20 PERVASIVEcomputing Published by the IEEE CS n1536-1268/10/$26.00©2010IEEE

L O C A T I O N - B A S E D S E R V I C E S

A lthough GPS technology provides accurate outdoor localization, it has drawbacks related to its sat-ellite line-of-sight requirements, increased power consumption,

and hardware cost, all of which have moti-vated studies in network-based localization.1–3 Among the various data types used for localiza-tion, received signal strength (RSS) is the most popular because it’s available in most wireless networks and requires no additional measure-ment capability in the network infrastructure or on mobile devices.

As you would expect, RSS-based localiza-tion systems must obtain accurate signal data

from many different points in an area of interest.4,5 In this context, we often collect signal patterns via wardriv-ing6—gathering RSS statis-tics for wireless networks via a measurement device in a moving vehicle. It’s used most often to optimize cellular network radio coverage, but pattern-matching localiza-tions also use it to construct signal pattern databases.7 An-other recent trend is its use in constructing maps of Wi-Fi access points (APs).

Our analysis of wardriv-ing data measured from

Wideband Code Division Multiple Access (WCDMA), Global System for Mobile Com-munications (GSM), and Wi-Fi networks re- and Wi-Fi networks re-and Wi-Fi networks re-

veals that signal data measured via wardriving should be processed carefully. That is, signal patterns measured at the same place can differ considerably depending on movement factors, such as the wardriving vehicle’s direction and speed. In particular, we observed signal drag-ging:8 when a wardriving terminal moves, the captured pattern tends to contain more signals from the reference points behind it (relative to the terminal’s moving direction) than from ref-erence points in front of it. (Reference points are base stations [BSs] in cellular networks and APs in wireless LANs, so we use either term, depending on the network discussed.)

In this article, we describe signal dragging’s causes and its impact on localization accuracy. Because using wardriving data without consid-ering this effect can result in large localization errors, we propose three postprocessing meth-ods to mitigate it.

The Signal-Dragging Phenomenon By analyzing various wardriving data, we can show the substantial signal-dragging phenom-enon that occurs when a terminal moves.

Wardriving DataA network measurement report (NMR) is an instance of RSS data measurement conveyed by a wireless terminal. An NMR usually con-tains multiple entries, each of which is the pair of a BS or AP identifier and the RSS of the BS’s pilot signal (for WCDMA/GSM) or the AP’s beacon frame (for Wi-Fi). The NMR’s original purpose was to support a network handoff de-cision by comparing RSS against a few thresh-

Signal measurement records in moving devices tend to contain outdated signal information that doesn’t correspond to the current location. The three postprocessing methods described here can mitigate this effect in wardriving efforts.

Jinyoung HanSeoul National University

Jeongkeun LeeHewlett-Packard Laboratories

Ted “Taekyoung” KwonSeoul National University

Daehyung JoSamsung Electronics

Taejoon HaRadiant Technologies

Yanghee ChoiSeoul National University

How to Mitigate Signal Dragging during Wardriving

JANUARY–MARCH2010 PERVASIVEcomputing 21

old parameters, but it’s also useful in network-based localization because it’s measured at short intervals (say, a few seconds). Here, we consider two use cases:

• a terminal’s radio module provides an NMR to its application program, which runs a localization algorithm in the terminal itself, and

• a terminal sends an NMR to entities in the network infrastructure over the air.

In light of these use cases, an NMR in this context refers to either the mea-surement data at a test terminal or an instance of wardriving data stored in a pattern-matching system’s signal pat-tern database.

We analyzed comprehensive NMR data from multiple locations in Seoul and Chicago, measuring WCDMA and Wi-Fi data in urban areas and GSM data in suburban zones. The wardriv-ing vehicles moved at approximately 10 to 30 kmh. Cellular service provid-ers provided accurate BS locations for the WCDMA and GSM networks we tested; we estimated the Wi-Fi AP loca-tions from wardriving data, as we de-scribe later.

