Provided by the author(s) and University College Dublin Library in accordance with publisher
policies. Please cite the published version when available.
Title Provisioning call quality and capacity for femtocells over wireless mesh backhaul
Authors(s) Olariu, Cristian; Fitzpatrick, John; Ghamri-Doudane, Yacine; Murphy, Liam, B.E.
Publication date 2013-09-11
Conference details 2013 IEEE 24th International Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC), 8 - 11 September 2013
Publisher IEEE
Item record/more information http://hdl.handle.net/10197/7518
Publisher's statement © © 2013 IEEE. Personal use of this material is permitted. Permission from IEEE must be
obtained for all other uses, in any current or future media, including reprinting/republishing
this material for advertising or promotional purposes, creating new collective works, for
resale or redistribution to servers or lists, or reuse of any copyrighted component of this
work in other works.
Publisher's version (DOI) 10.1109/PIMRC.2013.6666674
Downloaded 2022-08-09T13:30:42Z
The UCD community has made this article openly available. Please share how this access
benefits you. Your story matters! (@ucd_oa)
© Some rights reserved. For more information, please see the item record link above.
Provisioning Call Quality and Capacity for
Femtocells over Wireless Mesh Backhaul
Cristian Olariu
⇤
, John Fitzpatrick
§
, Yacine Ghamri-Doudane
†‡k
and Liam Murphy
†
⇤
TSSG, Waterford Institute of Technology, Cork Road, Waterford, Ireland
§
Openet, 6 Beckett Way, Park West Business Park, Dublin 12, Dublin, DB, Ireland
†
UCD, School of Computer Science and Information, Belfield, Dublin 4, Ireland
‡
ENSIIE, 1 Square de la r
´
esistance, 91025 Evry CEDEX, France
k
Universit
´
e Paris-Est, LIGM Lab, 75420 Champs sur Marne, France
colariu@tssg.org, johnfitzpat@gmail.com, ghamri@ensiie.fr, Liam.Murphy@ucd.ie
Abstract—The primary contribution of this paper is the design
of a novel architecture and mechanisms to enable voice services
to be deployed over femtocells backhauled using a wireless mesh
network. The architecture combines three mechanisms designed
to improve Voice Over IP (VoIP) call quality and capacity in a
deployment comprised of meshed femtocells backhauled over a
WiFi-based Wireless Mesh Network (WMN), or femto-over-mesh.
The three mechanisms are: (i) a Call Admission Control (CAC)
mechanism employed to protect the network against congestion;
(ii) the frame aggregation feature of the 802.11e protocol which
allows multiple smaller frames to be aggregated into a single
larger frame; and (iii) a novel delay-piggy-backing mechanism
with two key benefits: prioritizing delayed packets over less
delayed packets, and enabling the measurement of voice call
quality at intermediate network nodes rather than just at the
path end-points. The results show that the combination of the
three mechanisms improves the system capacity for high quality
voice calls while preventing the network from accepting calls
which would result in call quality degradation across all calls,
and while maximizing the call capacity available with a given set
of network resources.
I. INTRODUCTION
Wireless access technologies, such as WiFi, have increased
in popularity due to the widespread use of smart-phones. The
amount of data traffic has surpassed voice traffic driving both
the Mobile Network Operators (MNOs) and wireless access
standardization bodies to adapt accordingly. Widespread Long
Term Evolution (LTE) deployments are being undertaken in an
effort to cope with these increased traffic demands, while in
parallel, operators are looking to complementary technologies,
such as femtocells and WiFi, in an effort to reduce congestion
on the cellular radio access networks.
Reducing the radius of cell towers is a possible solution
to increase the performance of cellular networks and increase
frequency reuse [1]. Femtocells are small access-point-sized
devices with much smaller coverage areas when compared to
typical cellular base stations. Unlike traditional base stations,
femtocells are backhauled over normal Internet Protocol (IP)
connected networks such as residential Internet connections.
The focus of this work is to increase the cellular capacity in
a defined region where femtocells are deployed. This is par-
ticularly beneficial for transient deployments; such scenarios
include, but are not limited to, social events where attendees
need access to voice and data services at venues which are
used for a relatively short period of time. In such scenarios,
the networking infrastructure needs to be flexible and quickly
deployable.
A backhaul solution where each femtocell is provided with
wired infrastructure to forward its traffic to the Evolved Packet
Core (EPC) is not feasible as wired deployments tend to pose
serious logistical problems. This work proposes a new deploy-
ment scenario which utilises a WMN backhaul infrastructure
for femtocells, in order to overcome the deployment difficulties
imposed by wired solutions. WMNs have become increasingly
popular due to their capabilities: relatively large coverage areas
with minimal cabling requirements, quick deployment, fair
price, and ease of maintenance.
