Sharing low-cost wireless infrastructures with telecommunications operators for backhauling 3G services in deprived rural areas

Abstract

3G small cells are becoming a promising solution for inexpensive 3G coverage in rural small villages, but the high cost of backhaul infrastructures remains an obstacle for the worldwide adoption of this kind of solutions. On the other hand, broadband wireless infrastructures may be found in many rural areas for communications in specific sectors such as health or education services, environmental monitoring, etc. These sectorial broadband communications infrastructures are often multi-hop wireless networks using long-distance WiFi links or WiMAX. We propose the possibility of 3G operators sharing those third parties' infrastructures. It would permit them to have a low-cost backhaul solution for their rural 3G cells in those areas where the expected demand of 3G services does not ensure enough revenues to justify the deployment of dedicated infrastructures. This is a win-win solution because bandwidth-renting may help to achieve financial stability for such sectorial networks, while operators can extend their coverage to regions where this would not be economically sustainable otherwise. This paper studies the technical feasibility of the proposed solution. Different multi-hop broadband rural networks are studied, and a general architecture is derived from them. Then, additional requirements are introduced in order to make such networks support the QoS requirements of 3G voice and data traffic. Three techniques are considered in order to manage the network in such a way that 3G traffic and other traffics may be transported together in the network without compromising the quality of the backhaul: DiffServ, plain MPLS and MPLS-TE. Measurements taken on a real testbed in a laboratory allow us to compare the performance achieved with these three techniques, and the advantages and drawbacks of each of them are presented.

Full text

Sharing low-cost wireless infrastructures with
telecommunications operators for backhauling 3G
services in deprived rural areas
Javier Simo-Reigadas, Esteban Municio, Eduardo Morgado, Eva M. Castro and Andres Martinez
School of Telecommunications Engineering
Rey Juan Carlos University
Fuenlabrada, Spain
Email: javier.simo@urjc.es
Abstract—3G small cells are becoming a promising solution
for inexpensive 3G coverage in rural small villages, but the high
cost of backhaul infrastructures remains an obstacle for the
worldwide adoption of this kind of solutions. On the other hand,
broadband wireless infrastructures may be found in many rural
areas for communications in specific sectors such as health or
education services, environmental monitoring, etc. These sectorial
broadband communications infrastructures are often multi-hop
wireless networks using long-distance WiFi links or WiMAX.
We propose the possibility of 3G operators sharing those third
parties’ infrastructures. It would permit them to have a low-cost
backhaul solution for their rural 3G cells in those areas where the
expected demand of 3G services does not ensure enough revenues
to justify the deployment of dedicated infrastructures.
This is a win-win solution because bandwidth-renting may
help to achieve financial stability for such sectorial networks,
while operators can extend their coverage to regions where this
would not be economically sustainable otherwise. This paper
studies the technical feasibility of the proposed solution. Different
multi-hop broadband rural networks are studied, and a general
architecture is derived from them. Then, additional requirements
are introduced in order to make such networks support the QoS
requirements of 3G voice and data traffic. Three techniques are
considered in order to manage the network in such a way that
3G traffic and other traffics may be transported together in
the network without compromising the quality of the backhaul:
DiffServ, plain MPLS and MPLS-TE. Measurements taken on a
real testbed in a laboratory allow us to compare the performance
achieved with these three techniques, and the advantages and
drawbacks of each of them are presented.
I. INTRODUCTION
Urban areas currently make up more than 60% of the
world population, and technological solutions are commonly
focused on them. On the other hand, rural areas are much
larger extensions with only 40% of the population world-
wide, and providing this population with telecommunications
services is much more difficult and expensive due to lack
of accessibility and the low population density. In particular,
telecommunications operators in developing regions offer all
typical telecommunications services, both fix and mobile, in
cities, where users are concentrated and revenues compensate
the investments required to deploy telecommunications infras-
tructures. On the contrary, operators avoid deploying costly
infrastructure in rural areas because the expected revenues are
much lower.
