Location Estimation in Ad-Hoc Networks with Directional Antennas

Transcription

1 Location Estimation in Ad-Hoc Netwoks with Diectional Antennas N. Malhota, M. Kasniewski, C. Yang, S. Bagchi, and W. Chappell School of Electical & Compute Engineeing, udue Univesity 465 Nothwesten Avenue, West Lafayette, IN 4797 USA. {nmalhot, mkasnie, cyang, sbagchi, Abstact With the development of location awae senso applications, location detemination has become an inceasingly impotant middlewae technology. Numeous cuent technologies fo location detemination of senso nodes use the eceived signal stength fom senso nodes using omni-diectional antennas. Howeve, an inceasing numbe of senso systems ae now deploying diectional antennas due to thei advantages like enegy consevation and bette bandwidth utilization. In this pape, we pesent techniques fo location detemination in a senso netwok with diectional antennas unde diffeent kinds of deployment of the nodes. We show how the location estimation poblem can be solved by measuing the eceived signal stength fom just one o two anchos in a D plane with diectional antennas. We implement ou technique using Bekeley MICA senso motes and show that it is up to thee times moe accuate than tiangulation using omni-diectional antennas. We also pefom Matlab simulations that show the accuacy of location detemination with inceasing node density. Keywods: location estimation, diectional antenna, tiangulation, eceived signal stength, eceived signal angle. Intoduction Senso netwoks povide a pomising infastuctue fo gatheing infomation about paametes of the physical wold. Tiny wieless nodes equipped with diffeent kinds of sensos can be distibuted ove a field and can collect and tansmit the data to a data aggegation point, such as a cluste head o a base station. In ode to intepet the sensed data, it is often necessay to know the location of the node which is the souce of the data. In addition, position infomation is valuable fo optimizing the outing pocess, as shown in many position awae outing potocols [][]. In geneal, these stategies seek to avoid wasting valuable bandwidth by minimizing the contol taffic fo oute detemination. Most position-based outing schemes also emove the need to maintain outing tables at the nodes. Also, the node s location may change. Mobile senso netwoks ae becoming an impotant class in which the nodes may move in a contolled manne o though passive mobility. It is possible fo a node to have up-to-date infomation of its location if it contains location detemination hadwae, such as a GS eceive, mounted on it. Howeve, fom an economic standpoint, this would violate the equiement fo the deployments to be cost-effective. The economic consideations have been diving the cost of the individual senso nodes down to the point whee sub-$ nodes ae beginning to look achievable [4]. GS hadwae would incease the pice of senso netwoks substantially. Commecially available GS eceives come in a wide pice ange of $-$,. The eceives at the lowest end give poo accuacy, with inaccuacies of tens of metes possible [5]. Receives that give sub-mete accuacy, which may be needed fo many senso applications, ae moe than $5, in pice. The hadwae also adds to the weight of the unit with typical eceives anging upwads of 5 oz. Finally, the battey lives of the eceives ae much shote than that of the senso nodes themselves, e.g., tens of hous fo the typical GS eceives compaed to multiple months fo a epesentative senso node, the Bekeley mote. Thus, the combined unit of the senso node and the GS eceive will have to be eplaced fa too fequently fo it to be pacticable fo a lage class of deployments. Moe geneally, the eceived signal stength fo a GS can be as low as -3 dbm, odes of magnitudes less than the stength of taditionally eceived signals in teestial applications and lowe than the sensitivity of eceives on typical senso nodes (- dbm fo Bekeley motes). Theefoe, expensive eceives would be needed. Also, since elatively unobstucted views ae equied fo GS localization, in many senso netwok deployments, the GS measuements would need to be supplemented with anging data fom the local netwok. Though it may not be feasible fo all the nodes to be equipped with special pupose location detemination hadwae, it may be possible to equip a small faction of the nodes in the netwok with such hadwae. Such nodes, called ancho nodes, can act as efeence points fo location infomation and othe senso nodes, called taget nodes, can use infomation fom ancho nodes to estimate thei location. In the most commonly used technique called lateation the distance measuements ae equied fom