Signal Dragging In many cases, an NMR contains too many BS RSS entries to perform local-ization. We found that in GSM net-works, for example, the NMR included more than 15 BSs, even in suburban ar-eas, and in most WCDMA networks, it contained more than 10 BSs. However, the real problem is that an NMR re-

ported by a mobile terminal sometimes contains cached BS RSS entries that don’t correspond to the current mea-surement location.

Figure 1 shows signal dragging in action. An NMR from a mobile ter-minal contains signals from all three BSs while the terminal moves in a northeast direction (two signals from the backward BSs and one signal from the forward one). To illustrate signal dragging geometrically, we draw vec-tors from the terminal to the BSs as dotted arrows in Figure 1; a green ar-row shows the sum of BS vectors. If we reverse this sum, the resulting vector’s direction is somewhat close to the mo-bile terminal’s moving direction.

Figure 2 shows the areas in Seoul and Chicago in which we collected WCDMA, GSM, and Wi-Fi data sets. Red arrows from the start to finish po-sitions represent measurement termi-nal trajectory; blue arrows depict the reversed sum of BS vectors. The blue arrows are approximately pointing toward the mobile terminal’s moving direction, indicating sizeable signal drag.

The Signal-Dragging IndicatorA signal-dragging indicator (SDI) helps quantify the degree of the effect. We first calculate the angular difference (angularDiff) between the sum of BS vectors (sumOfBS) and the actual us-

er’s moving direction vector (user). Us-ing angularDiff, we can calculate the SDI as follows:

angularDiffsumOfBS user

=⋅−cos

| ||1 sumOfBS user

||

SDIangularDiff

= −−90

90

�

� .

The SDI’s value ranges from –1 to 1. If it’s closer to 1, the angularDiff is almost zero, indicating that most of the BSs in the NMR are behind the ground truth—the test terminal’s ex-act location—and thus exhibit strong signal dragging. If the SDI is near 0, it indicates little correlation between the terminal’s movement and the BSs’ geometric positions. If it’s closer to –1, the NMR contains the more forward-placed BSs, contradicting the signal dragging.

Figure 3 shows SDI distributions for the two cities in Figure 2; the values range only from 0 to 1 because the fre-quency of negative SDI values is negli-gible. Many of the frequently appearing SDI values were above 0.5, with the av-erage SDI at 0.57, 0.66, and 0.45, re-spectively, for the two cities. When we performed the same experiment for the WCDMA network over a 5-km × 5-km area in Seoul, the GSM network over a 2.8-km × 2.3-km area in Chicago, and the Wi-Fi network over a 1.6-km × 1.7-km area in Seoul, we found the

Sum of BS vectorsActual moving direction of terminalBS5

BS4

BS3

BS2

BS1

BS6

N

S

EW

Actual moving directionA BS vectorSum of BS vectors

Figure 1. Signal dragging. A network measurement report from a mobile terminal contains two signals from the base stations (BSs) behind the terminal and one signal from the one in front of it. The reversed sum of BS vectors are approximately pointing in the mobile terminal’s moving direction, indicating noticeable signal dragging.

22 PERVASIVEcomputing www.computer.org/pervasive

LOCATION-BASED SERVICES

average SDIs to be 0.17, 0.17, and 0.16, respectively. Although the average in-dicators over the larger areas in each network were smaller than the sample trajectories, many places exhibited high SDI values.

Causes of Signal Dragging When a mobile terminal moves around a cellular network, it receives all the pilot signals of nearby BSs. To make timely handoff decisions, it must maintain up-to-date RSS informa-tion measured from the serving BS, which serves the mobile terminal, and its surrounding stations, so it catego-rizes identified BS entries into multi-ple sets. We base our discussion here on the CDMA2000 specification’s BS set-handling policy,9 but other cellular networks (such as WCDMA or GSM) use similar approaches:

• The active set contains the serving BSs whose paging or data traffic a mobile terminal currently monitors or uses.