The TUCAN3G project [1] is aiming to propose low-cost
solutions for 3G/4G services in rural areas with acceptable
quality, mainly based on WiLD (WiFi over Long Distances) or
WiMAX in the backhaul and femtocells (HNBs) for the access
networks. The goal of this project is to significantly reduce
the investment needed for providing 3G services, ultimately
promoting 3G coverage in small localities where operators
are not offering their services yet. In this sense, two different
strategies are considered for the backhaul: (i) the deployment
of new communications infrastructures that are planned strictly
for backhaul purposes, or (ii) sharing pre-existent broadband
wireless infrastructures that are there for other purposes. The
second strategy is considered because there are more and more
broadband wireless infrastructures that have been deployed
for rural telemedicine, for Internet access in rural schools, or
for other specific purposes. These networks are often based
on low-cost technologies such as WiLD (WiFi over Long
Distances) systems or WiMAX systems. Regarding WiLD
systems, previous works have evaluated the performance of
real WiFi systems with the standard CSMA/CA medium access
control and also of modified systems such as Mikrotik NV2
that use TDMA instead of the standard MAC, all of them
resulting valid solutions for long point-to-point links covering
distances up to 50 km or even more. Even though these tech-
nologies may have capacities over 20 Mbps at long distances,
they are not the type of technology that operators use because
their quality and performance do not match their expectations
in conventional networks. Such infrastructures are commonly
deployed for strategic sectors, and public authorities may
decide to spend resources for their maintenance. However, it
would be beneficial for the owners to allocate some bandwidth
for renting virtual circuits to operators. This could lead to a
win-win solution in which the owners of infrastructures receive
funds that may help to achieve financial sustainability, and
operators initially spend only operational expenditures that are
proportional to their income.
In order to take these ideas into consideration seriously,
those existing WiLD or WiMAX multihop networks must
prove to meet the operators’ requirements for the backhaul
of 3G cells, which is not obvious. Furthermore, if such infras-
tructures are going to be shared, additional proof is required
to demonstrate that this can occur without compromising 3G978-1-4799-8461-9/15/$31.00 c
2015 IEEE

traffic’s quality of service. This paper presents a network
architecture that takes advantage of this kind of networks,
together with well known techniques for IP traffic control, to
achieve these objectives. Different techniques are tested and
compared for managing resources when 3G traffic shares the
infrastructure with other competing traffic: DiffServ, MPLS
and MPLS-TE. Results show that multihop WiLD or WiMAX
networks may be effectively used for 3G backhaul, and the
three techniques mentioned may be used to manage the traffic
in these networks. The advantages and drawbacks of each of
them are shown.
The paper is organized as follows: Section 2 presents a
real example that illustrates the type of networks we suggest
in our proposals. Section 3 presents the network architecture
proposed. Section 4 shows how DiffServ, MPLS and MPLS-
TE can be used to manage different traffic classes with differ-
ent QoS levels. Section 5 presents the experimental scenario
used for testing and the experiments that have been run on it.
Section 6 explains and discusses the results of the experiments.
Finally, Section 7 presents our conclusions.
II. A REAL EXAMPLE: BAL SA PU ERT O NET WORK
The Peruvian Amazon forest is one example of a vast
region in the world with large extensions in which most
human settlements lack access to communication networks for
reasons briefly explained in the introduction. In this context,
some networks have been deployed in the last years for
telemedicine [2], [3] in different parts of the forest, connecting
isolated health facilities to urban hospitals. One of these
telemedicine networks is Balsapuerto Network, which connects
a few health facilities with Yurimaguas’ hospital over WiLD
links with MIMO 2x2 antennas (see Figure 1). All links in
this network may have capacities largely over 20 Mbps (see
Table I for performance with different technologies considered
in [5]), which is much more than what is strictly necessary
for telemedicine purposes. Furthermore, Telefonica del Peru,
which is the incumbent operator in the region, is planning
the deployment of 3G femtocells in the same villages for 3G
services. TUCAN3G’s proposal in this case is to test whether
the same infrastructure can be successfully shared for both
telemedicine traffic and 3G traffic.
TABLE I. CA PACIT Y OFFE RE D BY DI FFE REN T TE CHN OL OGI ES I N THE
BAL SAP UE RTO NET WOR K WI TH A PE R-HO P DE LAY OF 5M S.
Link Distance Thr. WiMAX WiFi NV2
(km) req. (Kbps) (Kbps) (Kbps) (Kbps)
Balsap. 21.2 8012.8 31527.7 31200 70598
- S.G. (16QAM3/4) (MCS13) (MCS13)
S.G. 19.3 13382.4 42728 37440 70598
- S.J. (64QAM2/3) (MCS13) (MCS13)
S.J. 27.2 15110.4 31009.4 24960 52656
- Yuri. (16QAM3/4) (MCS12) (MCS12)
III. NETWORK ARCHITECTURE FOR A RURAL TRANSPORT
NE TWORK
The scenarios for which we are elaborating this proposal
are rural areas in developing regions where small villages
are far from well-connected towns or cities. The natural way
to bring broadband communications to several villages in
the same zone implies multihop networks, because once a
location has connectivity, another location that is close to it,
Fig. 1. Balsapuerto network is a simple chain network that extends from
the city of Yurimaguas to three villages. In each village, the health center
is connected to the local node, and the project exists for connecting a 3G
femtocell in each village additionally.