This paper considers the usage of a WiFi-based WMN
infrastructure, where the mesh nodes are hybrid stations fea-
turing multiple WiFi mesh interfaces and one LTE femtocell
embedded or co-located with the mesh node. Figure 1 depicts
the scenario this work is focused on. It can be seen that clients
roam with their User Equipment (UE) in an area serviced by a
grid of femtocells backhauled over a WMN network into the
femtocells’ network operator’s core network.
The WMN access-medium’s capacity saturates when traffic
demand is high, resulting in high latency and packet loss
causing the performance of services to degrade. A special case
of such services are voice calls. In LTE, both data and voice
traffic payloads are encapsulated in IP packets, hence the voice
calls in an LTE network are VoIP calls.
WMNs can be highly unpredictable in terms of Quality
of Service (QoS) offered to VoIP calls, hence the VoIP QoS
on WMNs degrades rapidly when only one call more than
the system’s capacity is added. In such situations detecting
the capacity barrier is critical to ensure ongoing calls are
guaranteed with satisfactory QoS levels. This work compares
the individual and combined effects of three mechanisms
designed to improve the call quality and capacity in a femto-
over-mesh deployment scenario: (i) a CAC mechanism which
uses samples of the calls’ quality to determine when to
restrict the addition of new calls into the network; (ii) the
frame aggregation feature of the 802.11e wireless standard,
allowing the aggregation of multiple small frames into a single
frame, thus reducing the delays induced by transport protocol’s
Figure 1: Topology of WMN-backhauled femtocells
overhead; and (iii) a proposed delay-piggy-backing mechanism
which attaches to every VoIP packet, while inside of the
WMN, the cumulative delay experienced as it propagates
through the network — this enables the scheduler to choose
from a Push-In-First-Out (PIFO) queue more delayed packets
first, and the WMN gateway to compute call quality samples
based on which the CAC mechanism takes actions regarding
new call requests.
The CAC mechanism resides in the Local Femto Gateway
(LFG) [2] which is an entity placed in the neighbourhood
of the WMN’s gateway. Simulation results show that a CAC
mechanism is necessary in such scenarios due to the fact that
uncontrolled additions of voice calls into the network will
negatively impact all existing calls. The overall call quality
and network VoIP capacity are the two performance indicators
analysed in this work.
The remainder of this paper is structured as follows. In
Section II, the relevant and related work is discussed. This
is followed by Section III which provides an overview of
the Mean Opinion Score (MOS) and E-Model concepts, a
description of the LTE femtocell architecture, and the solutions
proposed in this work. Section IV details the simulation
settings followed by the presentation of the simulation results
in Section V. This paper is then concluded in Section VI.
II. RELATED WORK
CAC is not a recent concept as it applies to any system
where reaching capacity is a concern. However, Wei et al. [3]
addressed the issues related to employing CAC in a WMN
serving VoIP as the main service for users. The possibility of
aggregating packets to improve the overall system’s capacity
is also presented in [3].
The concept of calculating an Intermediate Mean Opinion
Score (iMOS) was introduced in our previous work [4]. In
that work, the authors assumed the intermediate node is time-
synchronized with one of the end-points, hence the packet
delay could be measured in order to obtain the iMOS.
In [5] a possible implementation of PIFO queues is pre-
sented together with a performance analysis. As PIFO queues
require more computational power during the push-in phase
than normal First-In-First-Out (FIFO) queues, their perfor-
mance evaluation showed that current hardware is able to cope
with the increased processing power imposed by PIFO queues.
In [2] a solution to offload the mobile core network is
presented together with a proposed architecture addressing an
enterprise femtocell deployment. The work presents the con-
cept of the LFG entity where a proxy Mobility Management
Entity (MME) and a proxy Serving Gateway (S-GW) are used
to offload the core network by locally managing functions
related to mobility and data access.
This work combines all the solutions mentioned above and
further extends their scope by coupling them with a novel
method of prioritizing VoIP packets and obtaining the iMOS.
III. ARCHITECTURE DESCRIPTION
A. LTE Architecture and LFG concept
The LTE security requirements specify that all traffic ex-
change between a femtocell and the EPC needs to be trans-
ported over IPsec tunnels. The tunnel terminates at the entry
point of the EPC, which in Figure 1 is depicted as the Home
e Node B Gateway (HeNBGW).
Inside the EPC, the MME terminates the control-plane
signalling from femtocells and UEs. The S-GW handles the
data plane from the UEs. The PDN Gateway (P-GW) is the
anchor point of the UEs to the Internet and manages the IP
addresses domain.
One of the roles of the MME is to manage the bearers
and connections with the UEs, which further include the
acceptance of new calls and potentially dropping calls when
such action is required. Our proposed CAC mechanism needs
to have access to these functions in order to maintain the QoS
level for the VoIP calls inside the WMN.
