Ad Hoc Networks 11 (2013) 1484–1496

Contents lists available at ScienceDirect

Ad Hoc Networks

journal homepage: www.elsevier .com/locate /adhoc

Real-time wireless multi-hop protocol in undergroundvoice communication

Domenico Sicignano ⇑, Danilo Tardioli, Samuel Cabrero, José L. VillarroelRobotic, Perception and Real Time group, Aragón Institute for Engineering Research (I3A), Universidad de Zaragoza, Zaragoza, Spain

a r t i c l e i n f o

Article history:Available online 5 February 2011

Keywords:MANETReal-time networksUnderground environmentQoS

1570-8705/$ - see front matter � 2011 Elsevier B.Vdoi:10.1016/j.adhoc.2011.01.017

⇑ Corresponding author. Tel.: +34 976 762347.E-mail addresses: doxsi@unizar.es (D. Sicignano

(D. Tardioli), samuelcabrero@gmail.com (S. Cabre(J.L. Villarroel).

a b s t r a c t

The underground communication in tunnels and mines is very challenging due to the hos-tile nature of the environments and to the propagation issues that electromagnetic wavessuffer there. Communication is often unidirectional (e.g. in mines) or very costly (e.g. leakyfeeder in road tunnels) and hard to install and maintain. This work proposes the use ofmulti-hop ad-hoc networks to provide multimedia communication between mobile nodesin such a hostile environments, relying on a complete hardware/software, cheap and easy-to-setup platform that can be used both as temporary or fixed infrastructure or as commu-nication backbone in emergency scenarios like mine accidents or a tunnel collapse. Thecommunication is based on the Real-Time Multi-hop Protocol (RT-WMP) and its QoSextension executed over several nodes equipped with specific hardware. This protocolmanages delay sensitive messages and the node mobility across the network while theQoS extension is responsible for allowing the end-to-end voice communication. The spe-cific topology and situation have driven to a specialization of RT-WMP to better performin this type of environments, taking advantage of the a priori (partial) knowledge aboutthe topology. This proposal was tested in a real application in the Somport tunnel, theabout 8 km-long railroad linking Canfranc, Spain with Pau, France.

� 2011 Elsevier B.V. All rights reserved.

1. Introduction

Normal radio systems can provide at best a very limitedand ineffectual communications capability in confinedspaces. As a result, the problem of providing communica-tions capability in hostile underground environments suchas tunnels or mines, both for exploitation and emergencysituations has been an important issue in recent decadesspecially after a series of unfortunate underground disas-ters and accident.

Many researcher have started to investigate wave prop-agation in confined environments to discover the reason ofsuch a strange behavior of electromagnetic (EM) waves

. All rights reserved.

), dantard@unizar.esro), jlvilla@unizar.es

and several studies about EM wave propagation, channelcharacterization and measurements [1–4] have been car-ried out in the last decade.

These studies have driven to implement and offercommercial solutions to underground and tunnel communi-cation. As is well known, today the most widely used under-ground communication system is the Leaky Feeder (LF).However, LF based networks are very costly to deploy, main-tain and, in addition, lack standardization. Moreover, in anemergency scenario (a collapse or similar) a communicationsystem as LF based could be damaged and become useless.

Recently, wireless ad-hoc networks have been consid-ered as an alternative solution [5]. With its features of flex-ibility (nodes can be installed/moved/removed andreplaced easily) and low deployment costs, wireless sys-tems using the 802.11 standard can offer a suitable wayto interconnect nodes in underground mesh networks.However, conventional 802.11 networks can experience

D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496 1485

severe performance issues when supporting delay sensi-tive applications such as Voice over IP (VoIP), since itsMedia Access Control (MAC) layer is based on competitionmechanisms (CSMA/CA) and no priority-based accessschemes have been implemented (even if present in thestandard). Moreover the 802.11 protocol have not been de-signed to support multi-hop communication [6] andmobility is managed only within a single collision domain.

This work proposes the use of a real-time protocol forMobile Ad-hoc NETworks (MANETs) to carry out under-ground communication and guarantee voice communica-tion in this type of environments. In order to achieve thispurpose, we used the Real-Time Wireless Multi-hop Proto-col (RT-WMP) [7]. This is a protocol for MANETs that sup-ports real-time traffic while managing natively nodemobility. Recently, the protocol has been modified to addthe capability of managing multimedia and Quality of Ser-vice (QoS) traffic [8]. It is therefore capable of merging thereal-time traffic and human communication such as videoand voice. The proposal includes a complete hardware/software platform. The idea is to create a flexible infra-structure constituted by a set of backbone nodes placedin the confined environment at strategic points and a setof mobile nodes that can move maintaining the voice linkamong them.

The rest of the paper is organized as follows. In Section2 we present related works. Section 3 provides a brief re-view of the basic features of the RT-WMP protocol andits QoS extension. In Section 4, we show how the protocolhas been adapted to the special environment. Section 5provides a description of the environment, and the resultof the evaluation of the proposed scheme. Finally, Section6 sets out the conclusions and proposed future work.

2. Related works

The 802.11 networks have been extensively tested inoutdoor and indoor areas but very few performance mea-surements have been performed in confined environmentsuch as tunnels or mines. Literature offers some work thathave been focused on the 2.4 GHz band in order to allowcompatibility with WLAN systems.

