A Cluster-based TDMA System for Inter-Vehicle Communications *

Transcription

1 JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 30, (2014) A Cluster-bsed TDMA System for Inter-Vehicle Communictions * Deprtment of Electricl Engineering Ntionl Sun Yt-Sen University Kohsiung, 804 Tiwn E-mil: sheu@ee.nsysu.edu.tw This pper presents cluster-bsed TDMA (CBT) system for inter-vehicle communictions. In intr-cluster communictions, the proposed CBT uses simple trnsmitnd-listen scheme to fst elect VC (VANET Coordintor) nd it llows VN (VANET node) to rndomly choose time slot for bndwidth requests (BR) without limiting the number of VNs. In inter-cluster communictions, when two clusters re pproching, the CBT cn quickly resolve the collisions by re-llocting time slots in one of the clusters. To nlyze the performnce of the proposed CBT, we derive mthemticl equtions using probbility. The performnce metrics of our interests include the verge number of time slots for electing VC, the verge number of time slots required for BR, nd the totl number of time slots required before dt cn be successfully trnsmitted. The nlyticl results re finlly vlidted by simultion. Both the nlyticl nd simultion results show tht the proposed CBT spends less time to form smll-sized cluster thn IEEE p. Additionlly, when the number of joining VNs is incresed, CBT tkes less witing time before VN cn effectively trnsmit dt. Keywords: cluster-bsed, TDMA, MAC, inter-vehicle communictions, VANET 1. INTRODUCTION VANET (Vehiculr Ad-hoc Networks) inherited from MANET (Mobile Ad-hoc Networks) employs IEEE p s its MAC (Medium Access Control) lyer to support inter-vehiculr communictions. When there re not so mny vehiculr nodes (VN) in group, p cn resolve contention very quickly. However, s the nodes increses, collision nd unfirness my occur ccordingly. Different from the contention-bsed protocol s in , recently cluster-bsed TDMA ttrcts more ttention for VANET to improve the trnsmission efficiency s the number of VNs is incresed to certin lrge number. In cluster-bsed TDMA, cluster hed needs to be selected to serve s the network coordintor. The selected cluster hed is responsible for llocting time slots for dt exchnge mong its cluster members. Through creful scheduling of time slots for its members, collision cn be voided nd firness cn be chieved. Previous works on bsed VANET focused on two spects, trnsmission efficiency nd multi-chnnel techniques. In trnsmission efficiency, Wng et l. [1] proposed method to modify DCF. The size of contention window is reduced to hlf when the successfully trnsmitted pckets hve exceeded threshold. By pplying brgining-gme theory to VANET, Shresth et l. [2] proposed method tht cn solve the Received August 18, 2011; revised November 25, 2011 & Mrch 2, 2012; ccepted April 20, Communicted by Wnjiun Lio. * Prt of this work ws presented in IWCMC-VCS Conference, Istnbul, Turkey, July