but further from the city, will be easily connected through
the previous location in two hops. Extending this topology to
several localities, the most common scenario would be a tree
topology (see Figure 2). The tree root is a gateway system
which connects the rural wireless transport network with a
high capacity network. The leave nodes are user traffic sinks
and sources connected to ’edge nodes’ that become the starting
point on the way to the gateway. Those edge nodes may use
other nodes as downstream routers on the way to the gateway.
In general, wireless links will be point to point links dedicating
all the capacity to the communication between a pair of nodes,
although particular cases may rarely be found in which a point
to multipoint link can be used in this kind of network. As links
are closer to the gateway, they aggregate the traffic coming
from or going to different edge nodes.
The described architecture is not the only option, because
rings or incomplete mesh topologies may sometimes be used,
specially in networks where there are two or more gateways in
order to avoid single points of failure. However, in this paper
we are going to study the tree topology using a unique gateway
as the root node. Links may be WiLD, WiMAX or other
technologies that are considered appropriate for the context
in terms of costs, availability, maintainability, performance,
etc. As explained in [5], most of these technologies exhibit
an acceptable performance for all sorts of traffic when they
operate far below the saturation point, with low delays and
negligible packet-loss probability. As the operation point gets
closer to the saturation point, all QoS parameters are affected,
and the delay and the packet loss notably go beyond the
acceptable thresholds, especially for real-time traffic. Hence,
this kind of networks require that links are kept operating under
saturation, which in turn requires per-hop traffic control. This
is why the proposed architecture includes a node between any
pair of links that has traffic control capabilities.
However, the introduction of nodes for traffic control at
each hop implies: 1) the introduction of packet queues that
increase the delay and may eventually drop packets, and 2) the
application of policies that drop packets to prevent saturation.
Hence, the only way to keep the whole multi-hop network

operating with low end-to-end latency and negligible packet-
drop probability is an admission control system that limits the
traffic that enters the transport network. For the rest of the
document, we assume that edge nodes and the gateway node
apply ingress policies to limit how much traffic each user can
introduce into the network, according to the original planning.
Further works will study how all this can be dynamically
adapted as the traffic demand and/or the links’ capacities
change, but that is beyond the scope of this paper.
A. High-level architecture
The architecture represented in Figure 2 assigns a name to
each interface in order to avoid ambiguous references. Each
place in which communications equipment is installed is called
a Location Lk. Each location has a Router or Switch Rk
which controls all the traffic originated in / destinated to /
passing through a location Lk. All communications equipment
(femtocells Fk
ior WiFi/WiLD/WiMAX systems Wk
ior ’other
systems’ Xk
i) in Lkare connected to Rkthrough an Ethernet
link (Ethernet, Fast Ethernet or Gigabit Ethernet as needed;
this interface is not supposed to be a bottleneck in any
case). Interfaces between Wk
iand Rkare given a name Ik
W i,
and interfaces between Fk
iand Rkare called Ik
F i instead.
Wireless links are also given a name related to the upstream
location: lk,j . As explained in the introduction, we consider
edge routers to exchange traffic with two types of external
systems: 3G/4G femtocells, or ’other systems’. For example,
in the case presented in the previous section, the other users
would be the health facilities that exchange data traffic with
Yurimaguas hospital.
In the network planning phase the following requirements
are defined:
•Limited end-to-end delay. Let’s call Dk
F i the expected
maximum delay between the gateway (R0) and Fk
i,
Dkthe delay introduced in Rkcaused by queuing
and scheduling, and Dk,j
lthe delay introduced by the
communication between Wk
iand Wj
i(linked by lk,j ).
Then Dk
F i =Dk+Dk,j
l+Dj+Dj,i
l+... +Da,0
l+D0
•Enough capacity in every link. Let’s call Sk,j the
maximum throughput expected in link lk,j , and Sk
m
the maximum throughput admitted from Fk
m. Then,
Sk,j =P
∀x
Sx,k +P
∀y
Sk
y
•We have already seen that the studied wireless tech-
nologies may offer different delays and different ca-
pacities depending on the operation conditions. Pro-
vided that the real capacity of a link is Ck,j
R(t)and
the real delay introduced by that link is Dk,j
R(t),
the following conditions must be true at any time:
Ck,j
R(t)≥Sk,j , and also Dk,j
R(t)≤Dk,j
l. Addition-
ally, the real delay Di
Rintroduced by any system Ri
must also be Di≤Di
R,∀t.
