Crosslayer Function

7/27/2019 Crosslayer Function

1/21

Improving routing performance in wireless ad hocnetworks using cross-layer interactions

Erik Weiss a,*, Guido Hiertz a, Bangnan Xu b, Sven Hischke c,Bernhard Walke a, Sebastian Gross a

a Communication Networks, Faculty 6, RWTH Aachen University, Kopernikusstr.16, D-52074 Aachen, Germanyb SSC ENPS (Technologiezentrum), T-Systems, Am Kavalleriesand 3, D-64295 Darmstadt, Germany

c Deutsche Telekom AG, Friedrich-Ebert-Allee 140, D-53113 Bonn, Germany

Received 29 November 2004; received in revised form 3 March 2006; accepted 17 March 2006

Abstract

This article presents a combined layer two and three control loop, which allows prediction of link breakage in wirelessad hoc networks. The method monitors the physical layer transmission mode on layer two and exploits the gained knowl-edge at layer three. The mechanism bases on link adaptation, which is used in IEEE 802.11a WLAN to select the trans-mission mode according to the link quality. The process of link adaptation contains information that is useful to predict

link stability and link lifetime. After introducing the IEEE 802.11a Medium Access Control (MAC) and PHY layer, wepresent insight to the IEEE 802.11a link adaptation behaviour in multi-hop ad hoc networks. The link adaptation algo-rithm presented here is derived from Auto Rate Fallback (ARF) algorithm. We survey the performance gain of two newlydeveloped route adaptation approaches exploding the prediction results. One approah is Early Route ReArrangement(ERRA) that starts a route reconstruction procedure before link breakage. Hence, an alternative route is available beforeconnectivity is lost. Early Route Update (ERU) is a complementing approach that enhances this process, by communica-tions among routing nodes surrounding the breaking link. The delay caused by route reconstruction can be significantlyreduced if prediction and either of our new route discovery processes is used. 2006 Elsevier B.V. All rights reserved.

Keywords: Ad hoc network; Ad hoc routing; AODV; Link breakage prediction; Link adaptation; IPonAIR; ERRA; ERU; IEEE 802.11MAC

1. Introduction

While it is increasingly important to be connectedto the Internet world, the trend goes towards wire-less solutions, providing mobile access to the Inter-net at high data rates. Wireless Local Area Networks(WLAN) like IEEE 802.11a operating at 5 GHz can

1570-8705/$ - see front matter 2006 Elsevier B.V. All rights reserved.

doi:10.1016/j.adhoc.2006.03.003

* Corresponding author. Tel.: +49 241 802 8575; fax: +49 241802 2242.

E-mail addresses: [email protected] (E.Weiss), [email protected] (G. Hiertz), [email protected](B. Xu), [email protected] (S. Hischke), [email protected] (B. Walke), [email protected] (S. Gross).

Ad Hoc Networks xxx (2006) xxxxxx

www.elsevier.com/locate/adhoc

ARTICLE IN PRESS
mailto:[email protected]:[email protected]:[email protected]:[email protected]:walke@%20%20%20comnets.rwth-aachen.demailto:walke@%20%20%20comnets.rwth-aachen.demailto:Sebastian.gross@comnets.%20rwth-aachen.demailto:Sebastian.gross@comnets.%20rwth-aachen.demailto:Sebastian.gross@comnets.%20rwth-aachen.demailto:Sebastian.gross@comnets.%20rwth-aachen.demailto:walke@%20%20%20comnets.rwth-aachen.demailto:walke@%20%20%20comnets.rwth-aachen.demailto:[email protected]:[email protected]:[email protected]:[email protected]


2/21

support transmission rates up to 54 Mb/s. The highattenuation at 5 GHz limits the coverage area. Oneobjective is therefore, to extend the coverage byestablishing multi-hop routes. Since using wirelesstechnology enables the user to be mobile, the net-

work has to deal with effects introduced by adynamically changing network topology (Fig. 1).High throughput and limited transmission range

make WLAN systems adequate for areas with ahigh population density and users with the needfor high data rates. Such places, like airports or fair-grounds, are called Hotspots. Covering large areaswith WLAN is economically inefficient, since a con-siderable number of Access Point (AP) and routerswould have to be deployed due to the limited trans-mission range. The AP density can be reduced byincreasing the transmission power or enabling inter-

mediate terminals to forward the data to usersbeyond the Access Point range. High transmit pow-ers strain the batteries of the mobile devices andincrease the exposure of operators to radio waves,along with their yet undetermined health risk.

1.1. State of the art

Todays solution is to expand the fixed infra-structure using multi-hop connections. Mobilityand fast changing topologies are addressed by ad

hoc routing protocols, which can be subdivided intoproactive and reactive protocols. Reactive protocols[4,5] request a route when needed, whereas proac-tive protocols [15,16] permanently maintain routesto all known network members that can be usedon request, therefore minimizing packet delay.Reactive protocols avoid maintaining unneededroutes at the cost of a higher route discovery delayand packet delay. Therefore, hybrid approaches

[17] have been developed to reach a trade-offbetween the reactive and proactive approaches.

However, in case of a route break, all routingapproaches try to recover the connection. Mostrouting algorithms inform the source node, which

in turn starts a completely new discovery process,thereby flooding the network with a large numberof signalling messages. Some routing protocols per-form a local route discovery [4] around the breakageto limit the flooding. Moreover, all aforementionedapproaches solely only react when the link is alreadybroken. This leads to a high number of lost packetsas well as increased route rediscovering and packetdelay.

The new strategy proposed in this article is not towait until the link breaks but to act in advance.Based on link adaptation information we predict

the link state and start to rearrange the route beforethe link is interrupted. Lower layers, especially thelink adaptation (LA), provide information thatallows predicting the link conditions. We presentEarly Route ReArrangement (ERRA) and EarlyRoute Update (ERU), two new route rearrange-ment protocols based on link prediction. Since thetwo presented approaches prevent unnecessary sig-nalling, avoid packet loss and minimize packetdelays, both use the ad hoc network capacity moreefficiently than existing protocols.

A similar work based on legacy IEEE 802.11 waspresented in [2], although the prediction approachpresented in this paper employs the received signalstrength to estimate the likelihood of a link inter-ruption. The approach in [2] is limited to informingthe routes source about expected route breaks. Thesource node has to act upon this information andthe source node starts flooding the network toupdate the route using standard routing algorithms.

SourceSource

SourceSource

SourceDest.

SourceDest.

Multi-hopConnections

Transmission

Range

Fig. 1. Wireless ad hoc network.

2 E. Weiss et al. / Ad Hoc Networks xxx (2006) xxxxxx

ARTICLE IN PRESS


3/21

Our approach combines the idea of local routerearrangement on layer three and link predictionat layer two. We propose a prediction algorithmthat takes advantages of the link adaptation (LA)actions. We use the link adaptation like a pre-pro-

cessing function for the evaluation of channelconditions.The remaining parts of the article are organized

as follows: we start with a brief survey of IEEE802.11. The link adaptation fundamentals are cov-ered in Section 3, followed by a discussion on howthe link adaptation supports the necessary predic-tion information in Section 4. Simulations providefurther information on the efficiency of the pre-sented link adaptation. After giving a brief descrip-tion of the ERRA and ERU signalling procedures,we highlight the benefits of the new protocols in Sec-

tion 5. In Section 6 we present the simulation resultsof our new route rearrangement protocols ERRAand ERU. The last two sections conclude our con-sideration and discuss further research topics.

