Impact of AODV and OLSR routing protocols on the network ...Enfin, on a étudi é l’ampleur du...

Impact of AODV and OLSR routing protocols on the network cost of the Radio Environment Map Stephanie M. Faint

Defence R&D Canada – Ottawa

Technical Memorandum DRDC Ottawa TM 2011-001

March 2011

Impact of AODV and OLSR routing protocolson the network cost of the Radio EnvironmentMapStephanie M. FaintCommunications Research Centre

Defence R&D Canada – OttawaTechnical MemorandumDRDC Ottawa TM 2011-001March 2011

Principal Author

Original signed by Stephanie M. Faint

Stephanie M. Faint

Approved by

Original signed by J. Schlesak

J. SchlesakHead/Radio Communications Technologies

Approved for release by

Original signed by C. McMillan

C. McMillanHead/Document Review Panel

c© Her Majesty the Queen in Right of Canada as represented by the Minister of NationalDefence, 2011

c© Sa Majesté la Reine (en droit du Canada), telle que représentée par le ministre de laDéfense nationale, 2011

Abstract

Spectrum management using a radio environment map (REM) to dynamically allocate fre-quencies may facilitate the efficient exploitation of the spectrum in a heterogeneous en-vironment. The overhead costs of networking radios to create the map must be balancedagainst the impact of a poor REM due to low quality or quantity of sensor data. To studythe cost-benefit balance of two routing protocols, simulations were performed to evaluatethe transmission of REM data using Ad hoc On-Demand Distance Vector (AODV) routingand Optimized Link State Routing (OLSR). It was seen that both protocols caused signif-icant overhead in the system, effectively reducing the amount of data that could reach thedata collection centre, and thus potentially lowering the quality of the resulting REM.

Résumé

La gestion du spectre à l’aide d’une carte de l’environnement radio (CER) pour attribuer lesfréquences de façon dynamique peut faciliter l’exploitation efficace du spectre dans un en-vironnement hétérogène. Il faut comparer les coûts relatifs au surdébit de la mise en réseaude radios pour créer la carte avec les conséquences d’une mauvaise CER générée à partir dedonnées de détection insuffisantes ou de mauvaise qualité. Dans le but d’étudier les coûtset les avantages de deux protocoles de routage, on a effectué des simulations pour évaluerla transmission des données de CER à l’aide du protocole AOVD (Ad hoc On-DemandDistance Vector) et du protocole OLSR (Optimized Link State Routing). On a constatéque les deux protocoles causaient un surdébit important dans le système, ce qui réduisaitla quantité de données parvenant au centre de collecte de données et, conséquemment, laqualité de la CER produite.

DRDC Ottawa TM 2011-001 i

This page intentionally left blank.

ii DRDC Ottawa TM 2011-001

Executive summary

Impact of AODV and OLSR routing protocols on thenetwork cost of the Radio Environment Map

Stephanie M. Faint; DRDC Ottawa TM 2011-001; Defence R&D Canada – Ottawa;March 2011.

Background: Advances in wireless technology and the increasing demand on the avail-able spectrum require that radios function in a heterogeneous environment, necessitatingthe sharing of limited bandwidth over a large area while avoiding harmful interference. Itis therefore important to manage the spectrum dynamically by allocating frequencies basedon a user’s location and the current spectrum usage. One method of facilitating this sharedspectrum environment is the radio environment map (REM).

The REM contains information about the radio frequency (RF) environment. It indicatesthe areas where interference with or by other transmitters is possible. The intent of theREM is to create a map of the RF power levels of transmitters to be used by spectrummanagement throughout the region.

To create the REM, the collected RF data must be sent from each sensor to a single location.There are overhead costs to networking radios, such as loss of data, delay, and networktraffic; these costs must be balanced against the impact of a poor REM caused by lowquality or quantity of sensor data.

Principal results: To further study the cost-benefit balance of creating a REM, a series ofsimulations was performed in the OMNeT++ discrete event simulation engine using tworouting protocols: Ad hoc On-Demand Distance Vector (AODV) routing, and OptimizedLink State Routing (OLSR). These simulations were also performed both with and withoutTDMA, here modelled as start time delay.

The overall loss of data in each of the scenarios was studied. This loss was significant, asno sensor was able to communicate with HQ with more than 25% probability and in anyscenario, less than 50% of the sensors were able to connect. The AODV routing protocolwith start time delay allowed the most communication, while the same protocol withoutstart time delay performed the worst.

The delay in data traversing the network, measured in both time and number of hops, wasalso considered. Without start time delay, the AODV protocol had far less delay thanOLSR. However, adding the start time delay, as was necessary to promote communicationwith HQ, essentially eliminated this advantage. OLSR, on the other hand, was generallyable to connect to HQ with fewer hops than AODV required. While the AODV without start

DRDC Ottawa TM 2011-001 iii

time delay case was an exception to this, this exception only resulted due to the uniformlylow connection rate of that scenario.

Finally, the amount of traffic in the network was examined. Both routing protocols weresubject to a large number of collisions: with AODV this number increased as the number ofsensors increased, while with OLSR, the number eventually levelled off. OLSR, however,allowed an order of magnitude fewer messages to be successfully received at each sensorthan with AODV.

