND REQUIRED

SUPPLEMENT RESOURCES COMPARED TO PURE WIRED CUSTOMER
LR-PON S . W E REPORT ON THE PURE MACRO - CELL DEPLOYMENT AS
WELL AS ON A DEPLOYMENT WITH ADDITIONAL 8 OR 16 MICRO CELLS
PER MACRO CELL .

(b) Colours per LE Country Scenario PONs Feeder ODN Splitters OLTs ONUs
UK DRAN 0.2% 0.2% 0.2% 0.2% 7.5% 0.2%
CRAN(65%) 1.1% 0.9% 0.7% 1.0% 114.8% 0.3%
DRAN-8 1.3% 1.3% 3.4% 1.3% 20.9% 1.3%
CRAN-8(65%) 8.7% 6.9% 3.8% 7.9% 415.3% 1.5%
DRAN-16 2.4% 2.4% 6.5% 2.4% 36.5% 2.5%
CRAN-16(65%) 16.2% 12.8% 6.9% 14.7% 716.9% 2.7%

Spain DRAN 0.2% 0.2% 0.3% 0.2% 13.5% 0.2%

CRAN(52%) 0.5% 0.4% 0.8% 0.4% 97.1% 0.3%
DRAN-8 1.4% 1.4% 3.4% 1.4% 25.3% 1.4%
CRAN-8(52%) 6.0% 4.9% 3.9% 5.5% 335.2% 1.6%
(c) PON fill per LE
DRAN-16 2.6% 2.6% 6.5% 2.6% 42.1% 2.7%
CRAN-16(52%) 11.7% 9.5% 7.0% 10.7% 575.9% 2.9%
Fig. 8. UK solution, 73 MCs. The y-axis shows the number of LEs that
achieve the performance indicator on the x-axis. (a) Average number of PONs Italy DRAN 0.2% 0.2% 0.2% 0.2% 15.6% 0.2%
(b) Average number of colours (c) Average PON fill (1.0 = 512 ports; 0.8 = CRAN(70%) 0.6% 0.6% 0.7% 0.6% 117.6% 0.4%
410 ports)
DRAN-8 1.3% 1.4% 3.3% 1.3% 26.0% 1.5%
CRAN-8(70%) 6.5% 5.8% 3.7% 5.9% 415.3% 1.7%
DRAN-16 2.6% 2.8% 6.3% 2.6% 43.8% 2.8%
DP splitters are required. In contrast, in case of pure C RAN CRAN-16(70%) 13.2% 11.9% 6.7% 12.0% 716.2% 3.0%
PONs at most 40 ports can be used. We assume that the cabinet
splitter is omitted in this case. This gives a 4 · 16 = 64 split.
That is, if the number of C RAN PONs is k 0 then we have k 0
additional 4x4 amplifier and 4k 0 additional 1x16 DP splitters. average distance of 50% of the macro cell radius corresponding
to the same geo-type, that is, a distance between 125m (Dense
d) Approximating cabling: We will ignore the cost Urban) and 5km (Sparse Rural), see Table III. For clients
for ducts as we have not seen big differences in the cable without individual geo-type we assume the same geo-type as
deployment for the different macro cell scenarios (only wired most of the cells at the same LE. For the ODN section and
customers, additional D RAN deployment, mixed D RAN/C RAN wired clients we assume an average fibre distance of 100% of
deployment), see below. That is, we do not expect any sig- the corresponding geo-type. However for macro cells we can
nificant change in the duct infrastructure because of wireless obtain a more precise value since fibre distances to the local
integration if only macro cells are considered. To compute exchange can be obtained from the spanning tree mentioned
cable resources in the LR-PON we distinguish drop cables above. We follow the general assumption that the cabinet
to connect clients to the PON, ODN cables for the segment splitter is located close to the LE. Since we assume a 1x16
between first and last splitter, and feeder cables for cabling splitter at the DP we divide the client numbers by 16 to obtain
between LE and MC. For the latter we have precise numbers the number of fibres in the ODN. In the ODN we assume
for both the actual fibre routes and the number of fibres cables with 8 fibres on average.
following these routes (based on the number of PONs). We
use a feeder cable model with cables of different sizes having Table VI shows the resources required for an LR-PON
between 48 and 276 fibres, see Table VIII. For the Drop- deployment with only wired customers connected. It then also
section, the section between the client and the DP splitter, we reports on the additional resources (in %) for a deployment
use a single-fibre cable, see Table VIII. To approximate the with only D RAN cells and for a deployment where all feasible
required total Drop cable resources we assume Drop distances sites are centralised following the C RAN architecture. In this
that depend on the geo-type. For all clients we assume an case there are roughly 65% C RAN feasible cells, c. f. Table V.