2. IEEE 802.11a medium access control

The IEEE 802.11a standard describes an OFDMPHY layer at 5 GHz. The Medium Access Control(MAC) layer is equal to IEEE 802.11b and legacyIEEE 802.11. IEEE 802.11a mainly introduces

higher data rates [1,3]. IEEE 802.11a offers eightcoding and modulation schemes, so called PHY-Modes (cf. Table. 1). The MAC protocol used inIEEE 802.11 is called Distributed CoordinationFunction (DCF). IEEE 802.11 furthermoredescribes a Point Coordination Function (PCF).The PCF is used for centrally controlled access.To our knowledge, it is has never been implementedby any vendor. The DCF is based on carrier sensemultiple access with collision avoidance (CSMA/CA). As mobile nodes (MNs) are not able to

monitor the air interface while transmitting, theDCF uses backoff and request to send/clear to send(RTS/CTS) mechanisms to avoid collisions due tothe hidden station problem. Details of the IEEE802.11 MAC protocol are given in [3].

2.1. IEEE 802.11a transmission modes

The IEEE 802.11a standard does not specify anyrules for selecting the transmission mode. Eachvendor implements a proprietary link adaptationmethod. The first four bits within each packetpreamble specify the PHYMode chosen for codingthe data payload. Higher transmission modes arecapable of delivering higher data rates, but need aconsiderably higher carrier-to-interference ratio(C/I), thus limiting the bridged distance (cf. Figs.

2 and 3). In Table. 1 all available modes are listedwith their respective maximum data rate, appliedcode rate and bits per OFDM symbol [1,3].

Due to the channel attenuation the C/I secedes asthe distance between transmitter and receiverincreases. Thus, different areas can be dependablycovered with each PHYMode (cf. Fig. 2). IEEE802.11a allows stepping down the transmissionmode when the channel quality is decreasing, e.g.,due to a rising distance between source and destina-tion or in case of growing interference (cf. Fig. 3).

The IEEE 802.11a protocol allows choosing of anindividual PHYMode for each connection and eachdata packet. Every terminal attempts to determinethe PHYMode providing the best balance ofthroughput to packet error rate (PER). The Chairof Communication Networks, RWTH Aachen

Table 1PHYMode dependent parameters

Data rate (Mb/s) Modulation Codingrate (R)

Data bitsper symbol

6 BPSK 1/2 249 BPSK 3/4 3612 QPSK 1/2 4818 QPSK 3/4 7224 16-QAM 1/2 9636 16-QAM 3/4 14448 64-QAM 2/3 192

54 64-QAM 3/4 216

PHYModeRange Dest.

Dest.

Dest.

Dest.

Dest.

Dest.Dest.BPSK 1/2

BPSK 3/4

Source

64 QAM 3/4

64 QAM 2/3

Source

Fig. 2. PHYMode range.

E. Weiss et al. / Ad Hoc Networks xxx (2006) xxxxxx 3

ARTICLE IN PRESS


4/21

University [14] has developed a simulation tool toinvestigate IEEE 802.11a/e/g together with a com-plex IP layer. The simulation tool makes use of atwo-path propagation model over a reflecting sur-face [1,12] as the physical channel model. It com-bines calculated C/I with the measured values,Fig. 3 shows exemplarily some values [13]. Theresults from our simulation environment presentedin Fig. 4 evaluate the maximum throughput perPHYMode for a packet size of 2000 B. Since the

802.11 MAC handles small sized packages very inef-ficiently, our simulation results present a best-casescenario. The envelope of the throughput results isformed by the upper bound of all PHYModes andrepresents the optimum balance of PER andthroughput. Since terminals in a real system cannotdistinguish between signal power and interferencepower, it is impossible to measure the detailed C/Ifor each packet: Hence, each terminal only detectsa certain local power level.1

There are two ways to estimate the signal-to-

noise ratio: Terminals can either measure the inter-ference power within transmission pauses, or countthe successfully received and lost packets. The latterone is not intricate to implement as described in Sec-tion 3. The ratio between received and lost packets,combined with the measured packet error rate(PER) per PHYMode, leads to the prevailing C/I[12,13]. The IEEE 802.11a/g/e network simulator

bases on PER versus carrier-to-interference ratiomeasurements [12] shown in Fig. 3. Higher trans-mission modes are capable of delivering higher datarates. But nevertheless, they also need a remarkablyhigher C/I. Decrease of the channel quality has sev-

eral reasons, like fading, obstacles and interference.Fig. 4 presents the maximum achievable datathroughput of IEEE 802.11a with the DCF andRTS/CTS handshake.

As presented in Fig. 4 802.11a carries up to36 Mb/s when utilizing the 64-QAM 3/4 PHY-Mode. To improve the performance, a simple butefficient approach has been developed. After havingsuccessfully sent a predefined number of packets theLA of the transmitting station, steps up to a fasterdata rate. If a certain number of packets is lost later

on, a lower transmission mode is used. This linkadaptation algorithm is called Auto Rate Fallback(ARF). Further details can be found in [9]. How-ever, this algorithm is instable. Even if LA hasfound an optimum transmission mode it will tryto switch to a less robust but higher PHYModeleading to an increased packet error. Some otherLA approaches have been presented in [10,11]. Theyoutperform ARF but require changes to the IEEE802.11 standard [3].

3. Link adaptation for multi-hop ad hoc network

Since IEEE 802.11 does not define any rules foran internal link adaptation (LA), each vendorimplements a proprietary LA. Our predictionapproach and route rearrangement protocols relyon the behaviour of the LA, thus it is mandatoryto specify the used LA. Independently from the fla-vour of the LA, all algorithms are based on a similarprinciple that with degrading channel conditions alower but more robust PHYMode is chosen (or ahigher but less robust one, when the channel condi-

tions improve).

0.001

0.01

0.1

1

0 5 10 15 20 25 30 35

PacketErrorRate(PER)

PER)

Carrier to Noise Ratio C/I [dB]

BPSK 1/2BPSK 3/4QPSK 1/2QPSK 3/4

16-QAM 3/464-QAM 3/4

Fig. 3. Packet error rate versus C/I.

10

20

30

40

00 5 10 15 20 25 30 35

Link Adaptation

follows the max throughput

T

hroughput[Mb/s]

Signal to Noise Ratio (C/I) [dB]

BPSK 1/2

BPSK 3/4

QPSK 1/2

QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

Fig. 4. Reachable throughput per PHYMode.

1 This is the reason why routing protocols based on signalstrength stability operate on an uncertain basis. The receivedsignal strength changes very dynamical due to the burstiness of

the traffic.


ARTICLE IN PRESS


5/21

To evaluate the performance of 802.11 MAC in awireless multi-hop ad hoc network, our simulationtool has been extended with a fast and efficientLA. The developed LA can be described as anenhanced version of the Auto Rate Fallback

(ARF) [9] protocol. Fig. 5 shows the functionalprinciple of our LA algorithm. One LA instanceserves one destination only. The implemented ver-sion addresses the following design specifications.