2 (k+) neighbos in a k dimensional plane. The example of lateation in a -dimensional plane is called tiangulation in which the senso node needs to know the distances fom thee neighboing nodes. Seveal appoaches exist fo estimating distance fom a neighbo, e.g., signal attenuation and time of flight. In signal attenuation, the powe of eceived signal is measued by the senso node and knowing the signal stength emitted by the souce node and the attenuation elationship with distance (such as, / whee is the sepaation distance), the elative distance can be calculated. Typically fo indoo envionments o lage distances, the attenuation elationship becomes complex and difficult to epesent concisely due to multi-path effects and eflection of the adio waves. Othe techniques fo measuing elative distances, such as time of flight ([],[6]), ae less useful in ou envionment since the adio signal tavels at the speed of light and the distances taveled fo signals by the senso nodes ae elatively shot. Diectional antennas povide impotant benefits in senso netwoks. Diectionality can be used as a fom of divesity built into the senso node, which helps in coping with the vaiability in the communication channel and educe the link eo ate. The diectionality povides inceased tansmission anges compaed to omni-diectional antennas by focusing the tansmission enegy in the desied diection. They can also incease the secuity of communication by esticting the set of neighbos that can ovehea a communication [5]. Diectionality in expensive communication systems is commonly achieved though the ceation of a phased aay. Howeve, this is extemely expensive and is used pedominantly only in high cost militay applications. In addition, it is equied that the elements of the phased aay be an appeciable faction of a wavelength apat. This would not be possible in electically small fom facto senso nodes. This pecludes the use of a taditional aay to povide the desied beam scanning. Howeve limited diectionality can be cheaply integated into a small fom facto senso node. In this investigation, educed size patch antennas have been developed using standad patch aangements with high dielectic constant antennas. Multiple diectional antennas ae utilized and a simple switching netwok enables us to switch between polaization states ([7],[8]) and the diection of adiation. The solution to location detemination with omni diectional antennas is not applicable to diectional antennas since the adiation pattens ae diffeent and the eceived powe is dependant on angle as well as distance. In this pape, we use a model fo senso nodes equipped with fou diectional antennas. Diectionality povides elative angle measuements between ancho nodes and taget nodes with unknown positions and has been agued to impove localization estimates [3]. We demonstate an accuate method using knowledge of angle of eception. This knowledge is impecise due to channel andomness, and is based on measuements on a set of actual fabicated patch antennas. We show that even if the nodes have sizes much smalle than the wave length of the RF wave, then a senso node can estimate its location using infomation fom a single ancho. The tiangulation method, in contast, uses thee ancho nodes in the ideal case of no eos fo a two dimensional plane. The Eo Resilient Tiangulation (ERT) technique to cope with eos woks as follows. Given a edundant numbe of anchos, a edundant set of linea equations is set up and solved to minimize the least squae eo of the position estimate. In ou appoach, the multiple distance measuements, one fom each ancho node, ae aveaged to detemine the taget node s location. We pesent the algoithms fo position detemination fo diffeent kinds of deployment aligned and unaligned, single and multiple anchos. These algoithms localize nodes ove a single link, and then to apply this to a full adhoc netwok one can use existing algoithms such as [8] to disseminate location knowledge. The algoithms ae implemented on Bekeley senso motes commecially available though Cossbow Inc. of type Mica and expeiments caied out with them in an outdoo envionment in a eal senso netwok testbed. The expeimental data undelines the fact that simple dipoles do not have a tuly omni-diectional adiation patten and the patten fluctuates with time. This opens to question esults fo potocols developed using ideal adiation pattens. The expeimental esults show that adial distance estimates ae % bette when one of the motes (ancho o taget) have diectional antennas compaed to a completely dipole aangement (i.e. all motes having a unidiectional adiato). The aggegate position eo is minimized by employing two diectional anchos and switching between the antennas of the anchos fom which the tansmitted signal is gatheed. The aveage eo ove multiple ancho and taget positions in this case is.6% while that fo the omnidiectional system with 3 anchos the eo is 7.5%. The measuements with each node having diectional antennas lead to the insight that even if taditional tiangulation wee to be used, it is useful to employ diectional antennas to gathe the adial distance data. Even by discading the angle of tansmission and eception data, the tiangulation pocess is moe accuate with diectional antennas. We also build a simulation model whee the geneal case of ancho and taget nodes being unaligned with espect to each othe is simulated. Such oientations may esult fom some apid deployment of the nodes. The simulation shows esults fo vaying numbe of ancho node neighbos. We show though simulation that the aggegate eo is educed though the use of ou technique compaed to the ERT method by to 3.5 times. The est of the pape is oganized as follows. Section pesents backgound mateial on location detemination in ad-hoc wieless netwoks. Section 3 sets up the mathematical model fo ou solution appoach with diectional antennas. Section 4 pesents the simulation expeiments and esults. Section 5 concludes the pape with mention of futue wok.