2 214 problem of lrge pcket loss when nodes move very fst. To minimize multi-hop delivery dely, Yu et l. [3] proposed method to djust the trnsmission order ccording to the geogrphicl loctions of vehiculr nodes. In multichnnel techniques, Mk et l. [4] proposed n ccess control scheme to improve the communiction efficiency between AP (Access Point) nd VNs. Their proposed technique combines PCF (Point Coordintion Function) nd DCF (Distributed Coordintion Function) to mximize the chnnel utiliztion which not only gurntees the trnsmission of sfety messge but lso increses the throughput of non-sfety messges. In ddition to multi-chnnel, Su et l. [5] further utilized the concept of cluster-bsed to propose multi-chnnel protocol, with which non-rel-time nd rel-time trffic cn be delivered through contention nd contention-free periods, respectively. Some previous works on the cluster-bsed VANET studied how to use metrics to choose cluster hed. Among them, Goonewrdene et l. [6] nd Zydoun et l. [7] considered reltive moving speed nd direction to reduce link filure rte. By using reltive distnce, Wng et l. [8] proposed position-bsed prioritized clustering (PPC) to select cluster hed. Zhu et l. [9] selected cluster hed bsed on communiction efficiency, network connectivity, nd residul energy. Chu et l. [10] used the number of vehiculr nodes nd Guo et l. [11] used the trnsmit power nd coverge s metrics to choose cluster hed. Without using sophisticted method, Fn et l. [12] picked up cluster hed simply bsed on the smllest ID. In cluster-bsed TDMA, selected cluster hed is responsible for time slot lloctions. Stio et l. [13] proposed sptil slot lloction scheme to dynmiclly rellocte time slots bsed on node s priority. Similrly, Du et l. [14] nd Kosts et l. [15] respectively proposed slot-lloction schemes to fst deliver emergency nd rel-time trffic. For sensor networks, M et l. [16] proposed n NACPA (Nimble nd Adptive Control Phse Algorithm) consisting of four phses (frme synchroniztion, control, scheduling, nd dt trnsmission). By ssuming cluster hed is known to the cluster members, sensor node rndomly chooses time slot for bndwidth requests (BR) in control phse. If certin nodes filed in the previous round becuse they pick up the sme slot, the filed nodes hve to rise BR in the next round. Su et l. [17] proposed n integrtion of contention-free nd contention-bsed MAC protocol for VANET. Their scheme ssumed VN lwys hve dt to send nd the number of VNs is smller thn the number of time slots. When the number of VN exceeds the number of time slots, collisions cnnot be voided. Finlly, Ding et l. [18] employed self-orgnized cluster by ssuming two different chnnels, multiple control chnnels nd single dt chnnel. Unfortuntely, the uthors did not ddress time slots in TDMA. Unlike the previous works [16-18], our proposed cluster-bsed TDMA, nd therefter CBT, does not require unresonble ssumptions. First, in intr-cluster communictions, VC (VANET Coordintor) is not pre-ssigned, but it is elected through simple trnsmit-nd-listen scheme. Second, without confining the number of BR, the proposed CBT llows VN to rndomly choose time slot for BR. In inter-cluster communictions, when two clusters re pproching, the CBT cn quickly resolve the collisions by re-ssigning time slots in one of the clusters. To nlyze the performnce of the proposed CBT, we derive mthemticl equtions using the probbility model. The performnce metrics of our interests include the verge number of time slots required for electing VC, for presenting BR, nd the totl time slots required before dt cn be successfully

3 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 215 trnsmitted. Finlly, the nlyticl results re vlidted by n NS-2 simultion. The reminder of this pper is orgnized s follows. In Section 2, the proposed CBT is introduced. The formts of TDMA nd MAC frmes re subsequently defined. In Section 3, the CBT lgorithms for intr- nd inter-communictions re presented. In Section 4, mthemticl equtions re derived nd simultion is performed. In Section 5, we give the concluding remrks. 2. CLUSTER-BASED TDMA SYSTEM A cluster-bsed TDMA (CBT) system is designed for intr- nd inter-cluster communictions. The CBT ssumes every vehicle is equipped with GPS (Globl Positioning System) to synchronize TDMA slots mong vehicles prior to setting up cluster network. 2.1 Intr- nd Inter-Cluster Communictions Bsiclly, the proposed CBT system for inter-vehicle communictions consists of two different phses, intr- nd inter-cluster communictions. Initilly, when two or more thn two vehicles pproch together, the CBT of intr-cluster communictions is constructed in three steps. First, ll vehiculr nodes compete mong themselves for being elected s VANET coordintor (VC). Second, winner, the VC, is responsible for scheduling time slots for the rest of cluster nodes, referred to s VANET nodes (VNs). Third, dt trnsmissions over the designted time slots cn proceed without incurring ny collisions. Once VC is elected, the coverge of cluster is defined s the trnsmission rnge of the VC, i.e., meters, s shown in Fig. 1. On highwy, it is highly possible tht vehiculr nodes in cluster my move fster thn the ones in nother cluster, nd eventully two VCs of the two clusters my shre n overlpping re for certin time. As shown in Fig. 2, two VCs (VC 0 nd VC 1) of two clusters respectively coordinte the time slots for their own VNs (i.e., VC 0 for VN 01 nd VN 02, nd VC 1 for VN 11 nd VN 12). Once these two clusters pproch together, VNs belonging to different clusters my like to exchnge informtion s well. For exmple, VN 01 my send dt to VN 11. Thus, it is essentil to construct mechnism for inter-cluster communictions. Fig. 1. Intr-cluster communictions. Fig. 2. Inter-cluster communictions.