 We can clearly observe that the increase in required re- in the ODN and absent in the Drop section (1-fibre cable).
sources for wireless integration compared to existing LR-PONs We further note that if the duct build probability depends
connecting all households is insignificant. Even in case of a linearly on the number of cables an a trail, then the additional
C RAN deployment the number of PONs increases only by duct build in percent will be identical to the additional cable
1.14% (0.16% for D RAN only). Notice that almost all LEs resources in percent, e.g. 5.2% in the feeder area.
have a connected site in our solution for the UK. This means
that if D RAN cells where integrated into PONs independent VII. C OST CONSIDERATIONS
of PONs for wired customers, then the number of additionally
required PONs would be at least 5,449 (the number of LEs). So far we ignored the notion of cost. In particular, we
However, we only need 120 additional PONs for D RAN ignored the cost for the actual antenna sites. To get an idea of
deployment (0.16%). This means that most of the required the cost difference for D RAN versus D RAN deployments we
D RAN ports could be integrated into the existing residential use the techno-economical model summarised in Table VIII.
PONs. The same holds for the C RAN deployment. Since the Cost values are given in relation to the cost for one OLT port,
number of PONs does not increase significantly also all other which has a cost of 1.0. We assume that by centralising we
numbers have relatively moderate increases (splitters and fibre can save 50% operational and 20% capital expenditures, that
resources) as well. This shows that wireless convergence in is, the cost for a C RAN RRH plus the share at the centralised
LR-PONs is feasible with very low impact compared to the BBU is 80% of the cost of a D RAN macro site.
fixed access LR-PON deployment, both in C RAN and D RAN. TABLE VIII. H ARDWARE AND COST MODEL FOR LR-PON, C RAN ,
AND D RAN DEPLOYMENT. C OST VALUES ARE GIVEN RELATIVE THE COST
The only number that increases significantly in Table VI is OF ONE OLT PORT. C RAN SITE COST INCLUDES THE COST FOR THE RRH
the number of OLTs. For the D RAN deployments it increases AND THE BBU SHARE .
by 8% while for C RAN deployment it increases by even 115%.
Entity D RAN C RAN
Recall that residential customer PONs only occupy one colour
of the WDM PON while 32 D RAN sites already require its OLT port 1.00 1.00
ONU port 0.04 0.03
own colour and each sector in the C RAN deployment. Macro site (RRH plus BBU share) 6.42 5.13
Opex per macro site per year 11.43 5.72
It further turns out that these observations are relatively Cable kilometre ODN 8 fibres 0.38 0.38
independent from the actual country and MC node distribution. Cable kilometre Drop 1 fibre 0.07 0.07
In Table VII we summarise the results for the different Cable kilometre feeder 48 fibres 1.05 1.05
... ... ...