Nerguizian et al. [9] offer an extensive channel charac-terization through the measurements of the delay spreadand the coherence bandwidth in a mine. The results showthat channel does not follow a dual-slope relation with re-spect the distance. Benzakour et al. [10] offer a similarchannel measurements, analyzing both the 2.4 GHz and5.8 GHz bands. Their results show that indoor multipathcharacteristics can strongly depend on the nodes separa-tion and on the dimensions of the gallery.

Other works, based on 802.11 networks, analyze differ-ent aspects of the underground multimedia communica-tion. Beaudoin et al. [11] propose an approach for videotransmission in a mine providing statistics about packetlosses suffered. Moutairou et al. [12] present a work thatproposes a technique to optimize the mesh access pointslocation. Aniss et al. [13] proposes a hybrid solution thattries to adapt the 802.11 standard to underground commu-nications. The network is a combination of the wireless

standard with the DOCSIS data-over-cable standard usedas a backbone.

These solutions try to adapt the 802.11 standard to theunderground without taking into account, however, itslimitations in terms of multi-hop routing and Quality ofService (QoS) support.

This last issue has been deeply studied thanks to thegrowing interest on offering multimedia contents in MAN-ET networks. Protocols like 802.11e, for example, try tointroduce a certain degree of determinism at the MAC layerprioritizing traffic as a function of its type and class butagain in a single-hop scenario. Other researchers proposeinteresting solution to offer QoS guarantees over multi-hop networks modifying this protocol. Hamidian and Krner[14] add a QoS mechanism to the enhanced distributedchannel access (EDCA) scheme to allow a resource reserva-tion while [15] propose a scheme to prioritize the packetsusing a combination of the packet deadline and the num-ber of hops to the destination node to give higher priorityto the packets that have to traverse many hops.

Other solutions propose the use of token-passingschemes to support the transportation of time-sensitivedata, thanks to its deterministic access to the medium,robustness against single node failure or support for flexi-ble topologies.

The most part of them implement similar MAC opera-tions. The network generates a unique token to permit onlythe node that currently holds it to transmit data. Ergenet al. [16] show the potential of achieving higher channelutilization using a token scheme with respect to CSMAbased schemes. More recently, Wang and Zhuang [17] haveanalyzed the advantage of token based schemes with re-spect to contention based and to centralized pollingschemes to provide guaranteed priority for different trafficclasses in WLAN. A similar analysis is conduced by Zhanget al. [18] that shows how a token ring scheme applied invehicular ad-hoc networks can outperform IEEE 802.11DCF in terms of the average throughput.

Based on the ideas of 802.4 token bus protocol, Lee et al.[19] propose the WTRP, a token ring network in which eachnode can transmit for a fixed time when is token owner.Although multiple rings are allowed, a node can only com-municate with its neighbors, so the topology of the networkis limited. The WDTP [20] modifies the method to controlthe token transfer scheme of the WTRP. All nodes are clus-tered into subnets and the nodes of a subnets share a chan-nel. This improves the adaptability to network topology butthe number of used channels increase. Some proposal arebased on hybrid MAC Token CDMA policing mechanism.

Taheri and Scaglione’s proposal [21] is based on a ringnetwork where each token corresponds to a physical CDMAsubchannel which is guaranteed to have a certain averagerate and satisfy a probability of error bound by identifyingtwo classes of service to give priority to QoS traffic.

3. RT-WMP overview

The Real-Time Wireless Multi-hop Protocol (RT-WMP)is a protocol for MANETs. It works on top of the 802.11 pro-tocol and supports real-time traffic (see [7]).

Fig. 1. A hypothetical situation described by the network graph and the correspondent LQM. The hops sequence of the protocol is also shown.

Fig. 2. Time intervals used by the QoS Extension.

1486 D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496

The protocol works in three phases: the Priority Arbitra-tion Phase (PAP), the Authorization Transmission Phase (ATP)and the Message Transmission Phase (MTP). During the PAP,the nodes reach a consensus over which of them holds theMost Priority Message (MPM) in the network at that mo-ment. Subsequently, in the ATP, an authorization to trans-mit is sent to the node which holds the highest prioritymessage. Finally, in the MTP, this node sends the messageto the destination node (see Fig. 1).

To describe the topology of the network, RT-WMP de-fines an extension of the network connectivity graph add-ing non negative values at the edges of the graph. Thesevalues are calculated as functions of the radio signal be-tween pairs of nodes and are indicators of the link qualitybetween them. These values are represented in a matrixcalled the Link Quality Matrix (LQM).

To reach a consensus over the node that holds the high-est priority message, during the PAP a token travelsthrough all of the nodes. The token holds information onthe priority level of the MPM in the network. The last nodereceiving the token, which knows the identity of the MPMholder, closes the PAP and initiates the ATP. In this phase,the node calculates a path to the MPM holder using thetopology contained in the LQM and sends an authorizationmessage to the destination using the intermediate nodes.When the message reaches the destination node, theMTP starts. The node calculates the path to the destination,and sends the message in the same way as in the ATP. Thephases are repeated one after another.

The succession of the events that bring to the delivery ofa message (that is a succession of PAP, ATP and MTP or asuccession of a PAP and MTP or even a single PAP) is calledRT-WMP loop.

