Quality of experience of voice communication in large-scale mobile ad hoc networks

Abstract

Real-time voice communication is an essential requirement in first responder scenarios. While mobile ad hoc networks (MANET) already prove to be an appropriate communication substrate in small-scale real-world operations, questions regarding scalability limitations remain. In this paper, we identify major factors that affect the quality of experience of voice communication in MANETs. In a series of simulation studies, we show that voice transmission using MANETs is also feasible in large-scale scenarios, if appropriate settings are chosen.

Full text

Quality of Experience of Voice Communication in
Large-Scale Mobile Ad Hoc Networks
Christian Gottron∗, Andr´eK¨onig∗, Matthias Hollick†, Sonja Bergstr¨aßer∗, Tomas Hildebrandt∗, Ralf Steinmetz∗
∗Multimedia Communications Lab (KOM), TU Darmstadt, Germany
{christian.gottron, andre.koenig, sonja.bergstraesser, tomas.hildebrandt, ralf.steinmetz}@kom.tu-darmstadt.de
†Center for Advanced Security Research Darmstadt (CASED), TU Darmstadt, Germany
matthias.hollick@cased.de
Abstract—Real-time voice communication is an essential re-
quirement in first responder scenarios. While mobile ad hoc
networks (MANET) already prove to be an appropriate commu-
nication substrate in small-scale real-world operations, questions
regarding scalability limitations remain. In this paper, we identify
major factors that affect the quality of experience of voice
communication in MANETs. In a series of simulation studies,
we show that voice transmission using MANETs is also feasible
in large-scale scenarios, if appropriate settings are chosen.
I. INTRODUCTION
Offering features like self-organization and self-healing,
MANETs allow for the spontaneous establishment of a net-
work without infrastructure that is able to adapt to a constantly
changing topology, which makes them a suitable communica-
tion substrate for deployment in first responder scenarios. Yet,
due to the multi-hop wireless transmission, MANETs lack the
performance and reliability of wired and infrastructure-based
networks: this results in a reduced quality of service (QoS) in
terms of throughput, delay, loss and jitter. These characteristics
of MANETs are challenging for voice communication, which
is a basic demand of on-site units in emergency response
scenarios.
Although the applicability of MANETs in small-scale real-
world scenarios has been shown, questions regarding scalabil-
ity limitations still remain open. In small-scale scenarios, the
aforementioned challenges are negligible due to the usually
short route length. In large-scale scenarios, with the size
of cities and above, consisting of hundreds of nodes, the
increasing route length will strongly affect the quality of a
voice transmission.
Common QoS metrics such as throughput, delay, loss, and
jitter define a precisely measurable technical description of a
networks characteristics. The quality of a voice transmission as
it is experienced by the user can only be qualified limitedly in
these terms, but can be evaluated by a set of human listeners
rating the quality. Clearly, this requires a high effort and is
hardly realizable in our context. Instead, tools that are based
on human perception models can be utilized.
In this paper, we scrutinize the quality of experience (QoE)
of real-time voice streaming in large-scale MANETs. After
presenting related work that has motivated our research, we
shortly summarize the basics of digital voice transmission and
the method we use for the QoE analysis. Our main contribution
is the evaluation of the influence of network size and load as
well as that of voice codecs and their parameterization on the
perceived QoE of voice communication within MANETs.
II. RELATED WORK
In this section we present related work that has motivated
our research. We focus, in particular, on projects with com-
parable application scenarios, as well as on research on voice
transmission in general and voice transmission in MANETs.
A MANET-based communication network architecture for
large-scale emergency response scenarios is described in [6]
and in [7]. Amongst other services, voice communication is
one objective of the network design. Yet, to the best of our
knowledge, no systematic evaluation of the QoE of the voice
communication is performed.
The quality of VoIP streams when transmitted over a
MANET is analyzed in [16]. The authors compare the effects
of different MANET routing protocols and of node mobility
by means of simulation. The metrics used for the evaluation
are the technical QoS metrics delay and loss. The scenario
is a small-scale ad hoc network consisting of 21 nodes with
a low node degree of approximately 4neighbors per node
and an average route length of 2.7hops. Scalability issues are
not considered. A QoE analysis based on human perception
models is not performed.
