Experimenting an Indoor Bluetooth-based Positioning Service

Transcription

1 Expermentng an Indoor Bluetooth-based Postonng Servce G. Anastas, Computer Engneerng Dept., Unversty of Psa, Italy R. Bandellon, Computer Scence Dept. Unversty of Trento, Italy M. Cont, F. Delmastro, E. Gregor, and G. Manetto, IIT-Natonal Research Councl (CNR), Italy Abstract The Bluetooth wreless technology s an emergng technology orgnally desgned as a short-range connectvty soluton for personal, portable, and handheld electronc devces. Ths paper brefly presents the functonalty and the archtecture of an ndoor postonng servce based on ths technology. Most of the desgn choces for the servce have been strongly nfluenced by Bluetooth features. The effectveness of the ndoor postonng servce s crtcally analyzed. Expermental and smulaton results used for defnng the polcy of moble devce dscovery are shown. 1. Introducton In current dstrbuted systems, the noton of moblty s emergng n several forms and used n many applcatons. One of these forms s user moblty that arses naturally n wreless moble computng [1]. Here, when a moble user moves, hs pont of attachment to the fxed wred network changes. Current applcatons of wreless moble computng cover dverse applcaton areas, on dfferent spatal scales, rangng from the very small dmenson of sensor networks up to the worldwde sze of satelltebased networks [2]. In the scale represented by corporate and academc offce spaces, buldngs, and campuses, the ncreasng presence of Bluetooth wreless technology 1 [3] [4] s reflected n the wde avalablty of ths technology on a large number of Personal Dgtal Assstants and Portable PCs. In ths context, the moble users utlze the Bluetooth technology of ther portable devces n offce applcatons that are related to group work and to ndvdual access to corporate servces. Access to the ntranet nfrastructure and servces of a corporaton are granted by placng several Bluetooth access ponts nsde the buldng. Ths work has been partly funded by MIUR n the framework of the VICOM Project. 1 Bluetooth s a trademark owned by the Bluetooth SIG, Inc., USA. In ths paper, we present an experment that we performed usng Bluetooth and an Ethernet LAN as the enablng technologes of an ndoor postonng servce, named BIPS, coverng a buldng. BIPS offers a servce that allows a moble user to vsualze on hs portable devce the shortest path he has to follow n order to reach another moble user nsde the same buldng. The typcal buldng could be an academc department and the moble users can be students, vstors, professors, or staff members. The core of BIPS s devoted to trackng the moble users standng or walkng at a low speed nsde a buldng. To ths am, BIPS manages locaton nformaton about movng users. BIPS consders each room of the buldng as a granule of locaton nformaton. We defne a sngle room as a space that can ft nto a crcle of 10 meters radus snce ths s the dmenson of the greatest coverage area of a Bluetooth pcocell. BIPS has a staton archtecture consstng of a set of Bluetooth cells, one for every sgnfcant room of the buldng, nterconnected va an Ethernet LAN wth a central server machne. The locaton database of BIPS operates on the central server. 2. An overvew of BIPS BIPS s an ndoor postonng servce desgned for trackng moble users who may be n moton nsde a corporate buldng. Every moble BIPS user s represented by a handheld devce equpped wth a Bluetooth nterface for nteractng wth the statc part of the system. The statc part of BIPS s a locally dstrbuted system consstng of a centralzed server machne and a set of workstatons nterconnected va an Ethernet LAN. Every workstaton s provded wth a Bluetooth nterface that allows the statc part of BIPS to communcate wth handheld devces. The man task of every BIPS workstaton s dscoverng those moble users who enter the coverage area of the workstaton's Bluetooth nterface. Once a handheld devce has been dscovered, ts poston s communcated to the central server machne where the poston s stored n a database for the sake of