• The candidate set contains the BSs that aren’t currently in the active set but whose pilot signals were recently received by the mobile terminal with sufficient signal strength to demodu-late the signal.

• The neighbor set contains the BSs likely to be candidates for idle han-dover as specified in a control or paging channel message from the serving BS.

• The remaining set contains all other BSs not in the first three sets.

(a)

Long

itude

Latitude

600m

37.5270

37.5265

37.5260

37.5255

37.5270

37.5245

37.5240

37.5235

37.5230

37.5255

Trajectory of a mobile terminal

Reversed sum of BS vectors

Start

Finish

650 m127.033127.028 127.029 127.030 127.021 127.032

(b)

Long

itude

Latitude

41.8678

41.8676

41.8674

41.8672

41.8670

41.8668

41.8666

41.8664

41.8662

41.8660

Trajectory of a mobile terminal

Reversed sum of BS vectors

Start Finish

–87.664–87.668 –87.667 –87.666 –87.665

(c)

Long

itude

Latitude

37.5375

37.5370

37.5365

37.5360

37.5355

37.5350

37.5345

37.5340

37.5335

37.5330

Trajectory of a mobile terminal

Reversed sum of BS vectorsStart

Finish

127.025127.024127.02 127.021 127.022 127.023

620m

780 m

250m

640 m

Figure 2. Signal dragging in two cities: (a) WCDMA network trajectory in Seoul, Korea (155 measurement points); (b) GSM network trajectory in Chicago, Illinois (105 measurement points); and (c) Wi-Fi network trajectory in Seoul (95 measurement points). The blue arrows point toward the mobile terminal’s moving direction, showing that signal dragging notably occurs in the three networks.

JANUARY–MARCH2010 PERVASIVEcomputing 23

Because a mobile terminal often com-poses an NMR not only from its ac-tive and candidate set BSs but also from neighbor (and possibly remain-ing) sets, the NMR can contain old BS RSS entries not corresponding to the current measurement location. In particular, the old cached RSS values in the neighbor and remaining sets are the major culprit in signal dragging.

We observed that signal dragging happened in not only cellular net-works but also Wi-Fi networks (the av-erage SDI value of the Wi-Fi network in Figure 3 is 0.45). For example, we observed that Netstumbler, a popular Wi-Fi wardriving tool, tends to retain and report the RSS information of APs detected by a Wi-Fi laptop up to two minutes prior. Thus, we should also consider signal dragging when using wardriving data from Wi-Fi networks.

Signal Dragging’s Effect on Localization Signal dragging can significantly af-fect localization accuracy. For exam-ple, a pattern-matching localization system’s performance heavily depends on the signal pattern database’s accu-racy, so eliminating outdated BS entries from wardriving data is essential to construct and maintain an accurate da-tabase. Moreover, when a test terminal wishes to locate itself, it starts by ob-taining an NMR, which might contain outdated BS RSS entries. Reducing the localization errors caused by outdated data requires filtering out that outdated data.

Finally, many localization tech-niques require a network database (NDB) that contains the exact lo-

cations of BSs. However, obtaining NDBs from wireless network opera-tors isn’t easy because of confidenti-ality concerns. Cellular network op-erators might not maintain the most accurate or up-to-date BS location information because of maintenance

costs, which are hard to justify when there isn’t a strong need to maintain an accurate NDB. Personal Wi-Fi net-work users frequently deploy and relo-cate their APs in indoor and outdoor locations, so accurate Wi-Fi NDB information is unavailable or almost

(a)

(b)

(c)

0

35

30

25

20

15

10

5

00.1 0.2 0.3 0.4 0.5

Signal-dragging indicator (SDI)0.6 1.00.90.80.7

Freq

uenc

y

0

242220181614121086420

0.1 0.2 0.3 0.4 0.5SDI

0.6 1.00.90.80.7

Freq

uenc

y

0

20

18

16

14

12

10

8

6

4

2

00.1 0.2 0.3 0.4 0.5

SDI0.6 1.00.90.80.7

Freq

uenc

y

Figure 3. Signal-dragging indicator (SDI) distributions for the trajectories in Figure 2: (a) WCDMA in Seoul, (b) GSM in Chicago, and (c) Wi-Fi in Seoul. The average SDIs for the three areas were 0.57, 0.66, and 0.45, respectively, indicating noticeable signal dragging in these sample areas.