Significance of results: Neither of the two routing protocols performed satisfactorily, withor without start time delay. Thus, adding a routing protocol to the network is not sufficientto increase the amount of data available to the REM, and thus did not improve the qualityor timeliness of the REM. However, each of the two protocols had its advantages.

The selection of either AODV or OLSR would have to take into consideration the prioritiesof the system. If a higher number of connected sensors is the most important factor, theAODV protocol with start time delay should be used. If, however, minimizing networkdelay takes precedence, the OLSR protocol without start time delay would be the bestchoice. Of course, adjustments in the start time delay, or in the network’s communicationparameters, could yield different results.

Future work: One possibility to overcome the problems caused by having many nodesin the network is to combine one of the routing protocols with clustering. This wouldallow the routing protocol to be applied only to the cluster heads, minimizing the numberof routing messages that need to traverse the network, while the REM data from mostof the nodes would still reach HQ. This method would also have to use higher poweredtransmitters as the cluster heads, but could still use low powered sensors for most of thenodes, likely functioning on a different frequency from the cluster heads.

Further exploration of possible environments is also needed, particularly a non-flat earthsimulation, as well as experimentation with different simulated sensors and transmitters.Additionally, to better model a real world environment, mobility should be added to someor all of the sensors, and to the transmitters.

iv DRDC Ottawa TM 2011-001

Sommaire

Impact of AODV and OLSR routing protocols on thenetwork cost of the Radio Environment Map

Stephanie M. Faint ; DRDC Ottawa TM 2011-001 ; R & D pour la défense Canada– Ottawa ; mars 2011.

Contexte : Les progrès dans le domaine des technologies sans fil et la demande grandis-sante exercée sur les fréquences disponibles forcent les radios à fonctionner dans un envi-ronnement hétérogène exigeant le partage d’une bande passante restreinte sur de grandessuperficies tout en évitant le brouillage néfaste. Il est donc important de gérer le spectre defaçon dynamique en attribuant les fréquences selon l’emplacement de l’utilisateur et l’utili-sation actuelle du spectre. La carte de l’environnement radio (CER) est l’une des méthodesqui facilitent le partage du spectre.

La CER contient des renseignements sur l’environnement des radiofréquences (RF). Elleillustre les régions où peut survenir un brouillage causé par d’autres émetteurs ou uneinteraction avec ces émetteurs. Elle a pour but d’établir une carte des niveaux de puissanceradioélectrique des émetteurs à utiliser pour gérer le spectre dans une région donnée.

Pour créer la CER, il faut transmettre les données sur les RF recueillies par chaque capteurvers un seul endroit. La mise en réseau de radios comporte des coûts relatifs au surdébit,comme ceux liés à la perte de données, aux retards et au trafic du réseau. Il faut évaluer cescoûts en tenant compte des conséquences d’une mauvaise CER générée à partir de donnéesde détection insuffisantes ou de mauvaise qualité.

Résultats principaux : Dans le but d’approfondir l’étude comparative des coûts et desavantages de la création d’une CER, on a effectué une série de simulations à l’aide dumoteur de simulation d’événements discrets OMNeT++ et de deux protocoles de routage,soit le protocole AOVD (Ad hoc On-Demand Distance Vector) et le protocole OLSR (Op-timized Link State Routing). Ces simulations ont été faites avec et sans accès multiple parrépartition dans le temps (AMRT), modélisé ici sous la forme d’un retard du temps dedémarrage.

On a étudié la perte globale de données dans chaque scénario. Cette perte était impor-tante, et aucun capteur n’a pu communiquer avec le quartier général avec une probabilitésupérieure à 25%. Dans chaque scénario, moins de la moitié des capteurs ont pu établirune connexion. La combinaison du protocole AODV et du retard du temps de démarrage apermis le plus grand nombre de communications, mais ce même protocole sans retard dutemps de démarrage a obtenu le pire rendement.

DRDC Ottawa TM 2011-001 v

On s’est aussi penché sur le retard d’acheminement des données dans le réseau et on l’amesuré sur les plans de la durée et du nombre de bonds. Sans retard du temps de démarrage,le protocole AODV accusait moins de retard que le protocole OLSR. Toutefois, en ajoutantle retard du temps de démarrage, élément essentiel pour promouvoir la communicationavec le quartier général, cet avantage a presque complétement disparu. D’un autre côté, leprotocole OLSR pouvait généralement se connecter au quartier général en moins de bonsque le protocole AODV. Bien que le protocole AODV sans retard du temps de démarrageconstitue une exception à cette règle, cette exception s’explique par le taux de connexionuniformément bas du scénario.

Enfin, on a étudié l’ampleur du trafic dans le réseau. Les deux protocoles ont subi un grandnombre de collisions : avec le protocole AODV, ce nombre a augmenté avec le nombre decapteurs, alors que, dans le cas du protocole OLSR, ce nombre s’est finalement stabilisé.Toutefois, le protocole OLSR a permis la réception de moins de messages par chaquecapteur, comparativement au protocole AODV.

