Realistic propagation simulation of urban mesh networks q

Computer Networks 51 (2007) 3392–3412

www.elsevier.com/locate/comnet

Realistic propagation simulation of urban mesh networks q

Vinay Sridhara, Stephan Bohacek *

University of Delaware, Department of Electrical and Computer Engineering, Newark, DE 19716, United States

Received 24 March 2006; received in revised form 9 December 2006; accepted 22 January 2007Available online 12 February 2007

Responsible Editor: G. Morabito

Abstract

Simulation plays an important role in the verification of mobile wireless networking protocols. Recently, several cities haveeither begun deploying or are completing plans to deploy large-scale urban mesh networks (LUMNets). On the other hand,the networking research community has little expertise in simulating such networks. While the protocols are simulated rea-sonably realistically, the propagation of wireless transmissions and the mobility of nodes are not. Today, simulations typicallymodel propagation with either the free-space model or a ‘‘two-ray’’ model that includes a ground reflection. Such models areonly valid in open space where there are no hills and no buildings. Since wireless signals at the frequencies used for mobilewireless networking are partly reflected off of buildings and partly is transmitted into the building, the presence of buildingsgreatly influences propagation. Consequently, the open-space propagation models are inaccurate in outdoor urban areas.Indoors, the open-space models are not even applicable. This paper presents guidelines for simulating propagation in suchurban settings. Extensive background discussion on propagation is also included. The techniques for propagation are vali-dated against propagation measurements. The techniques discussed are implemented in a suite of tools that are compatiblewith protocol simulators and are freely available for use.� 2007 Elsevier B.V. All rights reserved.

Keywords: Wireless mesh networks; Modeling and simulation; Wireless propagation

1389-1286/$ - see front matter � 2007 Elsevier B.V. All rights reserved

doi:10.1016/j.comnet.2007.01.035

q This work was prepared through collaborative participationin the Collaborative Technology Alliance for Communicationsand Networks sponsored by the US Army Research Laboratoryunder Cooperative Agreement DAAD19-01-2-0011. The USGovernment is authorized to reproduce and distribute reprintsfor Government purposes notwithstanding any copyright nota-tion thereon. This work was support in part by NSF grantANIR-0322744.

* Corresponding author. Tel.: +1 302 831 4274.E-mail address: [email protected] (S. Bohacek).

1. Introduction

Recently, there has been interest in developingprotocols for urban mesh networks such as the onesto be deployed in Philadelphia [1], San Francisco [2],Taipei [3], Minneapolis [4], Anaheim, California, andTempe, Arizona. Simulation is the common tech-nique to determine the performance of these proto-cols. However, today, most simulations use thetrivial disk propagation model (i.e., the signal propa-gates exactly R meters and no further) or a highly ide-alized propagation model such as the free-space orthe two-ray model. Due to the presence of buildings,

.

mailto:[email protected]

V. Sridhara, S. Bohacek / Computer Networks 51 (2007) 3392–3412 3393

propagation in urban environments is far more com-plicated than the propagation presented by these sim-ple models. Consequently, channel variability, whichis a key aspect of wireless networking, is not wellmodeled in today’s simulations. The result is thatmany insights gained from the free-space environ-ment do not necessarily hold in the urban environ-ment, casting doubt on the applicability of theconclusions drawn from today’s simulations.

While the reasons that realistic propagation mod-els are typically not included into network simula-tions is not well documented, it seems that thetypical reasons include the belief that propagationmodeling is computationally intractable and thatpropagation cannot be realistically modeled due tosmall-scale fading and delay-spread. To the contrary,this paper demonstrates the feasibility of includingrealistic propagation into network simulation. Tothis end, a simulator has been developed in accor-dance with the guidelines set forth in this paper [5].This simulator is validated in three scenarios and itis shown that when used with packet level simulation,realistic propagation has a dramatic impact onsimulation results. Furthermore, while realistic prop-agation is computationally intensive, the computa-tional complexity of the approach taken here resultsin the propagation aspects of the typical simulationtaking approximately as long as the processing ofthe discrete events by the packet simulator. Conse-quently, this paper provides techniques and methodsfor greatly improving the quality of simulations.

In order to justify the simulation strategy used,an overview of the key features of propagation inan urban environment is presented. This non-math-ematical discussion of propagation also providesnetworking researchers with an intuition of propa-gation in urban environments. For example, someof the myths that can be found in the networking lit-erature are dispelled, and issues that are neglected inthe networking literature, such as the importance ofbuilding materials, are discussed.

The remainder of this paper proceeds as follows.In the next section, a brief discussion of urban meshnetworks is provided. Section 3 discussed stochasticmodels of propagation. While stochastic models areoften used in communication theory, their applica-bility to networking is limited, and, as will be dis-cussed, they should be used with care. In contrast,the propagation simulator discussed here is deter-ministic. Section 4 discusses several characteristicsof propagation. Specifically, the subsections ofSection 4 discuss transmissions through walls,

reflections off of walls, multipath fading, diffraction,scattering, time-varying channels, and delay-spread.Section 5 discusses computational issues, discussesdetails of the UDelModels’ propagation simulator,and presents validation of the simulator. Section 6provides some discussion on the impact that realis-tic simulation has on the performance of network-ing protocols. As one might expect, traditionalsimulations that use random way-point mobilityand free-space propagation yield very differentresults than the simulations carried out using realis-tic mobility and propagation. It is important to notethat realistic simulation of wireless networks is anongoing effort and that as the computational capa-bilities increase, more detailed and accurate simula-tions are possible. Some areas that will benefit fromfurther effort and improved accuracy are discussedin Section 7. Related work on propagation simula-tion is provided in Section 8 and finally, concludingremarks can be found in Section 9.

2. Urban mesh networks

A distinguishing feature of urban mesh networks(UMNets) is the presence of a large number ofinfrastructure nodes, each with relatively small cov-erage. This contrasts cellular networks that have farfewer infrastructure nodes, each with large regionsof coverage. UMNets also differ from traditionalMANETs in that UMNets have a fixed infrastruc-ture that includes some wired base stations. Onthe other hand, UMNets are similar to MANETsin that paths may cross multiple wireless links.However, a path in a UMNet typically includesmultiple infrastructure to infrastructure hops anda single infrastructure to mobile hop (see Fig. 1).Another difference between these networks is thatwhile cellular networks are carefully planned andMANETs are not planned at all, UMNets occupythe middle ground in that they are somewhatplanned. This lack of careful planning may resultin infrastructure nodes having highly overlappedcoverage areas, in contrast to cellular networks.

Besides architectural aspects, these networks dif-fer in terms of propagation. In cellular networks,the network performance is largely controlled bypropagation between the outdoor base station andthe mobile client. As shown in Fig. 1, in UMNets,propagation between the infrastructure and themobile client is not the only concern. For example,communication between infrastructure nodes is crit-ical. Also, in cellular networks, transmissions by

Fig. 1. Urban mesh networks are composed of wireless linksbetween infrastructure nodes and wireless links between infra-structure and mobile clients. Both types of nodes may be indoorsor outdoors and may be elevated or at street level. Communi-cation between mobile clients and the Internet is provided viamultiple wireless hops over the infrastructure (dotted lines).Many types of signal propagation are relevant. In the figureabove, the signals that reach the car are impacted by diffractionaround the corner of the building, propagation through theinterior of the building, transmission through an exterior buildingwall, and reflections off of a building and off of the ground.

3394 V. Sridhara, S. Bohacek / Computer Networks 51 (2007) 3392–3412

nearby clients are controlled with CDMA, FDMA,or TDMA, and hence, client-to-client interference isnot important (e.g., indoor propagation has littleimpact on performance). However, in UMNets,the coverage areas of infrastructure nodes maygreatly overlap, and hence clients communicatingwith different infrastructure nodes may experienceconsiderable interference. Consequently, indoor-to-indoor propagation is important as well as the prop-agation between street level nodes. In summary, theperformance of UMNets is impacted by

• propagation between mobile clients, which may beindoors or outdoor and, if outdoors, are at streetlevel,

• propagation between mobile client and infra-structure nodes, where both types of nodes couldbe either indoors or outdoors, but outdoor infra-structure nodes are typically elevated, and

• propagation between infrastructure nodes, where,again, these nodes can be indoors or outdoors.

1 For the remainder of the paper, we specify channel gains indB.

3. Stochastic models of propagation

One of the most important aspects of propagationis the channel gain (other characteristics are discussed

below). The objective of propagation simulation is todetermine the channel gain (and other channelcharacteristics). The physical layer model then usesthese channel characteristics to determine the proba-bility of transmission error. The relationship betweenthe received signal power Preceived and transmitter sig-nal power Ptransmitted is Preceived = K · H · Ptransmitter,where K is a constant that depends on the wavelengthand the antennas, and H is the channel gain. It is com-mon for this relationship to be specified in dB1 anddBm (dB milliWatts), i.e.,

P received½dBm� ¼ 10log10K þ H ½dB� þ P transmitter½dBm�:

To get an idea of common values, consider the typ-ical 802.11b transmissions. In this case, the trans-mission power is 15 dBm and 10 log10K = �41 dB.Furthermore, the sensitivity of commercially avail-able 802.11b receiver cards is approximately�93 dBm when receiving at 1 Mbps. Thus, if thechannel gain is lower than �67 dB, then it is notpossible to decode the transmission with marginalreliability. It is also common to require an extra gain

margin to allow for the signal to be decoded reliably(not simply marginal reliability), to account forlosses due to connectors and other analogue cir-cuits, and to allow for overly optimistic sensitivityspecifications. The gain margin depends on thetransmitter/receiver manufacturers and systemdesign, and can range from 3 dB up to 9 dB, whichrequire the 802.11b communication channel gains tobe above �64 dB and �58 dB respectively.