2 successve lookups. The network archtecture of BIPS s llustrated n Fgure 1 (accordng to Bluetooth termnology, a pcocell s called a pconet). An off-lne procedure has been mplemented for regsterng new BIPS users. The procedure assocates the name of a user wth a user dentfer (userd). In ths phase, a password and a set of access rghts are defned for enforcng securty and prvacy ssues. When a regstered user wth a moble devce enters a pconet of the buldng for the frst tme, the frst acton he should perform s loggng nto the BIPS system. The most mportant effect of ths operaton s the defnton of a one-to-one correspondence between a userd and the Bluetooth devce address (BD_ADDR). From ths moment untl the moble user log out of the system, BIPS wll constantly track all the movements of the moble devce nsde the buldng. Fgure 1. BIPS network archtecture Before generatng the query, BIPS verfes that the target moble user s logged n and that the queryng user has the rght to formulate ths queston. The formulated query s a very smple spato-temporal query that takes the followng form: "Select the target actual pconet of the moble devce BD_ADDR1 where BD_ADDR1 s assocated wth a userd1 and where userd1 s assocated wth the gven user name". Snce the poston of the moble devce changes n space and tme, the strategy of mantanng accurate poston nformaton s the most mportant techncal ssue of BIPS. As usual for ths knd of system, BIPS should provde a suffcent level of accuracy for poston nformaton, whch s a good tradeoff between the computatonal load and the precson n representng poston nformaton. BIPS does not mpose very strngent requrements on poston nformaton mantenance snce the moble users move at a maxmum speed of 2 meters per second through crcles wth a radus of 10 meter. In BIPS, every workstaton has the task of computng the presence of those moble devces postoned nsde the pconet. These presences are revealed at fxed ntervals of tme. In order to reduce the computatonal and communcaton burden of the system, a workstaton updates the central locaton database only when t reveals a new presence or a new absence n ts pconet. Once the current pconet of the target moble user has been dentfed, BIPS needs to determne the shortest path. To do ths, BIPS defnes a weghted undrected connected graph that reflects the topology of workstatons nsde the buldng. There s a node n the graph for every BIPS workstaton. An edge between two adjacent nodes s defned when there s a physcal path n the buldng that connects the rooms where the two correspondng workstatons are placed. Every edge s labelled wth a weght, a postve nteger that represents the dstance between the two workstatons. Thus the problem that BIPS has to solve s the classcal shortestpath problem,.e. to fnd the path n a weghted undrected connected graph connectng two gven vertces wth the property that the sum of the weghts of all edges s mnmzed over all such paths. BIPS mplements the Djkstra algorthm for solvng ths problem [5]. We underlne that the statc nature of BIPS wred network allows us to compute off-lne all the shortest paths that connect all the possble pars of two nodes. Hence the computaton of the shortest path has no mpact on BIPS onlne actvtes: gven two nodes, the answer s mmedate snce t has been prevously computed. 3. Programmng the Bluetooth technology for BIPS The basc concept n Bluetooth communcatons s the pconet, a collecton of Bluetooth devces that can communcate wth each other by sharng a common channel. A pconet s a star-shaped wreless network n whch one devce plays the role of the master and every other devce plays the role of the slave. Master and slaves of a pconet are synchronzed accordng to a channelhoppng sequence that s a functon of the master's BD_ADDR and clock. Bluetooth s a Frequency Hoppng System where the clock cycle lasts for s. Thus the clock rate s 3.2KHz. In BIPS, the role of master s always played from a fxed workstaton that s also connected to the wred network, and the role of slave s always played from the portable devces. Two sequental phases are needed for establshng a communcaton channel. The frst ntal phase s called nqury and t corresponds to devce dscovery from the master. The second phase s called pagng and t corresponds to ntal connecton setup. In both phases the master and the slaves perform dfferent actons, we wll brefly summarze them n the followng.