Fast reaction on channel condition changes isone of the major requirements for a LA, thoughslow variation must be considered as well. To fulfilthese requirements, our LA contains three differentlists (packet windows, PWs) of the sizes 5, 10 and25 to store the PHYModes used in the past, alongwith the information whether the respective packethas been acknowledged or not. The parameter val-

ues are the result of various tests to find optimizedparameters for a large range of possible test scenar-ios. As these PWs store short, medium, and long-term information they are referred to as short(PWs), medium (PWm) and long (PWl) packet win-dow (cf. (3.1)). The different window sizes (WSs)are labelled as short WSs, medium WSm, and longWSl. The latest packet determines the currentpacket number (CPN):

PWs fPCPN;PCPN1; . . . ;PCPNWSs g;

PWm fPCPN;PCPN1; . . . ;PCPNWSs ; . . . ;PCPNWSm g;

PWl fPCPN;PCPN1; . . . ;PCPNWSs ; . . . ;

PCPNWSm ; . . . ;PCPNWSl g:

3:1Each entry Pi 2 {0, 1} has two states: either thepacket was transmitted successfully (Pi = 0) or thepacket was not acknowledged (Pi = 1). Our proce-dure permanently calculates the ratio of successfulto unsuccessful transmissions (error ratio k, ERk)for each PWk (cf. (3.2)).

ERk 1

WSk

XCPNCPNWSk

PWk; k 2 fs;m; lg: 3:2

Eq. (3.2) shows how the ERk is calculated. When-

ever a packet is transmitted immediately after thePHYMode has been changed or if the link has beenidle for a certain period of time the respective LA in-stance is reset. All packet windows are deleted (en-tries are set to zero) and the new packet is labelledas the first one (CPN is set to 1).

Our LA algorithm is designed to differentiate twostates: the Init State and the Steady State. Afterthe creation or the reset of a LA instance, the LA

PC

Switch PhyMode

down

WS l

ERs := NACKs/ PWsERm := NACKm/ PWmERl := NACKl/ PWl

no

Idle

Reset

Steady State

CPN

Switch PhyMode

down

CPN >=

SRs := ACKs / WSsSRm := ACK m / WSmSRl := ACKl / WSl

no

Waiting

Reset

PC

Switch PhyMode

down

ERs := NACKs/ PWsERm := NACKm/ PWmERl := NACKl/ PWl

no

Idle

Reset

ACK or NACK

received

CPN

Switch PhyMode

down

CPN >=

ERs > ERLs

orERm > ERLm

yes

CPNWSs,m,l

NACKs,m,lERs,m,l,weightedERLs,m,weighted(up /down)V

Current Packet NumberSize of short, medium and

large packet window

Number of unacknowledged packets

Error Ratio

Error Ratio Limit

Weighting Vector

CPN < WS l

ERs := NACKs / WS sERm := NACKm/ WS mERl := NACK l / WS l

no

< ERL

Compute:

l

m

s

weighted

ER

ER

ER

*V=ER

Waiting

Reset

Switch PhyMode

up

ERweighted

Switch PhyMode

down

Init State

> ERLweighted,Down weighted,Up else

Fig. 5. Link adaptation working principle.


ARTICLE IN PRESS


6/21

starts in the Init State (cf. Fig. 5). When a new linkis established it is important to find an appropriatePHYMode before the IEEE 802.11 protocol startsdiscarding packets. Otherwise, the IEEE 802.11retry counter may expire [3]. Therefore, the LA

remains in Init State

until all lists PWk

are filled.In this Init State the LA converges fast from theinitial PHYMode to the actual link conditions, sinceonly PWs and PWm are considered for the decisionon switching to another PHYMode. Each PWk isassociated with a certain error ratio limit (ERLk).When the LA procedure operates in the Init Stateand the error ratio within PWs or PWm exceeds thecorresponding ERL, the PHYMode is immediatelydecreased (cf. (3.4) and (3.5)). The WSs has to becarefully selected. During our studies presentedhere, the initial PHYMode was kept equal to the

PHYMode used for layer three broadcast transmis-sions. Therefore, all layer three control packets aresent using QPSK 3/4. To further enhance the LAprocedure, after entering the Init State the LAinstance evaluates current or previous backwardslinks if available. If a backwards link exists and itsPHYMode is lower than the initial default PHY-Mode, the PHYMode of this backwards link is usedas the initial one, see Fig. 6. Node A transmits usingBPSK 1/2 to node B. Node A has already adaptedthe PHYMode to find the best suited one. Hence,

node B uses the same coding scheme of node A asthe initial PHYMode. This behaviour is based onelevated probability that both directions have com-parable propagation conditions. This educatedguess is used to improve the starting point for LA,as described above.

Once all lists have been filled the LA instanceswitches to the steady state where all lists are con-sidered for further calculations. To differentiatebetween long term and short term changes the threePWs,m,l are weighted using a weighting vector ~V (see(3.3)). The weighting vector contains a weight vi for

each PWk with vi 2 R \ 0; 1.

~V

v1

v2

v3

0B@

1CA with vi 2 R \ 0; 1 and X

i

vi 1:

3:3

The current PHYMode is taken as the current opti-mum as long it remains in a predefined range. Thebounds of this range are given by ERLWeighted,UP

and ERLWeighted,DOWN. If the weighted error ratiofalls below ERLWeighted,UP the PHYMode is in-creased and vice versa when the weighted error ratioarises over ERLWeighted,DOWN the PHYMode is de-creased. Our LA switching conditions are summa-rised in Eq. (3.4) and (3.5).

SwitchingDOWN

ERs > ERLs [ ERm > ERLm

for CPN < WPl;

vs

vm

vl

0B@1CA ERs; ERm; ERl > ERLWeighted;DOWN

for CPNPWPl:

8>>>>>>>>>>>>>>>:

3:4

SwitchingUP

non-switching up in Init State

for CPN < WPl;

vs

vm

vl

0B@

1CA

ERs; ERm; ERl < ERLWeighted;UP

for CPNPWPl:

8>>>>>>>>>>>>>>>:

3:5

BPSK (adapted)

QPSK (default)

BPSK (adapted)

BPSK (adapted)A

A B

B

YES

Equalize PHYModes

NO

YES

Link Adaptation

active

to Node A

Packet received fromTerminal A

IDLE

Fig. 6. Considering the bidirectional case.


ARTICLE IN PRESS


7/21

To be able to react to rapidly changing channelconditions, PWs should be highly weighted. Forthe simulation results presented here the weightsare set as follows: {v1 = 0.5; v2 = 0.3; v3 = 0.2}. Thismeans PWs weighted with 50%, PWm with 30%, and

PWl with 20%.The right choice for the WSs is a key requirementfor a sufficient LA operation. High data rate con-nections having small packets need larger WSs thana moderate data rate connection with large packets.Therefore, an additional function at each terminalhas to measure the data packet rate per link andto choose the packet window sizes independentlyfor each link. However, in addition to the packetrate, the WSs has to be defined small enough toavoid expiring the IEEE 802.11 retry counter [3].Therefore, during our studies a PWs of five packets

has been used to prevent packet loss in the InitState.

However, optimizing LA algorithm is out of thescope of this article. Our focus here is to increasead hoc routing performance using cross-layer infor-mation in support of ERRA and ERU. Neverthe-less, giving a detailed description of the LA isnecessary since they are none standardized LAs.Our LA turns out to be a suitable choice as basisfor a prediction algorithm. The next section showsthe LA behaviour and performance.