4 TDMA Time Slots To build intr- nd inter-cluster communictions for the proposed CBT system, first, we need to design TDMA time slots nd the ssocited MAC-lyer frme formt. As shown in Fig. 3 (), TDMA frme consists of n time slots (slot 0 to slot n 1). Among them, slot 0 serves s two different purposes in different TDMA frmes; (i) SYN (in the first TDMA frme): prior to the set up of CBT system, two or more thn two pproching vehicles will issue n 8-byte becon signl to synchronize to the strt of slot 1, nd (ii) SAM (in the remining TDMA frmes): once CBT system is formed, the elected VC begins to brodcst slot-lloction mp (SAM) to its VNs, so every VN hs designted time slot for trnsmitting dt. Slot 1 to slot n 1 of the first TDMA frme (the VC-elected stge) re used for the cluster nodes to compete for VC, while slot 1 to slot n 1 of the remining TDMA frmes (the slot-lloction stges) re designted time slots used for dt trnsmission. () Time slots in TDMA frmes. (b) MAC-frme formt in slot 0. (c) MAC-frme formt in slot 1 to slot n 1. Fig. 3. TDMA time slots nd MAC-lyer frmes.

5 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 217 Fig. 3 (b) shows the m-byte MAC-lyer frme formt in slot 0. It consists of three fields, 8-byte becon, two (m 8 4)/2-byte slot-lloction mps, nd 4-byte gurd bnd. The design of becon signl serves two purposes: (i) it synchronizes the strt of slot 1 during VC-elected stge, nd (ii) it llows one VC to detect the existence of nother VC, so tht the construction of inter-cluster communiction cn be initited. It is noticed tht with 8-byte becon length it cn support up to 300-meter trnsmission coverge nd up to 20-Mbps trnsmission rte, since 20 Mbps 8 bytes, where C is the speed C of light. The two slot-lloction mps (SAM) re designed for two VCs in two different clusters to exchnge slot-lloction informtion. A simple trnsmit-nd-listen scheme works in wy tht VC successfully received SAM from nother VC will reschedule its time slots to void ny collisions with nother cluster. Detil procedures to build intercluster communictions will be introduced in Section 3. Bsiclly, SAM consists of the following fields, introduced below. Fig. 3 TDMA Time Slots nd MAC-lyer Frmes F (1 bit): If F = 1, VN my ccess SAM; otherwise, it is for VCs to exchnge their SAMs. L (7 bits): The length of SAM (in bytes). VC MAC Address (6 bytes): The MAC ddress of VC. VN MAC Address (6 bytes): The MAC ddress of VN. Slot Number Allocted (1 byte): The ID (from 1 to n 1) of the llocted time slot. CRC (4 bytes): to protect SAMs. As n exmple, if time slot cn ccommodte m-byte MAC-lyer pylod, fter m 84 deducting the 4-byte gurd bnd, ech SAM cn occupy bytes, which cn 2 support up to K VNs in ech cluster, nd 2K VNs in both clusters. Fig. 3 (c) shows the m-byte MAC frme formt in slot 1 to slot n 1. Ech MAC frme consists of 16-byte MAC heder, (m 16 4)-byte dt pylod, nd 4-byte CRC. Except the source nd destintion MAC ddresses (6 bytes ech), the rest of fields in the MAC heder re defined s below: F (1 bit): The defult vlue is zero, indicting tht slot 1 to slot n 1 in the first TDMA frme re used for cluster nodes to compete for VC. Once VC is elected, F is set to one, indicting tht slot 1 to slot n 1 re used for dt trnsmission in the remining TDMA frmes. LO (1 bit): Est (LO = 1) or West (LO = 0) Longitude. LA (1 bit): North (LA = 1) or South (LA = 0) Ltitude. Exponent (3 bits): The exponent prt of 12-bit floting point number. Mntiss (9 bits): The mntiss prt of 12-bit floting point number. It is noticed tht LO, LA, Exponent, nd Mntiss re designed to locte vehiculr node moving on the street or highwy using the Longitude nd the Ltitude. For exmple, if vehiculr node is locted t Est-Longitude , then we hve LO = 1, Exponent = 111, nd Mntiss = , since = =