3.1. The QoS extension

The basic RT-WMP only supports real-time fixed prior-ities. However, QoS traffic needs variable priorities sincethe routing algorithm must take into account that a pack-ets priority varies as a function of the time. We developedtherefore a QoS Extension for the RT-WMP that, withoutaltering its real-time characteristics, allows the transporta-tion of variable priority traffic, using the time remainingwhen the RT-WMP loop is not a worst-case one. The basicconcept is shown in Fig. 2. As explained, the RT-WMPphases repeat one after the other as well as the loops. Beinga real-time protocol, all the phases (and thus the loops)

have a bounded and know duration. It implies that aworst-case loop tloop(wc) can be calculated. The worst-caseend-to-end delivery delay (that must be taken into accountat planning time) is strictly related with this value[7]. Theworst-case loops are, however, unlikely (but not impossi-ble) to take place in the reality since only can happen un-der certain circumstances and network topologies. The RT-WMP loop lasts in general tj 6 tloop(wc). The idea of theQoS extension is to take advantage of the free time Dj be-tween the worst-case loop and the actual loop to sendQoS data in this interval within worsening the RT-WMPworst-case end-to-end delay[8]. As the basic RT-WMP,the QoS Extension articulates in three phases: a QoS arbi-tration phase, a QoS Authorization Phase (QAP) and a QoSMessage Phase (QMP). The QAP and QMP can repeat oneafter the other for a limited numbers of times. The arbitra-tion phase is carried out during and in a similar way as thePAP, while the QAP and QMP are added to the basicprotocol.

In the arbitration phase, all the nodes which have a QoSmessage to send compete to gain the right to send it (dur-ing the PAP the token reaches all the nodes). One or moremessages can be selected for transmission depending onthe deadline and on the distance between the source nodeand the destination. The first QAP starts when the standardMTP ends (or after the PAP if there is no real-time messageto be sent). The node which ends the MTP (or the PAP), in-stead of restarting the successive PAP, sends an authoriza-tion to the owner of the first selected message, using thesame scheme used by the RT-WMP. The latter then startsthe QMP and sends the QoS message to the destinationnode. Successively, if there are other messages to be sent,it prepares an authorization and starts an additional QAPduring which the authorization reaches the selected mes-sage holder. The latter, in turn, sends its message duringa further QMP and so on. As anticipated, the QAP andQMP repeat one after another for a limited (and configura-ble) number of times, but in any case they stop when theworst-case loop time is reached. If during a QTP the worst-

D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496 1487

case loop time is reached, the transported message isstored in an intermediate node to be able to compete inthe successive PAP to be selected again.

3.2. Message priority policy

Timing guarantees can be considered as a fundamentalfeature for MANETs capable of supporting QoS applica-tions. QoS traffic is usually coupled with timing constraints(deadlines), and a delay bounded service allows the proto-col to know whether it is able to meet the deadlines or not.This constraint states that each message have to be deliv-ered before its deadline. If it is not possible, the packetmust be discarded. The deadline value is only needed untilthe packet is either successfully transmitted or discarded.

The mechanism to label and update the deadline onevery QoS message is quite simple. Messages are labeledwith their max admitted deadline, when generated. Dead-line values are usually 150 ms for voice and 400 ms for vi-deo traffic that correspond to maximum end-to-end delayadmitted for multimedia traffic [22]. When this parameterreaches the zero value, the message expires. So, duringmulti-hop transmission, this value has to be properly up-dated in every hop to take in account the time spent ineach frame pass.

The QoS extension implements a packet scheduler thatassigns a dynamic priority to a packet taking into accountthe deadline and the number of hops left to the destina-tion, in this order. Messages are sorted using the laxity(in a similar manner as in [15]), a parameter that combinesthe deadline and the number of hops left to the destinationdefined as:

laxity ¼ deadline=hopleft: ð1Þ

In fact, the laxity gives us an estimate of how much de-lay the packet can tolerate at each hop. Hence, the packetwith the lowest value of laxity is given the highest priority.If two packets have the same lowest value of laxity, we re-solve the conflict by sending the packet which has morehops to travel. If the laxity becomes zero, the packet is dis-carded since it is useless at the destination. This techniquegrants higher level of fairness since gives some possibilityof delivering to messages independently from relative po-sition of source and destination.

Fig. 3. Alternative paths to

4. Specialization on the environment

The RT-WMP has been designed principally to supportreal-time communication in teams of robots and thus togive support to any type of topology that can appear dueto node mobility, and its routing algorithm was developedto work in this situation. However, the environment con-sidered here is quite different. We have basically a n back-bone nodes and m (=2, at the moment) mobile nodes, andwe can take advantage of this previous knowledge to de-sign a more efficient routing algorithm and improve themobility characteristics of RT-WMP.

4.1. Using the minimum spanning tree

In the RT-WMP, frames are routed using the LQM takinginto account either the link quality of the nodes and thedistance to the destination node. To each link is assigneda cost depending on the corresponding value of the LQM.The node chooses the less costly path to reach the destina-tion. Depending on the link-quality topology of the net-work, the chosen path can be the shortest but even thelonger. The heuristics behind the routing algorithm, in fact,tries to reach a compromise between the options of send-ing frames over weak links and using a longer path. Letconsider Fig. 3I. We have two options to deliver a message:if we chose the first path, the destination node is only a hopaway while in choosing the second destination is 4 hopsaway. The probability of error is:

pp1 ¼ ð1� p10Þ; ð2Þ

using path #1, while:

pp2 ¼ ð1� p12Þ � ð1� p23Þ � ð1� p34Þ � ð1� p40Þ; ð3Þ

using path #2, pxy being the probability of transmission er-ror in the link between node px and py that, in turn, is cal-culated as a function of the RSSI.