4. Link adaptation behaviour

We use two scenarios to present the stability andperformance of the LA developed. The velocitiesused in our simulation are applicable in cases such

as a walking person or a slowly moving vehicle. Fas-ter speeds would mean frequent changes of linksand thus result in an inconstant link quality. Vehiclebased scenarios do not provide a promising applica-tion for stand-alone wireless local area networks.

First, we investigated a basic scenarios compris-ing two nodes. One node remains immobile. Weplace the fixed node at the point of origin. Thesecond node is placed at a distance of 60 m, it movesstraight towards the fixed node and stops after leav-ing the range of node 1. The aim of this setup is togain insights into the detailed LA reaction and to

observe the stability of the LA. The scenario isshown in Fig. 7 on the upper left corner. Node 1transmits to node 2. The offered data rate was100 kb/s constant-bit-rate (CBR) with a packet sizeof 512 B. Both nodes transmit with a power of100 mW. The propagation coefficient of the physicalmodel was assumed to be c = 3.0 [1]. Realistic val-ues for c are between 2 (free-space propagation)and 5 (strong attenuation, e.g., because of citybuildings). The simulation is repeated with varyingspeeds of node 2.

Distance between Node 1 and Node 2 [m]

-60 -40 -20 0 20 40 60

0.25 m/s

0.5 m/s

1 m/s

2 m/s

4 m/s

Node 2

600

Node 1

- 60

BPSK 1/2

BPSK 3/4

QPSK 1/2

QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

Fig. 7. Throughput results of node 2, under different velocities.


ARTICLE IN PRESS


8/21

Fig. 7 presents the used PHYModes for packetstransmitted from node 1 to node 2. The adaptationsteps can be observed for velocities from 0.25 m/sup to 4 m/s. It shows that the link adaptation iscapable of swiftly adapting the PHYModes. An

influence of the velocity can be stated for 2 m/sand 4 m/s, whereas all other velocities show roughlysimilar adaptation behaviour. The LA needs longerto increase the PHYMode when the destination sta-tion approaches faster. The downward curves arealmost identical in all cases showing that thedecreasing quality of the link is detected compara-bly at any simulated speed.

5. Predicting link breakages

Continuous switching down of PHYModes is a

hint for the network layer that the link may breaksoon. This information triggers the routing algo-rithm instance to reconsider the route and to adaptit according to the emerging situation. Being able topredict a link breakage has large benefits. Commonrouting protocols react after a link is broken, whileusing LA information our solution acts before thebreakage occurs. The channel dump in Fig. 12 pre-sents three LA behaviour examples extractingtypical downstairs-patterns. The next step is to eva-luate the relation between a downstairs-pattern and

a possible route break.

5.1. The rating function

To investigate the LA behaviour and detectdownstairs-patterns automatically, we defined a rat-ing function RF(t) shown in Fig. 8. The RF(t) ranksthe importance of PHYMode changes in relation toa breakage (cf. Fig. 8). A change from BPSK 3/4 toBPSK 1/2 (value 7) for example is more important

for the breakage prediction than switching from16-QAM 3/4 to 16-QAM 1/2 (value 3).

An important issue in the context of detectionrate and false alarms is the history of the previousLA decisions. A detection threshold (DFthreshold)

has to be determined in accordance to these infor-mation. The time window taken into account forthreshold determination is named detection interval(DI). We sum up all ratings within a limited time-frame. Switching down is valued as negative, switch-ing up as positive. Whenever a LA algorithmadvises the MAC to use the last feasible PHYMode(BPSK 1/2), the prediction method uses the RF(t) tomeasure the link behaviour in the elapsed time-frame. The prediction algorithm sums up all valua-tions done prior to the step to BPSK 1/2. To judgethe shape and speed of the downstairs pattern the

sum-up results are compared to a certain detectionthreshold.

We introduce a small scenario to show an idea-lised use case for our breakage prediction and routerearrangement algorithms. The scenario is presentedin Fig. 9 and includes seven nodes. It explains themajor advantages of breakage prediction based onthe LA for ad hoc routing. The source node 1 isstationary. The other nodes are arranged in a circleuniformly distributed with a distance of 35 m. Node1 is 10 m apart from the circle. Node 2 is the desti-

nation node and circularly moving with the othernodes at a predefined speed of 1 m/s. The offeredtraffic stream has a constant bit rate of 100 kb/s.The packet size is 512 B and simulation time is210 s. The circular setup was chosen to minimizesimulation time and number of nodes involved.

The movement causes all established routes todegrade in link quality and eventually (without pre-diction) to break between node 1 and its next hoppartner. Those breakages accompany packet lossand rerouting. Our results in Fig. 10 show thatroutes using standard AODV experiences a longperiod of time using BPSK 1/2 before the link ulti-mately breaks. After the actual route break, result-ing in a loss of data packets, AODV reroutes andconsequently the link quality increases again.

Breakage prediction enables the routing algo-rithm to start this rerouting process as soon asBPSK 1/2 is reached. The new route is found andcan be engaged long before the actual route breakoccurs. The link quality is considerably improvedwhen using breakage prediction and more stableroutes are found. Furthermore, in the prediction

based case no link breaks or packet losses occur.

5

1

2

3

4

6

7

11 Rating

Distance

Breakage

BPSK 1/2

BPSK 3/4

QPSK 1/2

QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4 + -

Fig. 8. Rating function.


ARTICLE IN PRESS


9/21

Besides the basic evaluations, we stress the LAand prediction with a larger scenario to investigatethe LA performance under more realistic traffic con-ditions. A second setup shown in Fig. 11 presents anad hoc network containing 40 nodes. Except for thesource and destination node of the three routesinvestigated here, all nodes move according to theRandom Waypoint Model (RWP) [7]. However itis important to note that a RWP [7] scenario is a

worst case for breakage prediction. Any predication

is based on the assumption that a measured devel-opment continues in the future. Human movementin real life situation is for the most part goal-ori-ented. Therefore, under normal circumstances thelink degradation follows the assumption and canbe predicted with certain reliability. A RWP mobil-ity model generates just the opposite of a goal-ori-ented behaviour. We use it in our investigationsbecause it is the commonly used model to simulate

movement behaviour for ad hoc networks and to

25 m

25m

35 m

25 m10 m

35m 25mNode 1

Node 7 Node 6

Node 5

Node 4Node 3

Node 2

25 m

25m

35 m

25 m10 m

35m 25mNode 1

Node 7 Node 6

Node 5

Node 4Node 3

Node 2

Fig. 9. Idealised demonstration use case.

0 20 40 60 80 100 120 140 160 180 200

BPSK 1/2

BPSK 3/4

QPSK 1/2

QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

Time [sec]

Standard AODV

AODV + Prediction

Fig. 10. Benefits of prediction based routing.


ARTICLE IN PRESS


10/21

assure the superior properties of ERRA and ERUeven under worst-case conditions.