If the three conditions are accomplished, the multi-hop
transport network will be a valid backhaul for the 3G cells.
If there is excess capacity in all links to accommodate the
other users’ requirements, then the network has the potential
to be shared for other uses (such as telemedicine services) and
3G backhaul.
Fig. 2. Network architecture. The gateway system connects the rural wireless
transport network with a high capacity network. The leaf nodes (femtocells or
external user systems) are connected to ’edge nodes’ that become the starting
point in the way to the gateway.
B. Description of the interfaces and interactions
The interfaces described are the following:
•Femtocell or other terminal system to router (Im
F n) -
This interface will be physically supported by a Fast
Ethernet cable connecting the system to an edge router
Rm. When IP traffic comes into the interface, it is
already marked using DSCP. The maximum admitted
traffic over this interface (Sm
F n) is defined in the
network planning phase. The Rmingress discipline
queue may dynamically discard traffic in excess or
may color the traffic, marking packets as needed, so
the Rmegress discipline queue can drop traffic in
excess in a controlled fashion.
•Router to wireless terminal (Im
W k) - The wireless
link may be WiLD, WiFi or WiMAX. In principle,
point-to-point links will always be used unless it
is absolutely clear that a point-to-multipoint link is
advantageous in a given scenario. In those cases where
one end acts as master and the other one as slave, the
master will always be the system closer to the gateway.
Router Rmand wireless terminal Wm
kwill be con-
nected with a fast-ethernet or gigabit-ethernet cable
as needed. The egress block in the router will ensure
that priorities are applied as required in the network
and the traffic is shaped so that the offered load to the
wireless terminal is always under Ck
m. On the other
hand, this traffic shaping might cause long queues
and eventually packet dropping. A mechanism that

monitors the state of queues must be used to prevent
HNB from generating more traffic when any router
in their path to the gateway is close to a congestion
status.
IV. TEC HN IQ UE S FO R TR AFFI C MA NAGEMENT
The 3G backhaul must support at least three types of QoS,
corresponding to three traffic classes: signaling, voice and data
traffic. More QoS types may be supported if necessary. We are
going to observe how ’external’ traffic exchanged by other user
systems (such as telemedicine traffic in the case of Balsapuerto
network) affects the performance of the backhaul. Each QoS
class requires a different QoS level, mainly defined by rate,
delay and jitter parameters, and some of them have strict QoS
requirements.
For the backhaul to be able to guarantee the appropriate
QoS for each service, three strategies are going to be evaluated
in this paper: 1) Differentiated Services (DiffServ), 2) a
combination of DiffServ and Multiprotocol Label Switching
(MPLS) [RFC3031], and 3) a combination of DiffServ and
MPLS-TE (MPLS Traffic Engineering [RFC2702]).
DiffServ is a network architecture that provides mecha-
nisms for classifying data flows and for providing different
QoS levels to them. Packets are marked using the DS field in
the IP header, which contains the 6-bit DSCP value. Several
classes of QoS behaviors and types of dropping priorities
are defined using those 6 bits. This architecture permits to
obtain a different per-hop behavior for elastic traffic (best
effort), expedited forwarding traffic (low-loss, low-latency),
and assured forwarding traffic (for which assurance of delivery
is provided as long as the negotiated rate is not exceeded). The
different QoS classes enable a way of define different types of
traffic in terms of priority and packet dropping probabilities.
Hence, it does not support quantitative QoS services.
MPLS avoids the need of routing in the network layer.
Virtual flows are identified with a label. Each packet with a
particular label will have a predefined path to the destination
(LSP, Label Switch Path), so no long routing tables are used
for routing the packets. Moreover, each flow can be associated
with a particular QoS set of parameters. Labels are distributed
across the network using the Label Distribution Protocol
(LDP) [RFC3036]. Additionally, MPLS-TE is a mechanism
for providing virtual links between end-point nodes in the
MPLS network. End-to-end tunnels restrict the amount of
resources that each virtual circuit may use, permitting that
several virtual circuits live together in the same network with
mutual protection, as long as the total amount of bandwidth
assigned to the tunnels is equal or lower to the real amount of
available bandwidth.