3 Backgound Tiangulation is a common method fo locating objects using othe objects which do know thei position. This is an applicable model fo ou envionment whee the positions of some senso nodes, possibly equipped with GS eceives, ae known. A nice oveview of tiangulation based location detemination techniques is to be found in [] and []. The tiangulation techniques can be subdivided into two categoies lateation, which uses distance measuements, and angulation, which uses angle measuements along with distance. If individual distance measuements ae completely accuate, lateation equies (n+) neighbos with knowledge of location to pinpoint the taget node in an n dimensional plane. An example of lateation in two dimensional space is shown in Figue (a) and is called tiangulation. Example use is in the Active Bat Location System []. Diffeent appoaches exist fo estimating the distance fom a neighbo, fo example time of flight, attenuation of signal stength, and diectionality ([],[3]). Measuing signal stength elies on the popety that adio waves attenuate in thei signal stength with inceasing distance between the tansmitte and the eceive. The eceive can calculate the distance if it knows the tansmission powe and the attenuation model [4]. d d 3 (a) d θ d (b) Ancho node Taget node Figue. Location detemination with neighboing ancho nodes. Lateation is in (a) and angulation in (b) The attenuation is often modeled as /, whee is a elatively shot distance outdoos. Indoos, eflection, and multi-path fading make the model and hence, the location estimate, inaccuate. The thid way of estimating location is to compute the angle of each efeence point with espect to the sensing node in some efeence fame. The position of the mobile node can then be computed using angulation. In pactice, the individual distance measuements ae inaccuate because the exact elation between the measuement of physical popeties, such as signal stength, and the inte-node distance is not known. Hence, infomation fom geate than (n+) nodes is needed fo pinpointing a taget node in an n dimensional plane. The wok in [6] pesents an appoach fo minimizing the θ aggegate eo by consideing measuements fom a edundant numbe of ancho nodes. A edundant set of equations is lineaized and solved to minimize the least squae eo. In [7], Savaese et al. popose an iteative potocol that diffuses the location infomation gatheed fom neaby ancho nodes though the netwokusing this technique, Bagchi et al. [8] show the elationship between the numbe of ancho nodes and the eos in the location detemination, given a cetain eo in one-hop neighbo distance estimation. Angulation is an altenate method to lateation fo computing location based on neighbo infomation, whee angles ae used in addition to distance. A schematic of the use of angulation is shown in Figue (b). Diectional antennas ae needed fo the angle measuements. evious wok has used phased antenna aays to use the angulation technique []. Sukhatme et al. show that using ange and appoximate secto infomation can impove localization accuacy at easonable node densities [3]. Niculescu discusses using the angle of aival (AOA) of the signal and node oientation adjustment to find node locations [8]. Thee is a class of location detemination techniques that do not ely on any popety of the eceived signal. Instead, they ely on the connectivity measue, i.e., if a node a is able to hea fom anothe node β, then a is connected to β and its location is constained to be within the tansmission ange of β ([3],[9],[]). This class of techniques based on connectivity measue povides location estimates which ae quite coase-gained. The ganulaity becomes coase with lage tansmission anges of the efeence nodes. An ovehead of beacon o hello messages is also incued and the convegence times of the algoithms ae often sensitive to the fequency of these messages []. Also, some of the potocols ([9],[]) equie centalized pocessing which limits thei scalability. Röme poposes a technique geaed to dust-sized senso nodes which only have passive optical communication capability and do not have active RF communication capability. It elies on a poweful base station that sends a photo beam and otates. Each senso node has a photo beam detecto and a clock and maks how long it sees the beam and the peiod of otation and detemines its location based on this. The method is only applicable if single hop communication is possible between all nodes and the base station. Also, as has been demonstated in [6] and appeas well accepted, distance measuements ove lage distances ae vey inaccuate. 3 Solution fo Diectional Antennas 3. Diectional Antenna Model One of the simplest semi-diectional antennas is the patch antenna. This antenna is used as a epesentative example of an antenna that may be used fo localization and which will still fit on a mobile fom facto. The ideal patch adiation 3