6 THE CBT ALGORITHMS For clrity, the proposed CBT lgorithms re divided into two different communiction types, Intr-cluster communictions (Intr-CC) nd inter-cluster communictions (Inter-CC). 3.1 Intr-Cluster Communictions Fig. 4 shows the stte trnsitions of Intr-CC, which describes the procedures of setting up CBT system. Intr-CC bsiclly consists of five sttes, Initil, Competition, VC, VN, nd Collision. In the Initil stte, vehiculr node, referred to s cluster node (CN), is serching for other CNs by issuing n 8-byte becon signl, s described in Section 2.2. Once collision of the becon signl is detected, CN remins idle till the strt of slot 1, which delibertely synchronizes ll the ner-by CNs to the strt of slot 1 in the first TDMA frme. Beginning from slot 1 to slot n 1 of the first TDMA frme, the trnsition enters Competition stte, where CN rbitrrily chooses to trnsmit-nd-listen competefor-vc (CFV) messge, formtted s shown in Fig. 3 (c), on slot-by-slot bsis. A VC is utomticlly elected if only one CN intends to trnsmit CFV nd ll others re in listening CFV. On the other hnd, if two or more thn two CNs intend to trnsmit CFV t the sme slot, the trnsition goes to the Collision stte, where ll the CNs will resume competition for VC t the next slot. Fig. 4. Stte trnsitions of intr-cluster communictions.

7 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 219 Fig. 5 shows the opertions of Intr-cluster communictions (Intr-CC). Bsiclly, Intr-CC consists of three phses. In phse 1, VC is rndomly selected mong ll the VNs through VC_elected (). In phse 2, ll the VNs present their Bndwidth Requests (BR) to VC through BR() nd Determine(). In phse 3, VN begins to use its designted time slots to trnsmit dt through Dt_trnsmit(). VC_elected() // VC is elected BR() // VNs issue bndwidth requests Determine() // to determine whether BR is successful Dt_trnsmit () // use designted time slots to trnsmit dt Slot i Input: K Input: n Input: N Rnd // Slot-ID // number of vehiculr nodes // number of time slots in TDMA frme // generte 1 to N Rnd time slots for dt trnsmission VC_elected(K) { Slot i = 0; VC = 0; // no VC is elected so fr While (VC = 0) { For i = 1 to K do { Every node rndomly chooses between 0 nd 1} If (more thn two nodes choose 1) // CFV collision Else if (ll nodes choose 0) // ll nodes re listening CFV Else // VC is successfully elected VC = 1; brek; Slot i = Slot i + 1;} Output: Slot i } BR(K) { ll_complete = K 1; // Initil to K 1 VNs incomplete = 1; Remining_slots = (Frme i n Slot i ); For i = 1 to K do { Num_of_slot i = Rnd[1, N Rnd ];} // generte multi-slots for dt trnsmissions While (incomplete = 1) { incomplete = 0; If (first_round = 1) // The first TDMA frme For i = 1 to K do { Slot i = Rnd[1, Remining_slots];} //Every VN rndomly selects slot for BR Determine(); // Determine whether ll BR re successfully first_round = 2; // Exit the first TDMA frme Else For i = 1 to K do { Slot i = Rnd[1, n];} // every VN rndomly selects slot for BR

8 220 Determine();} Output: Slot i } Determine() { For i = 1 to K For j = 1 to K If (j i nd Slot j = Slot i ) // determine whether two VN choose the sme slot for BR incomplete = 1; // BR collision If (incomplete = 0) //BR is successful ll_complete = ll_complete 1; // Decrement node number by one If (incomplete =1) // enter the next TDMA frme } Dt_trnsmit() { If (Num_of_slot i = 1) // using single-slot for dt trnsmissions Else // using multi-slots for dt trnsmissions } Fig. 5. Opertions of intr-cluster communictions. Fig. 6. Stte Trnsitions of inter-cluster communictions. 3.2 Inter-Cluster Communictions Fig. 6 shows the five-stte trnsition digrm of inter-cluster communictions (inter- CC). The trnsition begins with Becon-Issuing stte. Periodiclly, VC in cluster will issue becon signl in slot 0 of every TDMA frme. If no other becon signl is detected, the cluster remins in the intr-cluster communictions; else collision of becon signl implies tht there is nother VC of different clusters ner-by. The two clusters re cooperting through the VC-to-VC contct to build the inter-cluster communictions. First, these two VCs re synchronized t the end of becon signl. From there, two slot lloction mps (SAMs), s shown in Fig. 3 (b), require for exchnge. Two VC exchnge SAM to ech other using the simple trnsmit-nd-listen scheme from slot 1 to slot n 1. In other words, the one who successfully trnsmits SAM to the other is the winner, nd the

