Accelerating Bluetooth Inquiry for Personal Area Networks

Transcription

1 Accelerating Bluetooth Inquiry for Peronal Area Networ Gergely V. Záruba 1, and Imrich Chlamtac 2 1 Center for Reearch in Wirele Mobility and Networing (CReWMaN) Department of Computer Science and Engineering, The Uverity of Texa at Arlington Arlington, TX, U.S.A. zaruba@uta.edu 2 Center for Advanced Telecommucation Sytem and Service (CATSS) Department of Computer Science, The Uverity of Texa at Dalla Richardon, TX, U.S.A. chlamtac@utdalla.edu Abtract - The recent indutry tandard Bluetooth promie low cot replacement of commucation cabling with moderate ymbol-rate, hort-range wirele lin. The ame pecification alo addree the etablihment of point-to-multipoint piconet and the interconnection of everal of thee piconet into catternet, enabling Bluetooth to be ued a a technology for realizing peronal area networ. Etablihing Bluetooth piconet require node to dicover each other by completing an inquiry phae. Thi paper invetigate the inquiry phae; how the hortcoming of the current inquiry procedure in multi node PAN cenario, and outline and analye a bacward Bluetooth compliant modification to accelerate Bluetooth inquiry. Extenive imulation comparing the original and the propoed Bluetooth inquiry cheme how improvement of more than an order of magtude in device dicovery time. I. INTRODUCTION Wirele commucation gained igficant acceptance and importance in the lat decade of the previou century. The wirele revolution wa igted by 1 t and 2 nd generation cellular tandard, but at the end of the lat decade, dropping price of novel wirele technologie helped to create new wirele networing paradigm uch a wirele local area networing (WLAN) and hort range wirele peronal area networing (WPAN). It i generally predicted that thi and the next decade will bring an even higher wirele penetration mainly due to WLAN and WPAN. One of the emerging hort-range wirele networing technologie i the recent indutry tandard Bluetooth (BT) [1]. Bluetooth evolved from the need to replace wire in hort-range commucation, e.g., cable between mobile handet and their headet, or erial cable between computer and peripheral, with hort-range wirele lin. Derivable from the maximum tranmiion power and receiver enitivity of the pecification [1], a cla-3 (mot pervaive) BT ha a tranmiion range of approximately 10 meter in a free propagation environment at a nominal ISM band frequency of 2.4GHz. BT employ frequency hopping over 79 carrier frequencie 1, paced 1MHz away with a 1 Thi paper i retricted to the 79-hop ytem, yet the propoed approach can eaily be applied to the 23-hop ytem a well. pectral efficiency of 1bp/Hz. The main commucation tructure of Bluetooth called piconet i a point-to-multipoint tar topology, with a mater node in the center and lave node at the perimeter of the tar. Piconet can be interconnected to catternet by node taing on mater or lave role in more than one piconet concurrently. The above-decribed characteritic mae BT a viable candidate for etablihing inexpenive peronal area networ. Before BT device can exchange information among each other, they have to go through a three-phae lin etablihment procedure. In the firt phae, the device have to can and earch through a ubet of the hopping frequencie to determine whether there are other device in their tranmiion range. Thi firt phae erve a the neighbor dicovery proce, and i referred to a inquiry in BT terminology. In the ubequent tep, device that are already aware of each other proximity, itiate a handhae proce by which they exchange crucial information for piconet formation; thi phae i alo referred to a paging. The third phae deal with the etting up of a virtual channel for further control information exchange and negotiating commucation parameter relating to lin management iue. It i implied, by the above three-phae proce, that only becaue device may be in each other tranmiion-range they do not necearily have the mean to commucate with each other (unlie with Wi-Fi). Thi paper concentrate on the firt inquiry phae pointing out the limitation of the mandatory inquiry proce while propoing and decribing a techque to accelerate device dicovery. The ret of thi ection decribe the mandatory inquiry proce of BT in more detail, outline previou wor, and introduce our approach for accelerating the inquiry proce. A. The Bluetooth Inquiry Proce The purpoe of the BT inquiry proce i for device in each other tranmiion range to become aware of thi proximity, i.e., a ucceful inquiry handhae between two node reult in the itiating node acquiring nowledge on the reponding node identity and cloc offet. Such GLOBECOM /03/$ IEEE