4 model is a hemispheical adiato which allows fo semidiectional adiation. The typical gain of a patch antenna is on the ode of 3.5 to 6 dbi, depending on the dielectic substate used in the design. A epesentative angula vaiation of the gain fo a typical micostip antenna will be in the ange of βl βl θ G ( θ) θ < < cos sin( ) cos sin( ) whee β is the fee-space constant and l is the longest length of patch, assuming the lowest ode mode of opeation []. The gain is defined as the atio of the intensity, in a given diection, to the adiation intensity that would be obtained if the powe accepted by the antenna wee adiated isotopically. In the E-plane cut, the antenna s adiated e-field fom a βl standad patch adiato is ideally E = cos sin( θ ). This patten dependence is in elation to a coodinate system with the z-axis pependicula to the micostip patch adiato. This is the ideal solution fo a patch antenna with an infinite gound plane and is only slightly alteed using a finite size gound plane. The gound plane is used to shield the adiating field fom the est of the cicuity and the othe adiatos. Unshielded adiatos, such as those that ae standad with the motes, ae susceptible to paasitic adiating cuents which esult in asymmetic pattens. The eceived powe at an antenna is given by G ( Θ ) G ( Θ ) t t t λ =, whee Θ t and Θ ae the 4π tansmitting and the eceiving angles, espectively, and is the distance between the tansmitte and the eceive. λ is the RF wavelength of the caie fequency. Since (λ/4π) is a constant, we will exclude it fom futue expessions. It was howeve included in calculating the esults. A ealistic antenna adiation patten obtained fom the design of patch antennas in the HFSS simulation package is used to model the gain fo the expeiments. The simulations wee validated though actual expeiments in an anechoic chambe. Fo the senso netwok simulations in this pape we use an antenna model given βl by G ( Θ) = G ( Θ) = cos sin( Θ). This is the uppe t bound of the possible antenna gain given in Section 3 and is chosen in ou analysis and simulation so that an ancho has a lage numbe of taget nodes within its tansmission ange. Howeve, the poposed techniques ae equally valid fo any othe antenna model. 3. Aligned antennas In a numbe of pactical applications it is easonable to expect that the sensos will be manually deployed. Sensos set up to monito a bidge s health have to be placed by constuction wokes on the bidge fo example. In such scenaios even though it may not be possible to know the pecise location of the senso, it is possible to place these sensos in a pe-detemined oientation. Figue. Location detemination with aligned nodes If the antennas of a taget senso node ae aligned, then we can use the powe eceived at multiple eceiving antennas of the taget fom a single tansmitting antenna on an ancho fo position estimation. Without loss of geneality conside that an ancho node is placed to the south-east of the taget node as shown in Figue. The size of the senso would usually be much smalle than the tansmission distance. So d/=θ c. Then eceived powe at the two eceiving antennas of the taget node is given by equations () and () in two vaiables Θ and. Since, these ae nonlinea equations it is difficult to get a closed fom solution fo Θ and in tems of the input vaiables, and,. Howeve, these equations can be numeically solved by standad methods to obtain Θ and., Taget = G t B π t ( Θ ) G ( Θ ) = G Θ G ( Θ ) t, d Receive C D Θ,, Θ, t Θ t Θ t (), = G (,) (,) ( ) ( ) t Θt G Θ = G t Θ G Θ π d π d = G () t + Θ G + Θ Whee Θ = Θ, and Θ = Θ,. Altenatively, if the oientations of these sensos ae not pefect, Θ in () and () can be eplaced by Θ' = Θ Ф unaligned, whee Ф unaligned can be obtained fom a digital compass [3] o some othe simple algoithms [9]. A possible appoach is mounting an omni-diectional antenna with the fou diectional antennas on the same node and estimating Ф unaligned fom the diffeence of the eceived powe stength between the diectional antennas and the omni-diectional antenna. A baseline expeiment fo this is with the ancho node having omni-diectional dipole antennas. In this case the gain of the tansmitte, G t (Θ), is constant ove all Θ and Θ c Ancho Receive A 4

5 denoted G omni. Figue 3 shows this configuation. Now, since we know the distance as well as the elative diection of the taget with espect to the ancho, we can estimate its position. This estimate is based on measuements fom just one neighboing ancho node wheeas tiangulation equies measuements fom at least thee anchos. The estimates fom multiple anchos can be aveaged to obtain a bette estimate of the position. Altenatively, the infomation about Θ could be discaded and the ange measuements () can be used to tiangulate the position of the senso in a least squaes manne. Both these stategies have been evaluated in ou simulations and the aveaging stategy has yielded bette esults. Senso B Receive d θ C D Receive θ θ c θ y θ x A Ancho Figue 3. Location detemination with an omnidiectional tansmitte and diectional eceive 3.3 Genealization to Unaligned Antennas In cases whee it is not possible to ensue a global oientation of all nodes of a netwok, additional measuements can be used to estimate position. Received powe at two diffeent antennas of the taget node fom two tansmitting antennas of the ancho node is measued. Such an aangement is shown in Figue 4. Geometic elations between the vaious tansmission and eceiving angles can be deived fom the figue. π d Θ + Θ6 = Θ 4 + Θ8 = Θ3 + Θ5 = Θ + Θ7 = + + Θ + Θ + Θ = π,',' Θ 3 4 Gt ( π Θ Θ3 Θ4) * G ( Θ) = π d π d Gt ( + Θ3)* G ( + Θ ) = Let,ij denote the powe eceived by antenna i on the taget node when antenna j is tansmitting on the ancho node. We can use these equations to simplify the eceived powe equations as follows. (3) (4) Gt ( Θ3) * G ( Θ4),' = (6) Equations (3) though (6) in the fou vaiables Θ, Θ 3, Θ 4, and can again be numeically solved to estimate the location of the taget node. Senso Senso Ө N Ө 4 d Ө 3 Ө Ө Ө N Ancho π-ө N Ө N d Ө 6 Ө 8 Ancho Figue 4. Location detemination fo unaligned antennas This scheme equies that two taget antennas be able to simultaneously eceive tansmissions fom two ancho antennas. This would equie a tansmitte beam width of 8. This is non-optimal fo fou antennas coveing a 36 plane but is a tadeoff fo inceased degees of feedom in the oientation of the nodes. Besides, the inceased beamwidth will lend geate fault toleance to the system by poviding geate edundancy in the aeas eached by multiple tansmitting antennas. It will also make the antenna design easie since high diectionality, i.e. naow beam width is not needed. 