countries. We also tested solutions with a much larger MC Cable kilometre feeder 276 fibres 2.62 2.62
node density (not reported in the table) and we did not observe 4x4 splitter and amplifier 2.39 2.39
1x16 splitter 0.28 0.28
any difference in the D RAN deployment even when the number 1x8 splitter 0.18 0.18
of MCs increases drastically (we tested up to 600 MCs in the
UK). However, with increasing MC node numbers the C RAN
coverage increases which results in a moderate increase of the Based on this cost model we can easily evaluate the total
number of PONs and the corresponding resources for C RAN. cost for the LR-PON deployments in Table VI. For the UK
solution with 73 MCs we compute an up-front cost investment
It can be also be observed in Table VII that the distribution for wireline integration of 3 Mio OLTs (ignoring duct cost). In
of LEs and antenna sites is such that for Spain and Italy even Table IX it can be seen that most of this cost is consumed by
more D RAN and C RAN cells can be integrated into the wired ONUs and cable resources in the ODN/Drop. The increase in
PONs (the supplemental PON numbers are very small). In fact, cost for wireless integration is moderate if only macro cells are
the number of cells for the UK is relatively large compared considered. For D RAN we pay 11,155 and for C RAN 103,801
to the UK area and compared with Spain and Italy, which is OLT units. Most of this increase is in the OLT cost, 5,602
mainly due to a higher population density in some areas and the for D RAN and 85,408 for C RAN. The second largest absolute
resulting large number of (Dense) Urban sites, see Table IV. increase is in the total cost for cabling (Feeder plus ODN plus
We also studied a micro cell scenario where 8 or 16
small cells are deployed in addition to the macro site (except TABLE IX. D RAN AND C RAN DEPLOYMENT AND REQUIRED
for Sparse Rural sites), also reported in VII (rows: D RAN-8, ADDITIONAL CAPITAL EXPENDITURES COMPARED TO PURE RESIDENTIAL
C RAN-8, D RAN-16, C RAN-16). The micro cells are assumed LR-PON S (BASE ) – UK INSTANCE – BASE COST VALUES ARE GIVEN
W. R . T COST OF ONE OLT PORT. – W E REPORT ON THE MACRO CELL
to have one sector only and are treated similar to macro cells in DEPLOYMENT AND A DEPLOYMENT WITH 16 ADDITIONAL MICRO CELLS
terms of required resources. While the increase in additionally PER MACRO CELL .
required resources is still moderate in the D RAN case, it
is significant for C RAN, in particular in the 16 micro cell Supplement
Entity Base D RAN C RAN (65%) D RAN-16 C RAN-16(65%)
scenario. The number of PONs increases by 11.7% to 16.2%,
Feeder Cable 190,310 0.14% 0.66% 1.61% 8.91%
which results in a significant increase of the corresponding ODN Cable 107,433 0.42% 1.26% 6.93% 7.77%
resources such as cabling in ODN and backhaul and the Drop Cable km 968,341 0.21% 0.64% 6.41% 6.84%
number of splitters. We note that with an increase of 10% Amplifier nodes 177,982 0.16% 1.14% 2.41% 16.23%
and more of the fibre resources we can also not longer ignore Cabinet splitter 53,566 0.16% 0.16% 2.41% 2.41%
the additional cost for duct build. Clearly, the increase of cable DP splitter 351,551 0.16% 1.14% 2.41% 16.23%
kilometres is less than the increase in fibre kilometres, e.g. the ONUs 1,174,957 0.15% 0.30% 2.54% 2.47%
OLTs 74,397 7.53% 114.80% 36.49% 716.87%
12.8% increase in the feeder for the UK solution refers to an
increase of only 5.2% in required cables. The cables simply Total 3,098,536 0.36% 3.35% 4.63% 18.43%
contain more fibres. However this effect is not as significant
Drop). Recall that our numbers for the Feeder are accurate and WDM. We showed that wireless/wireline convergence is
estimations based on real fibre routes and a cable model with feasible as the supplement in required resources to include
different cable sizes. We do not expect a significant increase in antenna sites is marginal. This is in fact independent of whether
the cost for ducts for wireless macro cell integration. In fact, D RAN or C RAN approaches are considered. Clearly, it is
assuming a duct build probability of n · 3%, where n is the advisable to reserve capacities already when deploying PONs
number of cables on a trail, we estimate a duct base cost of for wired clients. With C RAN we observe a significant increase
67,800 OLTs (in addition to the 3 Mio up-front cost) in the in the number of required OLTs, which is larger than the
feeder area, which increases by only 56 for D RAN and 275 capital savings at the antenna site. However since centralising
for C RAN. should provide enormous savings in the operational cost, we
could show that C RAN amortises already in the first years after
In Table IX we ignore both the capital expenditures and the deployment compared to D RAN. The picture changes and
the operational expenditures for the antenna sites itself. We needs further studying if there is a significant portion of micro
have 44,309 macro sites, which results in an investment of cells deployed and integrated into the PON. In this case even
284,316 (D RAN) or 227,453 (C RAN), that is, we save 56,863 additional duct build might be necessary.