9 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 221 one who successfully receives SAM becomes the loser. The winner-vc will not lter its scheduled time slots, while the loser-vc hs to reschedule time slots for ll the VN under its supervision. Fig. 7 shows the opertions of Inter-cluster communictions (Inter-CC). As it cn be seen, Inter-CC consists of two phses. In phse 1, VCs re in the competition for sending SAM from slot 1 to slot n through Compete_SAM(). In phse 2, dt re successfully trnsmitted using designted time slots through Dt_trnsmit(). Compete_SAM() // ll VCs compete the sending of SAM Dt_trnsmit () // use the designted time slots for dt trnsmissions Slot i Input: K VC Input: n Input: N Rnd // Slot-ID // number of VC // number of time slots in TDMA frme // generte 1 to N Rnd time slots for dt trnsmissions Compete_SAM() { Slot i = 0; Success = 0; // SAM sending is not successful While (Success = 0) { For i = 1 to K VC do { Every VC rndomly chooses between 0 nd 1} If (more thn two VC choose 1) // SAM collision Else if (ll VC choose 0) // ll VC re listening SAM Else // one VC hs successfully sent out SAM Success = 1; The VC received SAM relloctes time slots for its VNs brek; Slot i = Slot i + 1;} Output: Slot i } Dt_trnsmit() { } Num_of_slot i = Rnd[1, N Rnd ]; If (Num_of_slot i = 1) // using single-slot for dt trnsmissions Else // using multi-slots for dt trnsmissions // generte multi-slots Fig. 7. Opertions of inter-cluster communictions. 4. ANALYSIS AND SIMULATION First, we use probbility model to derive mthemticl equtions. The performnce metrics of our interests include the verge number of time slots required for electing VC, for bndwidth requests (BR), for SAM brodcsting, nd the totl number of time slots required for VN to wit before dt cn be trnsmitted. To vlidte the mthemticl results nd to compre the performnce with IEEE p, we conduct simultion study using NS-2. In the simultion, we bsiclly implement the CBT opertions for intr- nd inter-communictions, the TDMA time slots, nd the MAC frme formts.

10 Mthemticl Anlysis Prior to introducing the mthemticl nlysis, we list ll the prmeters to be used in Tble 1. t n K K b K VC PERM d c m L SAM_ L SAM_b R T s Tble 1. Definition of prmeters. Prmeters Number of TDMA frmes Number of slots in TDMA frme Number of nodes in Cluster A Number of nodes in Cluster B Number of Clusters The permuttion of selecting d from c MAC frme size (bytes) SAM size of Cluster A (bytes) SAM size of Cluster B (bytes) Trnsmission bit rte (Mbps) Durtion of slot Intr-cluster nlysis For intr-cluster nlysis, we simply use cluster A s n exmple. Let P VC be the probbility of successfully electing VC. Recll to Section 3, in time slot, every competing vehiculr node is in either of the two sttes, trnsmit or receive CFV pcket. Therefore, there re totl of 2 K combintion possibilities. Among them, only K combintions cn be successful in electing VC; the rest of them fil. Hence, P VC cn be derived from Eq. (1), K P. (1) VC K 2 Next let x be the time slot in which VC is successfully elected. The probbility density function of P VC cn be expressed s f(x) = (1 P VC ) x-1 P VC, which implies tht the election of VC fils for the first x 1 time slots nd finlly succeeds t the xth time slot. Let E[X] intr be the verge time slots required for electing VC. We cn compute E[X] intr from Eq. (2). Note tht one extr slot is dded to the verge, since in our design it requires one slot for synchroniztion. K 1 2 EX [ ] intr xf( x) (2) P K x1 VC Let P 1 BR be the probbility of successful bndwidth requests (BR) issued from (K 1) VNs by using the first TDMA frme. Since in the first frme, it requires E[X] intr time slots to elect VC, only (n E[X] intr ) time slots remin for BR. For (K 1) VNs, there re totl of (n E[X] intr ) K-1 different combintions. An issuing BR cn be suc-