3.4 Aligned Antennas with Two Anchos The two location detemination methods descibed ealie ely on the diffeence in powe eceived at two antennas of a senso fom the antennas on the same ancho node. The eo in the powe eceived can become coelated due to the poximity of the two antennas, even if they ae pointed in sepaate diections. In a eal life scenaio the coelation can significantly educe the accuacy of the location estimate, especially fo a vey small senso node. To investigate the pefomance of the location detemination with inceasingly uncoelated channels, two tansmitted signals wee sent fom two motes substantially emoved fom each othe. This scheme is also useful in situations in which moe than one diectional antenna would not fit on a single mote. The aangement is shown in Figue 5. Ө 7 Ө 5,' d Gt ( + Θ = π π d + Θ3 + Θ4 ) * G ( + Θ4 ) (5) 5

6 Θ Θ 4 Θ 3 Θ Θ Ancho Senso 3 Θ 3 Θ Θ 5 Ancho Figue 5. Location detemination using measuements fom two anchos Since the location of the two anchos is known, the paametes 3 and Θ 3 can be detemined. Using geometic popeties of the system we get the following elations between the vaious angles π π Θ 5 = Θ Θ + Θ3 + Θ4 = The equations fo the eceived powe ae given by t t = G ( ) G ( ) (7) G ( ) G ( ) (8), t, t Θ Θ = Θ Θ Using the law of sines along with elations between the angles deived ealie we get two moe equations 3 = (9) cos( Θ +Θ ) sin( Θ +Θ ) 3 3 = Θ Θ Θ +Θ 3 cos( ) sin( ) () This gives us fou equations in fou unknowns,, Θ, and Θ, which can be numeically solved. Thus, the distance and the angle with espect to each of the two anchos ae detemined. The senso node s location can be estimated using eithe distance, angle pai and the final location estimated using aveaging of each estimate. In this section, we have povided the mathematical solution to the poblem of location estimation with diectional antennas in thee diffeent scenaios. The node specifications and the deployment conditions will detemine which scenaio is applicable. 4 Expeiments & Results 4. Expeimental Setup Cossbow MICA motes MR4CB opeating at 9 MHz and unning TinyOS as the pogamming envionment ae employed as the senso nodes fo ou testbed. Two kinds of omni-diectional antennas ae used - quate wave whip antennas (MMA4CA) (on the tansmitting ancho node, comes off-the-shelf with the motes) and quate wavelength monopole antennas with an expanded gound plane (on the eceiving taget nodes, fabicated fo easy intefacing and co-existence with the patch antennas). Diectional antennas ae fabicated and used on both the taget senso node, whose location is to be detemined, and the ancho senso nodes, whose location is assumed to be known. atch antennas fabicated on duiod substates, Roges RO3 whose dielectic constant is., ae chosen as the semi-diectional antennas in this testbed due to thei simple fabication, small size, and low-cost. The othe candidate designs such as Yagi antennas, hon antennas, and antenna aays, would have moe diectionality but ae lage and moe expensive and theefoe not applicable. Senso motes with fou diectional antennas contolled by a switching netwok accoding to the eceived signal stength indicato (RSSI) at the antennas ae employed to be the taget motes (functioning as eceives). The switching netwok is implemented by a GaAs MMIC S4T switch. Softwae executing on the taget motes monitos the eceived signal powe on the fou antennas and selects the equisite ones (the best, o the two best) fo its location computation. The motes ae tested in an outdoo envionment to obseve the pefomance of the location detemination system with diffeent kinds of wieless fading. The goal of the expeiments is to detemine the estimation eo of the location detemination system. Thee expeiments ae set up with tansmitting ancho motes that initially have omni-diectional (dipole) antennas (see Figue 7(a)). If necessay, aligned patch antennas ae mounted on the ancho motes (see Figue 7(b) and (c)) to have a complete diectional tansmitting and eceiving system. The taget motes, functioning as eceives, ae equipped with the switched diectional antennas and collect the data which is fowaded to a laptop though a Univesal Asynchonous Receive/Tansmitte (UART) Inteface. Matlab pogams ae witten to solve the equations given in Section 3 numeically. The softwae utilizes a gain patten deived fom the patch antenna fom Ansoft s High Fequency Simulation Softwae (HFSS) package. HFSS, a full wave electomagnetic simulation commecial softwae, is the default industy standad RF design simulation tool [4][5]. The adiation patten is simulated and compaed with the measuement of the fabicated patch (Figue 6). It is obseved that the two ae in close ageement and hence the HFSS model is used fo futhe analysis. The expeiments in Section 4. have 5 samples fo each position, taken in thee sepaated time intevals. These ae aveaged fo the epoted numbe. The lagest 95% confidence inteval fo these expeiments was % elative eo. Matlab simulation is used to iteatively solve the equations to minimize the eo in the adial distance and the angle of eception. If the iteative Matlab simulation is consideed too expensive to execute on the motes, a static lookup table of signal stength vesus adial distance and angle of eception can be ceated and uploaded into the motes. This cuts down on the latency of the location estimation as well as the 6