There could be therefore situations in which a shortestpath is considered safer than a longer one even if in the lat-ter, links are all stronger. This behavior is desirable in gen-eric networks (e.g. a team of robots uniformly distributed).However, in situations in which (part of) the topology isknown, the routing algorithm can be improved to obtaina safer behavior. In tunnels, for example, the backbonesnodes are placed in strategic points to guarantee a goodand, above all, symmetric link among them. Moreover,they use to have a high-gain antenna and the transmission

reach the same node.

Stable RT-WMP Link

QoS FlowUnstable RT-WMP Link

A B

A B

BB

Fig. 4. An illustration of mobility scheme.

1488 D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496

power is substantially greater than the power of the mo-bile nodes. It implies that the probability of communica-tion error between adjacent nodes is very low, almostnegligible.

We decided, thus, to force the protocol to use backbonenodes for backbone communication and mobile nodes tolink itself to the closest (in terms of link quality) backbonenode. In order to achieve that, we reduced the topology ofthe network to a spanning tree where only the best linksare selected or, more concretely, to a minimum spanningtree applying the Prims algorithm to the LQM. Fig. 3IIshows the results of this process. The backbone topologyof the network becomes a string and the node to commu-nicate using the best possible link with the backbone. Therationale is that the nodes see the network as a tree whosebranches are the best possible links. An example of howthis new scheme works is given in Fig. 4. As we can seein Fig. 4I, the link quality between nodes A and B enablesa stable connection between the mobiles nodes (A and B)and the backbone node while the weak link are ruled outby the Prim’s algorithm. While B is moving across the tun-nel (Fig. 4II), the protocol manages the link quality changeallowing multi-hop re-routing across the network andguaranteeing a connection all the time.

Since routing, as explained, is based on current and realsignal quality, in some situations (even if this is not com-mon) a mobile node can act as a bridge between two back-bone nodes.

5. Evaluation

The main experiment was performed in the Somporttunnel linking the old railway from Canfranc (Spain) toPau (France) situated in the Pyrenees. The tunnel (seeFig. 5) has a maximum height of 6 m, 4.7 m width (horse-shoe cross section) 7.7 km length and has a change in slopeat about 4.5 km from the spanish-side gate. The slope is

Fig. 5. An illustration of t

about 2% from the spanish gate to the change in slope (up-hill) and 3.4% from there to the French gate (downhill). Ithas several lateral galleries (about 400 m apart) along itsextension and the walls have roughness in the order ofabout 2 cm and some section of the tunnel has buried rails.The tunnel is closed to traffic, so we can assume the exper-iments were made in stationary conditions. Tests havebeen done along about 7.5 km.

Five nodes equipped with minimal embedded and ded-icated hardware (100 � 160 mm PcEngines ALIX3D3board, battery powered) and Atheros chipset-based wire-less cards and running the MaRTE OS [23] implementationof RT-WMP, were distributed along the tunnel and used asbackbone nodes. In addition, two laptop computers run-ning Linux OS were used as mobile nodes. The voice wassampled at 8 kHz and 16 bit per sample and was com-pressed using the speex [24] codec to obtain a full-duplexcommunication of 15 Kbps bandwidth for each flow. EachRT-WMP-QoS message contained four speex voice packets.The packets’ deadline was fixed to 150 ms following theITU-T recommendations [25]. The multi-hop chain topol-ogy is shown in Fig. 6 while Table 1 lists the parameter val-ues used in the tests.

Before performing the final test, however, we made aset of additional experiments to investigate the environ-ment in which the experiment had to take place and chosethe adequate parameters to obtain satisfactory results.

The first test was about the measurement of RSSI andDelay Spread along the tunnel to obtain information aboutthe best places where to put the backbone nodes avoidingzones affected by fading or interferences. The second set oftests have been performed to discover the best parametersto obtain a correct and effective voice communication interms of packet aggregation (the number of voice datapacket to be transmitted at a time) and reception queuesize. Finally we verified the new routing algorithm in an in-door test.

The next sections describe these preliminary test and itsresults.

5.1. RSSI and delay spread measurement

Several studies about EM wave propagation have shownthe substantial difference between tunnel and free-spacepropagation. Tunnel, in fact, acts as a waveguide with(imperfectly) reflecting walls: most of the power travelsin the lossless interior of the tunnel if this is sufficientlywide. As a waveguide, a tunnel has a cut-off frequency(that generally lies somewhere in the VHF band in roadtunnels) below which no effective propagation occurs[26]. In this case we are above such a cut-off frequency

he Somport tunnel.

Fig. 6. The environment.

Table 1Parameters used in the real tests.

Parameter Values

Scenario Frequency 2.412 GHzChannel rate 6 MbpsTx power 100 mW

Data Pkt size 160 ByteQoS flow rate 15 Kbps

Constraints Deadline 150 ms

D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496 1489

and is thus possible to take advantage of this effect to ob-tain greater communication ranges than in open space.

On the other hand, the communication is affected bythe multipath effect. The latter is the main factor responsi-ble for the fading phenomena that affects both the inten-sity and the quality of the signal that reaches a receiver.The first experiment was defined to evaluate the variationof the power sensed by the receiver in order to know theradio-signal and its variability along the tunnel and the ef-fect of the fading on the measurements. Thus we used aprocedure to measure the RSSI (Received Signal StrengthIndication) received by a mobile node while a fixed sourcewas transmitting. At the same time, we measured the RMSDelay Spread using the YellowJacket Tablet Wi–Fi analyzer(Berkeley Varitronics Systems).