24 PERVASIVEcomputing www.computer.org/pervasive

LOCATION-BASED SERVICES

impossible to obtain. Inferring Wi-Fi AP locations from wardriving data has become a popular way to circum-vent this problem—for example, Sky-hook Wireless provides Wi-Fi-based positioning services by constructing inferred Wi-Fi NDBs for many ma-jor cities. Because an inferred NDB’s accuracy highly affects localization performance, we should consider sig-nal dragging when constructing an in-ferred NDB from wardriving data.

Mitigating Signal DraggingTo mitigate signal dragging, we must differentiate outdated (or currently un-reachable) BSs from active ones in the NMRs, and then use only the active BSs for localization. We propose three methods for filtering outdated BS data:

• GT-based filtering. When a BS’s position is far from the mobile ter-minal’s ground truth (GT), we can treat the corresponding BS data as cached. To filter out the cached BS data, we can iteratively remove the BSs starting with the one farthest from the GT, but only when the mea-surement terminal’s GT is available. For example, when a localization service provider collects wardriving data to construct or update a pattern database, GPS instantly obtains the measurement terminal’s accurate lo-

cation and records it along with the associated NMR.

• WCL-based filtering. The weighted centroid localization (WCL) algo-rithm determines a test terminal’s position from the weighted average of the BS positions in the NMR by using the BSs’ RSS values as weights. If a BS’s position is far from the WCL result, we can deem the BS data as cached and then remove it. To filter out the cached BS data, we can itera-tively remove the BSs starting from the one whose position is the farthest from the WCL result. We recompute the reference WCL location every time we remove a BS.

• RSS-based filtering. This approach eliminates outdated data on the ba-sis of RSS values on the assumption that the RSS from the farther BSs will be weaker. To filter out the cached BS data, we iteratively remove the BSs starting with the one whose RSS is the lowest. This method is more useful when BS location information isn’t available.

Before filtering out any BSs, we first select the serving BS and two more sta-tions whose RSS values are the stron-gest (meaning they’re likely to be clos-est to the mobile terminal), but we don’t remove them during filtering. We preserve them because we need at

least three BSs to compute the user’s location via WCL or trilateration. In any of the filtering methods we just de-scribed, we can remove up to nine BSs based on a given criteria while keeping the three preselected BSs. To maximize the chosen filtering method’s effect, we seek the optimal number of removed BSs by considering various system pa-rameters such as network types (GSM versus Wi-Fi) and test areas (urban ver-sus suburban). We leave the particulars to future research.

Figure 4 shows GT-based filtering on the wardriving data from the WCDMA network in Seoul. The red arrow indi-cates the terminal’s trajectory; at its current position, the NMR includes six BSs. As we remove the farthest one, BS6, from the NMR, the SDI decreases from 0.75 to 0.14, indicating that the filtering effectively mitigated the signal-dragging effect.

Figure 5 shows the average SDIs for various data sets (the three sets from Figure 2 and the WCDMA2 set in Fig-ure 4) with and without the three fil-tering methods. In the WCDMA and WCDMA2 data sets, every filtering method eliminates the cached BS infor-mation effectively. However, the filter-ing methods aren’t as effective on the GSM and Wi-Fi sets.

Because the results in Figure 5 come from relatively small wardriving data

(a)

BS4

BS2

BS1

BS5

BS6StartStartFinishFinish

BS3

1,270

m

(b)

BS4

BS2

BS1

BS5

StartFinishFinish

BS3

Figure 4. Ground truth (GT)-based filtering. For the WCDMA scenario in Seoul, we used a 3.5 km2 area. (a) Before filtering, the SDI was 0.75; (b) after filtering, it was 0.14, indicating that our method mitigated signal dragging.