The propagation of wireless signals can be accu-rately modeled by Maxwell’s Equations. However,it is beyond today’s computational abilities to simu-late large urban regions using these equations. Forthis reason, there has been extensive research efforton developing techniques that provide computation-ally tractable estimates of the channel gain and othercharacteristics. In general, the techniques can beclassified into deterministic and stochastic models.Furthermore, they can be divided into site specificapproaches (i.e., detailed information about theenvironment is included into the propagation model)and non-site specific approaches. As discussed next,for simulation of urban mesh networks, determinis-tic site specific propagation models are appropriate.

We can easily rule-out deterministic non-site spe-cific approaches such as the Okumura–Hata model


[6] or COST 231 [7]. Such models provide a relation-ship between the channel gain and the distancebetween the receiver and transmitter. However,these models only provide an average behaviorand do not model the variability of the channel.

On the other hand, non-site specific stochasticmodels can account for channel variability andhence are widely used in communication theory.However, there are many pitfalls to stochastic mod-els which often make them not applicable to urbanmesh network simulation. Consider the well estab-lished lognormal shadowing model [8]. This modelspecifies that the received signal power obeys

P received ¼Kda � Y � P transmission; ð1Þ

where K is a constant, d is the distance between thetransmitter and receiver, a is the attenuation factor,Ptransmission is the transmitted power, and Y is a log-normally distributed random variable that accountsfor shadowing. This model dictates that if a trans-mitter and receiver are placed at randomly selectedlocations, then the relationship between distance,transmitted power and received power obeys (1).While the model has been verified through experi-mentation, the model focuses only on a single link.On the other hand, in networking, the interest ison a large number of links, specifically, the focusis on the graph formed by links. If the relationshipbetween received and transmitted signal strengthmerely obeys (1), then the graph formed by linkswould be a type of a random graph, whereas anurban setting induces a graph with deterministiccomponents. Consider the left and right frames in

Fig. 2. Left, channel gain generated by a correlated lognormal model. Rrectangle shapes are buildings and the shaded boxes inside the rectanbuilding. In both the left and right frames, the star marks the location

Fig. 2. The left-hand frame shows the channel gains(including correlations as described in [9,10]) when alognormal shadow fading model is used, while theright-hand shows the channel gains in an urban areathat is found through the techniques described inthe next sections. While it is reasonable to assumethat even in the urban setting, if two arbitrary loca-tions are selected, then the signal strength will fol-low the lognormal distribution (i.e., (1) is valid), itis quite clear that channel gain is not random.Rather, the channel gain is greatly influenced bythe map of the urban area. For example, the signalpropagates far down the streets, while it propagatespoorly into or through/over buildings. To see theimplication of this, consider the least-hop pathshown in Fig. 3. In this example, only the first andlast hop are able to propagate between indoor andoutdoor, while other hops follow streets (outdoorto outdoor propagation). Thus, we see that thepropagation results in routes in urban areas thatare not random, but are dependant on the structureof the city. For this reason, we do not use non-sitespecific models, and focus on site specific modelswhere the layout of the city and buildings influencesthe propagation. Currently, the UDelModels’ prop-agation simulator is deterministic. However, futurework will include appropriate stochastic elements.

4. Characteristics of urban propagation

4.1. Overview

As mentioned above and depicted in Fig. 1,there are several aspects of propagation that are

ight, channel gains generated by ray-tracing in an urban area. Thegles represent channel gain to locations on the first floor of theof the transmitter.

Fig. 3. Routing in an urban ad hoc network. The source anddestination are inside of two different buildings. The shortest hoppath travels along the road, it does not follow the shortestgeometric route (i.e., a straight line).


important for UMNets. Specifically, the relevanttypes of propagation include line-of-sight, reflec-tions off of buildings and the ground, transmissionsthrough exterior walls of buildings, indoor propaga-tion, and diffraction around and over buildings.While particular types of communication might bemore impacted by particular types of propagation(e.g., indoor-to-indoor communication is mostlyimpacted by indoor propagation), these aspects ofpropagation can impact any type of communication(e.g., indoor signals may exit the building, experi-ence a reflection, reenter the building, and impactthe communication between two indoor nodes). Inthis section, these types of propagation are brieflydiscussed. The objective is to provide intuition ofpropagation in urban environments. While thereare mathematical models of propagation, we donot provide any mathematical details. An excellentreference that contains mathematical details is [11].

Remark 1. We typically focus on the propagationfor a particular transmitter and receiver. However,due to reciprocity (e.g., [12]), the channel betweentwo antennas does not depend on which one is thetransmitter and which is the receiver. To somedegree, the symmetry of channels contradicts theobserved asymmetry of links (e.g., [13,14]). It isimportant to note that the observed asymmetry oflinks is not due to the channel but due to differencein the analogue electronics in the transmitters andreceivers. These differences may result in unequaltransmit powers and unequal receiver attenuation.In some cases, these differences can be reduced bycalibrating the transmit powers. It is also possiblethat one node experiences more interference thananother node. In this case, the probability of error-

free packet transmission depends on which node isthe transmitter.

4.2. Transmissions

In urban mesh networks, reflections off of objectssuch as buildings and the ground and transmissioninto buildings play a primary role in the magnitudeof the received signal strength. Diffraction and scat-tering are also important, but from numerical exper-iments it can be shown that they play a secondaryrole. When the object that the signal is hitting ismuch larger than the wavelength, then the behaviorof the signal can be accurately modeled as a ray thatis partially reflected off of the object, partially trans-mitted through the object, and partially absorbed bythe object.

If this ray model is accurate, then there are wellknown formulas that describe the behavior (e.g., see[11]). In general, the reflection can be written in termsof six components, H?TransmitðhIncidence; er; e00r ;w; f Þ,H?ReflectðhIncidence; er; e00r ;w; f Þ, H?AbsorbðhIncidence; er; e00r ;w; f Þ, H kTransmitðhIncidence; er; e00r ;w; f Þ, H kReflectðhIncidence;er; e00r ;w; f Þ, and H kAbsorbðhIncidence; er; e00r ;w; f Þ wherehIncidence is the angle between the ray and the material,e0r and e00r are the real and imaginary part of the relativedielectric constant of the material, w is the width ofthe material, f is the frequency of the signal, and thesuperscript denotes the polarization.

Instead of detailing how each of these factorsimpacts the propagation, we will consider an exam-ple. We examine the received signal strength on oneside of a wall with a transmitter on the other side asshown in left-hand frame of Fig. 4. We consider thereceived signal strength in two locations, specifi-cally, one at the same height as the transmitterand one elevated from the transmitter. We assumethat the signal strength decays according to free-space propagation and then intersects the wall andsuffers insertion loss H kTransmit (where we assume thatthe wireless signal is vertically polarized, hence the ksuperscript). We take the frequency to be 2.4 GHz,as in 802.11b/g. Three types of wall materials areexamined, namely, brick (e0r ¼ 3:9, e00r ¼ 0:001,w = 10 cm), concrete (e0r ¼ 5, e00r ¼ 0:2, w = 20 cm),and glass (e0r ¼ 5:8, e00r ¼ 0:003, w = 2.5 cm). Thebehavior for these materials is shown in the rightthree frames of Fig. 4.

Many aspects of transmission through materialcan be observed in Fig. 4. First, notice how in thecase where the receiver and transmitter are at the

d

15 m

wall

tran

smitt

er

rece

iver

0 10 20

-40

-30

-20

-10

0

distance from wall

brick concrete glass

elevatedsame height

chan

nel g

ain

(dB

)

30 0 10 20 30 0 10 20 30

Fig. 4. Transmission through a wall. The left-hand frame shows a side-view of the simulated scenario. The right-hand frames depict theattenuation of the signal when transmitted through walls of different kinds of materials. It also depicts the nature of attenuation due torelative heights of the transmitter and the receiver.


same height, the channel gain increases as the dis-tance between the transmitter and receiveddecreases. This agrees with the intuition from free-space propagation. The impact of the material canbe seen by noticing the variation in the received sig-nal strength for different materials. For referencepurposes, consider that without the wall, when thetransmitter is about 5 m from the wall, the channelgain is �14 dB. from the wall and a receiver that isat the same height as the transmitter. Thus, whenthe angle of incidence is 0, the wall reduces the chan-nel gain by 0 dB (for glass) to 6 dB (for concrete).The impact of the angle of incidence can be empha-sized by considering the channel gain when thetransmitter and receiver are not at the same height.As can be seen, this results in the signal strength firstincreasing and then decreasing as the distancebetween the transmitter and receiver increases; thisis in conflict with the behavior in free-space. Thekey characteristic of propagation at work here isthat when the ray hits the wall at a grazing angle,only a small amount of signal power is transmittedthrough the wall. The point where the signalstrength switches from increasing with the transmit-ter–receiver distance to decreasing with transmitter–receiver distance depends on the material, the thick-ness of the material, and the height of the receiver.

Remark 2. One interesting implication of Fig. 4 isthat if an urban mesh network uses relays mountedon lampposts, then the lampposts are quite close tobuildings. Thus, relays mounted on a lamppost onthe far side of the street will provide better coverageto a building than a relay mounted on a lampposton the near side of the street. We concluded that, insome cases, the angle of incidence can be moreimportant than the transmitter–receiver distance.

Fig. 4 shows variation of the channel gain due tothe different building materials. The dependence onmaterial poses a serious challenge for simulation.

First, walls are typically not made of a single mate-rial, but of layers of material. It is, however, possi-ble to extend the method to accommodate layersof different material (e.g., see [15]) at the expenseof computational complexity. A second and moreserious problem is that it is not realistic to knowthe construction material for each building. Thisdependence on construction materials emphasizesthe difficulty of accurate prediction of channel gainsand demonstrates why realistic simulation is morereasonable than accurate simulation. In the UDel-Models, it is possible to specify different buildingmaterials, but it is assumed that the material ishomogeneous for each building.