In [15] single-hop 802.11-based ad hoc networks are evalu-
ated subject to the network load in terms of simultaneous voice
streams, to node velocity, and to a packet prioritization based
on the size of the 802.11 congestion window. The evaluation
is performed in terms of the technical QoS metrics loss and
jitter. The packet prioritization is evaluated in a small-scale
scenario consisting of 40 nodes in an area allowing for direct
communication between any two nodes. Node mobility is
evaluated in a scenario consisting of 40 nodes with a node
degree of approximately 11 neighbors per node and an average
route length of 2.2hops. Neither scalability nor QoE are
subjects of the evaluation.
General challenges of voice communication in MANETs
are identified in [3]. Aspects of MAC and routing protocols
for MANETs are discussed. Different approaches of speech
coding are reviewed with respect to their applicability in
MANETs. A QoE analysis is not part of the work. Scalability
issues are not considered.
Christian Gottron, André König, Matthias Hollick, Sonja Bergsträßer, Tomas Hildebrandt, Ralf
Steinmetz:
Quality of Experience of Voice Communication in Large-Scale Mobile Ad Hoc Networks. In:
Proceedings of the second IFIP Wireless Days 2009, December 2009.
The documents distributed by this server have been provided by the contributing authors as a means to ensure timely
dissemination of scholarly and technical work on a non-commercial basis. Copyright and all rights therein are maintained by the
authors or by other copyright holders, not withstanding that they have offered their works here electronically. It is understood
that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These
works may not be reposted without the explicit permission of the copyright holder.

Besides 802.11-based ad hoc networks, Terrestrial Trunked
Radio - Direct Mode Operation (TETRA DMO) [4] offers a
solution for wireless ad hoc communication. However, this
standard is (to the best of our knowledge) not tailored to the
extensive multihop communication we consider in this work.
III. CODECS,PROT OCO LS &QOEANALYSIS
In this section we provide background information on the
relevant details of voice coding, voice transmission, and QoE
analysis as a basis for the remainder of this paper.
A. Voice Coding
The first step for voice communication in MANETs is the
application of a voice codec for conversion from an analog
signal to a digital signal. Voice codecs can be categorized
in (1) vocoders, (2) waveform codecs, and (3) hybrid codecs.
These types show fundamental differences with respect to their
architecture and their mode of operation. As a result, the band-
width required and the speech quality achieved differ strongly.
Vocoders require a bandwidth of only few (single-digit) kbit/s.
In consequence, the speech quality is poor. Waveform codecs,
on the other hand, can produce an excellent speech quality,
but have bandwidth requirements, of an order of magnitude,
higher than that of vocoders. Hybrid codecs can be classified
somewhere between vocoders and waveform codecs, regarding
the bandwidth requirements and speech quality achieved. For
further information, we refer to [5]. Due to the inherently poor
speech quality of vocoders, we focus on waveform codecs and
hybrid codecs with G.711 [11] and Speex [14] as particular
representations in this work.
G.711 is a well known waveform codec, which is based on
pulse code modulation. The uncompressed and high quality
codec requires a bandwidth of 64 kbit/s. An analog signal
is sampled at a rate of 8 kHz and quantized on an 8 bit
logarithmic scale resulting in a bandwidth requirement of
64 kbit/s. One of the most prominent fields of application
of G.711 is the ISDN telephone system. Speex is a hybrid
codec based on the code-excited linear prediction algorithm
[9]. A parameterization of Speex is possible regarding sample
rate, speech quality, and complexity of the encoding algorithm.
As a result, the bandwidth required varies from 2.15 kbit/s to
44 kbit/s.
B. Voice Transmission in MANETs
Transmission of encoded voice via a MANET can be seen as
a two-step process. In the first step, a route between sender and
receiver is established. This can be done either proactively or
reactively. In proactive routing protocols, each node constantly
keeps track on how other nodes of the network can be reached.