11 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 223 cessful (i.e., without encountering ny collisions) if nd only if every VN chooses differ- ( n E[ x] intr ) ent time slots to issue its BR. Thus, there re PERM K 1 cses for ll the (K 1) N s to chieve successful BR in the first TDMA frme. Let P i BR, i = 2, 3,, t, be the probbility of successful BR issued from the filed VNs by using the second, the third, nd up to the tth TDMA frme. It becomes more complicted to derive P i BR, i = 2, 3,, t, since filed bndwidth request in the first TDMA frme will continue to contend with other filed VNs in the second TDMA frmes, nd so on. By deducting the successful BR in the previous TDMA frme, the verge number of filed BR, N i F of the second (i = 2), nd of the follow-on TDMA frmes (i = 3, 4,, t) cn be expressed s N i F K 1 ne[ X ] intr PERM j K j n E[ X ] intr j 1 1, if i 2, i1 N F n i 1 PERM N j F j, if i 3, 4,..., t. n j 1 (3) Thus, we cn compute the probbility of successful BR issued from (K 1) VNs by using the ith TDMA frme, i.e., P i BR, i = 1, 2, 3,, t, from Eq. (4). P i BR n E[ X ] intr PERM K 1, if i 1, K 1 n E[ X ] intr n PERM i N F i, if i 2,3,..., t. N F n (4) Next, let E[Y] intr be the verge number of time slots required for the successful BR. We cn compute E[Y] intr from Eq. (5). EY [ ] ( ( int r P n [ ] ) 2 ( 1 )... BR E X intr PBR PBR 1 2 t1 t t(1 PBr )(1 PBR)... (1 PBR ) PBR) n t i1 1 i j ( intr ( BR BR BR i2 j0 ) P n E[ X] ) ( ip (1 P )) n. (5) After bndwidth requests, VC begins to brodcst SAM (slot lloction mp) to VNs. Let L SAM_ be the size of SAM (in bytes) for cluster A nd SAM intr be the number of slots required for VC in cluster A to brodcst SAM. Refer to the SAM formt in Fig. 3 (b), in ddition to 1-byte flg nd length, 6-byte VC MAC ddress, nd 4-byte CRC, vehiculr node lso requires 6-byte MAC nd 1-byte llocted slot number. Thus, we cn clculte SAM intr from Eq. (6), where R is the trnsmission bit rte (in Mbps) nd T S is the durtion time of time slot (in sec).

12 224 SAM intr L _ 8 7 ( K 1) 118 R R SAM 56 K 32 (6) T T T R S S S After the brodcsting of SAM, VN cn deliver its dt over the designted time slots. Here, we ssume two different types of slots re used for dt trnsmissions, single-slot nd multi-slots. Let E[Z] intr be the verge number of time slots required for witing before VN cn begin to trnsmit its dt. We cn compute E[Z] intr from Eq. (7). EZ [ ] intr K 1 K 1, if using single-slot, K N 2 Rnd ( ) K K 1 0 (1 N ) ( K 1) Rnd, if using multi-slots. 4 It is noticed tht in Eq. (7), for single-slot, node out of the K nodes hs to wit from K 1 from 0 to K 1 slots. Thus, the totl witing time of K nodes in single-slot is. For 0 For multi-slots, we ssume the requested time slots of VN follow Uniform distribution, where N Rnd is the mximum number of requested time slots of VN. Thus, on verge, every VN hs to wit ( 1 N Rnd ). As result, the totl witing time of K nodes in 2 K 1 1 N multi-slots is ( Rnd ). 2 0 Finlly, let S intr be the totl number of time slots counting from the time to elect VC to the time when node is redy to trnsmit its dt. We cn compute S intr from Eq. (8). S intr = E[X] intr + E[Y] intr + SAM intr + E[Z] intr (8) Inter-cluster nlysis Referring to Fig. 6, when two or more thn two clusters re pproching, VCs of these clusters will compete for sending SAM from slot 0 to slot n. The llocted time slots of the winner-vc remin no chnge, while the loser-vcs hve to reschedule time slots for their VNs to void ny possible collisions. Let K VC be the totl number of VC nd P SAM be the probbility of successfully trnsmitting SAM in mini-slot. We cn compute P SAM from Eq. (9), where the numertor is the number of cses to successfully trnsmit SAM nd the denomintor is the totl combintions if the number of VCs equls K VC. (7)