4.3. Reflections

Reflections are complementary to transmissions.However, it is important to note that some poweris absorbed by the material, hence, the power trans-mitted through a wall and the power reflected off ofa wall do not necessarily sum to the power of theincident signal and, depending on the material andangle of incidence, a significant amount of energycan be absorbed.

To get an understanding of the impact of reflec-tions, we consider propagation down a buildinglined street, as shown in Fig. 5. Here the signal isrepeatedly reflected off of walls and, in a sense,focused down the street. This effect is sometimesreferred to as propagation down an urban canyon.A similar effect can also arise in hallways and tun-nels. The result is that the received signal strengthis stronger when propagating down a street than itwould be with the same transmitter–receiver dis-tance in free-space.

Fig. 5 shows the channel gain down an urbancanyon for the same set of materials consideredabove and the free-space approximation. Note thatfree-space predicts a substantially smaller received

0 100 200 3000 100 200 300-50

-40

-30

-20

-10

-50

-40

-30

-20

-10

d

chan

nel g

ain

(dB

)receiver

transmitter

d

width

walls

width = 7 m width = 35 m

BrickConcrete

1/d 2Glass

1/d1.38

BrickConcrete

1/d 2Glass

1/d1.38

Fig. 5. Propagation down a street. The left-hand frame shows atop view of the simulated scenario. The right-hand frames depictthe effect of propagation down an urban canyon for different wallmaterials, different attenuation exponents, and for the free space.


signal strength than the simulated received signal.For example, when the receiver–transmitter distanceis 300 m, the difference between the free-space andthe model that accounts for reflection ranges from13 dB to 5 dB, depending on the material and thedistance between the walls. Fig. 5 also showsapproximations of the channel gain given by 1/da,where, for a canyon width of 7 m, a = 1.38 and acanyon width of 35 m, a = 1.47. It should bepointed out that it is often assumed that a 2 [2,4],however, as this simple example demonstrates, it ispossible that for some paths we have a < 2. Thatis, while buildings may block communication, theymay also enhance communication.

Note the lack of smoothness when the walls aremade of brick and the width is 35 m (e.g. d = 50).This is due to the wide variation in the reflectioncoefficient as the angle of incidence varies. Indeed,depending on the material and the width of thematerial, there may be some angles of incidencewhere no signal is reflected. Such angles are calledBrewster angles. In some settings, the wide variationof the reflected signals strength as a function of inci-

0 20 40 60 80 100-50

-30

-10

10

Frequency - 2.4 GHz (MHz)0 1

-60

-40

-20

0

20

Position

Sig

nal S

tren

gth

Sig

nal S

tren

gth

narrow bawideband

Fig. 6. Fast fading in narrow band and wide band communication scenarrow band signal propagating down an urban canyon. Middle: signalposition perturbation when propagating down an urban canyon. Right:function of bandwidth for when propagating down a urban canyon as

dent angle results in large fluctuations in the totalreceived signal strength for small movements ofthe receiver or transmitter antenna. This amountsto further sensitivity of communication on buildingmaterials.

4.4. Multipath fading

It is well known that wireless signals can experi-ence small-scale fading or multipath fading. Suchfading results from the constructive and destructiveinterference of signals that follow different pathsfrom the transmitter to the receiver. As a result,when the wavelength is small, a small displacementof the transmitter or the receiver will cause a changein the interference and hence a change in thereceived signal strength. While fast-fading is asignificant problem in many wireless systems, it istypically not relevant in wide bandwidth communi-cations such as those often used for data communi-cations (e.g., 802.11). The reason is that the receivedsignal strength is essentially averaged over thebandwidth, which, by definition, is wide. The left-hand frame in Fig. 6 shows an example of the signalstrength at various frequencies. Note that at somefrequencies, the signal strength is quite low, even40 dB less than the mean signal strength of 0 dB.A narrow bandwidth communication will experi-ence such degradations in signal strength. However,when averaged over a sufficiently wide bandwidth,the average signal does not experience degradation.For example, the middle frame in Fig. 6 shows howthe signal strength varies for small changes inposition. Note that the narrow bandwidth signalexperiences rapid and large fluctuations, while thewide bandwidth case experiences much smallervariations.

0 5 10 150

2

4

6

8

10

Var

ianc

e of

S

igna

l Str

engt

h

d=300 m width = 7 md=100 m width = 7 md=100 m width = 3 m

2 3 (m)

nd (20MHz)

Bandwidth (MHz)

narios. Left: the signal strength as a function of frequency for astrength of a narrow and wide band signal as a function of receivervariance in signal strength due to change in receiver position as ashown in Fig. 5.


In general, the wider the bandwidth, the less sus-ceptible to fast fading. However, variation in signalstrength also depends on the environment. Theright-hand frame of Fig. 6 shows the variance ofthe received signal strength over a circle with onemeter radius as a function of the bandwidth. Eachcurve is for a slightly different environment, but inall cases it is for propagating down the urban can-yon as shown in the left-hand frame of Fig. 5. Notethat in general, the variance decreases as the band-width increases. However, the variance of the nar-row bandwidth case and the rate that the variationdecreases with the bandwidth depends on the envi-ronment. In all cases, the variance is quite smallwhen the bandwidth exceeds 10 MHz, while802.11b has a bandwidth of 22 MHz. It should benoted that there are other factors besides multipathreflections that could cause large changes in signalstrength over small changes in position, e.g. changesin antenna orientation.

4.5. Diffraction

While reflections and transmission play animportant role in propagation, diffraction is alsosignificant and should not be neglected. Diffractionallows the signal to ‘‘bend’’ around corners andover/around buildings. However, the diffracted sig-nal will not be as strong as the original signal; thesharper the bend, the weaker the signal. Considerthe examples in Fig. 7, which use the Uniform Dif-fraction Theory [16,17]. The left-hand frames showthe diffraction around a corner. The second fromthe left frame compares free-space propagationaround the corner to free-space propagation withthe added loss due to diffraction. The right-handframes show diffraction over a building where twodiffractions are required. In both cases, as h

70

60

50

40

30

20

h

5mtransmitter

5m

5m

5m

h

transmitter

receiver

h

With dif

h (m)

Cha

nnel

gai

n (d

B)

0 10 20 30 40

Fig. 7. Diffraction around a corner (left) and over a building (right). Thof the simulated scenario. The other frames show the attenuation as adashed lines show the signal strength if the loss was only due to free-spathe loss due to diffraction is also included.

increases, the signal must make a sharper bend(i.e., the diffraction angle increases) and, hence,more loss is incurred. It is important to note thatthe signal strength decreases quickly as h increases.From the right-hand frames in Fig. 7, it can be seenthat diffracting over a building that is only 5 m.higher than the transmitter and receiver results ina channel gain that is too small for typical 802.11communication (recall the discussion in Section 3),while diffracting around a single corner (left-handplots of Fig. 7) might result in a sufficiently largechannel gains for most 802.11 communications.Indeed, our simulations indicate that for 802.11, dif-fracting around two corners results in loss that istypically too great for 802.11.

While diffracting around two or more cornersleads to large losses, 802.11 might be able to com-municate effectively around a single corner. Conse-quently, a significant portion of the coverage areaof 802.11 is due to diffraction. Using the UDelMod-els’ propagation simulator, it is possible to quantifythe impact of diffraction. Table 1 shows the numberof locations to which a particular transmitter is ableto communicate when only line-of-sight is used,when line-of-sight and reflections/transmissions areused, and when all the components are used,namely, line-of-sight, reflections/transmissions anddiffraction. The table also shows the difference inthe coverage when different numbers of iterationsare used. By the number of iterations, we meanthe maximum number of reflections, transmissions,or diffractions that each ray may experience. Wesee that about 22% of the locations are missed if dif-fraction is not used. However, if only line-of-sight isconsidered, then 78% of the locations are missed.For these calculations, a map of the Paddingtonarea of London was used and the transmitter andreceivers were located 1.5 m above the ground or

50m 5mreceiver

0 20 40120

90

60

30

h (m)

Cha

nnel

gai

n (d

B)

fraction Free-space only

e left-hand frame and second from the right frame show side viewsfunction of h. As h increases, the signal strength decreases. The

ce propagation while the solid line shows the signal strength when

Table 1Coverage and computation time

Experiment Coverage Time (s/transmitlocation)

Line of sight 937 561 iteration 2623 591 iteration without

diffraction1960 56

2 iterations 3927 622 iterations without

diffraction2616 57


diffraction2862 58


diffraction3065 63

All iterations withdiffraction

4265 122


above the floor. If the transmitter was higher (e.g., amobile phone base station), or transmissions withsmaller channel gains can be received, then it is pos-sible that diffraction could play a larger role than isillustrated in this example.

4.6. Scattering

Scattering is often used to refer to the impact ofsmaller unmodeled objects on the propagation, e.g.,lampposts, trees, vehicles, people, and office furni-ture. Scattering also accounts for the unevennessof building walls (e.g., windows, doors, andfacades). Without detailed knowledge of the loca-tion and dimension of small objects and withoutdetails of building walls, it is difficult to includethe effects of these types of scatterers into propaga-tion simulation. Furthermore, the inclusion of suchdetails greatly increases the computational complex-ity of propagation simulation. As a result, relativelylittle work has focused on including scattering intopropagation prediction. One exception is [18] where

5.4

m

Transmitter

Receiver

Fig. 8. Left, experimental setup for measuring the time-varying nature otime-series of the channel gain.

the impact of scatterers was examined. It was foundthat in dense urban environments, such scatteringdoes not dominate. More specifically, at short dis-tances, scatterers have little impact, but may havea some impact at greater distances (e.g., in somecases the difference between prediction and mea-surements decreased from 6 dB to 3 dB when scat-terers were included). On the other hand, in areaswhere building density is low (e.g., suburban areas),scatterers such as lampposts, tress, and vehicles maydominate the propagation.