For this, corresponding topology information is exchanged
periodically. Reactive routing protocols determine a route not
before it is required. Which class of routing protocol should be
deployed, depends on the MANET application area. Proactive
protocols produce a constant base-load of the network to keep
the routing tables of all nodes up to date. In our scenario, we
assume high node mobility combined with a relatively sparse
network. For this reason, we base our study on the reactive ad
hoc on-demand distance vector (AODV) routing protocol as
described in [8]. Here, to establish a route between a source
and a destination, a route request message is sent as broadcast
by the source. The request is forwarded by intermediate nodes
until it reaches the destination. Each intermediate node keeps
track on the predecessor from which the request was received,
thus establishing a reverse route. When the request reaches the
destination, a route reply message is generated and sent along
the reverse route to the source. Upon receiving the route reply
at the source, the route is established.
After a route is established, the second step conducts
the transmission of the encoded voice. For transmission, we
deploy the real-time transport protocol (RTP), as described
in [10]. The primary task of RTP is to ensure correct packet
ordering. Thus, RTP packets contain a sequence number and
a timestamp. The combination of RTP and UDP allows for a
reliable transmission of multimedia data with a minimum of
protocol related delay and overhead.
In order to transmit voice data over packet-oriented net-
works, the codec generates data frames of a specific length.
For G.711, this interval is variable and part of our evaluation.
For Speex, the interval is fixed to 20ms due to the compression
algorithm, which is part of the coding process.
The variance of the delay (jitter) of subsequent packets
demands usage of a jitter buffer. With this, the first packet
of a voice stream is buffered for a certain amount of time
before it is handed to the codec in order to compensate the
jitter of the following packets. The size of the jitter buffer
defines the maximum variance of the delay accepted. Packets
that exceed this maximum are dropped.
C. QoE Analysis
The aim of this paper is to evaluate the QoE of voice
transmissions. The QoE can be derived from the voice quality
as experienced at the receiver and from the transmission delay.
The International Telecommunication Union (ITU) defines
the voice quality by the Mean Opinion Score (MOS) [13]. The
MOS specifies five different categories of voice quality, from
1 (worst) up to 5 (best). In the best case, the G.711 codec has
a 4.4 MOS while the quality of Speex lies between 2 and 4,
depending on the codec settings.
The MOS can be determined subjectively, which requires a
sufficiently large number of listeners to rate the quality. Thus,
this subjective determination of the voice quality is not feasible
in our context. An objective evaluation of the MOS can be
performed automatically, based on human perception models.
For this, the ITU proposed the Perceptual Evaluation of Speech
Quality (PESQ) algorithm [12]. This algorithm is designed to
determine the quality of a narrowband voice stream. PESQ
compares the original voice stream with the one received. The
correlation between a subjective determination of the voice
quality and the corresponding PESQ result is 0.935.
IV. QOEANALYSIS IN MANETS
In this section, we present the analysis of the QoE of voice
communication in MANETs. After describing the experimen-

tal design, we evaluate the effects of (1) the network size, (2)
the frame size, (3) the network load and (4) the voice codec.
A. Experimental Design
For the simulation studies, we used an extended version of
the JiST/SWANS simulation tool [2]. We added a traffic model
for voice communication in the first responder scenarios. The
default simulation settings are given below. We used these if
not specified otherwise.
Transmission range 250m
Average neighbors
per node
8
Voice streams 16 different recordings as recom-
mended by the ITU [12]
Codec G.711
Network load 5simultaneous voice streams
Frame size 20ms
Jitter buffer 100ms
Average processing
delay per hop
1ms
Initial placement Random with uniform geographi-
cal distribution
Mobility Random waypoint with continuous
movement and minimum and max-
imum speed of 1m/s and 2m/s
Communication
model
Random selection of sender and
receiver.
Simulated time 600sper setup
The default network parameters and simulation settings are
chosen such that we obtain a moderately loaded network.
While transmission range may vary due to the environment, we
choseafixedvalueof250min an environment without obsta-
cles to reduce the influences on voice quality for the scope of
this paper. Further, the density of the network was chosen such
that a connected (unpartitioned) network is typically achieved.
The random waypoint mobility model can be considered as a
worst case scenario with respect to the predictability of node
movement. The simulated time per factor set was split up into
multiple simulation runs to reduce any unwanted side-effects
of the random waypoint model. While more realistic mobility
models for first-responder scenarios were proposed [1], for
our simulations we chose the random waypoint model as an
unpredictable worst case scenario.