7 computational expense on the simple pocessos of the senso nodes, at the expense of the accuacy of the solution. This appoach will be futhe investigated in the futue. diectional antennas, the taget node selects the two stongest signals on its eceiving antennas to execute the location estimation algoithm Taget mote node? Taget node? ?? HFSS si mul ation 6 7 Measu ement 8 (a) Ancho node (b) Ancho node Figue 6. Radiation patten of the patch antenna fom HFSS and the measuement in anechoic chambe oom The location eo is defined by the eo distance fom the estimated taget position to the actual taget ˆtaget position taget, divided by the known actual distance between the ancho mote and the taget mote ( R= ancho, taget, ) whee is a two-dimensional position vecto. Location eo R eo ˆ ˆ taget taget, taget taget, = = R = ancho, taget, ( xˆ x ) + ( yˆ y ) taget taget, taget taget, ˆ ancho taget, + ancho taget, ( xˆ x ) ( y y ) 4. Expeimental Results () Expeiment : Aligned case: Single omni-diectional antenna ancho, dual patch antenna taget In Expeiment, an ancho node is used with an omnidiectional o dipole tansmit antenna. The taget node has a dipole antenna and two diectional patch antennas. Test is the case with a dipole antenna on the taget node. Test has the diectional antennas on the taget node. The taget node is placed in the cente of a cicle of adius d = 8 feet and d = 4 feet. Received powe measuements ae made with the ancho node at diffeent points on the peiphey of the cicle. Calculations wee computed using equations () and () with G independent of Θ. The elative distance eo is shown in Figue 8 fo both tests. The x-axis is the angle of the ancho node with espect to the noth-south axis dawn fom the taget node, with the ancho node being moved on the cicumfeence of a cicle with the taget node at the cente. When using the Tx mote Taget node Θ Θ Ancho node atch_n atch_w Taget node atch_e atch_s ( c) Figue 7. Antenna configuations fo (a) Expeiment (b) Expeiment (c) Expeiment 3 When using the omni-diectional antenna, the position cannot be detemined since at least thee ancho nodes would be needed and hence, the distance estimate is used as the basis fo compaison. In this case the Fiis fomula is used with an expeimentally detemined paamete fo the exponent in the powe loss, R N, N=.89 in the measuement envionment. This paamete was detemined using the omni-diectional antenna and then genealized fo the expeiment setting and applied in all calculations. It is easonable that the value of N would be constant ove a section of the senso field and does not have to be calibated fo each souce-destination pai. This is classified as a line-of-sight and elatively multipath fee envionment and is applied to estimate ˆ R = ˆTX, RX,. The distance eo in the entiely omni-diectional antenna system is in the ange of -5% with a mean of 34%, while this eo is educed to -9% with a mean of 3% in the case of the diectional antennas. This esult can be explained by the fact that in omnidiectional antennas, thee ae moe multi-path effects leading to moe intefeence and moe fluctuation in eceived signal stength. Thus, even if tiangulation is used, diectionality of the antennas is useful since it leads to moe accuate estimates of individual distances. 7

8 Howeve, by using the angula infomation that is gained fom multiple eceiving antennas, the location of the taget node can be detemined with a single mote without the use of tiangulation (see Section 3.). The location eo of test in Expeiment is shown in Figue 9. The eo is lage than the adius eo because the angle estimation eo also contibutes to the location eo. On aveage, the location eos ae (a) 43.5% and (b) 55.75%. This motivates the need to use dissemination of location infomation though multi-hop communication fo lage distances. This eo can be futhe mitigated by using othe estimation anchos, as Expeiment 3 shows. E o (p e c e nt) Eo pecentage Distance Eo: dipole vs. patch antenna dipole patch Angle (degee) Figue 8. Relative eo in distance estimation fo Expeiment Location estimation eo Angle (degee) E o (p e c e n t) Location estimation eo Angle (degee) (a) (b) Figue 9. Location estimation eo fo Expeiment, Test fo (a) 8 feet, (b) 4 feet Expeiment : Single patch antenna ancho with dual patch antenna taget In this expeiment diectional patch antennas ae employed in the tansmission motes (see Figue 7(b)). The eceiving antennas ae all diectional. The analytical model is as shown in Section 3.. The eos in measued distance and angle ae shown in Figue and the maximum eos in the two metics ae seen to occu at diffeent positions. The location estimation eo calculated fom actual measuements is shown in Figue. These calculations wee made using equations () and (). The location eo gows lage as Θ inceases because the tansmission antenna is not oiented towads the taget node. Compaed to the expeiment with omni-diectional tansmitting antenna, the location eo is educed when the angle is smalle than 45 degees because the adiation patten of diectional tansmission antenna ovelaps with the eceiving antennas. Afte this coss-ove point, the patch #W on the