Each source node generates a traffic of 100 kb/susing a constant-bit-rate (CBR) traffic generator.The packet size is 512 B. The window sizes for theLA algorithm are set to WS = {WSs = 5; WSm =

10; WSl = 25}. Fig. 12 presents the corresponding

LA behaviour for the source with the node numberthree. Due to the mobility of the nodes, the numberof hops and the forwarding nodes change con-stantly. Fig. 12 presents the PHYMode of all pack-ets successfully sent by node 3. Node 3 as a source

node is not mobile. Several route breakages can beobserved along the routes (dashed lines). Thus,Fig. 12 presents the link breaks between station 3and its respective next hop-partner. Of course, linkbreaks may occur on different links along the wholeroute, too. Each link breakage results in a new routediscovery process.

We emphasise three particular situations inFig. 12 indicating a typical LA behaviour inadvance to a link failure. Breakages between node3 and its next hop on the route are marked with adashed vertical line. The node speed was set to

1 m/s for all moving nodes. Fig. 12 contains threesubfigures. The upper figure presents the PHY-Modes adjusted by the link adaptation process from650 s to 1350 s after the beginning of the simulation.Our LA algorithm switches several times to fasterPHYModes when link conditions improve and theother way when conditions decrease. Three typicalsituations are highlighted (by grey blocks) and pre-sented in detail below. Subplot Fig. 12(a) presentsthe typical channel pattern that can be observed inadvance of a link breakage. Fig. 12(b) highlights

the time interval [1000 s; 1040 s]. It presents an

100 m

Random Waypoint Mobility (1 m/s)

Destination (Fixed) Source (Fixed)

100m

Route (100kbit/s each Packet 512 Bytes)

Route 3

31 32 33 34 35

36 37 38 39

40

30

20

25

26

27

28

29

10 11

12 13

14 15

16 17 18

19

1 2

3 4

5 6

8

21

22

23

24

9

7

Route1

Route

2

Fig. 11. Ad hoc example.

1000670 12501010680 12701020690 12901030700 13101040710 1330

Time [s]Time [s] Time [s]

700 800 900 1000 1100 1200 1300Time [s]

BPSK 1/2BPSK 3/4QPSK 1/2

QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

Terminal 34

(c)Terminal 12

(a)

Terminal 29

(b)

Fig. 12. PHYMode dependent channel monitoring for multi-hop ad hoc networks.


ARTICLE IN PRESS


11/21

interfered pattern, although the pattern can be stillidentified. In addition, the third subplot Fig. 12(c)extracts one of the challenges our prediction algo-rithm has to deal with. The pattern of the first partof subplot (c) suggests an upcoming breakage. How-

ever, no break occurs in the highlighted situation.On the contrary, our LA algorithm increases thePHYMode. This can be observed several timesand depends on the mobility model used. Forinstances, with RWP mobility the node movesaway, reaches its drawn point and returns. There-fore, the link quality decreases when moving tothe far point and increase on the way back again.This is depicted in the first part of subplot (c).

As expected the link quality decreases withincreasing distances. Our LA procedure adaptswhenever necessary to avoid a connection interrup-

tion. Based on this behaviour, we are able to predicta link breakage.

Fig. 13 presents a histogram depicting the influ-ence of the detection interval and the detectionthreshold on the detection rate. The histogramshows the detection rate for the scenario presentedin Fig. 11. With an increasing detection intervalthe detection rate rises. Based on Fig. 13 one canalso state that a detection threshold lower than22 is not suitable. Applying a threshold of 18and a detection interval of around 40 s, allows pre-

dicting 70% of all occurring breakages. Apart fromthe detection rate, the number of needless routerearrangements is an important measure to evaluatethe prediction performance. Fig. 14 presents a histo-gram visualizing the percentage of correctlydetected breaks. Under worst-case conditions, upto 40% of all predictions were route breaks. The

remaining 60% represent false alarms identifyingand triggering the replacement of inefficient links.

Without link prediction, packet loss and inter-ruptions cannot be avoided. However, our predic-tion mechanism is able to avoid 70% of these

interruptions, even in this worst-case scenario.One has to keep in mind that there are numeroussituations, which cannot be acquired statistically.For instance, when an intermediate node leavesthe route and the prediction function regards oneof its hops as suspicious but a neighbouring hopbreaks first. In these situations, the update neverthe-less rearranges the route and inhibits the routebreakage. Hence, the breakage is bypassed but thisdoes not appear in Figs. 13 and 14, since the break-age was expected on a different link.

Our approach to weight the LA steps and to sum

up the weightings within a certain period is a veryelementary method. The analysis shown confirmsthe assumption that a prediction is possible andbeneficial. Approaches that are more complex mayraise the detection rate of the link breakages.

The detection interval (DI) is chosen accordingto the node speed and the movement direction. Itis very important to select a suited value for thedetection interval. However, DI strongly dependson the node speed and therefore cannot be deter-mined in advance. Consequently, we introduce an

adaptive windowing approach. Our LA starts tosum up all rating steps during a minimum detectioninterval (DImin). If the prediction threshold is notexceeded, the window size is enlarged until theupper limit (DImax) is reached. As long as theprediction results do not exceed the detection

1520

2530

3540

45

-28 -27-26 -25

-24 -23-22 -21

-20 -19-18 -17

-160

10

20

30

40

50

60

70

80

Detection thresholdDetectioninterval (DI) [s]

Percentageofpredictedbreakages

Fig. 13. Percentage of predicted breakages.

Detection

ThresholdDetection

Interval (DI) [s]15

2025

3035

4045

-22-21

-20-19

-18-17

-160

5

10

15

20

25

30

35

40

PercentageofCorrectlyDetected

Breaks

Fig. 14. Percentage of correct detections.


ARTICLE IN PRESS


12/21

threshold, no breakage is foreseen. If the thresholdis penetrated the network layer is informed thatthe link might break. The interrelationship betweenthe dynamic detection window, detection thresholdand the rating function is expressed in (5.1). Eq.

(5.1) sums up different intervals. Each interval startsat a different time in the past ranging from t-DIminto t-DImax and ends at the current time. Sinceswitching down is rated negative, the infimum is cal-culated over all sums. If the infimum falls short ofthe applied threshold Dthreshold a link breakage ispredicted.

Dthreshold 6 infDI2DImin;DImax

Xtt-DI

RFt

!: 5:1

The prediction is not only feasible at the sending

node, but also at the receiving node.Fig. 15 gives an overview of the different func-

tionalities and their ways of cooperation. The LAadapts the PHYModes to operate each link appro-priately. The rating function reviews the different

LA steps and the prediction function summarisesthe rating results and determines whether the linkis instable or not. In case the link is expected toterminate the route rearrangement algorithm isinstructed to look for an alternative route. Thus,

the prediction enables the route rearrangementalgorithms to circumvent numerous route break-ages.

6. Proper actions for upcoming link break

Assuming the link adaptation delivers the neces-sary information about the link state characteristics,this information triggers appropriate actions eitherto rescue the link and avoid a route rediscovery orto provide an improved link quality by finding anew route. In any case, several proper actions are

conceivable. The node that monitors incoming andoutgoing connections knows if one of these links,none or both of them have changing PHYModes.This enables the node to distinguish three differentcases:

Fig. 15. Cooperation between LA, rating function, prediction and network layer.


ARTICLE IN PRESS


13/21

1. The node recognizes that the LA for the outgoinglink is adapting towards lower transmissionmodes but the incoming link remains stable. Thisis a typical situation, where the next node on theroute wanders out.