Scattering can also occur when a wireless signalpropagates through vegetation. Indeed, if the vege-tation is large and dense, the scattered signal domi-nates over the direct, non-scattered signal. Evenwhen the vegetation is thin, as it usually is in urbanareas, the vegetation can cause loss. At 2.4 GHz,vegetation causes a loss of approximately 0.2 dBper meter of vegetation [19]. It is also common tomodel diffraction over vegetation [20].

4.7. Time-varying channel gain

While the variability of channel gain is greatlyaffected by the movement of the transmitter or recei-ver, the channel gain can also change when thetransmitter and receiver are fixed, but objects inthe environment move. The right-hand frame ofFig. 8 shows the channel gain as a function of timefor a receiver and transmitter as observed along asidewalk during rush-hour in the city center of Phil-adelphia. As shown in the left-hand frame of Fig. 8,the transmitter was 5.4 m above the sidewalk, whilethe receiver was only 1.2 m above the sidewalk. Thetransmitter was a Linksys BEFW11S4 and the recei-ver was the Yellowjacket [21] receiver made byBerkeley Varitronics Systems. While time-varyingpropagation is difficult to model, in [22], a diffu-sion-based model was developed that accuratelymodels the variability shown in Fig. 8. However,more work is required to develop models in general

30

40

50

60

0 50 100

observed data

Time (s)

-Cha

nnel

Gai

n (d

B)

1.2

m

f propagation in a busy street in Philadelphia. Right, an observed


settings. It should be noted that the time-varyingpropagation is due to mobility in the environment.

4.8. Delay-spread

As noted in Sections 4.2, 4.3 and 4.5, the wirelesssignal may follow several different paths from thetransmitter to the receiver. While one impact ofmultiple paths is that the signal may experience mul-tipath fading, another result is that the multiple cop-ies of the wireless transmission will be received atdifferent points of time. Essentially, these multiplecopies will interfere with each other. When thisself-interference is considered as noise, it is clearthat the effective SNR is decreased by the presenceof these delayed copies. Since the traditional calcu-lation of SNR considers delayed copies of the trans-mission as useful signal power, the traditional SNRmight not be a good indicator of probability oftransmission error.

A channel where multiple copies of the signalarrive at different times is said to have delay-spread.While there is no ideal metric of delay-spread, twocommon measures are the mean delay-spread andthe RMS delay-spread (see [8, p. 199] for defini-tions). To explore the impact of delay-spread, con-sider Fig. 9, which shows the relationship betweenSNR and bit-error probability for different amountsof delay-spread for a simple 802.11b receiver and fora 802.11a receiver. Note that the bit-error probabil-ity of 802.11b is greatly affected by the delay-spread.Also, it is can be seen that for an RMS delay-spread

-20 -10 0 10 20 3010-6

10-4

10-2

100

Es/No (dB)

Bit

Err

or P

roba

bilit

y

802.11b - 0.7ns802.11b - 13ns802.11b - 26.4ns802.11b - 58ns

802.11a -0.7ns to 220ns

Fig. 9. Bit-error probabilities as a function of delay-spread andSNR. Simulation of 802.11b at 1 Mbps (without a RAKEreceiver) and 802.11a at 6 Mbps. The different curves for 802.11bcorrespond to different RMS-delay-spreads. For a delay-spreadof 7–220 ns, 802.11a gave the same relationship between SNRand bit-error rate, hence this relationship is shown with a singlecurve.

of above 60 ns, which is quite common in an urbansetting, the impact of the delay is quite severe. Frommeasurements, it has been found that indoors, theRMS delay-spread is typically less than 50 ns, butoutdoors, it can exceed 500 ns. The dependence ofdelay-spread on the communication environmenthas been widely observed (e.g, [13,23]). It shouldbe noted that the delay-spread typically increaseswith the distance between the transmitter and recei-ver [24]. And therefore, roof-top configurationssuch as those studied in [13], which are able to prop-agate considerable distances due to the lack ofobstructions at higher heights, are likely to experi-ence large delay-spread values. As pointed out in[13], to predict the transmission error probability,in some cases, both delay-spread and SNR arerequired.

Incorporating the effects of delay-spread intowireless simulation poses significant difficulties.First, the mapping from RMS or mean delay-spreadto transmission error probability is not well defined.It is possible to have a channel with high RMSdelay-spread that yields lower transmission errorprobability than the one that has a lower RMSdelay-spread and conversely. Second, RAKE receiv-ers can be incorporated into receivers that cangreatly mitigate the effects of delay-spread. How-ever, the implementation of the RAKE receiverhas a significant impact on the performance. Forexample, [25] shows that high performance RAKEreceivers can mitigate the impact of delay-spread,while less effective RAKE receivers will performpoorly when the delay-spread is large. Thus, it is dif-ficult to make any conclusions about the effect ofdelay-spread on future mesh networks based on cur-

rent receivers.In anticipation of receivers more suitable for out-

door use, another alternative is to neglect delay-spread. Not only is such an approach reasonablewhen the inclusion of algorithms that mitigate theeffect of delay-spread into 802.11b receivers is antic-ipated, but also because other physical layers basedon OFDM greatly mitigate the impact of delay-spread. Consider Fig. 9, which shows the relation-ship between SNR and bit-error probability for dif-ferent values of RMS delay-spread when 802.11a isused. Note that the impact of delay-spread is notdiscernible. Schemes such as 802.11g and 802.16use OFDM and hence are also relatively immuneto delay-spread. In summary, while delay-spreadcan have a significant impact on transmission error,there are a number of physical layer schemes that

Fig. 10. Left: a top-view of the scene on the right. The lines


can mitigate these effects, and, considering the rapidmigration between physical layers (e.g., from802.11b to 802.11g), it seems likely that when thereis a need (i.e., when outdoor mesh networks arefinally deployed) such schemes will become wide-spread. Therefore, depending on the assumptionsmade about the network being simulated, realisticsimulation of future urban mesh networks doesnot necessarily require consideration of delay-spread.

shown are found from 2-D ray-tracing. Right: Two vertical planepaths (found from 2-D ray-tracing) and five ray paths.

5. Computational techniques

5.1. Overview

While much is known about propagation, amajor obstacle to simulation of propagation is thecomputational complexity. This complexity forcesa trade-off between computational complexity andaccuracy. The difficulty to specify the details ofthe environment (e.g., wall materials and exactdimensions and structure of buildings) also forcesa trade-off with accuracy. It is important to notethat the propagation must be determined from everylocation that a transmitter/receiver could be locatedto every other possible location that a transmitter/receiver could be located. The simulation of a1 km · 1 km urban region often requires hundredsof thousands of locations. Hence, the computationto determine the signal strength from a single loca-tion, which is in itself computationally difficult,must be repeated hundreds of thousands of times.In this section, techniques to efficiently simulatepropagation are discussed. The techniques describedare part of the UDelModels simulation package. Inthe next subsection, outdoor propagation is dis-cussed, and is followed by a discussion on indoorpropagation simulation.

5.2. Outdoor propagation

Once the map of buildings, bandwidth, andbuilding materials have been defined, propagationcan be estimated. As noted above, extreme caremust be taken to reduce the computation. A signif-icant computational savings can be achieved if it isassumed that all walls are vertical. In this case, the3-D ray-tracing problem reduces to a two stageproblem, each of which are much smaller than the3-D problem. The first stage is illustrated in theleft-hand plot in Fig. 10, where two vertical plane

paths are shown (the path that diffracts aroundthe smaller building is not shown). The first stageis a 2-D ray-tracing. Once the vertical plane pathsare found, the 3-D ray paths restricted to the verti-cal plane paths can easily be computed. The right-hand frame in Fig. 10 shows the paths of rays inthe vertical planes. One vertical plane path has threeray paths, one that diffracts over a building, onethat reflects off of the ground and passes througha building, and one that passes straight through abuilding. The other vertical plane path has tworay paths, one reflecting off of the wall of a buildingand one reflecting off of the wall of building andundergoing a ground reflection. While Fig. 10 showseach vertical plane paths having two or three raypaths, in general, in one vertical plane path thereare many ray paths that may include reflection offof the ground, and transmissions through buildingsor diffractions over buildings. The UDelModels’propagation simulator only considers three typesof ray paths, direct paths (no ground reflectionsand no diffractions over buildings), ground reflectedpaths with no diffraction, and paths that include notransmissions through buildings but rather diffractover buildings. Note that diffractions and transmis-sions through buildings result in significant loss, andhence neglecting ray paths with multiple diffractionsor diffractions and transmissions into buildings haslittle impact on accuracy. This method of 3-D ray-tracing is known as vertical plane launching [26],and is known to yield accurate propagation esti-mates in real cities, where all walls are not necessar-ily vertical [27].

A straightforward implementation of even 2-Dray-tracing is computationally difficult. Instead, atechnique that is more appropriately called beamtracing can be performed. Like ray tracing, the goalof beam tracing is to determine the paths from the

wall

wall tile B

wall tile A

1

2

3

4

virtual source

floor tiles

Fig. 11. Beam tracing. Suppose a beam strikes the wall tilemarked A. The reflection of this beam is modeled as a beamemanating from the virtual source as shown. This beam isbetween and is defined by the two rays marked 1 and 2. Theserays are used to select the floor tile ray-neighbors that this beamstrikes (since ray-neighbors are precomputed, this selection iscomputationally efficient). Similarly, the rays 1 and 2 are used toselect the set of A’s wall tile ray-neighbors that the beam strikes.One such wall tile is shown in the upper right and is marked B.The beam that strikes tile B is defined by the rays marked 3 and 4.These rays define the beam that is transmitted through tile B andcan be used to find the beam that is reflected off of B. This processof reflection and transmission is repeated until the beam’sstrength is small enough that it can be neglected (e.g., �110 dB).