In all cases, edge nodes apply traffic shaping, differenti-
ation and priorities at the IP layer. In the first strategy, the
rest of nodes still apply a PHB (per-hop behavior) at the IP
layer, therefore packets belonging to different traffic classes are
treated differently. In the second and third strategies, packets
are given a MPLS label at the edge node and MPLS switching
is used across the network, providing more efficient forwarding
than IP-based routing. In the third case, the use of end-to-end
tunnels across the network gives more control on the capacity
allocated for each LSP between user nodes and the gateway,
but this also avoids statistical multiplexing exchanging control
for performance.
The network scenario shown in figure 2 is valid for the
three alternatives. Routers connected only to backhaul wireless
links are core routers (R1) while those connected to user nodes
are edge routers (R11,R12 ,R13,R121 ,R131) . R0is a special
edge router acting as gateway between the backhaul and the
operator’s core network.
In the first case, Rinodes are DiffServ routers. All edge
routers may:
(a) apply traffic policy: limit incoming or outgoing traffic by
dropping packets when necessary.
(b) forward packets to the right interface according to the IP
routing table.
(c) provide different priority to telephony traffic, signaling and
data traffic.
(d) manage the bandwidth allocation between different traffic
classes.
(e) limit the traffic delay in each queue, if packets stay too
long in the queue they will be dropped.
In the first case, the rest of routers may provide (b), (c)
and (e).
In case MPLS or MPLS-TE are used, nodes Riare MPLS
switches. Those connected only to backhaul wireless links
are strictly MPLS switches (R1), while those connected to
user nodes (R11,R12 ,R13,R121 ,R131) are hybrid nodes
implementing both, MPLS and IP protocols. R0is a special
hybrid node acting as gateway between the backhaul and the
operator’s core network.
V. EXPERIMENTAL ENVIRONMENT AND EXPERIMENTS
In order to study the proposed alternatives, a real testbed
network has been deployed in a laboratory environment. The
network corresponds to the reliable model of the Balsapuerto
Network detailed in Section II and its scheme is illustrated in
Figure 3. The 6 nodes of the backhaul network are Mikrotik
RB493G with R52Hn chipsets, with MIMO 2x2 and IEEE
802.11n support and all have MPLS and MPLS-TE support.
Outside the backhaul network there are 4 PCs which inject uni-
directional traffic using the D-ITG software [6]. One of them
plays the role of Gateway to the core network. The other three
PCs play the role of the aggregated traffic injected by each
village, Balsapuerto, San Gabriel and San Juan respectively.
The wireless links WLAN1, WLAN2 and WLAN3 has been
configured with 802.11n using MCS4, MCS12 and MCS11
respectively in order to resemble the real network distance
restrictions.
Fig. 3. Scheme of the laboratory network used for the tests

It is assumed that in each village there is a health centre
and at least one femtocell. The traffic demand for each village
in both downlink and uplink is shown in Table II and has been
modelled according to a previous study of their population and
their usage expectations [7].
TABLE II. TR AFFIC E STIMATI ON F OR EAC H VI LLA GE IN
BAL SAP UE RTO NE TWO RK.
( Kbps ) Pop. Cells Voice UL+DL Data UL+DL Total
Balsapuerto 5139 2 1362.6 2740 4143.6
San Gabriel 948 1 302.8 520 831
San Juan 118 1 52 80 132.3
These values has been calculated considering the HNB
nano3G E16 [9] from IPAccess. For each town, 20% of people
in itinerancy have been considered in order to estimate a
realistic number of inhabitants, and the average subscribers for
the only ISP present in the area has been calculated as 53%
of the total population (based on available data from similar
regions). Voice traffic is calculated as the maximum voice
channels used in the busy hour given a block probability of
2%, where each channel has different requirements depending
on its direction, compression configuration and multiplexing
level. For the data traffic computation, 5% of potential users
for data services from the number of total subscribers has been
considered and the mean value for downlink and uplink in the
busy hour is 15 Kbps and 5 Kbps respectively according to
[8]. For the estimation of the total traffic, signalling traffic is
assumed to be a surplus of 1% of the total HNB traffic. The
guaranteed throughput for voice traffic and signalling is the
100% of the injected traffic, whereas the guaranteed throughput
for 3G Data traffic is only the 80% of the injected traffic.
In order to simulate accurate behavior when infrastructures
are shared, an additional flow is injected which play the role
of telemedicine traffic. This traffic is injected at each location
in a saturating rate which corresponds to 1/4 of its bottleneck
capacity for uplink and to 3/4 of its bottleneck capacity for the
downlink. The guaranteed throughput for telemedicine traffic
is only 10% of the injected traffic at for each location.