2. The node recognizes that the incoming as well asthe outgoing links are adapting towards lowertransmission modes. This indicates movementof the node itself. Thus, the node regards itselfas an instable intermediate node.

3. The LA for the incoming link has to decrease thePHYModes but the outgoing link remains con-stant. That is typical for a preceding node mov-ing away.

In the scenario depicted in Fig. 16 all the threecases2 are presented. Node 4 experiences case 1,node 5 faces case 2 and node 6 is exposed to case 3.

6.1. ERRA: Early Route ReArrangement

In the previous section, we explained how toforecast the degradation of a link. This sectiondescribes an approach that profits from early trig-gering. The Early Route ReArrangement (ERRA)is derived from the local repair idea, which is partof the Ad Hoc on Demand Distance Vector(AODV) routing protocol [4,8].

Unlike the local repair idea, ERRA does not waituntil the link is broken. Prior to a breakage, ERRArearranges the route to avoid disruption. In Fig. 16an example is given. The initial route starts fromsource node 2 to destination node 7. One intermedi-ate node (5) is going to wander off. Node 4 firstdetects the movement; since it has to adapt thePHYMode for the outgoing link to node 5. Oncenode 4 uses PHYMode BPSK 1/2 it starts lookingfor a downstairs-pattern in the elapsed timeframe.If node 4 finds an indication for such a pattern, ittriggers the ERRA procedure to rearrange theroute.

Node 4 locally broadcasts3 a rearrangementrequest (ERRA_REQ) with a PHYMode higherthan the last one used for the connection and conse-quently, reducing the coverage range for theERRA_REQ (cp. Fig. 2). This ensures that the for-mer link between nodes 4 and 5 is not selected. InFig. 16 the request is sent with QPSK 1/2. There-

fore, node 5 does not receive the request directly.Node 8 forwards the request to node 6, which isaware of a route to the destination. Node 6responds (ERRA_REP) and provides the alterna-tive route. Afterwards, node 4 compares the hopcounts of the old and the new route. If the newhop count is less or equal, the alternative route isused immediately. Otherwise, node 4 uses the newroute immediately after the old connection finally

BPSK

3 6

7

24

8

5

PhyM

ode>BPSK

BP

SK

BP

SK

Initial Route

Alternative Route

ERRA_REQ (

ERRA_REP (

BPSK

3 6

7

224

8

5

55

PHYM

ode>B

PSK

BP

SK

BP

SK

Initial Route

ERRA_REQ ( broadcast )Alternative Route

ERRA_REP ( unicast )

Fig. 16. ERRA signalling.

2 Obviously, the three cases can also occur when the neighbourterminals are moving; however the upcoming situation could be

handled equally.

3 The time to life (TTL) for the broadcast is calculated

according the AODV local repair rules [4].


ARTICLE IN PRESS


14/21


15/21

Fig. 19 which contains three curves. The first onehighlights the LA steps using standard AODV andshows several drawbacks. Primarily it is obviousthat the link keeps BPSK 1/2 until the connectivityis finally lost. During this long period the link bur-dens the network by operating with an inefficientPHYMode to transmit its data. Owing to the factthat BPSK 1/2 is exhaustively used, the opportunityto rearrange the route and to select a better-suitednode is missed. Consequently, the route breaksand a certain number of packets is lost and thesehave to be repeated. The next step is re-establishingthe route and choosing an intermediate node, whichis already fading away. Therefore, the maximumPHYModes are limited to QPSK 3/4 after the firstthree breaks. After the fourth route is switchedtowards transmitting via the incoming nodes (trans-

mitting clockwise). Hence, for a limited time the

next node approaches and the link quality increases.Having passed the sender, it fades away again. Thatdescribes a typical behaviour found while usingmost approaches.

The medium curve represents the handling of thesituation by ERRA. Immediately after BPSK 1/2 isengaged, the rerouting process is started and ERRAswitches to a new route. Adapting the route withERRA is fast, which results in short peaks towardsBPSK 1/2. ERRA lost no packets on the completeturn, implying five route adaptations.

The newly engaged routes are more stable andallow higher PHYModes on the described link.The higher PHYModes shorten the packet trans-mission time and disburden the network capacity.Similar to situation with AODV, a switch to aclockwise direction generates increasing link quality

due to an approaching next hop partner.

Standard AODV

AODV + ERRAAODV + ERU Break with Packet loss

Route Adaption

0 20 40 60 80 100 120 140 160 180 200


QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

AODV + ERU


QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

Standard AODV


QPSK 3/4

16-QAM 1/2

16-QAM 3/4

64-QAM 2/3

64-QAM 3/4

AODV + ERRA

Time [s]

Fig. 19. Comparison of different routing behaviour.


ARTICLE IN PRESS


16/21

ERUs mechanism is based on locally circumvent-ing the gap. ERU starts adapting the route from thenode behind the upcoming breakage. Consequently,the gap is filled by the mechanism with the nextavailable node. Route breaks and packet losses are

avoided. The tradeoff for this is increasing routelengths. The route is not switched to a clockwisetransmission at the same time like ERRA. Whenthe destination appears, in the neighbourhood ofthe breakage the initial situation is re-establishedand the direct route is chosen.

Figs. 20 and 21 highlight the handling of firstsituation with degrading link quality with AODVand ERRA. Fig. 20 displays the situation after60 s for AODV. The link finally breaks and twopackets are lost. After the break (60 s) node 7 isinserted as new intermediate node, whereas ERRA

in Fig. 21 reacts earlier and rearranges after 40 s.The PHYMode curve in Fig. 21 clearly displaysthe predicted downstair-pattern prior to routeswitching. We present no detailed figure for ERUbecause it shows the same fast reaction as ERRAin this situation.

In addition, we show the detailed figures for thesituation where AODV and ERRA switch to clock-wise routing (see Figs. 22 and 23). Fig. 24 showsthat ERU keeps the route direction and fixes thegap locally. Both proposed protocols guarantee

the route continuity, forfeit no packets and preventroute breaks.

After discussing the idealised setup, we completeour simulation analysis with the worst-case scenariomention in Section 5 and displayed in Fig. 11.

Three routes are loaded with constant-bit-ratetraffic. The offered traffic increases with every simu-lation. The packet size is set to 512 B. RTS/CTShandshake is used to reduce the hidden node prob-lem. Fig. 25 presents the percentage of average

throughput over all three routes. The figure com-pares ERRA and ERU with the AODV local repairmechanism. Generally, under high load conditions,the network is congested and cannot deliver allpackets. It is important to see in Fig. 25 that nothroughput degradation occurs using the ERRAor ERU algorithms. This is because packets are onlyaffected, if a breakage situation is foreseen. Thus,both approaches only delay a small fraction ofpackets.

In terms of additional signalling overhead ERRAand ERU are comparably efficient with local repair

from AODV. ERRA uses the same amount of sig-nals and bytes per signal as local repair. However,ERRA redirects the route in advance and disbur-dens the channel, by choosing a higher PHYModefor signalling, thus producing shorter packets.ERU requires knowledge about the actual neigh-bourhood. As proposed by AODV [4] as anoptional feature, nodes learn the neighbourhoodusing hello messages. Hello messages increase thesignalling overhead, but enable ERU to smoothlyadapt the routes and avoid breakages. ERRA and

ERU perform beneficially unless the network isuncongested, which represents the usual case. Assoon as the network cannot carry the offered traffic,routes break and the user satisfaction decreases.Neither of the two approaches can improve the sit-

806020 40

Node 1

Node 2

Node 3Node 7

Node 6

Node 7

From Source Node to

Node 2

Previous Route

Adapted Route

Lost Packets

Node 5

Node 4

Fig. 20. AODV handles the first situation with degrading link

quality.