2 Here, breadth first means that all the kth reflections are foundbefore the (k + 1)th reflections are found.


transmitter to receiver. Beam tracing begins with thesource broadcasting the signal in all directions. Thistransmission is not modeled as a large number ofrays, but as a small number of beams. When a beamintersects a building, two beams are generated, oneis reflected off of the building and one is transmittedinto the building. If only a part of the beam inter-sects the building, then multiple beams are gener-ated, in particular, one that reflects off of thebuilding, one that transmits into the building, onethat passes by the building, as well as several beamsthat account for the diffracted signal. Each dif-fracted beam models a particular range of anglesof diffraction. The strength of the diffracted beamis the average strength for the diffraction angles thatthe beam models. Note that smaller the range ofangles for each beam, the more accurate the diffrac-tion model is. But this generates more beams,increasing the computational complexity of the pro-cess. Finally, if the receiver is found to be includedwithin the span of a beam, the vertical plane pathis found and the 3-D rays from transmitter to recei-ver are determined. Once a 3-D ray is found, the3-D ray information, specifically, signal strength,phase, delay, angle of departure from the transmit-ter, and angle of arrival at the receiver are recorded.

The beam tracing computation can be furthersimplified by dividing the 2-D space into a gridand the determining the propagation between thecenter points of each square. Each square of the gridis called a floor-tile. Outdoors and indoors are dis-cretized in this manner. To reduce the number offloor tiles, the entire space is not discretized. Rather,floor-tiles are placed only along the center-lines ofsidewalks, roads, and hallways. For rooms, floortiles are placed in every location that a mobile nodecan be located. Similarly, the walls of buildings arealso divided into wall-tiles. Since the beam tracing isin 2-D, the wall-tiles are segments (1-D tiles). Thesmaller the size of these tiles (floor-tiles and wall-tiles) more accurate the simulation results, but alsothe more computationally expensive.

The computation is divided into two parts, pre-processing and beam tracing. During preprocessing,ray neighbors for each tile are found. A tile’s rayneighbors are all the tiles that could be directlyreached (i.e., without reflection, transmissionthrough a wall, or diffraction) by a ray emanatingfrom the tile. Once the ray neighbors are found,beam tracing can be performed efficiently. Fig. 11illustrates how the beam tracing with wall tiles isperformed.

The process of beam tracing Fig. 11 is carried outin a breadth first manner2 with each beam continuedto be transmitted, reflected/diffracted until either thebeam exits the modeled area or until the estimatedchannel gain of the beam falls below a threshold.Each time a beam is generated, the floor-tiles it cov-ers are determined and the 3-D ray information isrecorded. Note that the 3-D ray information allowsone to interpolate the signal strength between gridpoints as well as determine the delay-spread, theDoppler spread (once velocity is known), as wellas simulate directional antennas. It is possible todetermine the receiver power over a particularbandwidth by integrating the signal strengths in away that accounts for phase (i.e., accounting forconstructive/destructive interference). The 3-D rayinformation requires a significant amount of infor-mation be saved. A considerable amount of savingsis achieved by simply recording the received powerover a bandwidth and RMS delay-spread (which ismerely two integers). Of course, this approach can-not be used if directional antennas are simulated orDoppler spread included.

3 Recall that the floor tiles cover sidewalks, roads, hallways,and offices only.


Another important computational saving isachieved by recognizing that the communication isnot possible between most locations, hence thepropagation matrix is sparse. Computational sav-ings are possible by using techniques to store sparsematrices. However, since the resulting data is large,and since simulation requires the determination of alarge number of channel gains, it is important thatthe sparse matrix techniques support fast access.

5.3. Indoor propagation

Beam tracing can be performed indoors as well asoutdoors. However, the computational complexitydepends on the number of walls. Since building inte-riors have a large number of walls, beam tracinginside all the buildings within a large region of a cityexceeds today’s computational abilities. Fortu-nately, it has been found that a realistic estimateof indoor propagation can be performed withoutusing beam tracing. Specifically, the attenuation

factor (AF) model has been shown to provide real-istic channel gain estimates, with the error foundto be within 4 dB [8]. The AF model assumes thatcommunication indoors takes a straight line path(i.e., no reflections off of interior walls). Further-more, transmissions through each interior wall andtransmissions through each floor result in attenua-tion. While the amount of attenuation depends onthe building, a value of 4 dB per wall (for an officebuilding) has been shown to work well [8]. Realisticattenuation for passing through one floor is 30 dB,passing through two floors is 35 dB, passingthrough three floors is 39 dB and passing throughfour or more floors is 40 dB [8]. In summary, out-doors, rays make reflections off of buildings, diffrac-tions over and around buildings, and transmissionsinto buildings. Once inside a building, the ray willcontinue in the same direction, experiencing furtherattenuation for any interior wall or floor that itpasses through. When a ray strikes an exterior wallfrom the inside, it is both reflected back inside andtransmitted outside.

5.4. Computational complexity

Determining the propagation matrix of a regionof an urban area is a feasible but highly computa-tionally complex task. The complexity is both interms of memory usage and processing time. Pro-cessing times for a 1 km · 1 km urban region isoften on the order of tens of processor days. But

the process is highly parallelizable and nearly scaleswith the number of processors used (i.e., 75 proces-sor days takes about 5 days on 15 processors). Ofcourse, the entire channel gain matrix for each cityonly needs to be found once. Table 2 outlines thememory requirements and the processing time fortwo cities. These two cities represent two ends ofthe spectrum. Paddington is a dense city with rela-tively small buildings, whereas the University ofDelaware campus has much more open space andlarger buildings. Note that the campus map is signif-icantly larger than Paddington, but only has slightlymore buildings.

The memory requirement is dominated by thelists of ray neighbors. In Table 2, the wall tiles were2 meters long and floor tiles were 1 m · 1 m. Whilethe number of wall and floor tiles scales linearlywith the reciprocal of the size of the tiles (i.e., ifthe wall tiles are twice as long, there are half asmany)3, the total ray neighbors scale quadratically.For the simulations shown in Table 2, there arebetween 20,000 and 40,000 exterior floor tiles, and80,000–100,000 interior floor tiles. As expected,sidewalks and roads utilize little area outdoors,but indoors, hallways and offices fill large areas.

As shown in Table 2, there are on the order of10–100 million ray neighbors for a city of size about1 km · 1 km. Since each list entry requires 40B, thememory requirement approaches 4 GB. Assumingsufficient memory resources, the computation timefor the preprocessing stage is relatively short; thecities shown in Table 2 took a single day on anAMD Athlon 64 FX 55 with 8 GB RAM.

Once the ray neighbors are found, the propaga-tion characteristics between each pair of floor tilescan be found. From a single source, the propagationcharacteristics to all destinations can be found at thesame time. Table 2 shows that each source producesvertical planes that make approximately 500,000reflections, diffractions, or transmissions. Thesereflections, diffractions, or transmissions are sharedamong all destinations. However, for each destina-tion, all the vertical plane paths between the sourcefloor tile and destination floor tile are found. Hence,the total number of vertical plane paths foundgreatly exceeds the total number of reflections, dif-fractions, and transmissions.

Table 2Computational complexity

City Size (m) Numbuildings

Numfloortiles

Numwalltiles

Totalnum rayneighbors

Average numreflections persource

Ave. numdiffractionsper source

Average numvertical planepaths/source

Total comp.time(processordays)

Paddington 857 · 836 131 108K 17K 15M 194K 367K 1.3M 53UD

campus1768 · 1597 151 147K 19K 67M 84K 332K 946K 72

building

buildingbuilding

building

stre

et

street

Source

20 30 40 50 60 700

20

40

60

Distance in meters

Pat

hlos

s in

dB

data from Raytracingdata from Measurement

Fig. 13. Observed and estimated path loss at an intersection inPhiladelphia.


For each source it is necessary to find the allreflections. Thus, in the campus map, the total num-ber of reflections found was around 263 billion and12 billion for Paddington. Several optimizationsand tricks to efficiently distribute the programresults in the total processing time shown.

5.5. Validation

The goal of the propagation simulation for simu-lating mobile wireless networks is not to predict thesignal strength, but to produce signal strength thatbehaves in a realistic fashion. Nonetheless, it is use-ful to understand the accuracy of the propagationsimulation. To this end, three experiments were per-formed, two outside and one inside. In all cases, an802.11b access point was placed on a 1.5 m tripodand the Berkeley Varitronics Yellowjacket wirelessreceiver [28] was placed on a second 1.5 m tripod.The access point was placed at a fixed locationand the average receiver signal strength was deter-mined by receiving 600 802.11b beacons. Fig. 12bshows a part of the University of Delaware campusand Fig. 13b shows a street intersection in Philadel-phia. In Fig. 12b, the buildings were approximately14 m high while in Fig. 13b the buildings were atleast 40 m high. In both cases, the X-mark denotesthe location of the transmitter while the receiver ismoved along the indicated path. Fig. 12a shows

0 200 400 600 800 10000

20

40

60

80

100

Distance in meters

Pat

hlos

s in

dB

Pathloss data from RaytracingPathloss data from Measurement

A A

B

G

E F

I H

DC

Fig. 12. Observed and estimated path

the observed and modeled channel gain correspond-ing to the path starting from the transmission pointand moving along the path in the counter-clockwisedirection. Fig. 13a shows the model and observedchannel gain starting near to the transmitter, mov-ing down and then turning the corner.