13 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 225 KVC PSAM (9) K 2 VC Next, let x be the mini-slot in which SAM cn be successfully trnsmitted. The probbility density function of P SAM cn be expressed s f(x) = (1 P SAM ) x-1 P SAM, which implies tht the trnsmission of SAM fils for the first x 1 mini-slots nd eventully succeeds t the xth mini-slot. Let E[U] inter be the verge time slots required for VC to successfully trnsmit SAM. We cn compute E[U] inter from Eq. (10). Notice tht in our design there re two mini-slots in time slot. K VC EU [ ] inter xf( x) 2 2 P K (10) x1 SAM VC Next, let L SAM_ be the SAM size of cluster A, L SAM_b be the SAM size of cluster B, nd SAM inter be the number of time slots required for two clusters (sy, cluster A nd B) to brodcst their SAMs. Since L SAM_ = 7 (K 1) + 11 nd L SAM_b = 7 (K b 1) + 11, SAM inter cn be computed from Eq. (11). SAM inter L 8 L 8 SAM _ SAM _ b 56 K K 64 R R b T T T R s s s (11) Let E[W] inter be the verge time slots required for witing before dt cn be trnsmitted. Similr to Eq. (7), by replcing K with (K + K b ), we cn compute E[W] inter from Eq. (12). EW [ ] inter K K 1 K b 0 K K K 1 1 N 2 N K K Rnd K b K K b 0 b K 2 b 1, if using single-slot 1 1 Rnd b, if using multi-slots 4 (12) Finlly, let S inter be the totl number of time slots counting from the time when SAM is successfully trnsmitted by VC to the time when node is redy to trnsmit its dt. We cn compute S inter from Eq. (13). S inter = E[U] inter + SAM inter + E[W] inter (13) 4.2 Anlyticl nd Simultion Results As shown in Fig. 8, the verge number of time slots for electing VC is exponent-

14 226 Fig. 8. Averge number of time slots for electing VC. tilly incresed s the number of competing nodes increses from 2 to 10. We observe tht the mthemticl results, computed from Eq. (2), re very close to the simultion results; yet, simultion results re more relistic, since in the simultion, rndom number is used for VN to decide whether to trnsmit or listen. To compre with the CSMA/CA mechnism used in IEEE p, more simultions re performed using NS-2. As shown in Fig. 8, when the number of VNs is smller thn or equl to 6, the proposed CBT requires fewer time slots in electing VC. This is becuse using bckoff lgorithms my become too conservtive for smll number of VNs. Fig. 9 shows the verge number of time slots required for BR in CBT s function of incresing VNs. When the number of time slots in TDMA frme is smll (e.g., n = 30), mthemticl results re quite close to the simultion results. However, s n is incresed to 70, mthemticl results re incresingly fr prt from the simultion results; in fct, the difference increses from 2% to 26% s the node increses from 2 to 10. This phenomenon cn be understood from two spects: (i) mthemticl equtions, referring to Eq. (3), cn not ccurtely determine which nodes would successfully complete BR in TDMA frme, nd (ii) in the mthemticl model, it is no wy to determine which time slot is employed by which vehiculr node. However, the bove two uncertinty cn be esily conquered in simultion by ssigning n ID to ech VN. Referring to Eq. (7), in CBT the verge number of time slots required for witing before dt trnsmission, i.e., E[Z] intr, cn be computed for single-slot nd multi-slots, respectively. From Fig. 10, it is quite strightforwrd tht E[Z] intr would increse s the number of VNs increses. For single-slot trnsmissions, the percentge of increse is the smllest. For multi-slot trnsmissions, when N rnd, rndom number used to decide the number of time slots requested for BR, is slightly incresed from 4 to 7, we observe tht E[Z] intr increses very significntly. Additionlly, we observe tht mthemticl results cn mtch with the simultion quite well. Fig. 11 shows the totl number of time slots required before VN cn begin to trnsmit its dt on designted slots. In generl, s the number of time slots in TDMA frme (i.e., n) is incresed, S intr increses very significntly. From Fig. 11, we cn esily observe tht mthemticl curves re lmost in ccordnce with the simultion curves. For single-slot dt trnsmissions, the reson for n to dominte S intr is tht the verge time slots required for BR (i.e., E[Y] intr ) increses lrgely s n is incresed. On the other hnd, for multi-slot dt trnsmissions, the fctors tht cn distinguish S intr come from N Rnd s well. It is observed tht by enlrging N Rnd from 3 to 7, S intr cn gretly increse.