3 3.1. Inqury phase A master begns devce dscovery by enterng nto the nqury state and broadcastng messages n the followng way. The nqury hoppng sequence s splt nto two 16- hop parts, called tran A and B. A sngle slot lasts for 625 s. On every even slot, the master sends two ID packets swtchng between two frequences of the same tran every s. On odd numbered slots, the master lstens for slave answers. Snce a sngle tran wll last for 16*312.5 s and the slots for sendng and lstenng are nterleaved, two 10ms trans are defned. Each tran must be repeated for at least N nqury = 256 tmes before a new tran s used. In order to collect all responses n an errorfree envronment, at least three tran swtches must take place, so the nqury state may have to last for 10.24s. A slave that wants to be dscovered enters n nqury scan state. In ths state, t lstens for ID packets on the same 32 dedcated frequences used from the master. The slave changes lstenng frequences every 1.28s. The T w_nqury_scan tme, durng whch the slave lstens on the same frequency, must last enough to completely scan one tran. T nqury_scan s the tme between the begnnng of two consecutve nqury scan cycles and t wll last at most 2.56s. Both ntervals can be changed. Default values are: T nqury_scan = 1.28s T w_nqury_scan = 11.25ms 3.2. Page and connecton phases Known the dentty of the devces n ts proxmty, the master of a pconet explctly pages them to jon ts pconet. Frst the master selects the devce to page and then t sends page messages on every frequency belongng to the two trans. Also n ths phase two ntervals are defned to establsh the amount of tme n whch the slave lstens for page messages. Default values for page scan parameters are equal to nqury scan default values: T page_scan = 1.28s T w_page_scan = 11.25ms After several nteractons, the master and the slave enter nto a substate n whch the clock nput to the hop selecton mechansm s frozen on the same value, and they can communcate usng the same hoppng sequence. At the end of the page, the master enters nto the connecton state and asks to open a connecton. If the slave agrees, t acknowledges the request, the connecton s fnally establshed and the two devces can begn to exchange data on ths connecton The Slave-Master trackng problem In BIPS a slave and a master must be programmed n such a way that they can dscover when and how the context n whch they operate s changed. Ths means that a slave must be able to open a connecton wth a new master when t enters nto a new pconet and that a master must dscover a change n the composton of ts pconet. The amount of tme and resources that slaves and masters dedcate to the trackng actvty should be mnmal n order to have a small mpact on the other computatonal actvtes of the system. 4. Intal measures and expected results In ths secton, we provde some ntal expermental and smulatve results. The results concern the tme and related probablty of devce dscovery. The knowledge of these results s mportant for defnng the schedulng polcy of a master, whch must dedcate a certan percentage of ts workng tme to devce dscovery and the remanng tme to serve slave applcatons. 4.1 The experments The whole system was developed on Lnux platforms. The master role s played by a workstaton equpped wth a Pc-Pcmca adapter from Texas Instruments and a 3COM Bluetooth card. The slave role s played by one laptop equpped wth the same card. Lnux operatng system contans the Offcal Bluetooth Protocol Stack (BlueZ), necessary to program card drvers. In all the experments we have used the ftme routne of the Lnux operatng system for measurng the tme spent by the master machne n dscoverng a new moble devce. In all these experments the master s completely dedcated to executng the dscovery procedure,.e. t s always n the nqury phase. Ths stuaton s the most favourable polcy of devce dscovery. It should be consdered an upper bound that allows a frst verfcaton of the feasblty of trackng moble users wth BIPS. In the varous experments we have only changed the workng hypothess of the slave. The tme that we measured n ths frst set of experments s an nterval that begns just before the master enters nto the nqury state and that ends when the master receves the answer from the slave to the nqury message. Due to space lmtatons, we only report the results of the most sgnfcant experment. In ths experment the slave alternates the perods of nqury scan and page scan. In both cases the slave lstens for nqury/page messages for an nterval of 11.25ms, as long as the default values specfed n the Bluetooth standard. The table llustrates the outcome of the experment. Startng Tran Case No. T average Same s Dfferent s Mxed s