Portée des résultats : Aucun protocole de routage n’a offert un rendement satisfaisant,avec ou sans retard du temps de démarrage. Ainsi, l’ajout d’un protocole de routage auréseau ne suffit pas à accroı̂tre le nombre de donnés disponibles pour la CER et, conséquem-ment, n’a pas permis d’améliorer la qualité ni l’actualité de la CER. Toutefois, chaqueprotocole comporte des avantages.

Pour choisir entre les protocoles AODV ou OLSR, il faut tenir compte des priorités dusystème. Si le nombre élevé de capteurs connectés est le facteur le plus important, alorsil faut utiliser le protocole AODV avec retard du temps de démarrage. En revanche, s’ilfaut plutôt mettre l’accent sur la réduction des temps d’attente du réseau, alors le protocolesans retard du temps de démarrage constitue le meilleur choix. On peut, bien évidemment,la modification des paramètres du retard du temps de démarrage ou de communication duréseau peut produire des résultats différents.

Recherches futures : L’une des solutions visant à surmonter les problèmes causés par legrand nombre de nœuds du réseau consiste à joindre l’un des protocoles de routage avecune grappe. Cela appliquerait seulement le protocole de routage aux têtes de grappe etréduirait donc au minimum le nombre de messages de routage devant traverser le réseau,tout en permettant aux donnés de CER de la plupart des nœuds d’atteindre le quartiergénéral. Cette méthode nécessite aussi des émetteurs plus puissants comme têtes de grappe,mais elle peut toujours recourir à des capteurs de faible puissance pour la plupart desnœuds, lesquels utiliseront une fréquence différente de celle des têtes de grappe.

Il faut aussi approfondir l’étude des environnements possibles, surtout une simulation ter-restre non plate, et mener des expériences avec différents capteurs et émetteurs simulés.En outre, pour mieux modéliser le monde réel, il faudrait utiliser quelques capteurs etémetteurs mobiles.

vi DRDC Ottawa TM 2011-001

Acknowledgements

The author would like to thank Geoff Colman, Tricia Willink and Oktay Üreten for theirhelp with this work.

DRDC Ottawa TM 2011-001 vii


viii DRDC Ottawa TM 2011-001

Table of contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

Résumé . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

Executive summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Sommaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Table of contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

List of figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Routing protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 AODV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.2 OLSR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.1 Loss of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.2 Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.3 Hops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.4 Network Traffic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

DRDC Ottawa TM 2011-001 ix

List of figures

Figure 1: Examples of locations of transmitters and sensors in the five scenariosstudied. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Figure 2: Cumulative distribution of number of sensors connected to HQ. . . . . . 10

Figure 3: Connectivity of each sensor to HQ. . . . . . . . . . . . . . . . . . . . . 11

Figure 4: Cumulative distribution of average (for each run) total delay for asensor to report to HQ using AODV and OLSR, with send time delay. . . 12

Figure 5: Cumulative distribution of average (for each run) delay, aftersubtracting each sensor’s send time delay, for a sensor to report to HQusing AODV and OLSR. . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Figure 6: Cumulative distribution of average (for each run) total delay for asensor to report to HQ using AODV and OLSR, without send time delay. 13

Figure 7: Average number of hops required for a sensor’s data message to reachHQ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Figure 8: Cumulative distribution of the average number of messagessuccessfully received at each sensor. . . . . . . . . . . . . . . . . . . . . 16

Figure 9: Cumulative distribution of the average number of collisions at each sensor. 17

x DRDC Ottawa TM 2011-001

1 Introduction

Advances in wireless technology and the increasing demand on the available spectrumrequire that radios function in a heterogeneous environment, necessitating the sharing oflimited bandwidth over a large area while avoiding harmful interference. It is thereforeimportant to manage the spectrum dynamically by allocating frequencies based on a user’slocation and the current spectrum usage. One method of facilitating this shared spectrumenvironment is the radio environment map (REM) [1],[2, Chapter 11].

The REM contains information about the radio frequency (RF) environment. It indicatesthe areas where interference with or by other transmitters is possible. The intent of theREM is to create a map of the RF power levels of transmitters to be used by spectrummanagement throughout the region.

To create the REM, the collected RF data must be sent from each sensor to a single location,referred to here as the headquarters (HQ). As noted in [3], there are overhead costs tonetworking radios, such as loss of data, delay, and network traffic. These costs must bebalanced against the impact of a poor REM caused by low quality or quantity of sensordata [4]. In [5], this cost-benefit balance was studied using only flooding and basic TDMAas the network routing protocol.

In order to study this cost-benefit balance, a series of simulations was performed. Thesesimulations used two routing protocols, Ad hoc On-Demand Distance Vector (AODV) [6]routing, and Optimized Link State Routing (OLSR) [7].

The two routing protocols are outlined in Section 2 of this paper. The scenarios simulatedare introduced in Section 3. The cost of too few sensors will be considered in Section 4.1.In Sections 4.2, 4.3 and 4.4, the costs of the REM will be approximated by the delaybefore a message is received by HQ, the number of hops each message takes to get to theheadquarters sensor, how many messages are sent throughout the network and the numberof collisions that occur. A brief summary, as well as suggested areas of further study, isgiven in Section 5.

DRDC Ottawa TM 2011-001 1


2 DRDC Ottawa TM 2011-001

2 Routing protocols

Two routing protocols were studied: Ad hoc On-Demand Distance Vector (AODV) [6]routing, a reactive protocol, and Optimized Link State Routing (OLSR) [7], a proactiveprotocol.