However, these functions are not accessible to the WMN
operator. Zdarsky et al. [2] have shown in their proposed
architecture that these functions could be assigned to a LFG
entity. In their proposal, the LFG intercepts control messages
using its Proxy-MME function. This enables the LFG to take
actions regarding new or existing calls, based on the iMOS
reported by the WMN.
B. iMOS-based CAC
The International Telecommunication Union-
Telecommunication Standardisation Sector (ITU-T) report
P.800 [6] introduces a general scoring system to assess the
quality of a speech transmitted via telephone lines, on a range
from 1 to 5. Human subjects grade the speech quality using
this scale, the result being the Mean Opinion Score (MOS).
This work uses an estimative model which does not involve
human subjects in rating speech quality. The most popular
model for the estimative speech assessment is the E-Model
(ITU-T G.107 [7]) which is based on transmission parameters
and is widely accepted as an accurate tool for transmission
network planning.
The E-Model was initially not designed for real-time mea-
surements, but its adaptation to VoIP calls [8] allows the
calculation of MOS scores on-the-fly. The E-Model was de-
signed to measure the MOS on an end-to-end basis where
the packet delay can be estimated as the half of the Round
Trip Time (RTT) obtained form Real-Time Control Transport
Protocol (RTCP) packets. When trying to measure the MOS
at an intermediate point in the path, even the rough RTT delay
estimation is not possible. Time synchronization protocols
such as Network Time Protocol (NTP) can synchronize the
intermediate point with the end-points, hence the intermediate
nodes would be able to precisely estimate the network delay.
However, the synchronization solution is not viable, as usually
the end-points do not support synchronization from unknown
network entities.
In this work, we enable the WMN gateway to obtain an
accurate measurement of the packet delay from the delay-
value attached to each VoIP frame coming from the WMN.
This accurate delay measurement enables the WMN gateway
to measure the call quality. Since the WMN gateway is an
intermediate point in the calls’ path, we name the MOS
obtained in the WMN gateway as iMOS.
The WMN gateway keeps track of all ongoing VoIP calls
and their corresponding iMOS scores. The iMOS scores used
by the mechanisms presented in this work are obtained only
from the up-link traffic. However, the mechanism can be
employed also on the down-link, but would require each
WMN node to be iMOS-aware and to have the means to
inform the WMN gateway about possible issues. Both of these
assumptions unnecessarily increase the load on the WMN
nodes and on the network traffic itself.
The iMOS values obtained from the up-link voice packets
are the basis of the CAC mechanism. CAC can be invoked
automatically or manually, allowing a network operator to
prevent extra traffic from being injected in the network
when certain conditions occur. Usually, the conditions revolve
around maintaining a pre-established QoS level.
The CAC mechanism is controlled by the LFG based on
call quality reports received from the WMN gateway, i.e. the
WMN node in Figure 1 equipped with a wired interface.
This work assumes the usage of a LFG placed in the archi-
tecture as indicated by Figure 1. The LFG is able to intercept
and interpret signalling messages between the femtocells and
the EPC. In this way the CAC mechanism can take actions
regarding new and existing calls.
As VoIP applications generate tens of packets per second,
it would be infeasible to take a CAC decision based on
every VoIP packet traversing the WMN gateway. However, the
WMN gateway calculates the iMOS for each traversing VoIP
packet and places the value into an accumulator. Periodically
the CAC mechanism, residing on the LFG, will request
the WMN gateway to provide one aggregated value which
represents the overall iMOS of all ongoing calls during the
last period. If that value is under a certain pre-established
threshold, then CAC is activated and new call requests are
rejected.
We define the network’s capacity as being identified when
the addition of the most recent call causes the average iMOS
value across all calls to drop below some threshold. In addition
to activating the CAC mechanism when the overall iMOS
value falls under the threshold, the CAC mechanism in our
scenario, drops the call with the worst iMOS. A few calls are
dropped in this manner until the network reaches a steady state
after it was destabilized by the extra call accepted above the
capacity.
As the CAC mechanism periodically checks the aggregated
iMOS value, it is important to determine a proper rate at which
this checking is done. According to [8], human subjects are
able to determine a quality drop within 5 seconds. Taking that
into account, we used 1 second for the CAC loop interval,
allowing the mechanism to drop a few calls, hence restoring
the actual call quality before the user can detect it.
C. Delay-Piggy-Backing based Priority Scheduler
This work uses a packet scheduler which prioritizes more
delayed VoIP packets over less delayed VoIP packets. In con-
junction with the frame aggregation, the proposed scheduler
levels the delay distribution over all VoIP calls so that the
quality variation between calls decreases thereby increasing
fairness.
In order to achieve this delay-based prioritization, the
queueing delay, expressed in time units, is attached to each
WiFi frame. In case of multiple aggregated frames, the delay
value is attached to each individual frame. The value is
additive, thus next hops will add to it the queueing delay
suffered while waiting in their queues. When packets are
placed in the queue, the delay value is used to find the proper
location of insertion, assuming the WMN node uses PIFO
queues.