The traffic model used for voice traffic has been constant
packet rate and constant packet size (from 121 bytes to 736
bytes depending of the multiplexation level). Data traffic is
usually estimated as a Poisson process MMPP (Markov Modu-
lated Poisson Process)[10] or using FTP model 2 [7]. However
these approaches have short term dependency, as we consider
only bursty behavior in small time scales and are focused
on the access networks. Other models for traffic aggregation
consider a larger dependence in the long term. These models
assume bursty behaviors independently of the time scale, more
suitable for backhaul networks where traffic is aggregated. This
effect is described with the self-similarity property or fractality
and can be modelled with different stochastic processes [11].
Self-similarity level of any traffic pattern is defined by the
Hurst parameter H, which is defined as follows in Eq. 1 and
is related with the shape with Eq. 2:
D(Xm) = m2(H−1) ·D(X)(1)
H=3−α
2(2)
where D(Xm)in the aggregated variance of the transmission
interval given a specific distribution D(X). Data traffic shows
different values for H depending on the traffic type. For the
experiments, a Hurst parameter H=0.8 has been considered,
which according to the Q.3925 ITU-T recommendation [12]
correspond to HTTP traffic. Using this value for H, the inter-
departure time of each packet of 3G Data, signalling and
telemedicine traffic is modelled with the long-tailed Pareto
distribution. The packet size has been considered constant with
a value of 1400 bytes at IP level.
Backhaul nodes are configured so that voice is the highest
priority traffic, signalling is the second highest priority traffic
and 3G Data traffic and telemedicine traffic are both lower
priority traffic and at the same level. This traffic differentiation
is made at different levels depending on the configuration
chosen:
- DiffServ: Traffic is shaped in each node using HTB
queuing disciplines at IP level and using the WMM
specification in the the MAC level. The QoS informa-
tion of each packet is transported in the DHCP field
and is read en each backhaul node.
- MPLS: Traffic is shaped only in edge nodes using
HTB queuing disciplines at IP level and using the
WMM specification in the the MAC level. Due to
DSCP is not read inside the backhaul since MPLS
is configured, the QoS information is transported in
the EXP field of the MPLS header.
- MPLS-TE: Traffic is shaped only in edge nodes using
HTB queuing disciplines at IP level in the tunnels
entrance and using the WMM specification in the the
MAC level. As well as before, the QoS information
for link level is transported in the EXP field of the
MPLS header.
Policing rules are applied in order to ensure proper sharing of
the backhaul network fulfilling the QoS requirements of each
traffic, as shown in Table III for each configuration.
All values have been calculated in order to set each link
under saturation conditions. When bandwidth is available and
minimum values of throughput are guaranteed for each traffic,
both traffic 3G Data and telemedicine compete to get the
surplus resources under equal conditions.
Finally, with this testbed configuration, the 8 different types
of tests executed are detailed in Table IV. They are two sets
of similar tests; the first 4 are when the backhaul is not shared
with telemedicine traffic and the last 4 are when the backhaul
is shared with telemedicine traffic, which is always trying to
saturate the network. The values for Voice in tests number 1
and 4 are the voice traffic injected by the femtocell in normal
performance according to the initial network dimensioning. For
the rest of the tests, values for Voice MAX corresponds to the
voice traffic when a maximum number of channels in each
cell are used, ie. 16 voice channels. In a similar way, 3G Data
values correspond to those traffic values for what the network
was designed for, and 3G Data MAX when up to 16 3G Data
channels are used per cell. Since only the 80% of the 3G Data
traffic is assured, 3G Data MAX is equivalent to a saturation
condition for 3G Data traffic.

TABLE III.