2.1 AODVAODV is a routing protocol for mobile and other wireless ad-hoc networks. It is reactive,establishing a route from source to destination only on demand, that is, when a messageneeds to be sent. It achieves this by forwarding Routing Request (RREQ) messages throughthe network until the destination node is reached. The destination node then sends a RouteReply (RREP) along the path followed by the RREQ. As this RREP makes its way back tothe source node, the path to the destination node is updated at each intermediate node. If aRREQ is received at any of these intermediate nodes for the same destination node before apre-determined expiry time, this path information can be sent in reply to the RREQ withouthaving to reach the destination node.

Once the RREP is received by the source node, the original message is sent through theresulting path to the destination. If it arrives successfully, a received response is again sentback to the source. If a RREQ, the message, or their received responses, do not arrive attheir target nodes, the source node will continue to send either the Routing Request or themessage until a response is received, or a pre-determined number of re-tries is exhausted.

2.2 OLSROLSR is also a routing protocol for mobile and other wireless ad-hoc networks. It is proac-tive, meaning that it sends Hello and Topology Control messages in advance of demand,in order to create a routing table for the network. Hello messages are sent to “introduce” anode to nearby nodes and identify a subset of its single-hop neighbors that will be used forretransmitting messages. Topology Control messages are sent to disseminate the list of allnodes a node is in communication with, and the “last-hop” destination on the path to thatnode. From the information in these two types of messages, a routing table is created withavailable destination nodes and their “next-hop” addresses. In a large network, creating acomplete routing table can take many iterations.

Additionally, in OLSR, the link information in the table expires over time, and must thusbe updated periodically, which requires the sending of further messages. Routing infor-mation expires so that links that are made obsolete by node mobility or power failures canbe replaced, and so that new nodes can be added to the routing table. However, neither ofthese situations occur in the experiment, so it might make sense to remove or increase the


link expiry interval. Unfortunately, in OLSR, link expiry times are tied to route creationintervals, and thus, to create a longer lasting routing table, longer route creation times arerequired. Retaining the link expiry intervals will also allow a more representative compar-ison to future experiments when mobility and power constraints are included.

When a data message is to be sent, the routing table is checked to determine if the des-tination node is reachable from the source node. If it is, the message is broadcast, withthe next-hop node as the current target and the destination node as the final target. If thedestination is not on the table, the message is not sent. As with AODV, if a response is notreceived from the destination node, the message is re-sent until a response is received, or apre-determined number of re-tries is exhausted.


3 Scenarios3.1 ObjectiveTo compare the costs and benefits of the two routing protocols in the creation of the REM,five different scenarios were used. In each scenario, three transmitters, N sensors and anHQ sensor were placed in a 15 km × 15 km square flat area. The N sensors were randomlyplaced in the area, while HQ was placed in the lower right corner, (15000 m, 15000 m). Thethree transmitters were located at (6000 m, 1000 m), (1000 m, 11000 m) and (14000 m,4000 m). Each scenario comprised 15, 25, 35, 45, or 55 sensors. Examples of theseplacements are shown in Figure 1.

Every transmitter had a transmission power of 100 W and a 512 MHz carrier frequency.The sensors also used a 512 MHz carrier frequency for their transmissions, with a transmis-sion power of 10 W. While having different transmitter and sensor frequencies would havebeen more realistic, the limitations of the simulation software prevented this. To compen-sate, all transmissions from the transmitters occurred before transmissions from the sensorscommenced.

A simple pathloss model [8, Chapter 4] was used, such that the received power at a dis-tance d from the transmitter was

PR = PT

(λ4π

)2(d−α) (1)

where PT is the transmitted power, λ is the wavelength and the path loss coefficient isα = 3.

In all scenarios, the three transmitters broadcast their signals omni-directionally. Wheneverthese signals were detected by a sensor, the total received power and the sensor locationwere recorded to be forwarded to HQ.

3.2 ImplementationEach of the five scenarios described in Section 3.1 was run 2500 times with random sensorplacement, using the OMNeT++ discrete event simulation engine [9]. The INET-MANETsimulation framework [10], in particular the code for the AODV and OLSR routing proto-cols, was incorporated into the scenario.

To simulate the broadcast signals from the transmitters, a data message was sent fromeach transmitter. The use of this discrete event simulator meant that multiple simultane-ously received signals were not correctly handled. As a workaround, the transmitters eachbroadcast a single data message, with a delay interval of 0.3 s. This allowed all data mes-sages from one transmitter to reach all sensors in range before the next transmitter started


its broadcast. As the sensors received the transmissions, the received power values weresummed, in order to simulate simultaneous transmissions.

Once the total received power was calculated by the sensor, this value, along with the sensorlocation, was forwarded to HQ, using one of the two routing protocols. In addition to thetotal received power and the sensor location, the number of hops taken to reach HQ and thetime that the message was received by HQ were recorded by HQ.