5

Node 1

10 20 30 50 60

Node 2 Node 7

From Source to Node

Previous Route

Adapted Route

Lost Packets

Node 2Node 3

Node 4

Node 5Node 6

Node 7

Fig. 21. ERRA handles the first situation with degrading link

quality.


ARTICLE IN PRESS


17/21

Previous Route

Adapted Route

Lost Packets

Node 1

Node 2

Node 7

Node 5

Node 4

160

Node 4

From Source Node to

120 140

Node 5

Node 3

Node 6

Fig. 24. Fourth route adaptation by ERU.

Node 1

Node 5

Node 3

From Source Node to

140 160140

Node 2

Node 5

Previous Route

Adapted Route

Lost Packets

Node 3

Node 4

Node 5

Node 6

Node 7

Fig. 22. AODV handles fourth route break.

Node 1

120 140 160

From Source Node to

Node 5

Node 3

Node 2

Node 4

Node 5Node 6

Node 7

Previous Route

Adapted Route

Lost Packets

Node 3

Fig. 23. Fourth route adaptation by ERRA.

0.3 0.45 0.6 0.75 0.9 1.5 2.25 3.0 3.75 4.5

Offered load [Mb/s]

Percentageofthroughput[%]

20

30

40

50

60

70

80

90

100

10

0

AODV + Local RepairAODV + ERRAAODV + ERU

Fig. 25. Throughput percentage.

100

101

102

103

104

Packet delay [ms]

10-6

10-5

10-4

10-3

10-2

10-1

100

P(X>t)

Offered Load: 0.6 Mb/s


Fig. 26. Complementary cumulative distribution function

(CCDF) of the packet delay.


ARTICLE IN PRESS


18/21

uation. However, both do not worsen the situation.Under overload conditions, ERRA and ERU per-form similar to local repair. Figs. 25 and 27 leadto the same conclusions.

Fig. 26 presents the packet delay for all threeroutes. It shows the results for both proposedapproaches in comparison with the AODV local

repair procedure.Apart from the typical delay characteristics orig-

inated from the reactive routing protocol, it is obvi-ous that both approaches outperform the AODVlocal repair mechanism. When the offered trafficincreases, ERRA and ERU reveal the same perfor-mance as local repair. The involved signalling doesnot negatively influence the throughput (cf. Fig. 25).

Under low load conditions, ERU shows the low-est delay. With the increase of the traffic loadERRA and ERU still achieve smaller delay than

local repair (cf. Fig. 27). We can conclude that linkprediction improves the route continuity, avoidspacket losses, and decreases packet delay becauseof the timely and accelerated route discoveryprocedures.

9. Conclusions

This article explains the Early Route ReArrange-ment (ERRA) and the Early Route Update (ERU)approaches based on link breakage prediction. Wepresent the idea to predict the link state based uponthe link adaptation behaviour. We describe a suit-able link adaptation (LA) and show its functional-ity. Based on link adaptation, we introduced aprediction algorithm that rates the PHYModechanges in terms of their importance towards anupcoming route interruption. The prediction

100 101 102 103 104

Packet delay [ms]100 101 102 103 104

4

Packet delay [ms]

100 101 102 103 10

Packet delay [ms]

10-6

10-5

10-4

10-3

10-2

10-1

100

P(X

>t)



10-6

10-5

10-4

10-3

10-2

10-1

100

P(X

>t)



10-6

10-5

10-4

10-3

10-2

10-1

100

P(X>t)



Fig. 27. Packet delay (CCDF) for an increased traffic offer.


ARTICLE IN PRESS


19/21

enables our route rearrangement protocols to acttimely and prevent route breaks and packet losses.In particular, ERRA and ERU are designed to ben-efit by being triggered early from prediction algo-rithms, acting early enables shorter and more

stable routes. Better PHYModes result in shorterpacket and less transmission duration, thus reduc-ing network congestion and disburdening networkcapacity. Higher layer transport protocols notice-ably benefit from reduced packet loss and routebreaks.

This article presents an efficient control loopbased upon cross-layer information shared betweenmedium access and network layer. Thereby, thisarticle gives a first glimpse to the potentials ofcross-layer interaction.

10. Discussion

Evidently, our LA algorithm presented here can-not distinguish between packet loss originated fromtransmission errors or collisions. In a congestedchannel, the LA procedure detects unacknowledgedpackets and decreases the PHYMode. However, thisincreases packet transmission time and worsens thesituation. In the scenarios described here, thisbehaviour is unfavourable. In contrast, our currentwork monitors the MAC queue length and enhancesthe LA efficiency thereby. If the channel busy timeincreases, monitoring MAC transmission queuesreveals that congestion is more probable. Therefore,the LA algorithm decisions could be improved withMAC queue length information taken into account.While our current LA algorithm will not increasethe PHYMode in case the channel is congested, anenhanced LA might advise to do so. With increasedPHYMode, congestion could be reduced. Con-gested links using the fastest PHYMode need todetect and establish new routes around the local

congestion.

Acknowledgements

This work was supported by T-Systems, Deut-sche Telekom and the German research project IPo-nAir, funded by the German Federal Ministry ofEducation and Research. The authors would liketo thank the members of the project for the valuablediscussion. The contributions of our colleague Mi-chael Fuhrmann are highly appreciated.

References

[1] B. Walke, Mobile Radio Networks, first ed., Wiley, NewYork, 1999.

[2] T. Goff, N.B. Abu-Ghazaleh, D.S. Phatak, R. Kahvecioglu,Preemptive routing in ad hoc networks, in: Proceedings of

ACM/IEEE MobiCom, July 2001, pp. 4352.[3] IEEE LAN/WAN Standards Committee, Part11: WirelessLAN Medium Access Control (MAC) and physical layer(PHY) specifications: high-speed physical layer in 5 GHzband, IEEE Std 802.11a-1999, IEEE, New York, November1999.

[4] C. Perkins, E. Belding-Royer, S. Das, Ad hoc on-demanddistance vector (AODV) routing, RFC 3561, IETF NetworkWorking Group, July 2003, Category: Experimental.

[5] C.E. Perkins, Ad Hoc Networks, Addison-Wesley, NewYork, 2000.

[6] E. Weiss, M. Frewel, B. Xu, G. Hiertz, Improving ad hocrouting for future wireless multihop networks, in: Proceedingof European Wireless 2004, Barcelona, Spain, February 2003.

[7] C. Bettstetter, H. Hartenstein, X. Perez-Costa, Stochasticproperties of the random waypoint mobility model, in:ACM/Kluwer Wireless Network, Modelling and Analysis ofMobile Networks, March 2003 (special issue).

[8] E. Weiss, G. Hiertz, S. Hischke, B. Xu, Performance analysisof AODV on top of IEEE 802.11a and the impact of the5 GHz channel, in: Proceedings of Wireless 2003, Calgary,Canada, July 2003.