Diffserv MPLS MPLS-TE
Node 1 min ceil min ceil Tunnel min ceil
Voice 700 3457 700 3457 700 1832
Sig 100 3457 100 3457 1832 100 1832
3G-Data 916 3457 916 3457 816
Tele 435 3457 435 3457 1625 x x
Node 2 min ceil min ceil Tunnel min ceil
Voice 1600 9918 x x x x
Sig 100 9918 x x x x x
3G-Data 2745 9918 x x x
Tele 1305 9918 x x x x x
Node 3 min ceil min ceil Tunnel min ceil
Voice 1400 5313 599 1856 350 966
Sig 200 5313 100 1856 966 100 966
3G-Data 1373 5313 458 1856 457
Tele 1105 5313 670 1856 890 x x
Node 4 min ceil min ceil Tunnel min ceil
Voice 2500 15160 x x x x
Sig 200 15160 x x x x x
3G-Data 4120 15160 x x x
Tele 3315 15160 x x x x x
Node 5 min ceil min ceil Tunnel min ceil
Voice 2000 6980 350 1663 350 966
Sig 300 6980 100 1663 966 100 966
3G-Data 1830 6980 458 1663 458
Tele 1775 6980 670 1663 700 x x
Node 6 min ceil min ceil Tunnel min ceil
Voice 3700 19800 3700 19800 1700 5042
Sig 300 19800 300 19800 5042 100 5042
3G-Data 5500 19800 5500 19800 2750
Tele 5325 19800 5325 19800 4875 x x
Voice x x x x 1000 2572
Sig x x x x 2572 100 2572
3G-Data x x x x 1373
Tele x x x x 2650 x x
Voice x x x x 1000 2572
Sig x x x x 2572 100 2572
3G-Data x x x x 1373
Tele x x x x 2090 x x
TABLE IV. I NJ ECT IO N CON FIG URATI ON F OR EA CH TE ST
Test Femtocell Traffic Shared
1 Voice + Signalling No
2 Voice MAX + Signalling No
3 Voice MAX + 3G Data + Signalling No
4 Voice MAX + 3G Data MAX + Signalling No
5 Voice + Signalling Yes
6 Voice MAX + Signalling Yes
7 Voice MAX + 3G Data + Signalling Yes
8 Voice MAX + 3G Data MAX + Signalling Yes
VI. RESULTS AND DISCUSSION
Several tests were run in the experimental environment
described in the previous section:
1) The network is dedicated exclusively to be the backhaul
for the 3G femtocells. This is the most favourable case,
as there is not any external traffic competing for the
resources. Firstly, 3G traffic injected in the network is
adjusted to the initial planning. Results are shown in
Figure 4 for both downlink and uplink. Then, 3G data
traffic is increased up to saturation and results are shown
in Figure 5 for both downlink and uplink.
2) The next set of experiments represents the situation in
which the network is shared for 3G backhaul and another
use (such as telemedicine), which is given the same
priority as 3G data traffic. In this case, telemedicine traffic
injected from the gateway and to all nodes has enough
intensity to saturate the network by itself. Figures 6 and
7 represent the total throughput supported in the network,
distinguishing traffic coming or going to each village.
The figures represent the addition of downlink and uplink
traffic, injecting only the planned traffic or saturating with
3G data traffic, respectively. Also, delay is represented as
in the previous set of experiments: 3G traffic injected in
the network is adjusted to the initial planning. Results are
shown in Figure 8 for the downlink and uplink. Then, 3G
data traffic is increased up to saturation. Results for the
downlink and the uplink are shown in Figure 9.
Fig. 4. Uplink and Downlink latency when the network is not shared and
3G traffic is not saturating.
Fig. 5. Uplink and Downlink latency when the network is not shared and
3G traffic is saturating.
The results of the first set of experiments show that the
multihop network works perfectly as a backhaul solution for
rural femtocells. When the offered traffic load does not saturate
the network (see Figure 4), the performance is very good for
all sorts of traffic, although it is better for voice and the delay

Fig. 6. Aggregated traffic Uplink+Downlink when 3G traffic is not saturating
the network.
Fig. 7. Aggregated traffic Uplink+Downlink when 3G traffic is saturating
the network.
increases for flows that cross more hops (e.g. Balsapuerto
experiments more delay than San Juan in all cases). The
three solutions tested for traffic management are good, with
insignificant differences between MPLS and DiffServ, and a
clearly better behaviour for the case of MPLS-TE. As the data
traffic offered to the network increases, Figure 5 shows that
3G data traffic obviously suffers from the effects of saturation
(in the tunnel created for that traffic, not in the whole link),
but signalling and telephony continue experimenting the same
QoS.
When telemedicine traffic (or any other external traffic in
other cases) is present, we can avoid impacts on the quality
perceived in telephony and signaling as well, as seen in Figures
8 and 9. Additionally, Figures 6 and 7 also show that the band-
width guaranteed for 3G data traffic is respected. Particularly,
for the case of MPLS-TE even 3G data is protected and keeps
low delays no matter whether telemedicine traffic is present or
not. On the other hand, MPLS-TE imposes hard restrictions to
the use of bandwidth for each type of traffic, and avoids when
one type profits while another one is not consuming its share.
VII. CONCLUSIONS AND FURTHER WORK
The transport network architecture presented in this paper
for broadband rural communications has proven valid for
backhauling rural 3G cells. In previous works WiLD, WiFi-
based TDMA solutions for long distances, and WiMAX had
been validated and compared. Now we have proposed a way
to make a multihop network, with links based on these tech-
nologies, work efficiently with effective performance. Next,
Fig. 8. Uplink and Downlink latency when the network is shared and 3G
traffic is not saturating.