Because the two routing algorithms used created traffic on the network, two different starttime scenarios were used. In the first scenario, a 2 s delay interval was used to avoidcollisions and lost data due to network congestion. This modelled the Time Delay MultipleAccess (TDMA) channel access method, which requires some centralized coordination. Inthis case, coordination was achieved by scheduling the broadcast times in advance, withno updates to the schedule based on actual performance. Thus, this delay interval didnot guarantee that, in all cases, a sensor’s data message would reach HQ before anothersensor’s data message was broadcast, but did allow this to occur in most cases. In thesecond scenario, no delay is used, and all sensors start their data message transmissionprocedure randomly within a 0.5 s window. In both cases, no data message is broadcastuntil 30 seconds after the simulation start.

At the completion of each run, the total received power at, and location of, each of thesensors connected to HQ was recorded. Additionally, the number of hops the first datamessage from each sensor took to reach HQ and the time at which it reached HQ, as wellas the total number of transmissions received and collisions at each sensor were reported.These data were then imported into Matlab for further analysis.


(a) 15 sensors (b) 25 sensors

(b) 35 sensors (b) 45 sensors

(c) 55 sensors

Figure 1: Examples of locations of transmitters and sensors in the five scenarios studied.


4 Results

The experiment looked at four network costs: the loss of data, the delay in receiving datafrom the sensors, the number of hops taken to reach HQ and the network traffic. To studythese costs, Matlab was used to analyze the results from the OMNeT++ output.

4.1 Loss of dataLoss of data occurs when messages sent from the sensors do not reach HQ, and thus thereceived power and other information from that sensor cannot be used in the REM. Theproblems with this, and the poor quality, possibly misleading REM that results, are illus-trated in [5].

If there aren’t enough sensors, or if they are not optimally placed, the network may not befully connected to the HQ, and thus some messages will not reach their destination. Sensorsmay also be disconnected from HQ because of collisions due to the overhead caused by therouting protocol: so many routing messages have to be sent to set up the network thatthe messages interfere with each other, and in the end, none of the routing messages getthrough. Section 4.4 discusses this topic more thoroughly.

In [5], the location of the sensors played a much larger part in the connectivity of thesensors; if a path existed to HQ, then the sensor’s data would get there, and vice versa.With the additional overhead of the routing protocol, a path to HQ is no longer sufficientto guarantee connectivity. Even sensors within one or two hops of HQ may be unable toconnect due to the collisions caused by the added messages traversing the network.

Figure 2 shows the cumulative distributions of the number of sensors able to communicatewith HQ for both the AODV and the OLSR routing protocols, with and without send timedelay. In all cases, a significant portion of the runs have no sensors able to report to HQ.Not surprisingly, this happens more often when there are fewer sensors in the network,as distant sensors have fewer possible paths through which to communicate with HQ, andthere may in fact not be any sensors in range of HQ to form a path.

Figures 2 (a) and (c) show the benefit to the AODV protocol of the start time delay: withoutit, very few sensors can communicate with HQ, because of the large number of routingmessages being sent at one time, causing collisions. On the other hand, Figures 2 (b) and(d) show that the start time delay actually makes the OLSR protocol perform slightly lesswell: fewer sensors can communicate with HQ with the start time delay, likely because therouting table expires during the delays, so re-use of paths is not possible, requiring morerouting messages to be sent.

Figure 3 shows the connectivity to HQ of each of the sensors. That is, it shows for eachof the sensors the percentage of runs for which that sensor’s data message reached HQ. In


(a) AODV, with start time delay (b) OLSR, with start time delay

(c) AODV, without start time delay (d) OLSR, without start time delay

Figure 2: Cumulative distribution of number of sensors connected to HQ.

none of the cases does any sensor have more than a 25% chance of communicating withHQ.

Figure 3 (a) shows that with the AODV protocol and start time delay, the probability ofa sensor being connected to HQ increases until there are 35 sensors in the network, thendecreases for more sensors. This is explained by the increasing likelihood of a path to HQas the number of sensors grows, allowing more messages the chance to reach HQ, whilealso increasing the number of routing messages being forwarded through the network, andtherefore also collisions.

In Figure 3 (c), AODV without start time delay, the probability that a sensor is connected toHQ is about 5% for all sensors in all scenarios. This is due to the large number of routingmessages that are sent all at once, making route creation very unlikely for all sensors.




Figure 3: Connectivity of each sensor to HQ.

Figure 3 (b) indicates that for the OLSR protocol with start time delay, the connectivity ofa sensor to HQ depends more on when the message is sent than the number of sensors inthe network. Except for the 15 sensor case, which gives each sensor about a 10% chance ofconnecting to HQ, the probablity that a given sensor is connected to HQ decreases with alater start time. This is because, as time passes, the pre-determined paths to HQ expire, androuting messages are required to maintain the links. This, again, causes more messagesto be in the system, and thus causes more collisions with both the data and the routingmessages.

Finally, in Figure 3 (d), the likelihood that a sensor will be able to communicate withHQ when the OLSR protocol is used without start time delay increases until there are 35sensors in the system, and then remains stable. This is because the chances of a path to HQbeing available increases as the number of sensors increases, while the number of routing


messages forwarded throughout the system also increases. This behaviour is similar to thatseen in Figure 3 (a); the differences are explained in Section 4.4.