[9] A. Kamerman, L. Moneban, A high-performance wirelessLAN for the unlicensed band, in: WaveLAN II, Bell LabsTechnical Journal (1997) 118133.

[10] G. Holland, N.H. Vaidya, P. Bahl, A rate-adaptive MACprotocol for multi-hop wireless networks, in: Mobile Com-puting and Networking, 2001, pp. 236251.

[11] B. Sadeghi, V. Kanodia, A. Sabharwal, E. Knightly,Opportunistic media access for multirate ad hoc networks,in: Mobicom 2002, September 2002.

[12] S. Mangold, S. Choi, N. Esseling, An error model for radiotransmissions of wireless LANs at 5 GHz, in: Proc. AachenSymposium, 2001, pp. 209214.

[13] J. Khun-Jush, P. Schramm, U. Wachsmann, F. Wenger,Structure and performance of the HIPERLAN/2 physicallayer, in: Proc. IEEE VTC Fall-1999, Amsterdam, TheNetherlands, September 1999, pp. 26672671.

[14] Chair of Communication Networks, RWTH Aachen Uni-versity, Kopernikusstrae 16, 52074 Aachen, Federal Repub-lic of Germany. Available from: .[15] C.E. Perkins, Pravin Bhagwat, Highly dynamic destination-

sequenced distance vector routing (DSDV) for mobilecomputers, in: Proc. SIGCOM94 Conference on Commu-nication Architectures, Protocols and Applications, August1994, pp. 234244.

[16] S. Murthy, J.J. Garcia-Luna-Aceves, An efficient routingprotocol for wireless networks, in: Routing in MobileCommunication Networks, ACM Mobile Networks andApplications Journal (1996) (special issue).

[17] Z. Haas, M. Pearlman, The performance of query controlschemes for the zone routing protocol, in: Proc. SIG-COMM98 Conference on Communications Architectures,Protocols and Applications, September 1998.


ARTICLE IN PRESS
http://www.comnets.rwth-aachen.de/http://www.comnets.rwth-aachen.de/http://www.comnets.rwth-aachen.de/http://www.comnets.rwth-aachen.de/


20/21

Erik P. Weiss received his diplomadegree in Electrical Engineering fromAachen University, RWTH, Germany,in 2001. After his studies he joined theChair of Communication Networks(ComNets) at RWTH Aachen Univer-

sity, where he is working towards hisPhD degree. He participates in theIPonAir project in cooperation withT-Systems. His working areas are theintegration of heterogeneous systems, IP

mobility, Cross-Layer Communication, and Routing in Ad HocNetwork. His current research interests are the development of acommon inter-system architecture, performance analysis of ver-tical handover mechanisms between GSM/GPRS/UMTS andIEEE 802.11a/g/b. At present he is involved in the Europeanproject MYCAREVENT for vehicular diagnosis and mainte-nance and leads the work package for mobile communication(WP4). He is a student member of IEEE, inventor/co-inventor ofseveral patents and has authored/co-authored several papers at

IEEE conferences.

Guido R. Hiertz studied electrical engi-neering at RWTH Aachen University.During his studies he focused on com-munication networks, computer scienceand engineering. In his diploma thesis(MS) he included 802.11e functionalityin an advanced simulation tool and sur-veyed QoS supporting procedures. SinceApril 2002 he is with the Chair ofCommunication Networks (ComNets) atRWTH Aachen University, working

towards his PhD. His main research fields are support for QoSWLAN (802.11e), protocols for Gigabit (802.11n) and Meshwireless LAN (802.11s) and PAN (802.15.5). He has beeninvolved in and the editor of different research projects. Atpresent, he is involved in the research project Wireless Gigabitwith advanced Multimedia support (WIGWAM) that is fundedby the German Federal Ministry of Education and Research. Hehas been involved in the design of the high speed WPAN MACprotocol of WiMedia Alliance that became ECMA standard in2005. Since 2004 he is a voting member of IEEE 802.11, where hesubmitted several presentations and proposals. He is chartermember of Wi-Mesh Alliance that submitted a joint proposal to802.11s. He is the inventor/co-inventor of several patents. As

student member of IEEE, he has authored/co-authored severalpapers at IEEE conferences.

Bangnan Xu received his BSc and MScdegrees in electrical engineering fromDalian Maritime University, Dalian,China in 1986 and 1989, and PhD (Dr.-Ing.) degree in information technologyfrom Aachen university of Technology in

2002, respectively. From 1996 to 2001,he was a research assistant at the chair ofcommunication networks, Prof. Dr.-Ing.B. Walke, Aachen University of Tech-nology. Since 2001, he has been with T-

Systems, Deutsche Telekom, working as R&D Fellow and Pro-ject manager in the department of Mobile and Wireless Solutions.

Sven Hischke received the diplomadegree in Electrical Engineering from theUniversity of Applied Sciences Giessen-Friedberg and the PhD degree from CityUniversity London in 1996 and 1999,

respectively. From 1999 to 2003 he waswith T-Systems in Darmstadt where hewas responsible for different researchprojects in the area of broadband wire-less systems and mobile networks focus-ing on the design of IP-based

architectures. In October 2003 he joined the corporate innovationdepartment of Deutsche Telekom as a R&D program managerfor broadband wireless access and global seamless network pro-jects. Since May 2005 he is a senior manager innovation man-agement of Deutsche Telekom and is working on theidentification and development of group wide products andbusiness models in the area of seamless services.

Bernhard H. Walke is running the Chairfor Communication Networks (Com-Nets) at RWTH Aachen University,Germany, where about 30 researcherswork on topics like air-interface design,development of tools for stochastic eventdriven simulation and analytical perfor-mance evaluation of services and proto-cols of XG wireless systems. Most of thiswork continuously has been funded fromthird parties grants. He is author of the

2002 book Mobile Radio Networks Networking, Protocols

and Traffic Performance and co-author of the 2001 book UMTS The Fundamentals and the forthcoming 2006 book IEEE 802


ARTICLE IN PRESS


21/21

Wireless LAN/PAN/MAN Systems: Standards, Models andTraffic Performance. He has been a board member of ITG/VDEand is Senior Member of IEEE. He has served as ProgrammeCommittee and Steering Committee Chair of various conferenceslike the European Wireless (EW) conference that he co-founded.In 2005, he was the Scientific Chair of IEEE-PIMRC 2005,Berlin. His group has substantially contributed to the develop-ment of standards like ETSI/GPRS, ETSI/BRAN HiperLAN2,CEN TC 278 DSRC (electronic fee collection), IEEE 802.11e,802.16 and 802.15.3. From 2001 to 2003 he was an elected Chairof Working Group 4 (New Technologies) of the Wireless WorldResearch Forum. Prior to joining academia, he worked for 18years in various industry positions at AEG Telefunken (nowEADS AG). He holds a Dr. degree (1975) in information engi-neering from University of Stuttgart, Germany.

Sebastian Gross studies towards hisdiploma degree in Electrical Engineeringat the RWTH Aachen University. Hehas taken a strong interest in wirelesscommunication and networking. Afterfinishing his student thesis at the Chairof Communication Networks at RWTH

Aachen University, he plans to write adiploma thesis in the field of wirelesscommunication networks.


ARTICLE IN PRESS

Documents

Crosslayer Function