The protocols used in the network are as follows.
Physical layer Segmental calculation of the signal
power and the SNR
MAC Layer IEEE 802.11 DCF with a transmis-
sion rate of 11Mbit/s
Network Layer IPv4 with AODV routing service
and buffers up to 127 packets
Transport Layer UDP with RTP
In the evaluation, we focus on the metrics (1) voice quality,
(2) packet loss, and (3) transmission delay. The voice quality
was determined using the PESQ evaluation tool available at
0
20
40
60
80
100
50 100 200 500
loss rate [%]
number of nodes
(a) Packet loss for the variation of the network
size
1
2
3
4
5
50 100 200 500
voice quality [MOS]
number of nodes
(b) Voice Quality for the variation of the network
size
Fig. 1. Variation of the network size
[12]. The results were converted to the MOS scale to be
compared to related work. The delay is measured in two
different ways. The overall end-to-end delay is calculated as
the average of the delay of all packets, including the delay
that results from the route discovery process of the reactive
AODV routing. While the routing delay is a major part of the
delay, it only affects the very beginning of a voice stream (like
the waiting time after dialing in a traditional telephone call).
Thus, we also present the transmission delay of the network,
which does not include the routing delay.
B. Effects of the Network Size
Firstly, we illustrate the dependency between the net-
work size and the voice quality. For this, we evaluated a
small (50 nodes in a 1050m·1050marea), medium (100
nodes in a 1500m·1500marea) and large (500 nodes in a
3300m·3300marea) scenario. The area was chosen such that
the average number of neighbors per node (i.e. the probability
for a connected network) is not affected.
With an increasing network size, in terms of nodes, we
observed an increased route length and with this, an increased
number of packet collisions and mobility-induced route breaks.
As a result, the packet loss rate also increased, as shown in
TAB LE I
ROUTE LENGTH,DELAY,AND JITTER FOR THE VARIATION OF THE
NETWORK SIZE
Nodes Route length Overall delay Trans. delay Jitter
50 2.53 76.5 ms 24.9 ms 4.5 ms
100 3.50 160 ms 32.8 ms 11.5 ms
500 7.40 584 ms 73.2 ms 48 ms

0
20
40
60
80
100
20 40 100 160
loss rate [%]
frame size [ms]
100 nodes
500 nodes
(a) Variation of the frame size
0
20
40
60
80
100
5 10 15 20
loss rate [%]
number of parallel streams
100 nodes
500 nodes
(b) Variation of the network load (20ms frames)
0
20
40
60
80
100
5 10 15 20
loss rate [%]
number of parallel streams
100 nodes
500 nodes
(c) Variation of the network load (100ms frames)
Fig. 2. Packet loss for the variation of the frame size and the number of parallel streams
1
2
3
4
5
20 40 100 160
voice quality [MOS]
frame size [ms]
100 nodes
500 nodes
(a) Variation of the frame size
1
2
3
4
5
5 10 15 20
voice quality [MOS]
number of parallel streams
100 nodes
500 nodes
(b) Variation of the network load (20ms frames)
1
2
3
4
5
5 10 15 20
voice quality [MOS]
number of parallel streams
100 nodes
500 nodes
(c) Variation of the network load (100ms frames)
Fig. 3. Voice Quality for the variation of the frame size and the number of parallel streams
Figure 1(a). Due to this, the voice quality rapidly deteriorates
as the network size increases. As shown in Figure 1(b), a
good voice quality (3.7 MOS) was achieved in the small-scale
scenario. The medium-scale and large-scale scenarios showed
a heavily reduced voice quality due to the increased loss rate.
The delay and jitter were affected in a similar way as shown
in Table I. The overall end-to-end delay of the large-scale
scenario was above the 500ms limit that was considered ac-
ceptable for voice communication: yet, the transmission delay
of the network was acceptable for voice transmission. The
small and medium-scale scenarios showed both an acceptable
overall delay and transmission delay. For all scenarios, the
jitter was between 4.5ms and 48ms and could, thus, be
completely compensated by the jitter buffer.