Figs. 12a and 13a show that the model and obser-vations match well both qualitatively and quantita-tively (within 5 dB in most locations). To gain moreinsight into propagation modeling we examine thepropagation prediction quality at different loca-tions, especially the location where the predictionquality is low. In the area between B and C, thereis an unmodeled archway that is indicated inFig. 12b. Similarly at location F, there is a bridgeas depicted with the rectangle. Ignoring theseobjects impacts the accuracy of the propagation pre-diction. In the locations marked E and G, there are

AAB BC

F

E

DG

trees

archway

large airconditionersbridge

archway

H

bushesI

loss in the campus environment.

1 1915105

21

23

22

20

24X

4 8 12 16 20 240

20

40

60

80

Measurement Points

Pat

hlos

s in

dB

Data from Model

Data from Measurement

Fig. 14. Observed and estimated path loss in an indoor environment. The measurement points shown on the x-axis correspond to thenumbered locations in the map. The map is of the third floor of Evans Hall in University of Delaware campus.

4 By open space, we mean the 2-ray model, where the receivedsignal power decays as C2

2=d2 for d < 200 m and as C4/d4 ford P 200 m, where C2 and C4 are constants.


several moderate sized unmodeled objects (large airconditioners at E and trees at G) that partiallyblocked the signal. As mentioned above, sometimessuch objects are called scatterers. We see that scat-terers can slightly decrease the received signalstrength. On the other hand, in the areas wherethere is purely line-of-sight (marked As), line-of-sight with reflections (marked Bs) and reflectionswith diffraction (marked D), there is good agree-ment between the model and the observations. See[29] for further discussion on the relationshipbetween accuracy of propagation prediction andaccuracy of the map. Furthermore, it is importantto note that due to mobility of objects, propagationmeasurements made at different times will be differ-ent. For example, in [30] it was found that identicalmeasurements made at different times can differ by5 dB.

Fig. 13a also shows a good fit. Again, the influ-ence of scatters can be observed. In this case thescatterers includes things such as mailboxes, parkedcars, and irregularity of the walls (e.g., doors thatare set back from the wall). Nonetheless, the modeland observations are within a few dB. Note that thesignal strength decreases quite slowly as the distancefrom the corner increases. This is due to the urbancanyon effect discussed in Section 4.3 and depictedin Fig. 5.

Finally, Fig. 14b shows the layout of a buildinginterior and Fig. 14a compares the modeled andobserved signal strength for the points indicated inFig. 14b. Again, we see that there is reasonablygood agreement between the model and theobservations.

In summary, it is clear that accurate propagationprediction requires more detailed knowledge of theenvironment. However, coarse knowledge (e.g.,location of buildings alone) provides realistic prop-agation, both qualitatively and quantitatively.

6. Impact on mobile wireless simulations

In much of the mobile wireless networking litera-ture, random way-point with open-space4 propaga-tion (RWP/OS) or random way-point with two-ray(RWP/TR) are used. It is clear that these simplepropagation models is quite different from the onediscussed here. A natural question is how the simu-lation strategy impacts the conclusions drawn fromthe simulations? However, a direct comparison ofopen-space propagation and realistic urban propa-gation is difficult. One problem is that randomway-point mobility makes little sense in urban area.For example, since urban areas are in three dimen-sions, a random way-point model would have nodesflying through the air between buildings. Hence, inorder to perform sensible simulation, an urbanmobility model is necessary. To this end, the UDelUrban Mobility Model is used [31]. Briefly, the UDelUrban Mobility, models the realistic urban mobility.Nearly all aspects of the model are based on surveys.For example, the model includes the time people(nodes) arrive at work, take lunch breaks, etc. Thedata for this part of the model is from the US Bureauof Labor and Statistics Time Use Study [32]. Themobility of office workers is based on surveys ofoffice meetings (including two-person meetings)(e.g., [33]). And vehicle traffic is from the City ofSan Francisco [34]. In all, over 25 different types ofsurveys are incorporated into the model.

Another problem with comparing open-spacepropagation to realistic urban propagation is thatthe resulting networks are very different and a hencemeaningful comparison is difficult. For example, inan open-space environment, distant nodes are able


to communicate whereas in an urban setting, thenodes separated by very small distances might notbe able to communicate (e.g. when one of thecommunicating nodes is indoors and the other isoutdoors), or, as shown in Fig. 5, nodes that arefar apart might be able to communicate (e.g., whencommunicating down an urban canyon). To see theimpact of the variability in propagation, considerFig. 15. These plots show the mean number ofnodes that a node can communicate with (i.e., themean degree) and the coefficient of variance (CoV)of the degree. These plots shown are from threedifferent scenarios, namely, urban mobility withrealistic urban propagation, urban mobility withopen-space propagation, random way-point mobil-ity with open-space propagation. In all cases, thesimulated region is generated from a 9-block(�0.5 km · 0.5 km) region of Chicago. The case ofrandom way-point also uses the Chicago map, butnodes move according to the random way-pointmobility and propagation is given by the 2-raymodel. While the random way-point mobility is dis-tinct from urban mobility, we selected the nodespeeds to be similar, i.e., Gaussian with mean1.34 m/s and standard deviation 0.26 [31]. In allcases, a mesh infrastructure was placed on 29 uni-formly distributed streetlights. The statistics of thedegree distribution of the infrastructure of the meshand the mobile client nodes are shown for severaldifferent numbers of mobile nodes in the network.It can be seen that the degree distributions are differ-ent in a number of ways. The most obvious differ-ence is that the open-space propagation results ina much larger mean degree. On the other hand,the coefficient of variation is smaller, indicating lessvariation in the degree, i.e., the degree is more stable

500 1000 150010

0

101

102

103

Number of Mobile Nodes

Mea

n D

egre

e

Client Nodes - Urban Mobility/Urban Prop.Client Nodes - Urban Mobility/Open-Space Prop.Client Nodes - RWP Mobility/Open-Space Prop..

Fig. 15. Statistics of degree distribution in random way-point with openand urban mobility with realistic urban propagation. The mesh infrastruare mobile. The coefficient of variance (CoV) is the standard deviation

in open-space propagation environment than in theurban environment. Furthermore, comparing thestatistics of the degree distribution of the infrastruc-ture nodes to the client nodes, we see that the urbansetting shows a larger difference between these typesof nodes than does the open-space.

To further examine the differences between therealistic propagation and open-space propagation,consider the plots shown in Fig. 16. These perfor-mance statistics are also generated from the 9-blockregion of Chicago as described above. Twenty UDPconnections were initiated from a centrally locatedbase station to randomly selected client nodes. Eachconnection attempted to send a packet once every250 ms. An adapted version of AODV [35] wasused. Specifically, when a mobile node receives aroute request (RREQ) from an infrastructure node,the mobile node will not relay the RREQ until100 ms has past. In this way, routes tend to followthe infrastructure, but mobile-to-mobile relaying isstill possible. Three environments were considered,namely, urban mobility with urban propagation,urban mobility with open-space propagation, andrandom way-point with open-space propagation.

As expected, Fig. 16 shows dramatic differencebetween urban propagation and open-space propa-gation. As alluded to above, realistic node densitiesfor urban environments result in very high densitieswhen open-space propagation is used. Such high den-sities result in a large number of collisions which leadto route searches failing and even route failures.Indeed, for 750 or more mobile nodes, very few pack-ets are successfully delivered in the open-space prop-agation setting. Note that there is little differencebetween urban mobility with open-space propa-gation and random way-point mobility with

500 1000 15000

0.2

0.4

0.6

0.8

1

Number of Mobile Nodes

CoV

of th

e D

egre

e

Mesh Nodes - Urban Mobility/Urban Prop.Mesh Nodes - Urban Mobility/Open-Space Prop..Mesh Nodes - RWP Mobility/Open-Space Prop.

-space propagation, urban mobility with open-space propagationcture nodes are fixed on top of streetlights, while the client nodesdivided by the mean.

0

1000

2000

3000

4000

Rou

te r

eque

sts

forw

arde

d

by in

term

edia

te h

ops

0

2

4

6

8

10

12

Num

ber

of r

oute

failu

res

per

conn

ectio

n

0

0.2

0.4

0.6

0.8

1

Pac

ket d

eliv

ery

ratio

500 1000 15000

0.4

0.8

1.2

1.6

End

-to-

End

del

ay (

sec.

)

500 1000 1500500 1000 1500

500 1000 1500Number of nodes Number of nodes

Number of nodes Number of nodes

UDel ModelsRandom waypoint/open space propagationUDel Mobility/open space propagation

Fig. 16. Performance of a network under different types of mobility and propagation.

5 Currently, linear interpolate is used to determine the propa-gation between updates.


open-space propagation. However, this does notindicate that mobility is unimportant, but indicatesthat mobility is unimportant when used with open-space propagation. The impact of mobility modelswhen realistic propagation is employed is beyondthe scope of this paper.

As discussed in Section 5.4, determining thepropagation matrix is computationally intensive,however, it only must be done once for each city.Once the propagation matrix is found and a mobil-ity model determines the locations of nodes, it isnecessary to select the entries from the propagationmatrix corresponding to each pair of node loca-tions. This channel information is saved into a prop-

agation trace file, which the protocol simulatorreads during packet simulation. Fig. 17 shows theexecution time, which includes the time to constructthe propagation trace file and the time to run thepacket simulation for a one minute of simulatedtime. Fig. 17 demonstrates that realistic propaga-tion does increase the execution time. This increaseis less than a factor of two for high node densities,but is quite large for low node densities. On theother hand, the right-hand frame in Fig. 17 showsthat for 250 mobile nodes, the packet simulationtime makes up the majority of the execution time,and at all node densities, the time to construct thepropagation trace file is similar to the packet simu-

lation time. Furthermore, since the time to performa packet simulation depends on the number of dis-crete events that must be processed, the propagationenvironment will have a significant impact on thepacket simulation time. Hence, comparing packetsimulation times of open-space propagation andrealistic propagation is difficult.