2 handhae require ome of the node to be in an inquiry (itiating) tate while other node to be in an inquiry can (reponding) tate. In the inquiry tate, device tranmit very hort (68µ) ID pacet every 312.5µ. The hort duration and uque bit pattern of thee ID pacet doe not only enable a receiving node to efficiently correlate it receiver to the ID pacet but mae the diviion of regular 625µ lot into two 312.5µ half-lot poible. Conequently, in even numbered lot the inquiring node will end out two ID pacet at different frequencie, while in an odd numbered lot, the ame node will tune it receiver to the correponding repone carrier. The number of frequency carrier ued for inquiring i reduced from 79 to 32 inquiry and 32 inquiry repone carrier for increaed dicovery efficiency. Thee 32 carrier are divided into two 16-carrier train, i.e., A- and the B- train. According to the mandatory inquiry cheme a ingle train ha to be repeated for at leat 256 time before changing to the other train. In the inquiry can tate a node i liteng for at leat 2048 lot (1.28) on one of the 32 different inquiry-can frequencie waiting to overhear an ID pacet from an inquiring node. If an ID pacet i overheard, the node generate a uform random number b from [0,1023] and upend the inquiry proce for a duration of b lot to reduce the chance of colliding inquiry repone from different node liteng on the ame frequency. Once thi timeout ha expired, the device reenter the inquiry can tate and repond to the firt ID pacet it overhear. The repone pacet i a o-called FHS (frequency hopping equence) pacet containg the ID and cloc value of the reponding node. The inquiring node can now mae a note of the cloc and ID value of the reponding node and can either continue the inquiry proce or itiate the paging proce (or aume it original tate). It ha been hown [1] that proper inquiry and inquiry can tate holding time enure that two device dicover each other in le than 10 econd, yet that doe not imply any performance metric for cae where there are more device competing. In a PAN cenario, where all node are in each other tranmiion range, we can identify five problem lowing down the inquiry proce. The firt problem correpond to the fact that each time an ID pacet i overheard the node pend a random time drawn from a tatic interval in a bac-logged tate. The econd problem correpond to the infrequent change of inquiry train, which reduce the probability that an inquiring node i tranmitting on a frequency a canng node i liteng to. The le obviou problem include: i) the poible cae where by the time a node return to the can mode after bacing-off, the correponding inquiring node ha changed it tate and i not tranmitting ID pacet anymore; ii) when a node return from the bac-off to the inquiry can tate it will reply to the firt ID pacet it overhear, which may be tranmitted by a different device than earlier, and iii) ince the ID pacet i a uque bit pattern, it i impoible for device to detect who the originator of the tranmitted pacet wa, and thu it i poible that a canng node will reply to the ame inquiring node again and again. B. Previou Wor Previou reearch wor on Bluetooth ha mainly focued on ditributed catternet formation in PAN and ad hoc environment (e.g. [2,4] repectively), on cheduling policie of mater node in pico- and catternet, and on performance and interference meaurement in the preence of other piconet or interfering electro-magnetic force (e.g., Wi-Fi networ, or microwave). Some catternet formation approache aume [4], that device dicovery ha already been taen place, thu all node are aware of the identitie of all neighboring device, further motivating our wor. The author in [3] derived trategie baed on different tate holding probability ditribution for the cae when there i one inquiring and one canng node preent auming that inquiry train change after each train. At the time of thi writing, the author are not aware of any other wor publihed addreing BT device dicovery. C. Our Approach In the following ection we decribe a novel approach to reduce device dicovery time baed on meauring the number of idle lot during inquiry can and etimating the number of contending node by thi meaurement. The number of contending node i important metric for maing the random bacoff election adaptive, thu reducing the number of wated lot/time during inquiry. The ret of the paper i orgazed a follow: in the next ection we preent our accelerated ervice dicovery techque uing a top-down approach; in the ucceeding ection we will how uing imulation, what effect