ancho node which is oiented to the west can be used as the tansmission antenna. Hence, the location eo can be contolled to be less than 4% in a single ancho. The aveage eo is 9.8%. Estimated distance (mete) Distance estimation Angle (degee) Estimated angle (degee) Angle estimation theta theta Angle (degee) Figue.Distance and angle estimation eo fo Expeiment. [The solid line is the estimation and the dashed line is actual measuement.] Eo (pecent) Location estimation eo Cossove point to anothe antenna pai patch #S patch #N Angle (degee) Figue. Location estimation eo fo Expeiment Expeiment 3: Two single patch antenna anchos and single patch antenna taget Two anchos can be used to povide additional data fo location estimation, as shown in the analysis of Section 3.4. The configuation of Expeiment 3 is shown in Figue 7(c). In this expeiment only the two patches eceiving the stongest signals ae used while the othes ae tuned off. Calculations fo this expeiment wee made using equations (8), (9), and (). The outdoo measuement esult is shown in Figue. Fo this expeiment, measuements ae taken 8

9 individually fom each diectional ancho node and not at the same time. This does not affect the validity of the esults since the stochastic model fo the channel is time invaiant. When an additional ancho povides anothe powe measuement, the estimation eo educes fom 9.8% in Expeiment to an aveage of 7.78% in Expeiment 3. The fist cause of this impovement is the incease in the sepaation of the two tansmission patch antennas in Expeiment 3 compaed to the elatively small distance of the two patches in Expeiment. This means that the eo in the adial distance R and the eception angle Θ ae now uncoelated. Second, the fading channels ae moe uncoelated in Expeiment 3 because the distance between the two anchos is much geate than the wavelength of the RF wave. Thus, we conclude that significant location estimation impovement can be obtained by using multiple anchos with diectional antennas. Figue. Location estimation eo fo two ancho nodes with diectional patch antennas We pefom statistical analysis on the aw measuements to quantify the eos using infomation fom thee ancho nodes. Fo test of expeiment (pue omni-diectional), the eo is 7.5%, fo expeiment %, and fo expeiment 3.6%. The impovement is moe stiking with moe neighbo anchos, as the simulation esults show. 4.3 Simulation Results We ae inteested in evaluating the accuacy of the location detemination potocols with vaying numbe of ancho nodes. The accuacy is measued fo between and 3 anchos of diffeent kinds: omni-diectional, diectional, unaligned, and the two ancho case. This ange of expeiment would be beyond the hadwae esouces of ou testbed and hence a simulation methodology is used. These simulations ae only intended to highlight the behavio of the schemes with inceasing node densities. The distotion in eceived powe is simulated using a Rician fading channel. Each data point is based on the aveage of 5 diffeent samples. Fo each sample, the ancho nodes ae placed andomly within a x mete gid. The tansmitted powe is - db, equivalent to that used in hadwae, and the size of the senso node is set at cm. Figue 3 is shown with espect to the inceasing numbe of neighbos. Fom top, the diffeent cuves coespond to: single ancho unaligned, thee omnidiectional anchos and omni-diectional taget node, single ancho aligned antenna with least squae eo aggegation, single ancho aligned antenna with aveaging fo aggegation, two anchos with aligned antenna. Except fo the omni-diectional case, all the othes have patch antennas on the taget node. The fist thee fom one goup and the last two fom anothe. The eos in the fist goup ae noticeably highe than those in the second goup. The estimates in both goups show a hoizontal tend beyond 4 ancho neighbos indicating that highe density is not equied fo location estimation puposes. The genealized oientation, howeve, equies up to neighboing ancho befoe stabilizing. The highest eo is obseved fo the genealized oientation, indicating the impotance of appoximate alignment at the least. The next highest eo is fo the omni-diectional ancho antenna, followed by the single ancho with least squae eo aggegation. This indicates the value of diectional antennas and aveaging as the method fo aggegation. The two ancho case with aligned antenna and using aveaging as the aggegation technique outpefoms the single ancho case at lowe densities but is statistically equivalent at highe node densities. The eo bas coesponding to 95% confidence inteval ae shown. We obseve that aveaging as the method fo aggegating gives much bette esults than least squae estimation. One of the pimay easons fo this phenomenon is that aveaging cancels the eos in estimates in individual measuements while the effect of eo in least squaes estimation is additive. ecentage Eo osition Estimation Eo Two Anchos One Ancho LSE One Ancho Unaligned Numbe of Neighbos One Ancho Aveaging OmniDiectional Antenna Figue 3. Evaluation of estimation eo fo vaying numbe of neighboing anchos The esults shown in Figue 3 have slightly lowe eo ates than those found in hadwae. This is most likely due to a limited ability to account fo all multipath popagation, fading, and othe envionmental affects. Moeove, the aligned cases wee pefectly aligned in the simulation, 9