The measurements were repeated every 25 m over3.2 km of the tunnel and the results are shown in Fig. 7.

0 500 1000 150

RSS

I (dB

m)

0 500 1000 150

distan

Del

ay S

prea

d [u

s]

−80

−70

−60

−50

−40

20

25

30

35

40

Fig. 7. RSSI and Delay Spread va

As we can see, it is possible to appreciate some typicaltunnel propagation effects, characterized by the fadingeffect.

As expected, the mean radio-signal decreases with dis-tance but the fading has a strong presence both in terms ofRSSI and Delay Spread that oscillates between 25 ns and38 ns aproximately, showing similar values to that calcu-lated in experiments carried out in similar environments(see [27] and [28], for example) and slight higher thanthe usual values that are around the 20 ns in empty tun-nels [29].

The RSSI is influenced also by the presence of lateralgalleries that affect the received signal, producing a sharpfall in the signal intensity in correspondence of themouths. On the other hand, the waveguide effect allowshigher RSSI values along the tunnel than in open space.

These aspects have been taken in account in the deploy-ment operations since we wished to provide an efficientmulti-hop coverage of the communication. The idea is todeploy backbone nodes in order to optimize the transmis-sion in the tunnel taking advantage of one of the peaks thatare visible in the figure and avoiding, instead, the valleys.

5.2. Parameters choosing

To obtain a continuous flow without cuts or interrup-tion, voice packet must be exchanged among mobile nodeswith and adequate frequency and within its deadline. Itmeans that if we are able to deliver a message each50 ms, for instance, the packet must contain at least50 ms of voice. On the contrary, if the packet containsinsufficient data (e.g. 20 ms of voice) the listener will hearsilence until the arrival of the subsequent packet (seeFig. 8). We have, thus, to consider the Inter-Arrival Time(IAT) that the communication network is able to provideto decide how much voice data have to be transportedwithin a single packet. We made a first indoor experimentto determine this. We arranged a seven nodes chain net-work (providing the nodes with a fake LQM) and saturatedthe network with two end-to-end QoS flows. Fig. 9 pre-sents both the distribution and the temporary shape of

0 2000 2500 3000

0 2000 2500 3000

ce (m)

lues sensed from receiver.

1490 D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496

the IAT of the voice packet. The image shows three majorpeaks, the last of which around the 70 ms and a very smallone at about 150 ms. It means that in general we must beable to send at least 70 ms of voice every 70 ms. Moreoverthe peaks tell us that in some rare circumstances we’ll re-ceive packets 150 ms apart. Since the speex codec, in theconfiguration used in these experiments, generates datapackets containing 20 ms of voice, we should send:

npackets ¼7020

� �¼ 4; ð4Þ

packets in each loop. On the other hand to absorb sporadichigh IATs, we should have a queue of:

nqueue ¼15020

� �¼ 8; ð5Þ

packets. This queue will introduce a delay of:

delayqueue ¼ nqueue � 20 ms ¼ 8 � 20 ms ¼ 160 ms; ð6Þ

Fig. 8. Relation between voice data and inter-arrival time.

0 20 40 60 8

Time

Occ

urre

nces

0

50

100

150

200

250

300

350

400

450

500

0 200 400Sam

Tim

e (m

s)

0

20

40

60

80

100

120

140

160

Fig. 9. Distribution and raw data of Inter-Arr

that must be added to the mouth-to-hear end-to-end delayof the packets. This value is however assumable as guaran-teed by the ITU-T recommendations [25].

We tested the goodness of these parameters with an-other indoor experiment, using two real voice flows(128 Kbps each one before compression, about 15 Kbpsafter speex compression). This time, a movement of oneof the nodes (node 0) was simulated through the dynamicmodification of the fake LQM. The RSSI provided to thenodes was calculated as a function of the simulated dis-tance and perturbed with a 20% of noise to obtain a similarsituation to the real. Moreover all the nodes ignored all theframes that, in a real situation, would not have receiveddue to the excessive distance.

Fig. 10 presents the results of the test. The histogramshows now several peaks which one most important is atabout 70 ms. The plot shows that the distribution approx-imately constant along the whole experiment even if thisgraph give us the information that the peak at 130 ms inthe previous figure corresponds principally to the first halfof the experiment while the one at 42 ms to the secondhalf. It is due to the reconfiguration of the network duringthe movement that promotes different delivery paths.

On the other hand, the analysis of the end-to-end delay(see Fig. 11) suggests that the packets honor its deadlinesince the most part of the packets are delivered in an inter-

0 100 120 140 160

(ms)

600 800 1000ple #

ival Time (IAT) for two saturated flows.

0 20 40 60 80 100 120 140 160 180

Time (ms)

Occ

urre

nces

0

200

400

600

800

1000

1200

0 1000 2000 3000 4000 5000 6000 7000

Tim

e (m

s)

Position (m)

0

50

100

150

200

250

Fig. 10. Distribution and raw data of Inter-Arrival Time (IAT).

0 20 40 60 80 100 120 140Time (ms)

Occ

urre

nces

0

100

200

300

400

500

600

700

800

Fig. 11. End-to-end delay distribution.

D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496 1491

val between few milliseconds and 100 ms guaranteeing thecorrect playback of the voice at the destination node. Asexpected, no packets were delivered beyond its deadline(150 ms).