the propoed meaurement reolution ha on the etimation of bacoff and what performance improvement can be expected from our cheme. Finally we draw our concluion and identify future wor. An early, reearch in progre verion of thi paper ha appeared in [5] identifying the problem and outling a olution but not howing performance evaluation reult on the propoed acceleration techque. II. ACCELERATING INQUIRY Looing at the problem with the BT inquiry procedure, we imply that it inefficiency come mainly from the tatic maximum bacoff period B that i et to In order to accelerate the inquiry proce, B hould reflect the number of node actually contending for the attention of the inquiring node. In thi ection we are going to how how thi number can be made adaptive and how it hould be calculated to achieve lower device dicovery time. We will conider typical WPAN cenario where all node are in each other proximity; we will aume that no other interfering ource are preent. GLOBECOM /03/$ IEEE

3 A. Calculating the Bet Bacoff Value - B Let u aume that we already poe the nowledge on the number of node that are in inquiry can mode liteng on the ame frequency, and denote thi population by n f. All thee node are going to overhear an inquiring node tranmitting on the correponding carrier frequency, thu all of them are going to generate a uform random number from [0,B]. In order to mimize the colliion while reducing the number of wated replying opportutie, it can be hown [5] that B hould be et to n f. B. Number of Node Scanng on the Same Frequency In the acceleration cheme we will change the train equence after every train, thu inquiring not only on 16 but on all I =32 deignated inquiry frequencie in a row. Node in the inquiry can mode can liten on any of thee I frequencie, the exact frequency determined only by their native cloc. Thu the probability P i (), that there are exactly node liteng on the ame frequency, auming the nowledge on the number of node n in the inquiry can mode, can be derived from an (n,1/i ) parameter binomial ditribution: n n 1 1 P ( ) 1 i =, where I = 32 I I The mot liely value of an (n,p)-parameter binomial ditribution i (n+1)p. Thu, a good approximation for the number of liteng node and the maximum bacoff B i: n + 1 B = nf = I C. Determing the Number of Node in Inquiry Scan State Let u denote the number of node in the inquiry tate by n i, and aume that all device wor according to the ame inquiry trategy and aume that by nowing thi trategy a probability ditribution function P IQ () can be derived that correpond to the probability that =n /n i. Let u alo aume that P IQ () ha a fite expected value E(P IQ ())=Q. If node in the inquiry can mode can determine or etimate the number of node in the inquiry mode - n i, then they can etimate the number of node in the inquiry tate mode they are contending with by: n =Q*n i, thu: Q * + 1 B = I D. Determing the Number of Node in the Inquiry State Let u mae the aumption that node change from inquiry to inquiry can tate and vice vera le frequently than they change between train, i.e., a change in number of node in the inquiry tate while two conecutive inquiry train tae place i negligible (i.e., the number of node performing a tate change i mall in a t TR =625µ*32=20m period). Appropriate mean value for inquiry and inquiry can duration can validate thi aumption. If node that have jut tarted the inquiry can operation pend the firt 32 lot by liteng at the channel only, and auming that there are no colliion among inquiry tranmiion of inquiring node, then they could count the number of ID pacet received, which would determine n i. Yet, thi latet aumption i liely not to hold, thu a good etimation on n i allowing colliion hould be ought. To be able to reduce the complexity of thi problem, let u aume that all node are ynchrozed, i.e., there i no drift between cloc and cloc tic are ynchrozed. Later on we will how how thi aumption may be relaxed. If a node i liteng for t TR duration on a given frequency, then bit-zero of it cloc will change =64 time. Device in the inquiry mode will tart their tranmiion exactly on thee half-lot boundarie. The liteng node will be able to tell in how many of thee lot there wa radio energy preent on the channel, thu be able to determine how many of thee lot were idle lot (denoted by i ). The quetion i: can the number of inquiring node n i be etimated by nowing and meauring i? For the ynchrozed cae a relatively imple etimation can be given: let u revere the problem and aume nowledge on n i and thu attempting to determine i. The probability (P e ()) that a lot wa ued for exactly