While Fig. 17 shows that the time it takes to per-form packet simulation is similar to the time it takesto generate the propagation trace file, it is possible togenerate the propagation trace file once and reuse thesame trace file for multiple simulations. In fact, alarge set of propagation trace files are availableonline [5]. If this approach is used, then the only com-putation related to realistic propagation is loadingthe propagation trace file. However, since the tracefile specifies the propagation between all pairs ofnodes, if a large number of nodes are simulated, thenthe trace file can be quite large. On the other hand,the trace file only specifies the propagation betweennode pairs when the propagation changes; hence, ifnodes move slowly, then the propagation trace fileis quite small. Moreover, the trace file only providesthe propagation once every second.5 Therefore, if

0 500 1000 15000

200

400

600

800

1000

1200

1400

Tot

al e

xecu

tion

time

(sec

.)

Propagation computation timePropagation matrix loading timeReading mobilityPacket simulation with qualnet

0 500 1000 15000

200

400

600

800

1000

1200

1400

Number of nodes Number of nodes

Exe

cutio

n tim

e fo

r qu

alne

t sim

ulat

ion

(sec

.)Urban mobility/Urban propagation

Urban Mobility/Open-spacePropagationRandom way-point mobility/Open-space propagation

Fig. 17. Simulation execution time. The left-hand frame shows the simulation execution time for three simulation techniques. The right-hand plot shows a break-down of the execution time when realistic propagation is included. The propagation computation time includesthe time to identify which floor tile a node is nearest and the time to look-up the propagation between nodes from the propagation matrix.


high data rates are achieved, then the computationalcomplexity of processing packet transmission/recep-tion events is higher than the computational com-plexity of loading the propagation matrix. Thus,when the propagation trace file is precomputed, theimpact of including realistic propagation might berelatively small. In the right-hand frame of Fig. 17,it can be seen that the simulation time grows slowlywith the number of nodes. More specifically, in theworst-case, the propagation trace file grows quadrat-ically with the number of nodes. However, Fig. 17shows that the simulation time does not grow qua-dratically except for 1500 nodes, which shows a largejump in simulation time. Interestingly, at this nodedensity, the simulation with realistic propagationtakes approximately as long as the simulations withopen-space propagation.

7. Future work

While the propagation simulation discussedabove is significantly more realistic than the 2-raymodel commonly used in mobile wireless networksimulation, there are several areas that require fur-ther effort. For example, as discussed in Section4.7, propagation is time-varying, and is linked tothe mobility of objects, specifically people and vehi-cles. Since the integrated mobility and propagationsimulator include the locations of people and vehi-cles, these locations can be included into the propa-gation calculations. However, more effort isrequired to model the impact of people and othermobile objects. One approach is to incorporate sto-chastic models. In this case, the simulation strategy

would be site specific and mixed deterministic andstochastic. See [36] for mixed deterministic and sto-chastic propagation modeling.

Modeling propagation from indoors to outdoorsand visa versa is an area of ongoing research [37].One difficulty is that accurate propagation throughnon-homogeneous walls is difficult to compute. Asshown in Sections 4.3 and 4.4, propagation isgreatly influenced by the material. However, thesesections only considered propagation through wallsmade from a single material, whereas most walls aremade of several types of material. While [38] devel-ops a technique to model propagation through com-plicated walls, it is computationally complex.Furthermore, even if more computational power isavailable, there remains the difficulty of determiningthe materials and wall structure of realistic urbanbuildings.

Throughout the discussion and in the computa-tions above, it was assumed that transmitter andreceiver antennas are vertically polarized. Further-more, we also assumed that they are ideal. Whileit is straightforward to include the model of an idealdipole antenna, in many cases the antenna is not anideal dipole and is not vertically positioned. Forexample, see [39] for an example where the angleof inclination of the antenna played a critical rolein connectivity.

8. Related work

Currently, free-space and two-ray propagationare the most popular propagation models forMANETs research. For example, ns-2 [40,41] only


supports free-space and 2-ray propagation models.On the other hand, QualNet [42] supports open-space propagation as well as stochastic propagationmodels such as Rayleigh, Rician and Lognormalfading. Qualnet also supports channel gain tracefiles. The UDelModels simulation package gener-ates trace files that are compatible with Qualnet.Furthermore, OPNET [43] supports open-spacepropagation models as well as an enhanced open-space model that accounts for hills, foliage andatmospheric affects.

There has been little effort on integrating realisticpropagation into mobile wireless network simula-tion. Some examples where propagation modelingis considered include [44–46], and [47] where obsta-cles were included in the simulated environment andpropagation was limited to line-of-sight. In [45], theobstacles were randomly placed buildings. But thesehave limited applicability because, as is shown in theTable 1, most of the communication in an urbanarea is non-line-of-sight. Other examples of propa-gation simulation in MANETs are [48], where ray-tracing is used to enhance ns-2’s propagation model,and [49], where indoor MANET simulations makeuse of the attenuation factor model similar to whatis discussed in Section 5.3.

Within the communications area, propagationprediction continues to be an active area ofresearch. Ray-tracing (e.g., [50,51]) and the verticalplane method [26] discussed in Section 5.2 are oftenused. See [37] for a detailed review of work in prop-agation prediction.

There are several commercial packages that canbe used to predict coverage of a single or a smallnumber of mobile phone base stations or wirelessaccess points (examples include [52–54]). Thesetools have limited applicability to LUMNet simula-tion. Specifically, due to different goals (realistic vs.prediction), propagation predictions are typicallymore computationally intensive than is requiredfor realistic network simulation. Furthermore,many tools focus on outdoor coverage for mobilephones, or indoor coverage for wireless base sta-tions; they neglect mixed indoor/outdoor simulationscenarios.

9. Conclusion

Issues related to propagation simulation forurban wireless networking have been explored.Also, the design of a propagation simulator hasbeen discussed. This propagation simulator is inte-

grated with a mobility simulator discussed in [31].This simulation package, which is available fordownload [5], demonstrates that it is possible toinclude realistic propagation into simulation ofurban wireless networks such as the ones beingdeployed in several cities. As has been shown, tradi-tional simulators that use random way-point mobil-ity and the 2-ray propagation model are not suitablefor urban environments. It is important to note thatthe simulation strategies discussed here and imple-mented in the simulation tools focus on realisticsimulation, not accurate prediction. Thus, they aresuitable for understanding the performance ofurban wireless networking protocols, but not forplanning specific networks. However, as computa-tional resources increase, we expect that more com-plicated and accurate techniques will be integratedinto the simulation packages, and will, perhaps,allow for accurate prediction as well.

This paper only provide details on realistic prop-agation of LUMNets. It does not describe realisticmobility and nor how to simulate the relevant appli-cation traffic.

Disclaimer

The views and conclusions contained in this doc-ument are those of the authors and should not beinterpreted as representing the official policies,either expressed or implied, of the Army ResearchLaboratory or the US Government.

References

[1] Philadelphia Mayor’s Office, Government leaders teleconfer-ence – wireless: the 21st Century technology, 2004.<http://www.phila.gov/mois/press/multimedia/pdfs/WirelessPhiladelphia092004.pdf>.

[2] City and C. of San Francisco, Request for proposals 2005-19– TechConnect: community wireless broadband network.<http://www.sfgov.org/site/techconnectindex.asp>, 2005.

[3] T. Times, Ma says Taipei to be wireless in two days, 19December 2005.

[4] T.C. of Minneapolis, Request for proposal to providebroadband IP data access services, 2005.

[5] S. Bohacek, V. Sridhara, J. Kim, UDel models. <http://udelmodels.eecis.udel.edu>.

[6] M. Hata, Empirical formula for propagation loss in landmobile radio services, IEEE Transactions on Vehicle Tech-nology 29 (3) (1980) 317–325.

[7] European Cooperation in the Field of Scientific and Tech-nical Research, Urban transmission loss models for mobileradio in 900 and 1800 MHz bands, The Hague, TechnicalReport, Revision 2, 1991.

http://www.phila.gov/mois/press/multimedia/pdfs/WirelessPhiladelphia092004.pdf

http://www.phila.gov/mois/press/multimedia/pdfs/WirelessPhiladelphia092004.pdf

http://www.sfgov.org/site/techconnectindex.asp

http://udelmodels.eecis.udel.edu

http://udelmodels.eecis.udel.edu


[8] T. Rappaport, Wireless Communications – Principles andPractice, Prentice Hall, 2002.

[9] F. Graziosi, F. Santucci, A general correlation model forshadow fading in mobile radio systems, IEEE Commun.Lett. 3 (2002) 102–104.

[10] M. Gudmundson, Correlation model for shadow fading inmobile radio systems, Electron. Lett 27 (1991) 2145–2146.

[11] H.L. Bertoni, Radio Propagation for Modern WirelessSystems, Prentice Hall, New Jersey, 1999.

[12] C.A. Balanis, Advanced Engineering Electromagnetics,Wiley, New York, 1989.

[13] D. Aguayo, J. Bicket, S. Biswas, G. Judd, R. Morris, Link-level measurements from an 802.11b mesh network, in:Proceedings of the 2004 Conference on Applications, Tech-nologies, Architectures, and Protocols for Computer Com-munications (SIGCOMM), ACM Press, New York, 2004,pp. 121–132.

[14] M. Yarvis, K. Papagiannaki, W.S. Conner, Characterizationof 802.11 wireless networks in the home, in: 1st Workshopon Wireless Network Measurements (WiNMee), 2005.