5.3. RSSI and prim based routing

The same experiment gave us information about theeffectiveness of the RSSI and Prim based routing. Fig. 12

shows its behavior. The figure is referred to mobile node0 and shows the identity of the last-hop sender (that is,the identity of the node that delivered the message to node0) and the RSSI, considered as indicator of the link qualityin this article, with which the first listens the latter (seeFig. 13).

The two mobile nodes (0 and 6) start closely each otherand to the node 5. Several frames are exchanged directlyamong the node 0 and node 6. Then, node 0 starts moving

0 1000 2000 3000 4000 5000 6000 7000Position (m)

RSS

I (dB

m)

0

1

2

3

4

5

6

Nod

e Id

−95

Fig. 12. RSSI and Prim based routing simulation.

Fig. 13. Identity of the last-hop sender.

Table 2Position and relative RSSI of the nodes.

Position (m)

Nodes 1–2 �2000Nodes 2–3 �2000Nodes 3–4 �1200Nodes 4–5 �1500

Table 3Main testbed results.

Flow 1 Flow 2

PDR (%) 98.2 97.9MOS >3.5 >3.5

1492 D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496

toward the other end of the backbone. The link qualitywith node 6 falls and node 0 begins to exchange frameswith node 5 following the rules marked by the Prim’s algo-rithm. The same occurs with the subsequent nodes. Asexpected, the link quality with the sender is always main-tained at acceptable values. The algorithm suffers fromshort oscillations in the switching point due to RSSI noisethat are, however, completely assumable by the system.In some situation, moreover, node 0 acts as a bridge be-tween adjacent nodes again due to the RSSI fluctuation.

5.4. Real experiment

The real experiment consisted in the deployment of thecited five nodes. Table 2 shows where the backbone nodeshave been deployed along the tunnel in order to grant anadequate inter-backbone node RSSI value. The third nodeposition, however, was chosen in the peak nearest to thetunnel slope change. In this way we guaranteed the pres-ence of Line-of-Sight (LoS) between each couple of nodes.Especially, the third node act as relay between the left-sideand the right-side of the chain that have not LoS betweenthem due to the change in slope.

Two laptops running Linux OS were used as mobilenodes. The sampling of the voice signal was performedaccessing directly the/dev/dsp device and compressing320 byte of data (160 samples of 16 bit) to a 40 bytes speexpacket (20 ms of voice). Packets were aggregated in groupsof four and sent to the other mobile node and as in the lab-oratory experiment, the QoS extension was configured to

transport up to two QoS messages in each protocol loop(see [8] for details). One of the mobile nodes was main-tained still at about 400 m apart from one of the backboneend while the other was moved, within a car, toward theother end of the backbone maintaining in any momentthe voice link. The movement speed was about 40 km/h.

The most notable parameters for evaluating voice trans-mission are the Packet Delivery Ratio (PDR), the end-to-end delay and the variance of the voice packet inter-arrivaltime (jitter). The following sections show the result of theexperiments.

5.4.1. PDR and MOSTable 3 lists the main results related to the characteris-

tics of the voice transmission obtained in the real experi-ments when the two mobile nodes were communicatingwith each other. Packet Delivery Ratio (PDR) is a measureof the percentage of packets that reach the destination.The PDR is calculated as the ratio of the number of packetsreceived within the deadline by the destination applicationlayer, and the number of packets sent by the applicationlayer at the source node. In our tests, we registered valuesaround 98% during the whole duration of the test. With

0 50 100 150 200 250 300

Time (ms)

Occ

urre

nces

0

100

200

300

400

500

600

700

800

900

1000

0 1000 2000 3000 4000 5000 6000 7000

Tim

e (m

s)

Position (m)

0

50

100

150

200

0

50

100

150

200

0

Fig. 14. Distribution and raw data of Inter-Arrival Time (IAT) in the real experiment.

D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496 1493

this level of PDR, speex audio codec guarantee a MeanOpinion Square (MOS) greater than 3.5, which was approx-imately the MOS level that we achieved during the test.This value is considered fair (imperfections can be per-ceived but the sound remains clear).

5.4.2. Delay and jitterThe IAT in the real experiment (see Fig. 14) is quite sim-

ilar to the one obtained in the indoor experiment even ifwe can notice a little widening of the distribution due tothe presence of a little percentage of discarded packet(about 2% as anticipated earlier). The analysis of the sameparameter as a function of the time presents also a similarbehavior but again in a wider range due to the fact thenodes that were not in a virtual chain but were free ofcommunicating among them following the routing algo-rithm based on real link quality.

Fig. 15 shows the distribution of the end-to-end (frommouth-to-hear) delay obtained during the real experiment.Again, the shape is a little wider due to the movement ofthe node along the tunnel but conserve the behavior ofthe simulation experiments. The most part of packets weredelivered within 100 ms of its creation, honoring comfort-ably its deadlines.

5.4.3. RSSI and prim based routingFig. 16, tries to illustrate the effectiveness of the Prim

and RSSI based routing algorithm. As in Fig. 12, the redline shows which (backbone) node have delivered themessage containing the voice data to the mobile node(node 0) and the link quality among them. As can beseen, at the beginning frames where directly exchangedbetween the mobile nodes 0 and 6 (due to the fact thatthey were close each other) or through node 5. Whennode 0 started to move towards the end of the backbone(node 1), the routing algorithm adapted itself to pro-vide always a good delivery path. This is reflected bythe fact that the last-hop was executed by differentnodes during the movement along about 7.5 km of thetunnel.