JANUARY–MARCH2010 PERVASIVEcomputing 25

sets, we ran the same filtering experi-ments over the larger areas mentioned earlier; Figure 6 shows the results. Al-though the average SDIs in the origi-nal data sets (without filtering) aren’t high (roughly 0.17), the three filtering methods noticeably reduce the SDI values in the WCDMA and GSM net-works. However, the RSS-based filter-ing doesn’t reduce the SDI in the Wi-Fi network. One possible reason is that most Wi-Fi APs are indoors, so RSS values exhibit large fluctuation and don’t indicate a distant or old AP entry. Similarly, the GT- and WCL-based fil-tering methods didn’t show significant SDI reductions in the Wi-Fi network; we believe this is because we used an inferred NDB affected by signal drag-ging for the AP locations.

Filtering’s Effect on Pattern MatchingIn pattern-matching localization sys-tems, operators collect signal patterns from several selected points in the ser-vice area and store them in a pattern database.2,7 Based on the assumption that the signal pattern at a certain point is temporally and spatially sta-ble, we can determine a terminal’s po-sition by comparing its current signal pattern (NMR) to the database entries and finding the best-matching location.

To see the impact of signal dragging and our proposed filtering methods on pattern-matching systems’ localization accuracy, we collected signal patterns by wardriving in both directions on the same course of trajectories in Figure 2’s test areas. For each of the WCDMA, GSM, and Wi-Fi data sets in the fig-ure, we measured signal patterns on the same road using the same device in two opposite driving directions: we call one direction initial and the other reverse.

We first considered two reference cases in which we applied no filtering methods. In one case, we constructed the signal pattern database by war-driving in the initial direction. The test terminal for localization moves in the reverse direction, so we call this

case no-filter-reverse. The other case is when the test terminal moves in the same initial direction as the pattern database wardriving vehicle (no-filter-same). As Figure 7 shows, the average localization errors in no-filter-same for the three networks were much smaller than those in no-filter-reverse, clearly indicating that the signal pat-tern will be very different depending on the wardriving direction.

We applied the three filtering meth-ods to both the signal pattern database and the NMRs from the test terminal of no-filter-reverse. Figure 7 shows av-erage pattern-matching localization errors in the three networks. Com-pared to no-filter-reverse, the filtering methods can reduce pattern-matching errors depending on network types and filtering methods. The degree of error reduction in Figure 7 is strongly

0.6

0.5

0.4

0.3

0.2

0.1

0WCDMA GSM

WCL-based

WCDMA2Wi-FiAv

erag

e SD

I

RSS-basedGT-basedNo �ltering

0.18

0.16

0.14

0.12

0.10

0.08

0.06

0.04

0.02

0

WCL-based

Aver

age

SDI

RSS-basedGT-basedNo �ltering

WCDMA5 km × 5 km,

31,360measurement points

GSM2.8 km × 2.3 km,

9,444measurement points

Wi-Fi1.5 km × 1.7 km,

3,091measurement points

Figure 5. Average SDIs for the test areas of Figures 2 and 4. The WCDMA data set in Figure 4 is denoted as WCDMA2. Every filtering method effectively reduces the SDI values in the WCDMA and WCDMA2 data sets of Figures 2 and 4.

Figure 6. Average SDIs in the larger test areas. In general, filtering methods effectively mitigate signal dragging in large areas.

26 PERVASIVEcomputing www.computer.org/pervasive

LOCATION-BASED SERVICES

related to the reduction of the SDI values in Figure 5. For instance, the WCDMA network shows the largest reduction in SDI values and, conse-quently, localization errors (roughly 40 percent with the RSS-based filter-ing). In contrast, the GSM network

exhibits little reduction in SDI values, resulting in little gain from using these filtering methods.