[15] S.V. Savov, M.H.A.J. Herben, Modal transmission-linemodeling of propagation of plane radiowaves throughmultilayer periodic building structures, IEEE Trans. Anten-nas Propagat. 51 (9) (2003) 2244–2251.

[16] R.G. Kouyoumjian, P.H. Pathak, A uniform geometricaltheory of diffraction for an edge in a perfectly conductingsurface, Proc. IEEE 62 (11) (1974) 1448–1461.

[17] J.B. Andersen, UTD multiple-edge transition zone diffrac-tion, IEEE Trans. Antennas Propagat. 45 (1997) 1093–1097.

[18] K. Rizk, J. Wagen, J. Li, F. Gardiol, Lamppost and panelscattering compared to building reflection and diffraction, in:COST 259 TD, Turin, Italy, 1996, pp. 158–167.

[19] D.M.L. Vinc, M.A. Karam, Dependence of attenuation in avegetation canopy on frequency and plant water content,IEEE Trans. Geosci. Remote Sens. 34 (5) (1996) 1090–1096.

[20] T. Kurner, A. Meier, Prediction of outdoor and outdoor-to-indoor coverage in urban areas at 1.8 GHz, IEEE J. SelectedAreas Commun. 20 (3) (2002) 496–506.

[21] BVSystems,http://www.bvsystems.com/Products/WLAN/Yellowjacket/yellowjacket.htm.

[22] V. Sridhara, H.-C. Shin, S. Bohacek, Observations andmodels of time-varying channel gain in crowded areas, in:2nd International Workshop on Wireless Network Measure-ment, WinMee’06, Boston, Massachusetts, USA, 2006.

[23] C. Steger, P. Radosavljevic, J.P. Frantz, Performance ofIEEE 802.11b wireless LAN in an emulated mobile channel,in: The 57th IEEE Semiannual Vehicular TechnologyConference (VTC 2003-Spring), 2003, pp. 22–25.

[24] L.J. Greenstein, V. Erceg, Y.S.Y.M.V. Clark, A new path-gain/delay-spread propagation model for digital cellularchannels, IEEE Trans. Vehicular Technol. 46 (2) (1997) 477–485.

[25] M.V. Clark, K.K. Leung, B. McNair, Z. Kostic, OutdoorIEEE 802.11 cellular networks: radio link performance, in:IEEE International Conference on Communications (ICC),IEEE, New York, NY, 2002, pp. 595–599.

[26] G. Liang, H.L. Bertoni, A new approach to 3-d ray tracingfor propagation prediction in citiee, IEEE Trans. AntennasPropagat. 46 (1998) 853–863.

[27] K. Rizk, R. Valenzuela, S. Fortune, D. Chizhik, F. Gardiol,Lateral, full-3d and vertical plane propagation in microcells

and small cells, in: 48th IEEE Vehicle Technology Confer-ence, 1998, pp. 998–1002.

[28] Berkeley Varitronics. <http://www.bvsystems.com/>.[29] K. Rizk, J.F. Wagen, F. Gardiol, Influence of database

accuracy on two-dimensional ray-tracing-based predictionsin urban microcells, IEEE Trans. Vehicle Technol. 49 (2000)631–642.

[30] K. Rizk, J. Wagen, F. Gardiol, Two-dimensional ray-tracingmodeling for propagation in microcellular environments,IEEE Trans. Vehicle Technol. 46 (1997) 508–517.

[31] J. Kim, S. Bohacek, A survey-based mobility model ofpeople for simulation of urban mesh networks, in: Mesh-Nets, 2005. <http://www.eecis.udel.edu/bohacek>.

[32] U.D. of Labor Bureau of Labor Statistics, American timeuse survey (ATUS), 2003. <http://www.bls.gov/tus/>.

[33] N.C. Romano, J.F. Numamaker, Meeting analysis: findingsfrom research and practice, in: Proceedings of Teh 34thHawaii International Conference on Systems Science, 2001.

[34] City of San Francisco Department of Parking and Traffic,Daily traffic volume. <http://www.sfgov.org/site/uploaded-files/dpt/Volumes>.

[35] C.E. Perkins, E.M. Royer, Ad hoc on-demand distancevector routing, in: Proceedings of the 2nd IEEE Workshopon Mobile Computing Systems and Applications, 1999, pp.90–100.

[36] A. Domazetovic, L.J. Greenstein, N.B. Mandayam, I.Seskar, Propagation models for short range wireless channelswith predictable path geometries, IEEE Trans. Commun. 53(7) (2005) 1123–1126.

[37] M.F. Iskander, Y. Zhengqing, Propagation prediction mod-els for wireless communication systems, IEEE Trans.Microwave Theory Tech. 50 (3) (2002) 662–673.

[38] C.L. Holloway, P.L. Perini, R.R. DeLyser, K.C. Allen,Analysis of composite walls and their effects on short-pathpropagation modeling, IEEE Trans. Vehicle Technol. 46(1997) 730–738.

[39] S. Jadhav, T.X. Brown, S. Doshi, D. Henkel, R.G.The-kkekunnel, Lessons learned constructing a wireless ad hocnetwork test bed, in: 1st International Workshop on WirelessNetwork Measurements, 2005.

[40] Network Simulator - ns-2. <http://www.isi.edu/nsnam/ns/>.[41] Wireless and mobility extensions to ns-2. <http://www.mon-

arch.cs.cmu.edu/cmu-ns.html>.[42] Scalable Network Technologies, The QualNet simulator.

<http://www.qualnet.com/>.[43] OPNET. <http://www.opnet.com>.[44] P. Johansson, T. Larsson, N. Hedman, B. Mielczarek, M.

Degermark, Scenario-based performance analysis of routingprotocols for mobile ad-hoc networks, in: MobiCom, 1999.

[45] A. Jardosh, E.M. Belding-Royer, K.C. Almeroth, S. Suri,Towards realistic mobility models for mobile ad hocnetworks, in: MobiCom, 2003.

[46] A.P. Jardosh, E.M. Belding-Royer, K.C. Almeroth, S. Suri,Real-world environment models for mobile network evalu-ation, IEEE J. Selected Areas Commun. 23 (3) (2005) 622–632.

[47] S. Marinoni, H.H. Kari, Ad hoc routing protocol perfor-mance in a realistic environment, in: Proceedings of theInternational Conference on Networking, InternationalConference on Systems and International Conference onMobile Communications and Learning Technologies

http://www.bvsystems.com/Products/WLAN/Yellowjacket/yellowjacket.htm

http://www.bvsystems.com/Products/WLAN/Yellowjacket/yellowjacket.htm

http://www.bvsystems.com/

http://www.eecis.udel.edu/bohacek

http://www.bls.gov/tus/

http://www.sfgov.org/site/uploadedfiles/dpt/Volumes

http://www.sfgov.org/site/uploadedfiles/dpt/Volumes

http://www.isi.edu/nsnam/ns/

http://www.monarch.cs.cmu.edu/cmu-ns.html

http://www.monarch.cs.cmu.edu/cmu-ns.html

http://www.qualnet.com/

http://www.opnet.com


(ICNICONSMCL’06), IEEE Computer Society, Washing-ton, DC, USA, 2006, p.96.

[48] J.-M. Dricot, P.D. Doncker, High-accuracy physical layermodel for wireless network simulations in NS-2, in:IWWAN, 2004.

[49] A.L. Cavilla, G. Baron, T.E.H.A.L. Litty, E. de Lara,Simplified simulation models for indoor MANET evaluationare not robust, in: Proceedings of SECON, 2004.

[50] R.A. Valenzuela, L.J. Greenstein, Performance evaluationsfor urban line-of-sight microcells at 900 MHz using a multi-ray propagation model, in: Global TelecommunicationsConference GLOBECOM, 1991, pp. 1947–1952.

[51] S.Y. Seidel, T.S. Rappaport, A ray tracing technique topredict path loss and delay spread inside buildings, in:Global Telecommunications Conference, 1992, pp. 649–653.

[52] WirelessValley, LANPlanner. <http://www.wirelessvalley.com>.

[53] S.J. Fortune, D.M. Gay, B.W. Kernighan, O. Landron, R.A.Valenzuela, M.H. Wright, WISE design of indoor wirelesssystems: practical computation and optimization, Comput.Sci. Eng. 2 (1) (1995) 58–68.

[54] WaveCall, WaveSight. <http://www.wavecall.com/>.

Vinay Sridhara received his B.E. inComputer Science and Engineering fromR.V. College of Engineering, BangaloreUniversity, in 1999. He received in M.S.in Computer Science from University ofSouthern California, Los Angeles, in2002. He is currently working towardshis Ph.D. degree in Electrical Engineer-ing at the University of Delaware,Newark. In summer and fall of 2006, hewas an Intern with Qualcomm R&D,

San Diego, CA, working on integration of 802.11 based systems

and 3G Cellular systems. His current research interests lie in thearea of modeling and performance evaluation of Wireless MeshNetworks. Specifically he is interested in developing models andmethodologies for simulating realistic propagation mechanismsfor 802.11 wireless systems. He is also interested in developinghandoff mechanisms between cellular and 802.11 systems.

Stephan Bohacek received the B.S. inelectrical engineering from the Univer-sity of California at Berkeley in 1989. Hereceived the Ph.D. in electrical engi-neering from the University of SouthernCalifornia in 1999 in Control Theory. Heis currently an Assistant Professor in theDepartment of Electrical and ComputerEngineering at the University of Dela-ware. His research focuses on the design,analysis, and control of data networks.

His current interests include congestion control, capacity, androuting for wireless and wireline networks, modeling mobile
wireless networks, and cross-layer design for wireless networks.
http://www.wirelessvalley.com

http://www.wirelessvalley.com

http://www.wavecall.com/

Realistic propagation simulation of urban mesh networks q

Documents

Transcript of Realistic propagation simulation of urban mesh networks q