The graph shows, despite the high level of noise that isusual in RSSI measurement, the good work of the routingalgorithm specially thanks to the introduction of the Primalgorithm. In fact, it promotes the exchange of dataamong the mobile nodes and the closest (from the linkquality point of view) backbone node. As can be seen,the RSSI is maintained above the value of �60 dBm. Thisvalue is considered high enough to guarantee a reliablelink.

0 50 100 1500

100

200

300

400

500

600

700

800

Occ

urre

nces

Time (ms)

Fig. 15. Distribution of end-to-end delays in the real experiment.

0 1000 2000 3000 4000 5000 6000 70000

1

2

3

4

5

6

−95

−60

−35

RSS

I (dB

m)

Position (m)N

ode

IdFig. 16. Identity of the node that delivered the data packet to the mobile node and the RSSI value with which destination node has received the frame.

1494 D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496

5.4.4. Traffic influenceThe most part of the experiments have been carried out

in an empty tunnel in which the only vehicle involved wasthe one that transported the mobile node. However, to ver-ify the influence of the traffic on the communication, wecarried out an additional experiment simulating the pres-ence of a light traffic. We positioned two backbone nodes1.2 km apart each other (with LoS among them) and gener-ated a simulated voice flow. After recording in both nodesthe frames exchanged during two minutes without obsta-cles, we sent a first car from the position of the first nodeto the position of the second one at a speed of approxi-mately 35 km/h. Successively two cars, few meters aparteach other, traveled the same route at the same speed.

The cars never closed the LoS between that backbonenodes since the letters were positioned close to the tunnelwall. It is assumable since in a hypothetical permanentinstallation they would be secured on the top of the tunnelwithout possibility of losing LoS.

The data collected showed that, in this configuration,the traffic only has a little influence on the communication.

The PDR fell from 99.86% in the obstacle free experiment tojust 99.84% with one car and 99.832% with two cars.

6. Conclusions

This work addresses the problem of allowing multime-dia QoS communication between mobile nodes in under-ground environments (tunnels, mines, etc.). The schemeproposes the deployment of a set of backbone nodes alongthe confined area that act as relay for QoS data that mobilenodes exchange during its movement.

It is based on a complete software/hardware networkarchitecture using low-cost commercial hardware and run-ning the RT-WMP protocol and its QoS extension.

Thereby, the latter has been specialized to takeadvantage of the a priori knowledge of the networktopology.

The whole system was finally tested in the Somporttunnel (linking the old railway from Canfranc, Spain toPau, France) at the presence of the tunnel crew, the

D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496 1495

director of the road unit of the Huesca province andrepresents of the Spanish Ministry of Public Works withsatisfactory results, in August, 23, 2009 and again in July,19, 2010.

The real experiments show that it is a valid, flexible andeasy-to-setup solution for supporting QoS flows in tunnels,mines or disaster zones where the use of an infrastructurenetwork is impossible or too expensive.

Acknowledgments

This work has been funded by the projects TESSEODPI2009-08126 (MCYT, Spanish Gov.), 2008/0486 (CyT,Aragn Gov.) and by the agreement between the IAF (AragnGov.) and the I3A (University of Zaragoza) regarding theWALQA research laboratory, 2008/0574.

References

[1] P. Delogne, Em propagation in tunnels, IEEE Transactions onAntennas and Propagation 39 (1991) 401–406.

[2] M. Lienard, P. Degauque, Natural wave propagation in mineenvironments, IEEE Transactions on Antennas and Propagation 48(2000) 1326–1339.

[3] D. Dudley, Wireless propagation in circular tunnels, IEEETransactions on Antennas and Propagation 53 (2005) 435–441.

[4] D. Dudley, M. Lienard, S. Mahmoud, P. Degauque, Wirelesspropagation in tunnels, IEEE Antennas and Propagation Magazine49 (2007) 11–26.

[5] S. Yarkan, S. Guzelgoz, H. Arslan, R. Murphy, Underground minecommunications: a survey, IEEE Communications Surveys &Tutorials 11 (2009) 125–142.

[6] S. Xu, T. Saadawi, Does the ieee 802.11 mac protocol work well inmultihop wireless ad hoc networks?, IEEE CommunicationsMagazine 39 (2001) 130–137

[7] D. Tardioli, J.L. Villarroel, Real-time communications over 802.11:RT-WMP, in: The Fourth IEEE International Conference on MobileAd-hoc and Sensor Systems, 2007.

[8] D. Sicignano, D. Tardioli, J.L. Villarroel, Qos over real time wirelessmultihop network, in: J. Zheng et al. (Eds.), ADHOCNETS 2009, vol.28, LNICST, 2009, pp. 110–128.

[9] C. Nerguizian, C. Despins, S. Affes, M. Djadel, Radio-channelcharacterization of an underground mine at 2.4 GHz, IEEETransactions on Wireless Communications 4 (2005) 2441–2453.

[10] A. Benzakour, S. Affes, C. Despins, P.-M. Tardif, Widebandmeasurements of channel characteristics at 2.4 and 5.8 GHz inunderground mining environments, in: IEEE 60thVehicularTechnology Conference, 2004 (VTC2004-Fall), vol. 5, 2004, pp.3595–3599.

[11] J. Beaudoin, G. Tran, P.-M. Tardif, P. Fortier, Undergroundexperiments of video transmission over an ieee 802.11infrastructure, in: IEEE 60th Vehicular Technology Conference,2004 (VTC2004-Fall), vol. 5, 2004, pp. 3610–3614.