C. Effects of the Frame Size
The G.711 codec enables a flexible selection of the frame
size, as no compression is performed. For the following, we
varied this parameter from 20ms to 160ms for both the
medium (100 nodes) and large (500 nodes) scenarios.
Up to the size of 100ms, an increased frame size positively
TAB LE II
DELAY FOR THE VARIATION OF THE F RAME SIZE
Frame 100 nodes 500 nodes
length Overall delay Trans. delay Overall delay Tran s. delay
20 160 ms 32.84 ms 584 ms 73.2 ms
40 163 ms 89.3 ms 686 ms 82.1 ms
100 269 ms 109.9 ms 726 ms 128.9 ms
160 548 ms 171.4 ms 1102 ms 224.9 ms
affected the loss rate, as can be seen in Figure 2(a). While the
loss rate in the large-scale scenario with 20ms frames was
about 54%, a frame size of 100ms reduced the loss by more
than 35%. Due to the modified loss rate, the MOS rate was
improved by 1.4points in the large-scale scenario as shown
in Figure 3(a). A similar effect was observed in the medium-
scale scenario. Altogether, adapting the frame size resulted in
a reduced loss rate and thus, in an increased voice quality.
On the downside, an increased frame size caused higher, yet
tolerable overall and transmission delays, as shown in Table
II.
D. Effects of the Network Load
As shown in the previous section, reducing the network load
by increasing the frame size resulted in a reasonable QoE
of voice communication in all scenarios. Thus, we directly
analyzed how the network load, in terms of parallel voice
streams, affected the QoE. Starting from the 5parallel streams
of the previous experiments, we increased the load up to 20
parallel streams. Again, we focused on the medium-scale and
on the large-scale scenario. For each scenario, we determined
TABLE III
TRANSMISSION DELAY FOR THE VARIATION OF THE NETWORK LOAD
Streams 100 nodes 500 nodes
20 ms 100 ms 20 ms 100 ms
532.8 ms 109.91 ms 73.2 ms 158.4 ms
10 87.6 ms 113.01 ms 113.4 ms 191.7 ms
15 151.7 ms 231.43 ms 135.2 ms 231.4 ms
20 76.60 ms 235.60 ms 127.2 ms 235.6 ms

0
20
40
60
80
100
G.711 Speex
loss rate [%]
codec
100 nodes
500 nodes
(a) Variation of the network size
0
20
40
60
80
100
20 ms 100 ms
loss rate [%]
frame size [ms]
100 nodes
500 nodes
(b) Variation of the frame size
0
20
40
60
80
100
5 10 15 20
loss rate [%]
number of parallel streams
500 nodes
100 nodes
(c) Variation of the network load
Fig. 4. Packet loss for varied network parameters with the Speex codec
1
2
3
4
5
G.711 Speex
voice quality [MOS]
codec
100 nodes
500 nodes
(a) Variation of the network size
1
2
3
4
5
20 ms 100 ms
voice quality [MOS]
frame size [ms]
100 nodes
500 nodes
(b) Variation of the frame size
1
2
3
4
5
5 10 15 20
voice quality [MOS]
number of parallel streams
500 nodes
100 nodes
(c) Variation of the network load
Fig. 5. Voice quality for varied network parameters with the Speex codec
the influence of the network load for both the standard frame
size of 20ms of G.711 and the 100ms optimum.
Increasing the network load led to an increased packet loss,
due to collisions, in both scenarios. The effect was amplified
for a small frame size of 20ms, while 100ms frames provided
low packet loss even in scenarios with increased network load,
as shown in Figures 2(b), 2(c), 3(b) and 3(c). The large-scale
scenario still showed a tolerable loss rate and a fair voice
quality for up to 10 parallel streams and a frame size of
100ms. In the medium-scale setup, with 100ms frames, the
network load could be increased up to 15 parallel streams,
while maintaining a feasible packet loss and a good voice
quality. What stands out here was that the packet loss caused
by the network for 15 parallel streams in the large-scale
scenario was about the same as the loss for 5parallel streams
in the medium-scale scenario. Yet, the voice quality for these
settings differed strongly, whereas within the experiments
regarding the frame size presented in the previous section,
a comparable loss rate for the medium-scale and large-scale
scenarios resulted in a comparable voice quality. This could be
explained by the fact that the jitter in the large-scale scenario
TAB LE IV
OVERALL DELAY FOR THE VARIATION OF THE NETWORK LOAD
Streams 100 nodes 500 nodes
20 ms 100 ms 20 ms 100 ms
5160 ms 189 ms 583 ms 652 ms
10 285 ms 220 ms 673 ms 765 ms
15 473 ms 259 ms 750 ms 1009 ms
20 490 ms 578 ms 782 ms 1015 ms
exceeds the jitter-buffer of 100ms. Thus, in addition to the
packet loss that is caused by the network itself, packets have to
be discarded due to the jitter, which results in the deterioration
of voice quality.