In real localization scenarios, we can’t apply GT-based filtering to a mobile terminal’s NMRs. So, we could use GT-based filtering for the pattern

database and WCL-based filtering for the terminals that request localization services.

Filtering’s Effect on Inferred NDB Accuracy By using the same data sets from Fig-ure 6, we generated inferred NDBs for the WCDMA and GSM networks; we excluded the Wi-Fi network from this evaluation because we don’t have a real Wi-Fi NDB with which to com-pare our results. We identified 154 and 54 BSs from the WCDMA and GSM data, respectively, and inferred each BS’s location by averaging the GT locations. Before constructing the inferred NDBs, we applied the three filtering methods to the raw wardriv-ing data to mitigate signal dragging. We calculated the Euclidean distance between the estimated BS location and the actual one provided by network operators to measure the estimation error (the inferred NDB error).

Figure 8 shows the WCDMA and GSM networks’ inferred NDB errors, respectively; for comparison, we also show the operator-provided NDB (the left-most bars in each subgraph). All three filtering methods effectively re-duced the inferred NDB error com-pared to the original inferred NDB with no filtering applied.

Figure 8 also compares WCL errors. The inferred NDBs we constructed from filtered wardriving data pro-duced fewer localization errors com-pared to the case without filtering for both the WCDMA and GSM net-works. Interestingly, localization with filtered inferred NDBs performs better than that with the real NDBs from net-work operators, signifying the impor-tance of signal-dragging mitigation for more accurate localization.

T o the best of our knowl-edge, this is the first study that presents concrete methods for refining war-

driving data for location-based ser-

180

160

140

120

100

80

60

40

20

0

WCL-based

Aver

age

patte

rn-m

atch

ing

loca

lizat

ion

erro

r (m

)

RSS-basedGT-basedNo- lter-reverseNo- lter-same

WCDMA GSM Wi-Fi

1,200

1,000

800

600

400

200

0

RSS-based WCL-basedGT-basedNo �lteringNDB

Averageinferred NDB

error

AverageWCL error

Averageinferred NDB

error

AverageWCL error

WCDMA GSM

Infe

rred

NDB

and

term

inal

loca

lizat

ion

erro

rs (m

)

Figure 7. Average pattern-matching localization errors (in meters) for WCDMA, GSM, and Wi-Fi networks. The filtering methods can reduce pattern-matching errors depending on network types and filtering methods. The degree of pattern-matching error reduction is co-related with the reduction of the SDI values in Figure 5.

Figure 8. Comparison of inferred NDB errors and WCL errors among filtering methods for WCDMA and GSM networks. Filtering improved the accuracy of inferred NDB and localization performance based on the inferred NDB. We used the same WCDMA and GSM data sets from Figure 6: average inferred NDB error, average WCL error, average inferred NDB error, and average WCL error.

JANUARY–MARCH2010 PERVASIVEcomputing 27

vice applications. In our own work, we’re studying ways to further opti-mize the filtering methods discussed in this article by considering various system parameters such as network types (GSM versus Wi-Fi) and test ar-eas (urban versus rural). We plan to devise another filtering method con-sidering the spatial or temporal rela-tion of RSS data among adjacent mea-surement points and to study how to further refine the wardriving data by considering the speed and direction of wardriving vehicles.

ACKNOWLEDGMENTSThisworkwassupportedpartlybytheITR&DprogramofMKE/IITA(2007-F-038-03,Funda-mentalTechnologiesfortheFutureInternet)andinpartbytheKoreaResearchFoundationgrantfundedbytheKoreanGovernment(MOEHRD,BasicResearchPromotionFund;KRF2007-331-D00267).SeoulNationalUniversity’sInstituteofComputerTechnology(ICT)providedresearchfacilitiesforthisstudy.

REFERENCES 1. M.Y. Chen et al., “Practical Metropoli-

tan-Scale Positioning for GSM Phones,” Proc. Ubicomp, Springer, 2006, pp. 225–242.