Overall, neither of the routing protocols performed satisfactorily, with or without start timedelay. Less than half of the sensors could communicate with HQ in any of the scenarios,and in most cases, significantly less than half. However, the AODV routing protocol withstart time delay allowed the most communication, while the same protocol without starttime delay performed the worst.

4.2 DelayThe network delay, the average time it takes for data to get from a sensor to HQ, is a sig-nificant form of overhead for any network. When, in addition to sending data, the networkmust also find a path from the source to the destination, this delay can be even greater (see[5] for effects on the REM). Delay can be caused by the distance and number of hops asensor is from HQ, both when sensing the message and finding a path, and also by waittimes before retries when collisions have occurred.

Figures 4 and 5 show the cumulative distribution of the delay in the send time delay sce-narios for both the AODV and the OLSR routing protocols. Figure 4 gives the total delay,which includes the start time delay, while Figure 5 shows delay adjusted for the staggeredsend times of each sensor. Figure 6 shows the cumulative distribution of the total delay forthe scenarios without send time delays. The distributions in the figures do not reach one assome runs had no sensors connected to HQ, and thus returned no results for total delay.

(a) AODV (b) OLSR

Figure 4: Cumulative distribution of average (for each run) total delay for a sensor toreport to HQ using AODV and OLSR, with send time delay.


(a) AODV (b) OLSR

Figure 5: Cumulative distribution of average (for each run) delay, after subtracting eachsensor’s send time delay, for a sensor to report to HQ using AODV and OLSR.

(a) AODV (b) OLSR

Figure 6: Cumulative distribution of average (for each run) total delay for a sensor toreport to HQ using AODV and OLSR, without send time delay.

Figure 4 shows that in the start time delay case, the total delay with OLSR appears to beapproximately the same as with AODV. This is not surprising, as most of the total delayderives from the start time offsets. Because of this, when the start time delay is subtractedfrom the data, as in Figure 5, a very different picture emerges.

Looking only at the time taken from when the data message was sent, when there are 35 ormore sensors in the network, AODV produces much less delay than OLSR. This is because,with AODV, most messages in the system during any delay interval are working to create


a route to enable one sensor to communicate with HQ (some messages might be left overfrom earlier sensors), or to send that sensor’s data to HQ. This decreases the chances thatrouting messages will interfere with data messages. With OLSR, however, route mainte-nance messages are unrelated to the data messages, meaning both that collisions betweenrouting and data messages are more likely, and that a required route may have expired bythe time it is needed. These two circumstances contribute to reporting delay.

Figure 6 shows that the total delay in the no send time delay case is similar to the adjusteddelay in the send time delay case, particularly for the OLSR protocol. This indicates thatwhile the send time delay may affect the number of sensors that can communicate withHQ, it does not affect how quickly they are able to do so.

The total delay with start time delay was unacceptably high: a mobile vehicle could have asignificantly different current position than when it was originally reported. While the de-lay without start time was much lower for both protocols, the delay in the OLSR scenarioswas still high, compared to AODV.

4.3 HopsThe network delay can also be represented using the number of hops that a sensor’s datatakes to reach HQ. As a data message requires time to be received, processed and forwardedby each sensor, a higher number of hops means that data may be more out of date by thetime it reaches HQ, and that more sensors are involved in passing on the data, meaning thatthere is more overhead. This can lead to an accurate but out-of-date REM, or even to a lessaccurate REM as data may be subject to collisions and never reach HQ.

Figure 7 shows the cumulative distribution of the average number of hops each sensor’sdata took to reach HQ, over each simulation run. It can be seen that, when at least onesensor is able to report to HQ, of the data messages that reach HQ, most take three or fewerhops to do so, regardless of routing protocol or start time delay.

Figures 7 (a) and (c) show the effect that the start time delay had on the data messagesreaching HQ with the AODV routing protocol. With start time delay, a portion of thedata messages take four or more hops, with higher numbers of hops when there are lowernumbers of sensors, due to the greater difficulty in finding a direct path to HQ. Without starttime delay, almost no data messages take more than two hops, because the large numberof routing messages in a small time interval make it very difficult for any message to reachits destination, and only those data messages close to HQ afford few enough collisionopportunities to overcome this.

Figures 7 (b) and (d) show that start time delay had very little effect on the hop performanceof the OLSR routing protocol, particularly for the higher number of sensors. The 15 to 35sensor cases skew a little more towards a single hop when start time delay is not added,




Figure 7: Average number of hops required for a sensor’s data message to reach HQ.

again because of the very small number of sensors able to communicate with HQ in thesescenarios, all of which are close to HQ. For the 45 and 55 sensor cases, because, in total,more data messages reach HQ when no start time delay is used, more data messages areable to reach HQ from a greater distance and thus a larger number of hops is needed.

4.4 Network TrafficThe network traffic created by building an REM can be represented by the number of dataand routing messages each of the sensors in the system received, and by the number ofcollisions that occurred. Figure 8 gives the cumulative distribution of the average numberof messages that are successfully received by each sensor. Figure 9 gives the cumulativedistribution of the average number of collisions that occur at each sensor. As the number


of messages in the system increases, the possibility of collision, and thus loss of data, alsoincreases.



Figure 8: Cumulative distribution of the average number of messages successfully re-ceived at each sensor.