tranmiion can be determined by a (n i, 1/) parameter binomial ditribution: 1 Pe 1 ( ) = 1 Thu the probability that a randomly elected lot i empty i P e (0)=(1-1/). The probability P (i) that exactly i lot are idle can be modeled again by a binomial ditribution with parameter: (,P e (0)): 1 P ( i) = 1 i i* The expected value of thi function i the product of it parameter: E(P (i))=*(1-1/). Thu it i expected that if n i node have to contend for lot, there will be *(1-1/) lot empty, providing with a good etimation for i. Now let u olve the above equation for n i, thu calculating a good etimate on the number of inquiring node employing channel meaurement with ynchrozed cloc: 1 = ln( i / ) / ln1 The ynchrozation aumption can be relaxed by enhancing node to meaure the time t E in which energy i preent on the channel. Depending on the ophitication of the node hardware, the reolution r e of the meaurement of t E can be from a couple of microecond up to the duration of an ID pacet t ID =68µ. We argue, that a meaurement reolution of 1µ<r e <10µ can be eaily achieved without generating a major cot increae of Bluetooth chip. In thi cae node meaure energy on the channel during lot with r e reolution, i.e., in t TR /r e micro-lot. If the radio energy on the channel during a micro-lot doe not go above i GLOBECOM /03/$ IEEE

4 a given threhold, then i will be incremented. The election of the reolution time r e ha a igficant impact on the accuracy of n i etimation a we are going to ee in the next ection. The etimated value of n i with r e reolution of meaurement and m idle micro-lot can be given a: ln( m n i = * ln 1 r / t t ( e TR ) ID Now the optimal bacoff B can be determined by: ln( m Q * * + 1 ln( 1 r ) e tid B = I Leaving Q a the only variable to be evaluated. E. Proportion of Node in the Inquiry Scan State Obviouly, the inquiry policy will have a igficant impact on the proportion of node in the inquiry tate veru node in the inquiry can tate. Thu the value of Q will depend on thi policy. Here we will calculate Q for a probabilitic inquiry cheme relying on more or le uform ditribution of tate holding time. In [3] it ha been hown that determitic tate holding time will reult in device dicovery time with an infite expected value. Let u determine the value of Q for a imple tate holding time ditribution. It i neceary for the previou aumption to hold, that each node ha a lower bound on the number of lot it pend in each of the inquiry tate. Obviouly, the more time they pend in the individual tate at a time, the more liely that they dicover each other. On the other hand, ince the device dicovery procedure i aymmetric (i.e., node i dicovering node j, will not provide node j with nowledge on node i' identity), node have to change tate frequently. Additionally, the only way to avoid continuouly overlapping tranmiion of different inquiring node at the ame frequency (of which the lielihood i growing with the population) i to mae ure they eventually and independently witch tate. The determination of the bet tate holding time and ditribution i beyond the cope of thi paper. In our imple probabilitic model each node i required to pend 0 time lot in inquiry tate and then an additional 1 lot that i randomly choen from the interval [0, 2K]. Furthermore every node ha to aume the inquiry can tate for l 0 time lot and then an additional l 1 lot that i randomly choen from the interval [0, 2L]. Another implified way of decribing thi model i that after pending 0, and l 0 time lot in the appropriate tate, each node i changing to the other tate with a probability of K and L into the inquiry can/inquiry tate repectively. Thu in average a node pend 0 +K-1 lot in the inquiry and l 0 +L-1 lot in the inquiry can tate. Conequently the etimated value for the ratio of the tate holding time can be given a: Q=(l 0 +L-1)/( 0 +K-1). Therefore the optimal bacoff value can be etimated by: l0 + L 1 ln( m * * ( ) 0 K 1 ln 1 re tid B = [1] I Note, that if all node wor according to the decribed trategy, then all variable are nown in Equation 1 after the firt time a device pend t TR time in the inquiry can tate. III. SIMULATION RESULTS Simulation effort to the evaluation of the propoed cheme are two fold: i) determination of the accuracy of the prediction of the inquiry tate population of node with different lot meaurement reolution; ii) evaluation of the performance gain in the inquiry acceleration cheme compared to the original BT inquiry cheme. The firt imulation effort wa carried out uing Matlab cript, while for the econd we have implemented a C++ baed dicrete event imulation, modeling