Fig. 9. Uplink and Downlink latency when the network is shared and 3G
traffic is saturating.
DiffServ, MPLS and MPLS-TE have been tested as alternative
techniques for network-wide QoS traffic management.
After testing these alternatives in a laboratory testbed, it is
shown that all three are valid, but MPLS is the most efficient in
terms of throughput (it gets the most from the network), while
MPLS-TE obtains the best delays at the cost of misusing the
capacity of the network. Also DiffServ obtains a performance
very similar to MPLS, which suggests that any of the two
may be applied with similar results depending on the level of
support for these solutions in the systems chosen for routers.
Hence, in cases where there is excess capacity and MPLS-TE
may be used with satisfactory results, this would be the best

choice. However, if it is important to benefit from statistical
multiplexing, any of the other two options are valid.
In conclusion, it has been proved that a multihop network
based on WiLD or WiMAX links may be a valid solution
for rural 3G backhaul. Furthermore, if there is a network of
this kind that is being used for other purpose and there is
excess bandwidth that can be allocated for 3G backhaul, we
have shown that this can be done with good quality for the
operator.
Further works should study mechanisms for dynamic adap-
tation of the network, as the static proposal is based on the
stability in the traffic demand and the links capacities, and
both suppositions are not very likely to exist for a long time in
any real network. Finally, it is advisable to investigate possible
mechanisms for the backhaul to give feedback to the access
network when edge nodes perform severe packet-drops, as this
could help the access network to perform an intelligent access
control.
REFERENCES
[1] TUCAN3G Project. URL: http://www.ict-tucan3g.eu/
[2] J. Simo-Reigadas, P. Osuna-Garcia, S. Lafuente-Sanz, A. Martinez-
Fernandez, and J. Seoane-Pascual, “The Design of a Wireless Solar-
Powered Router for Rural Environments Isolated from Health Facilities,
IEEE Wireless Communications Magazine, vol. 15, no. 3, pp. 24-30,
June 2008.
[3] C. Rey-Moreno, I. Bebea-Gonzalez, I. Foche-Perez, R. Quispe-Tacas,
L. Li˜
n´
an-Benitez, and J. Simo-Reigadas, ”A telemedicine WiFi network
optimized for long distances in the Amazonian jungle of Peru“, Proc.
of the 3rd ACM Extreme Conference on Communication: The Amazon
Expedition, Manaus, Nov 2011.
[4] J. Simo-Reigadas, A. Martinez-Fernandez, J. Ramos-Lopez, and J.
Seoane-Pascual, ”Modeling and optimizing ieee 802.11 DCF for long-
distance links”, IEEE Transactions on Mobile Computing, Vol. 9, N. 6,
2010, pp 881–896.
[5] J. Simo-Reigadas, E. Morgado, E. Municio, I. Prieto-Egido and A. Mar-
tinez, “Assessing IEEE 802.11 and IEEE 802.16 as backhaul technologies
for rural 3G femtocells in rural areas of developing countries”, Proc. of
EUCNC’14 Conference, Bologne, Italy, Jun 2014.
[6] A. Botta, A. Dainotti, A. Pescap, “A tool for the generation of realistic
network workload for emerging networking scenarios, Computer Net-
works, V. 56, n. 15, Oct 2012, pp 3531-3547.
[7] A. Garca, I. Prieto, E. Snchez, O. Muoz , J. Vidal, A. Agustn,
A. Pascual, J. Sim, J. Paco, Y. Castillo, G. Ramrez, O. Tupay-
achi. Socio-economic scenarios, technical specifications and architecture
for the proof of concept. 2013 http://www.ict-tucan3g.eu/documents/
deliverables/TUCAN3G D21EHASd.pdf
[8] SCF 047.01.01,Small Cell Forum. Extending Rural and Remote Coverage
Using Small Cells. Feb 2013
[9] IPAccess SoHo SME Access Points http://www.ipaccess.com/
[10] D. Bertsekas and R. Gallager, Data Networks (2nd Ed.). Prentice-Hall,
Inc., Upper Saddle River, NJ, USA, 1992.
[11] B. B. Mandelbrot, Long-run linearity, locally gaussian processes, h-
spectra and infinite variances,” Intern. Econom. 1969
[12] Recommendation ITU-T Q.3925. Traffic flow types for testing quality
of service parameters on model networks. ITU-T, 2012.