with C RAN. However already the increase in the OLT cost
exceeds these savings. Considering the pure total investment
ACKNOWLEDGEMENT
(LR-PON cost plus site cost) a C RAN solution for the UK
costs 36,079 more than a D RAN solution. However, this is not To authors would like to thank Thomas Pfeiffer and Rene
a huge difference and since C RAN outperforms D RAN in terms Bonk for many fruitful discussions regarding wireless inte-
of operational cost (50% savings assumed), C RAN amortises gration. We would also like to thank Marco Schiano and
already in the first year (compared to D RAN, see Fig. 9(a)). Andrew Lord for providing realistic data from Telecom Italia
If we decrease both the capital and operational savings by and British Telecom.
centralising to only 10% per site C RAN still amortises already
in the second year, see Fig. 9(b). R EFERENCES
[1] “D ISCUS (FP7 Grant 318137), Deliverable 2.3, Updated reference
architecture,” 2014.
[2] M. R. et al., “Discus: An end-to-end solution for ubiquitous broadband
optical access,” IEEE Com. Mag., vol. 52, no. 2, 2014.
[3] “40-Gigabit-capable passive optical networks (NG-PON2): General re-
quirements,” ITU-T, Series G: Transmission Systems and media, digital
systems and media, Tech. Rep. G.989.1, 2013.
[4] “D ISCUS (FP7 Grant 318137), Deliverable 4.2, System specifications
for LR-PON implementation ,” 2014.
[5] C. Raack, N. Ascheuer, and R. Wessäly, “A nation-wide optimization
study on the consolidation of local exchanges using long-reach passive
optical networks,” in Proc. ICTON, Graz, Austria, 2014, pp. 1–5.
[6] ——, “Nation-wide deployment of (long-reach) passive optical net-
works: Computing the location and number of active nodes.” in Proc.
ITG Fachtagung, Leipzig, Germany, 2015.
(a)
[7] OSM, “Open Street Maps, http://www.openstreetmap.org,” 2015.
[8] “D ISCUS (FP7 Grant 318137), Deliverable 2.6, Architectural optimiza-
tion for different geo-types,” 2014.
[9] L. C. Wang, K. Chawla, and L. J. Greenstein, “Performance studies
of narrow-beam trisector cellular systems,” International Journal of
Wireless Information Networks, vol. 5, no. 2, pp. 89–102, 1998.
[10] J. Itkonen, B. Tuzson, and J. Lempiäinen, “Assessment of Network
Layouts for CDMA Radio Access,” EURASIP Journal on Wireless
Communications and Networking, vol. 2008, no. 259310, 2008.
[11] M. U. Sheikh and J. Lempiäinen, “A flower tessellation for simulation
purpose of cellular network with 12-sector sites,” IEEE, Wireless
Communications Letters, vol. 2, no. 3, pp. 279–282, 2013.
[12] M. Shimrat, “Algorithm 112: Position of point relative to polygon,”
Communications of the ACM, vol. 5, p. 434, 1962.
[13] B. Guo, W. Cao, A. Tao, and D. Samardzija, “LTE/LTE-A
(b) signal compression on the CPRI interface,” Bell Labs Technical
Journal, vol. 18, no. 2, pp. 117–133, 2013. [Online]. Available:
Fig. 9. Additional capital and operational expenditures of wireless in- http://dx.doi.org/10.1002/bltj.21608
tegration including LR-PON resources and antenna sites. UK instance. a) [14] P. Chanclou, A. Pizzinat, F. Le Clech, T.-L. Reedeker, Y. Lagadec,
capital/operational expenditures as in Table VIII b) capital/operational savings F. Saliou, B. Le Guyader, L. Guillo, Q. Deniel, S. Gosselin, S. Le,
per site by centralising only 10%. T. Diallo, R. Brenot, F. Lelarge, L. Marazzi, P. Parolari, M. Martinelli,
S. O’Dull, S. Gebrewold, D. Hillerkuss, J. Leuthold, G. Gavioli,
and P. Galli, “Optical fiber solution for mobile fronthaul to achieve
VIII. C ONCLUSION cloud radio access network,” in Future Network and Mobile Summit
(FutureNetworkSummit), 2013, July 2013, pp. 1–11.
In this paper we studied wireless integration into an access
architecture based on long-reach passive optical networks