the inquiry behavior of BT device. For both imulation approache we have evaluated the reult tatitic on the fly, thu executing an adequate number of imulation intance to claim that the confidence interval of our imulation are 95-5, i.e., we are 95% ure, that our reult have le than 5% error. A. Accuracy of Prediction Intuitively, the higher the reolution of the meaurement i (i.e., the maller t r i), the better the etimation of the population of inquiring node (n i ) will be. To tudy the effect of the reolution on n i and to evaluate the etimation given by Equation 1, we have imulated N number of node acceing a channel randomly during t TR =20m duration. Figure 1 and 2 draw the real and etimated number of node in function of the empty micro-lot: m with the dotted line repreenting imulation, while olid line repreenting calculated reult (for more reult and dicuion conult [5]). Figure 1. Simulated and calculated population - 1µ reolution Figure 1 depict the cae where t r wa et to 1µ, a value igficantly le than the duration of an ID pacet t ID, the figure clearly how that Equation 1 provide a good GLOBECOM /03/$ IEEE

5 etimation in thi cae. Figure 2 how the reult for t r =10µ concluding that the etimated value doe not igficantly differ from the real population even for a meaurement reolution an order of magtude higher than before. Figure 1 and 2 imply that the meaurement method will limit the population to below two thouand node. Figure 2. Simulated and calculated population - 10µ reolution B. Improvement to the Original BT Inquiry Scheme Figure 3 and 4 how our reult of device dicovery time uing the original BT cheme (dotted line), our accelerated cheme with 1µ reolution (olid line) and a third curve repreenting a cenario where a train equence change i enforced after each train with the original BT inquiry. Figure 3 depict the average time between two conecutive non-redundant dicoverie in function of the population, while Figure 4 provide inight in the total time required for all peer to dicover each other. It can be oberved that more than an order of magtude of improvement in the device dicovery time can be achieved uing our cheme compared to the original BT cheme, and that the improvement i igficantly due to the adaptive bac-off not only to the immediate train change. IV. CONCLUSIONS AND FUTURE WORK In thi paper we preented an approach to reduce the device dicovery time in Bluetooth PAN, where all node are aumed to be in each other proximity. Our techque i baed on the obervation that in the original inquiry pecification, the bacoff value doe not reflect the number of contending node on the channel, but i et to a tatic high value. We have provided with a top-down analytical approach to mae the bacoff value adaptive requiring only a mall modification to the node: enabling them to meaure the RF channel. We have demontrated with imulation that a good approximation on the number of contending node can be achieved with appropriate meaurement reolution. Furthermore, we have hown by imulation that the accelerated inquiry approach outperform the original BT inquiry cheme igficantly in mean of dicovery time,. Future wor include i) conidering channel change of liteng node, where they have to eep trac of device they already have replied to (another ituation where meaurement can provide with reduced device dicovery time); ii) evaluation of the impact of ymmetric dicovery, where node completing an inquiry handhae in turn itiate a connection et-up, ha on device dicovery time. dotted curve: original cheme dahed curve: modified orig. cheme olid curve: our cheme dotted curve: original cheme dahed curve: modified orig. cheme olid curve: our cheme Figure 4. Total dicovery time for all peer. Figure 3. Average inter-dicovery time of device. REFERENCES [1] Bluetooth Special Interet Group, Specification of the Bluetooth Sytem, Core Verion 1.1, Specification Volume 1, February [2] T. Salodi, P. Bhagwat, L. Taiula, and R. LaMaire, Ditributed Topology Contruction of Bluetooth Peronal Area Networ, Proc. IEEE INFOCOM 2001, pp , Anchorage, AK, April [3] T. Salodi, P. Bhagwat, L. Taiula, and R. LaMaire, Proximity Awarene and Ad Hoc Networ Etablihment in Bluetooth, Intitute of Sytem Reearch, Uv. of Maryland Tech. Report, [4] G.V. Záruba, S. Baag, and I. Chlamtac, Bluetree-Scatternet Formation to Enable Bluetooth-baed Ad Hoc Networ, Proceeding of the IEEE International Conference on Commucation, ICC 2001, vol. 1, pp , Helini, Finland, June [5] G.V. Záruba, Neighbor Dicovery in Bluetooth Baed Peronal Area Networ, IASTED PDTPA 2002 Worhop, June GLOBECOM /03/$ IEEE