10 while the expeiments had elatively aligned antennas, but within a magin of eo that is vey difficult to detemine. 5. Conclusions In this pape we have pesented vaious techniques fo location detemination in ad-hoc netwoks using diectional antennas. The combination of the schemes is designed to meet the vaiety of equiements and degees of feedom fo eal life applications. The vaious tadeoffs between these schemes, such as feedom of oientation vesus beam width and node size vesus numbe of ancho nodes, have also been evaluated. The solution appoach can fom the foundation fo location detemination potocols fo the inceasingly popula diectional antennas. Ou esults, shown though expeiments on actual senso motes and simulations, bing out the fact that location estimation with omni-diectional antennas equies at least thee anchos and is less accuate than with diectional antennas. Also, using uncoelated communication channels though two geogaphically sepaated ancho nodes poduces bette esults than multiple antennas on the same ancho node. The eo is educed fom 7.5% in an omni-diectional system to.6% with two diectional ancho nodes. In futue, we intend to expeimentally detemine the extent of coelation in channel noise at the diffeent antennas of a senso. This will be applied to a sensitivity analysis of ou schemes. We also popose to analyze location estimates deived fom measuements fom multiple tansmitting antennas to multiple eceiving antennas. Finally, we will seek to solve the genealized antenna case with multiple ancho nodes. Refeences [] Y.-B. Ko and N. H. Vaidya, Location-Aided Routing (LAR) in Mobile Ad Hoc Netwoks, ACM/BaltzeWieless Netwoks (WINET) J., vol. 6, no. 4,, pp [] Jög Widme, Matin Mauve, Hannes Hatenstein, Holge Füßle, osition-based Routing in Ad-Hoc Wieless Netwoks, In Mohammad Ilyas (ed.): The Handbook of Ad Hoc Wieless Netwoks, CRC ess, Boca Raton, FL, U.S.A.,. [3] hilo Juang, Hidekazu Oki, Yong Wang, Magaet Matonosi, Li-Shiuan eh, and Daniel Rubenstein, Enegy-efficient computing fo wildlife tacking: design tadeoffs and ealy expeiences with ZebaNet, ASLOS, pages [4] J. M. Kahn, R. H. Katz and K. S. J. iste, Mobile Netwoking fo Smat Dust, in ACM MOBICOM 99, Seattle, WA, August 7-9, 999. [5] Gamin s etex, at [6] N. B. iyantha, A. Chakaboty, and H. Balakishnan, The cicket location-suppot system, In Mobile Computing and Netwoking, pages 3-43,. [7] L.-C. Kuo, 3-D FDTD Design Analysis of A.4 GHz olaization-divesity inted Dipole-Antenna with Integated Balun and olaization-switching Cicuit fo WLAN and Wieless Communication Application, IEEE Tans. Micowave Theoy and Techniques, vol.5, no., pp ,feb 3. [8] J. eiss et al., olaization divese antenna fo potable communication devices, U.S. atent ,. [9] M. A. Jensen, and Y. Rahmat-Samii, efomance Analysis of Antennas fo Handheld Tansceives Using FDTD, IEEE Tans. Antennas and opagation, 4, 994, pp [] Jeffey Hightowe and Gaetano Boiello, Location sensing techniques, Technical Repot of the Univesity of Washington CS Depatment, UW-CSE--7-, July. [] J. Hightowe and G. Boiello, Location systems fo ubiquitous computing, IEEE Compute, pages 57-66, August. [] Andy Hate, Andy Hoppe, ete Steggles, Any Wad, and aul Webste, The anatomy of a context-awae application, in ACM MOBICOM 99, pages 59-68, Seattle, WA, August 999. [3] N. Bulusu, J. Heidemann, and D. Estin, GS-less low cost outdoo localization fo vey small devices, IEEE esonal Communications Magazine, pages 8-34, Octobe. [4] J. Hightowe, R. Towe, and G. Boiello, SpotON: An indoo 3d location sensing technology based on RF signal stength, Technical Repot of the Univesity of Washington, CS Depatment, Febuay. [5] A. D. Wood and J. A. Stankovic, Denial of sevice in senso netwoks, In IEEE Compute, 35():54 6, Octobe. [6] Jan Beutel, Geolocation in a icoradio Envionment, M.S. Thesis, ETH Zuich, Decembe, 999. [7] Chis Savaese, Jan Rabaey, Koen Langendoen, Robust ositioning Algoithms fo Distibuted Ad-Hoc Wieless Senso Netwoks, USENIX Technical Annual Confeence, Monteey, CA, June. [8] S. Cabuk, N. Malhota, L. Lin, S. Bagchi, and N. Shoff, Analysis and evaluation of topological and application chaacteistics of uneliable mobile wieless ad-hoc netwok, in th acific Rim Dependable Computing Confeence, (RDC 4), Mach 4. [9] N. Sundaam and. Ramanathan, "Connectivity based location estimation scheme fo wieless ad hoc netwoks," oceedings of Globecom vol., pp.43-47, Nov.. [] L. Dohety, L. El Ghaoui, K. S. J. iste, Convex osition Estimation in Wieless Senso Netwoks, Infocom, Anchoage, AK, Apil. [] Kay Röme, The Lighthouse Location System fo Smat Dust, In MobiSys 3. [].J.B. Claicoats, Y. Rahmatt-Samii, J.R. Wait, Handbook of Micostip Antenna, Volume, IEEE Electomagnetic Waves Seies 8, 989. [3] M. Vossiek, L. Wiebking,. Gulden, J. Wieghadt, C. Hoffmann,. Heide, Wieless local positioning, IEEE Micowave Magazine, pages 77-86, Dec. 3. [4] HFSS simulation tool. Available at: [5] L. Boccia, G. Amendola, G. Massa, ope modeling using an FEM electomagnetic simulato leads to the design of a low-cost, lightweight GS patch antenna capable of excellent multipath ejection, Micowave and RF, Jan. 3. [6] Cossbow Technology Inc., MICA Mote: [7] TinyOS Community Foum: [8] D. Niculescu, B. Nath, Ad Hoc ositioning System (AS) Using AOA, IEEE Infocom 3, pages [9] L. Fang,. J. Antsaklis, et. al, Design of a wieless dead eckoning pedestian navigation system The NavMote expeience, Technical Repot, ISIS Lab, Univesity of Note Dame. [Online]. [3] DMC-SX Digital Magnetic Compass Opeato Manual, Leica Vectonix AG, Switzeland. [3] K. Chintalapudi, A. Dhaiwal, R. Govindan, G. Sukhatme, Ad- Hoc Localization Using Ranging and Sectoing, IEEE Infocom 4, pages