[12] M. Moutairou, H. Aniss, G. Delisle, Wireless mesh access pointrouting for efficient communication in underground mine, in:Antennas and Propagation Society International Symposium 2006,IEEE, 2006, pp. 577–580.

[13] H. Aniss, P.-M. Tardif, R. Ouedraogo, P. Fortier, Communicationsnetwork for underground mines based on the ieee 802.11 and docsisstandards, in: IEEE 60th Vehicular Technology Conference, 2004(VTC2004-Fall), vol. 5, 2004, pp. 3605–3609.

[14] A. Hamidian, U. Krner, Providing QoS in ad hoc networks withdistributed resource reservation, in: 20th International TeletrafficCongress (ITC-20), Ottawa, Canada, 2007.

[15] T.B. Reddy, J.P. John, C.S.R. Murthy, Providing mac qos formultimedia traffic in 802.11e based multi-hop ad hoc wirelessnetworks, Computer Networks 51 (2007) 153–176.

[16] M. Ergen, D. Lee, R. Sengupta, P. Varaiya, Wireless token ringprotocol-performance comparison with ieee 802.11, in: Proceedingsof the Eighth IEEE International Symposium on Computers andCommunication, 2003 (ISCC 2003), vol. 2, 2003, pp. 710–715.

[17] P. Wang, W. Zhuang, A token-based scheduling scheme for wlansand its performance analysis, in: IEEE International Conference onCommunications, 2007 (ICC ’07), 2007, pp. 3716–3721.

[18] J. Zhang, K.H. Liu, X. Shen, A novel overlay token ring protocol forinter-vehicle communication, in: IEEE International Conference onCommunications, 2008 (ICC ’08), 2008, pp. 4904–4909.

[19] D. Lee, R. Attias, A. Puri, R. Sengupta, S. Tripakis, P. Varaiya, Awireless token ring protocol for ad-hoc networks, in: AerospaceConference Proceedings, IEEE, 2002.

[20] H. Zhai, J. Wang, Y. Fang, Ducha: A new dual-channel mac protocolfor multihop ad hoc networks, IEEE Transactions on WirelessCommunications 5 (2006) 3224–3233.

[21] S. Taheri, A. Scaglione, Token enabled multiple access (tema) forpacket transmission in high bit rate wireless local area networks, in:IEEE International Conference on Communications, 2002 (ICC 2002),2002, pp. 1913–1917.

[22] B. Vlaovic, Z. Brezocnik, Packet based telephony, in: InternationalConference on Trends in Communications (EUROCON’2001), vol. 1,2001, pp. 210–213.

[23] M.A. Rivas, M.G. Harbour, Marte os: An ada kernel for real-timeembedded applications, in: Proceedings of the InternationalConference on Reliable Software Technologies (Ada-Europe-2001)2001.

[24] J. Valin, The speex codec manual the speex project, March 2003,<http://www.speex.org>.

[25] One-way transmission time, in: ITU-T Recommendation G.114,2003.

[26] S. Mahmoud, Electromagnetic Waveguides: Theory and Application,IEE, 1991.

[27] P. Aikio, R. Gruber, P. Vainikainen, Wideband radio channelmeasurements for train tunnels, in: 48th IEEE VehicularTechnology Conference, 1998 (VTC 98), vol. 1, 1998, pp. 460–464.

[28] M. Lienard, S. Betrencourt, P. Degauque, Theoretical andexperimental approach of the propagation at 2.5 GHz and 10 GHzin straight and curved tunnels, in: IEEE VTS 50th VehicularTechnology Conference, 1999 (VTC 1999-Fall), 1999.

[29] F. Molisch, Wireless Communications, IEEE, 2011.

Domenico Sicignano received the Engineer ofElectronic degree from the Universitá degliStudi di Salerno, Italy, in 2005 and the Com-puters Science master degree from the Uni-versidad de Zaragoza, Spain, in 2009. In 2007he joined the Group of Technology for hostileEnvironments (GTE), where he is currently aPhD student. His professional research activ-ity lies in the field of wireless communica-tions, mobile Ad-Hoc Networks and Quality ofServices.

Danilo Tardioli received the computer sci-ence engineer degree from the University ofBologna, Italy, on 2004 and the Ph.D. from theUniversity of Zaragoza in October 2010. Hisresearch activity are mainly focused in wire-less real-time and QoS communication inmobile ad-hoc networks in confined andhostile environments. He is member of theRobotics, Perception and Real-Time andTechnology in Hostile Environments groups ofthe University of Zaragoza.

1496 D. Sicignano et al. / Ad Hoc Networks 11 (2013) 1484–1496

Samuel Cabrero received the Engineeringdegree in Computer Science in June 2010. Hecompleted the final year project within thegroup of Robotics, Perception and Real-Timeat the University of Zaragoza. His researchactivity are mainly focused in wireless com-munication and embedded systems develop-ment.

José Luis Villarroel was born in Huesca,Spain, in 1961. He received the M.Sc. and thePh.D. degree in industrial engineering fromthe University of Zaragoza, Spain, in 1985 and1990, respectively. Since 2003 he has beenwith Aragón Institute for EngineeringResearch (I3A), Zaragoza, Spain, and since1997 he is head of the Group of Technologyfor hostile Environments (GTE). His researchinterests include Real-Time Systems, Under-ground Communications and Ad-Hoc Net-works. He is currently an Associate Professor

of Real-Time and Embedded Systems in the Department of ComputerScience and System Engineering, University of Zaragoza, Zaragoza, Spain.