Figure 8 shows that the routing protocol had more impact (an approximately ten-fold dif-ference between the number of messages received by sensors using OLSR and those usingAODV) on the number of messages successfully received at each sensor than did the starttime delay. The start time delay had little impact because the same functions were per-formed over the same time period. For AODV, slightly fewer messages got through with-out the start time delay, as the increase in network traffic at one time would have causedmore collisions, interfering with the reception of the messages. For OLSR, there was someincrease when no start time delay was used, in the 35 or more sensor cases, due to theexpiring route information in the start time delay case.




Figure 9: Cumulative distribution of the average number of collisions at each sensor.

Figure 9 again shows the lack of effect of the start time delay. However, in this case, thethe effect of the routing protocol, while present, is not as striking. What is more prevalent,but not surprising, is the effect of the number of sensors. With AODV, the increase in thenumber of collisions that occur is very sensitive to the increase in the number of sensors;with OLSR, this sensitivity is very evident until 35 sensors are reached, when it becomesmuch less pronounced. This is due to the nature of each protocol: AODV requires a largenumber of routing messages each time a data message needs to be sent, while the routingmessages in OLSR are sent regardless of the data messages. Thus, the number of routingmessages in the system at any given time is more dependent on the number of sensors withthe AODV protocol than it is with the OLSR protocol.

The collision data also explains the slightly different behaviours seen in Figures 3 (a) and(d). While in Figure 9 (a), the number of collisions accelerates as the number of sensors


is increased, in Figure 9 (d) the number of collisions levels off after 35 sensors. Thus theeffect of the number of collisions outpaces the effect the increased number of sensors hason route availability, as seen in Figure 3 (a), while, in Figure 3 (d), the two effects canceleach other out.


5 Conclusion

Neither of the two routing protocols performed satisfactorily, with or without start timedelay modelling TDMA. The loss of data from the sensors was high: no sensor was able tocommunicate with HQ with more than 25% probability and in any scenario, less than 50%of the sensors were able to connect, and generally significantly fewer. The AODV routingprotocol with start time delay allowed the most communication, while the same protocolwithout start time delay performed the worst.

Without start time delay, the AODV protocol had far less delay than OLSR. However,adding the start time delay, as was necessary to promote communication with HQ, essen-tially eliminated this advantage. OLSR, on the other hand, was generally able to connect toHQ with fewer hops than AODV required. While the AODV without start time delay casewas an exception to this, this exception only resulted due to the uniformly low connectionrate of that scenario.

Both routing protocols were subject to a large number of collisions: with AODV this num-ber increased as the number of sensors increased, while with OLSR, the number levelledoff after 35 sensors. OLSR, however, allowed an order of magnitude fewer routing anddata messages to be successfully received at each sensor than with AODV.

Overall, the selection of either AODV or OLSR would have to take into consideration thepriorities of the system and those of the potential REM. If a higher number of connectedsensors, and thus REM accuracy, is the most important factor, the AODV protocol withstart time delay should be used. If, however, minimizing network delay, and thus increasingthe timeliness of the REM, takes precedence, the OLSR protocol without start time delaywould be the best choice. Of course, adjustments in the start time delay, or in the network’scommunication parameters, would likely yield different results.

One possibility to overcome the problems caused by having many nodes in the networkis to combine one of the routing protocols with clustering. This would allow the routingprotocol to be applied only to the cluster heads, minimizing the number of routing messagesthat need to traverse the network, while the REM data from most of the nodes would stillreach HQ. This method would also have to use higher powered transmitters as the clusterheads, but could still use low powered sensors for most of the nodes, likely functioning ona different frequency from the cluster heads.

Further exploration of possible environments is also needed, particularly a non-flat earthsimulation, as well as experimentation with different simulated sensors and transmitters.Additionally, to better model a real world environment, mobility should be added to someor all of the sensors, and to the transmitters.


References

[1] Zhao, Y., Raymond, D., da Silva, C., Reed, J. H., and Midkiff, S. F. (2007),Performance Evaluation of Radio Environment Map-Enabled CognitiveSpectrum-Sharing Networks, In Proc IEEE MILCOM 2007, pp. 1–7.

[2] Fette, B. A. (2009), Cognitive Radio Technology, Second ed, Academic Press.

[3] Willink, T. J. and Rutagemwa, H. (2009), Framework for Performance Evaluation ofCognitive Radio Networks in Heterogeneous Environments, In Proc IEEE CCECE2009, pp. 199–203.

[4] Hanif, M. F., Smith, P. J., and Shafi, M. (2009), Performance of Cognitive RadioSystems with Imperfect Radio Environment Map Information, In Proc AusCTW2009, pp. 61–66.

[5] Faint, S. M., Ureten, O., and Willink, T. J. (2010), Impact of the Number of Sensorson the Network Cost and Accuracy of the Radio Environment Map, In Proc IEEECCECE 2010, pp. 199–203.

[6] Perkins, Charles E. and Royer, Elizabeth M. (1999), Ad hoc On-Demand DistanceVector Routing, In Proc IEEE WMCSA 1999, pp. 90–100.

[7] Jacquet, P., Muhlethaler, P., Clausen, T., Laouiti, A., Qayyum, A., and Viennot, L.(2001), Optimized link state routing protocol for ad hoc networks, In Proc IEEEINMIC 2001, pp. 62–68.