The overall delay and the transmission delay for the varia-
tion of the network load are shown in Tables IV and III. Both
increased along with the network load. Yet, the transmission
delay was tolerable for all levels of network load.
E. Effects of the Voice Codec
Up to now, we used the waveform codec G.711 for voice
coding. In this section, we compare G.711 to the hybrid
codec Speex. In contrast to G.711, Speex uses a compres-
sion algorithm, thus reducing bandwidth requirements and
also reducing the maximum voice quality achievable. Speex
allows for adaption of this compression algorithm in order to
further reduce the bandwidth requirement. Since our goal is to
optimize the voice quality, we used the highest quality setting
of Speex. With this, the bandwidth required is 25.6kbit/s
which is still less than half the bandwidth requirement of
G.711 (64kbit/s). Due to the compression algorithm, Speex
TAB LE V
TRANSMISSION DELAY FOR THE VARIATION OF THE NETWORK SIZE AND
THENETWORKLOADWITHTHESPEEX CODEC
Streams 100 nodes 500 nodes
5106.7 ms 137.0 ms
10 109.2 ms 142.8 ms
15 110.9 ms 167.0 ms
20 117.0 ms 188.0 ms

requires a fixed frame size of 20ms. For comparison to the
increased frame size of G.711 we thus sent multiple frames per
packet when we used the Speex codec. We compared G.711
and Speex for frame sizes / multiple frames per packet of
20ms and 100ms. We also compared both codecs subject to
the network load for the 100ms frames.
Since the maximum voice quality achievable is higher for
G.711 than for Speex, the voice quality of G.711 exceeded that
of Speex for the medium-scale scenario as shown in Figure
5(a). Yet, in the large-scale scenario, the reduced bandwidth
requirement of Speex led to a better voice quality compared
to G.711, due to a lower packet loss as shown in Figure 4(a).
The qualitative results for a variation of the network’s size
and workload when using Speex were comparable to the
results for G.711, as presented in the previous sections. For
both codecs, the increased packet loss and thus, the reduced
voice quality could be compensated by increasing the frame
size or the number of frames per packet, respectively, as shown
in Figures 5(b) and 4(b).
Also, due to the low bandwidth requirement of Speex, a
higher network load could be applied. As shown in Figure
5(c), the voice quality was always between fair and good
for the medium-scale scenario. For the large-scale scenario,
a fair voice quality could be maintained for a load of up to
15 parallel streams. The packet loss, as shown in Figure 4(c),
supports this result. Again, it stands out that a comparable
packet loss did not lead to a comparable voice quality in the
different scenarios. As for the results presented in the previous
section, we ascribe this to a jitter that exceeded the 100ms
jitter buffer. The effect was also slightly amplified by the fact
that sending multiple frames per packet resulted in a more
bulky packet loss compared to the statistically distributed loss,
which was observed if one frame was sent per packet. With
respect to the resulting voice quality, Speex was affected to a
greater extent by this than G.711, due to the different modes
of operation of the codecs.
The transmission delay, shown in Table V, when using
Speex was slightly lower than for G.711: this is due to the
reduced network load leading to a lower congestion related
delay.
V. C ONCLUSION
In this paper, we analyzed major factors influencing the QoE
of voice communication in MANETs. We focused on: (1) the
network size in terms of nodes and geographical expanse, (2)
the length of voice frames, (3) the network load in terms of
parallel voice streams and (4) the voice codec deployed.