4 The table shows the average dscovery tme that has been measured on 500 nqury trals. The frst column of the table shows the stuaton when the startng frequency trans used by master and slave are the same or not. The last row ndcates the average result of all trals. It s worth notng that the probablty that the master and the slave start on the same tran s close to 50%. We note that the average dscovery tmes are close to the prevous results. Snce the slave cannot be completely dedcated to beng dscovered but t should also spend some tme exchangng data wth the master, the stuaton descrbed n the last experment represents the way n whch the slaves are programmed. The stuaton measured n the frst set of experments s not realstc. The stuaton that we would lke to measure s one n whch BIPS tracks several moble devces so that we could evaluate the real master s performance n devce dscovery. For defnng the master schedulng parameters before the actual mplementaton of BIPS, we used a smulaton envronment to obtan some results for the dscovery phase wth more than one slave. Ths s llustrated n the next subsecton. 4.2 Smulaton envronment and results To establsh a correct hypothess for master behavour n BIPS, we have developed a Bluetooth extenson to the VINT project network smulator ns2 [6]. Our extenson s based on BlueHoc, the ns2-based smulator released by IBM [7]. We have enrched BlueHoc wth a mechansm for handlng collsons that mght arse durng the establshment of a lnk. In the smulated scenaros, we consdered a sngle pconet n whch only one devce always assumes the role of master, whle the others assume the role of slaves. In partcular, slaves are always n nqury scan mode and they start lstenng on frequences of tran A. Snce dscovery tme depends on the number of slaves n the pconet, we performed our smulatons varyng ths number n the range [2,20]. In the unque smulaton that we report here for space lmtaton, the master alternates devce dscovery and connecton management and t transmts nqury messages usng only tran A. It s reasonable to hypothesze that the master s dedcated to devce dscovery for 20% of ts operatonal cycle. For ths reason we have fxed the length of the nqury phase to 1 second and the total perod to 5 seconds (Fgure 4). l b a b o r P y r e v o c s D tme (seconds) 2 slaves 4 slaves 6 slaves 8 slaves 10 slaves 15 slaves 20 slaves Fgure 2. Inqury and connecton management. We can note that, when the maxmum number of slaves n the pconet reaches 10, on the average the master succeeds n dscoverng about 90% of these slaves n the frst 1 second. Only n the second operatonal cycle t dscovers 100% of the slaves. When the number of slaves ncreases by 5 (.e., 15-20), usually the slaves are all dscovered n 2 cycles. (see Fgure 2). In a real scenaro, the startng trans for master and slaves cannot be defned by the programmer, so they have a 50% probablty of startng wth the same tran. For ths reason, we have to fx the length of the dscovery phase to more than 3.84s. Indeed, the master spends 2.56s on the frst tran, dscoverng all the slaves lstenng on the same tran. Then, as we have seen n the frst smulaton, 1.28s are enough to dscover 90% of the remanng slaves lstenng on the other tran (f they are less than 10). In summary, when there are 20 slaves n a master coverage area, f the master spends 3.84s on devce dscovery and the slaves are programmed for enrollng n the pconet, then the percentage of moble devces dscovered s about 95% (19 slaves). In fact 50% of the slaves belongng to the frst tran are completely dscovered, and 90% of the remanng 50% wll be dscovered wth the second tran. 5. Conclusons As shown n the prevous secton, f the master spends 3.84s on devce dscovery and the slaves are programmed to enroll n the pconet, then n the average case 95% of the slaves wll be dscovered. Ths represents a percentage that s satsfactory for BIPS. To establsh the percentage of workng tme that a master dedcates to devce dscovery, we have to estmate the average tme that an average moble user spends n crossng a pconet. Consderng that a moble user

5 normally walks wth a speed n the range [0, 1.5] meters per second and that the dameter of the coverage area s about 20m, we can estmate that the average walkng user wll spend 20m : 1.3m/s = 15.4s n the pconet. Ths tme represents the length of a complete operatonal cycle of a master. Hence, the master wll dedcate a contnuous slot of 3.84s for devce dscovery and the remanng 11.56s for servcng the slaves. The average load of trackng servce n BIPS s about 24% of the operatonal cycle. 6. References [1] T. Imelnsk and B.R. Badrnath, Wreless Moble Computng: Challenges n Data Management, Comm. ACM, vol. 37, no. 10, Oct. 1994, pp [2] R. Mallad and D.P. Agrawal, Current and Future Applcatons of Moble and Wreless Networks, Comm. ACM, vol. 45, no. 10, Oct. 2002, pp [3] C. Bsdkan, An Overvew of the Bluetooth Wreless Technology, IEEE Comm. Magazne, vol. 45, no. 10, Dec. 2001, pp [4] Bluetooth web ste: [5] E. W. Djkstra, "A note of two problems n connexon wth graphs", Numersche Mathematk, 1 (1959), pp [6] The VINT Project, The ns Manual, [7] IBM, BlueHoc: Bluetooth Ad Hoc Network smulator, Verson 1.0, developerworks/projects/bluehoc, June 2001.