2. Y.-C. Cheng et al., “Accuracy Charac-terization for Metropolitan-Scale Wi-Fi Localization,” Proc. Mobisys, ACM Press, 2005, pp. 233–245.

3. J. Lee et al., “Distributed and Energy-Effi-cient Target Localization and Tracking in Wireless Sensor Networks,” Computer Comm., vol. 29, nos. 13–14, 2006, pp. 2494–2505.

4. D. Fox et al., “Bayesian Filtering for Loca-tion Estimation,” IEEE Pervasive Com-puting, vol. 2, no. 3, 2003, pp. 24–33.

5. A. Lamarca et al., “Place Lab: Device Positioning Using Radio Beacons in the Wild,” Proc. 3rd Int’l Conf. Pervasive Computing, Springer, 2005, pp. 116–133.

6. M. Kim, J. Fielding, and D. Kotz, “Risks of Using AP Locations Discovered through War Driving,” Proc. Pervasive, Springer, 2006, pp. 67–82.

7. P. Bahl and V.N. Padmanabhan, “Radar:

An In-Building RF-Based User Location and Tracking System,” Proc. IEEE Info-com, IEEE Press, 2000, pp. 775–784.

8. D. Jo et al., “Signal Dragging: Effects of Terminal Movement on War-Driving in CDMA/WCDMA Networks,” Proc. Location- and Context-Awareness

(LOCA), Springer, 2007, pp. 211–227.

9. “Upper Layer (Layer 3) Signaling Stan-dard for CDMA2000 Spread Spec-trum System, 3GPP2 C.S0005-D v2.0,” 3GPP2, 2005; www.3gpp2.org/Public_html/specs/C.S0005-D_v2.0_071012.pdf.

theAUTHORSJinyoung Han isworkingtowardaPhDincomputerscienceandengineeringatSeoulNationalUniversity’sSchoolofComputerScienceandEngineering.Hisresearchinterestsincludepeer-to-peernetworks,wirelessmeshnetworks,andubiquitouscomputing.Contacthimatjyhan@mmlab.snu.ac.kr.

Jeongkeun Lee isaresearchscientistatHewlett-PackardLaboratoriesinPaloAlto,California.Hisresearchinterestsincludemobilenetworkingsystemsandenterpriseanddatacenternetworking.LeehasaPhDincomputerscienceandengineeringfromSeoulNationalUniversity.He’samemberoftheIEEEandtheACM.Contacthimatjklee@hp.com.

Ted “Taekyoung” KwonisanassociateprofessoratSeoulNationalUniver-sity’sSchoolofComputerScienceandEngineering.Hisresearchinterestsaresensornetworks,wirelessnetworks,IPmobility,andubiquitouscomputing.KwonhasaPhDincomputerscienceandengineeringfromSeoulNationalUni-versity.Contacthimattk@snu.ac.kr.

Daehyung Jo worksinSamsungElectronics’DigitalMediaandCommunica-tionsR&Dcenter.Hisresearchinterestsincludenodetracking,meshandadhocnetworks,andwireless/wirednetworkconnectivity.JohasanMSincom-puterscienceandengineeringfromSeoulNationalUniversity.Contacthimatdaehyung.jo@samsung.com.

Taejoon Ha hasaseniormanagementroleinRadiantTechnologies’engineer-ingdivision.Hisresearchinterestsincludeprojectandprocessmanagement,softwarearchitecture,softwarequalitymanagement,andprocessimprove-ment.Contacthimattjha@radiantech.net.

Yanghee Choi isdeanofSeoulNationalUniversity’sGraduateSchoolofCon-vergenceScienceandTechnology,presidentoftheAdvancedInstitutesofCon-vergenceTechnology,andchairofKorea’sFutureInternetForum.HehasaPhDincomputersciencefromEcoleNationaleSuperieuredesTelecommunicationsParis.Contacthimatyhchoi@snu.ac.kr.