Our results show that increasing the network’s size and
workload leads to a reduced voice quality due to an increased
packet loss rate. Since the network’s size and load are specified
by system requirements and user behavior, these parameters
cannot be adapted or limited unrestrictedly. To improve the
voice quality in large-scale and/or heavily loaded networks,
measures such as changing the codec deployed and increasing
the frame size can be taken. The choice of the codec directly
affects the maximum achievable voice quality. Due to the
individual bandwidth requirements of different codecs, the
network load is also affected. Increasing the frame size or
the number of frames per packet reduces network load and
protocol related overhead. To a certain extent, an adaptation of
the codec and the frame size can thus counterbalance possible
negative effects of large-scale and heavily loaded networks,
allowing for voice communication with a reasonable quality of
experience, in these scenarios. Furthermore, our studies show
that the statistical packet loss can only be used as an indicator
for the QoE. While packet loss always results in a weaker
QoE, scenarios presenting the same packet loss rate may differ
greatly in terms of voice quality.
REFERENCES
[1] N. Aschenbruck, P. Martini, and M. Gerharz. Human Mobility in
MANET Disaster Area Simulation. - A Realistic Approach. In Pro-
ceedings of the 29nd Annual IEEE International Conference on Local
Computer Networks (LCN 04), 2004.
[2] R. Barr. An efficient, unifying approach to simulation using virtual
machines. PhD thesis, Cornell University, 2004.
[3] H. Dong, I. Chakares, C. Lin, A. Gersho, E. Belding-Royer, U. Madhow,
and J. Gibson. Speech coding for mobile ad hoc networks. In
Proceedings of the Thirty-Seventh Asilomar Conference on Signals,
Systems and Computers (ACSSC 03), 2003.
[4] ETS 300 396-1. Terrestrial Trunked Radio (TETRA); Technical require-
ments for Direct Mode Operation (DMO); Part 1: General network
design. ETSI, 1998.
[5] L. Hanzo, C. Somerville, and J. Woodard. Voice and Audio Compression
for Wireless Communications. John Wiley & Sons, Inc., second edition,
2007.
[6] K. Kanchanasut, A. Tunpan, M. Awal, D. Das, T. Wongsaardsakul, and
Y. Tsuchimoto. A Multimedia Communication System for Collaborative
Emergency Response Operation in Disaster-affected Areas. Technical
report, Internet Education and Research Laboratory (intERLab), Asian
Institute of Technology (AIT), 2007.
[7] W. Lu, E. C. Peh, W. Seah, and Y. Ge. Communications Support for
Disaster Recovery Operations using Hybrid Mobile Ad-Hoc Networks.
In Proceedings of the 32rd Annual IEEE International Conference on
Local Computer Networks (LCN 07), 2007.
[8] C. E. Perkins, E. M. Belding-Royer, and S. R. Das. Ad hoc On-Demand
Distance Vector (AODV) Routing. IETF RFC 3561, 2003.
[9] M. R. Schroeder and B. S. Atal. Code-excited linear prediction(CELP):
High-quality speech at very low bit rates. In Proceedings of the IEEE
International Conference on Acoustics, Speech, and Signal Processing
(ICASSP 85), 1985.
[10] H. Schulzrinne and et al. RTP: A Transport Protocol for Real-Time
Applications. IETF RFC 3550, 2003.
[11] SERIES G. G.711: Pulse Code Modulation (PCM) of Voice Frequencies.
ITU-T, 1988.
[12] SERIES P. P.862: Perceptual Evaluation of Speech Quality (PESQ).
ITU-T, 2001.
[13] SERIES P. P.800.1 Mean Opinion Score (MOS) terminology.ITU-T,
2006.
[14] J. Valin. The Speex Codec Manual Version 1.2 Beta 3. Xiph.org
Foundation, December 2007.
[15] P. Velloso, M. Rubinstein, and O. Duarte. Analyzing voice transmission
capacity on ad hoc networks. In Proceedings of International Conference
on Communication Technology (ICCT 03), 2003.
[16] H. Zhang, J. Homer, G. Einicke, and K. Kubik. Performance Comparison
and Analysis of Voice Communication over Ad Hoc Network. In
Proceedings of the 1st Australian Conference on Wireless Broadband
and Ultra Wideband Communications (AusWireless 06), 2006.