In a PIFO queue, packets are enqueued in the push-in phase
at a specific location based on a comparison criteria. In our
work, the criteria is the piggy-backed delay value. Specifically,
the current packet’s cumulative delay is compared against that
of each packet’s in the queue, starting with the head of the
queue. The newly arrived packet will take the position in the
queue where its cumulative delay is for the first time bigger
than that of the packet’s used for comparison.
In Figure 2, a 35 seconds sample of a voice call shows the
absolute difference between the actual and the estimated delay
0
5
10
15
40 50 60 70 80
Absolute Delay Difference (ms)
Simulation Time (sec)
Figure 2: Validation of delay measurement precision.
values. The actual values were obtained from the simulator,
by computing the difference between the arrival time and the
sending time (only possible in a simulation environment). The
estimated delay is extracted from the value attached to every
packet. There is a high degree of correlation between both val-
ues with a variation typically lower than 5 milliseconds which
is an insignificant amount when related to its influence on
the MOS. This demonstrates the ability of the piggy-backed-
delay solution to provide accurate estimations of network
packet delay without requiring any time synchronisation, and
furthermore it enables the WMN gateway to calculate accurate
iMOS values for ongoing calls.
QoS support can be activated by enabling the 802.11e
[9] protocol. This protocol defines four Access Categories
(ACs) as follows: AC BK for Background traffic, AC BE for
Best Effort traffic, AC VI for Video traffic, and AC VO for
Voice traffic. In this work, we enabled the delay-piggy-backing
prioritization only in the AC VO queue.
IV. SIMULATION SETTINGS
The parameters of the simulations carried out in this work
are presented in Table I. We simulated a WMN grid of 16
nodes using NS-3.10 [10]. The WMN nodes are equipped with
two 802.11a interfaces for backhauling femtocell traffic. For
simplicity we placed the UE applications directly on the WMN
nodes.
The queues used in our simulation have a maximum capac-
ity of 50 packets, as this is the queue size used by the most
widespread wireless drivers, i.e. MadWiFi [11] and ath5k [12].
The impact of the packet delay on the MOS score is
significant when its value is higher than 150 ms [7], and values
higher than 400 ms render a conversation as non-interactive.
Therefore the AC
VO queue in our simulations employs an
early-dropping mechanism by removing packets when their
queueing delay on a node becomes larger than 250 ms.
The routes are statically assigned in order to being able
to draw statistically significant conclusions from the results.
However, it is worth noting that Optimized Link State Routing
Protocol (OLSR) [13] was used for initial route discovery.
We injected 100 calls into the network which is higher than
the expected capacity. Calls are injected sequentially and the
inter-call arrival rate is exponentially distributed with a mean
of 1 second. Figure 3 depicts the call duration and start times
0
10
20
30
40
50
60
70
80
90
100
40 60 80 100 120 140 160 180 200
Simulation Time (sec)
Call ID
Accepted
Rejected
Dropped
Figure 3: Call duration and status; dropped calls are the calls
which were initially accepted but the CAC decided later to
drop, as their corresponding iMOS was low.
Parameter Value
Simulator NS-3.10 [10]
Topology Grid 4x4
Distance between nodes 100 m
Number of interfaces 2
WiFi Mode 802.11a
WiFi Data Rate 6 Mbps
Network Access Method CSMA-CA
Propagation Model LogDistancePropagationLossModel
Error Rate Model YansErrorRateModel
Remote Station Manager ConstantRateWifiManager
WiFi interfaces queue size 50 packets per AC
Early-drop threshold 250 milliseconds
Routing Algorithm Fixed routes, pre-discovered by OLSR
No. of injected calls 100
Call Duration Exp. dist. between 160 and 40 seconds
Call Direction Full-duplex
Voice codec AMR 12.20 & AMR SID modes only
Speech model ITU-T/P.59 [14]
iMOS threshold R=70 (MOS=3.6)
CAC loop interval 1 second
Number of simulation epochs 10
TABLE I: Simulation Setup
of all calls injected in the simulation. For consistency of results
only the MOS scores obtained during the time span of the
shortest accepted call are considered.
The calls are full duplex and use the AMR codec. For
simplicity we implemented only the 12.20 kbps and Silence
Indicator (SID) modes. We implemented the speech model de-
fined in [14] to mimic realistic conversations. During an active
period of the speech model, i.e. when someone speaks, AMR -
12.20 packets are sent, and during silent periods AMR SID
packets are sent.
A threshold value of 3.6 for the iMOS is used by the CAC
mechanism to detect if it is necessary to activate CAC, as on
the MOS scale it represents the border between Some users
dissatisfied and Many users dissatisfied.