[8] Molisch, A. F. (2005), Wireless communications, John Wiley & Sons Ltd.

[9] Varga, A., OMNeT++ users manual: OMNeT++ version 4.0. Available athttp://www.omnetpp.org/doc/omnetpp40/manual/usman.html.

[10] INET Framework for OMNeT++/OMNEST. Available athttp://inet.omnetpp.org/doc/INET/neddoc/index.html.


DOCUMENT CONTROL DATA(Security classification of title, body of abstract and indexing annotation must be entered when document is classified)

1. ORIGINATOR (The name and address of the organization preparing thedocument. Organizations for whom the document was prepared, e.g. Centresponsoring a contractor’s report, or tasking agency, are entered in section 8.)

Defence R&D Canada – Ottawa3701 Carling Avenue, Ottawa ON K1A 0Z4, Canada

2. SECURITY CLASSIFICATION (Overallsecurity classification of the documentincluding special warning terms if applicable.)

UNCLASSIFIED

3. TITLE (The complete document title as indicated on the title page. Its classification should be indicated by the appropriateabbreviation (S, C or U) in parentheses after the title.)

Impact of AODV and OLSR routing protocols on the network cost of the Radio EnvironmentMap

4. AUTHORS (Last name, followed by initials – ranks, titles, etc. not to be used.)

Faint, S. M.

5. DATE OF PUBLICATION (Month and year of publication ofdocument.)

March 2011

6a. NO. OF PAGES (Totalcontaining information.Include Annexes,Appendices, etc.)

38

6b. NO. OF REFS (Totalcited in document.)

10

7. DESCRIPTIVE NOTES (The category of the document, e.g. technical report, technical note or memorandum. If appropriate, enterthe type of report, e.g. interim, progress, summary, annual or final. Give the inclusive dates when a specific reporting period iscovered.)

Technical Memorandum

8. SPONSORING ACTIVITY (The name of the department project office or laboratory sponsoring the research and development –include address.)

Defence R&D Canada – Ottawa3701 Carling Avenue, Ottawa ON K1A 0Z4, Canada

9a. PROJECT OR GRANT NO. (If appropriate, the applicableresearch and development project or grant number underwhich the document was written. Please specify whetherproject or grant.)

15bx

9b. CONTRACT NO. (If appropriate, the applicable number underwhich the document was written.)

10a. ORIGINATOR’S DOCUMENT NUMBER (The officialdocument number by which the document is identified by theoriginating activity. This number must be unique to thisdocument.)

DRDC Ottawa TM 2011-001

10b. OTHER DOCUMENT NO(s). (Any other numbers which maybe assigned this document either by the originator or by thesponsor.)

11. DOCUMENT AVAILABILITY (Any limitations on further dissemination of the document, other than those imposed by securityclassification.)( X ) Unlimited distribution( ) Defence departments and defence contractors; further distribution only as approved( ) Defence departments and Canadian defence contractors; further distribution only as approved( ) Government departments and agencies; further distribution only as approved( ) Defence departments; further distribution only as approved( ) Other (please specify):

12. DOCUMENT ANNOUNCEMENT (Any limitation to the bibliographic announcement of this document. This will normally correspondto the Document Availability (11). However, where further distribution (beyond the audience specified in (11)) is possible, a widerannouncement audience may be selected.)

13. ABSTRACT (A brief and factual summary of the document. It may also appear elsewhere in the body of the document itself. It is highlydesirable that the abstract of classified documents be unclassified. Each paragraph of the abstract shall begin with an indication of thesecurity classification of the information in the paragraph (unless the document itself is unclassified) represented as (S), (C), or (U). It isnot necessary to include here abstracts in both official languages unless the text is bilingual.)

Spectrum management using a radio environment map (REM) to dynamically allocate frequen-cies may facilitate the efficient exploitation of the spectrum in a heterogeneous environment. Theoverhead costs of networking radios to create the map must be balanced against the impact of apoor REM due to low quality or quantity of sensor data. To study the cost-benefit balance of tworouting protocols, simulations were performed to evaluate the transmission of REM data usingAd hoc On-Demand Distance Vector (AODV) routing and Optimized Link State Routing (OLSR).It was seen that both protocols caused significant overhead in the system, effectively reducingthe amount of data that could reach the data collection centre, and thus potentially lowering thequality of the resulting REM.

14. KEYWORDS, DESCRIPTORS or IDENTIFIERS (Technically meaningful terms or short phrases that characterize a document and couldbe helpful in cataloguing the document. They should be selected so that no security classification is required. Identifiers, such asequipment model designation, trade name, military project code name, geographic location may also be included. If possible keywordsshould be selected from a published thesaurus. e.g. Thesaurus of Engineering and Scientific Terms (TEST) and that thesaurus identified.If it is not possible to select indexing terms which are Unclassified, the classification of each should be indicated as with the title.)

radio environment mapdynamic spectrum managementspectrum sensingrouting protocolsAODVOLSR

Date post:	09-Feb-2021
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

Impact of AODV and OLSR routing protocols on the network ...Enfin, on a étudi é l’ampleur du...

Documents