15 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 227 Fig. 9. Averge number of time slots for BR in CBT. Fig. 10. Averge number of time slots witing before dt trnsmissions.. Fig. 11. Totl number of time slots required before dt trnsmissions. This phenomenon revels tht in intr-cluster communiction the mjor dominting fctor for VN to wit before its dt cn be trnsmitted is the number of time slots requested during BR. To compre the performnce between CBT nd IEEE p, more simultions re conducted using NS-2. From the simultion results s shown in Fig. 11, CBT consumes fewer time slots thn p, no mtter in single-slot or multi-slot trnsmissions. This is becuse CBT employs TDMA nd it efficiently utilizes pre-scheduled time slots to trnsmit dt, while using bckoff lgorithm in p will exponentilly increse the witing time, prticulrly when the number of joining VNs is incresed. Consequently, the ltter wits more time slots before VN cn effectively trnsmit dt.

16 228 Finlly, Fig. 12 shows the verge number of time slots required in inter-cluster communiction. In this nlysis, for simplicity, we ssume only two clusters (Clusters A nd B) exist. When the totl number of VNs in these two clusters increses from 4 to 32, we observe tht ll curves of S inter increse lmost linerly. Among them, for single-slot dt trnsmissions, S inter increses very smoothly. However, for multi-slot trnsmissions, S inter increses very rpidly no mtter when N rnd is equl to 7 or 12. By referring to Eq. (10) to (12), since mthemticl results of E[U] inter nd E[W] inter do not mke big difference, it is observed tht mthemticl results re only slightly different from the simultion results. To mke performnce comprison, we run simultions for CBT nd IEEE p using NS-2. From Fig. 12, we cn observe tht multi-slot trnsmissions in p consume the mximum number of time slots, nd single-slot trnsmissions in p tke the second. Even though CBT hs n overhed tht requires the exchnge of SAM between two clusters, it significntly outer performs p, prticulrly when the number of joining VNs to the two clusters is incresed. The min reson for CBT outer performing p is quite strightforwrd; i.e., the ltter employs bckoff lgorithm which exponentilly increses the witing time long with the increse of joining VNs. Fig. 12. Averge number of time slots required in Inter-CC. 4.3 Protocol Overhed As compred to p, the proposed CBT my incur extr cost nd protocol overhed. First, to form cluster, CBT must synchronize time slots mong VNs. Thus, to enble CBT, VN must be equipped with GPS. Second, s shown in Figs. 3 () nd (b), the first slot (i.e., slot 0) in ech TDMA frme is reserved for SYN nd SAM; it cnnot be used for trnsferring user dt. It is noticed tht this overhed is reduced when the number of time slots in TDMA frme (i.e., n) increses. Furthermore, the overhed does not increse long with the increse of VNs. This is becuse in our design the number of VNs supported in cluster (i.e., K) is fixed when the size of MAC frme (i.e., m) is determined. In fct, fter simple clcultion, we cn esily derive K from m; i.e., K = ( m 34). 14 For inter-cluster communiction, in ddition to the bove-mentioned overhed, CBT will further require the exchnge of SAM between ny two clusters. The size of SAM is m 84 equl to bytes, s it is defined in Section 2.2 nd illustrted in Fig. 3 (b). 2

17 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS CONCLUSIONS In this pper, we hve presented cluster-bsed TDMA (CBT) system for inter-vehicle communictions. One of the novelties of the proposed CBT is right in tht it uses reltively simple trnsmit-nd-listen scheme to timely elect VC nd to quickly resolve the collisions when two clusters re pproching together. Performnce of the proposed CBT ws nlyzed by deriving mthemticl equtions to compute the verge number of time slots for electing VC, for presenting BR, nd the totl number of time slots required before dt cn be effectively trnsmitted. The nlyticl results were finlly vlidted by simultion using NS-2. From the nlyticl nd simultion results, we hve observed tht in forming smll-sized cluster the proposed CBT spends less time thn IEEE p. Additionlly, when the number of joining VNs is incresed, CBT tkes less witing time before VN cn effectively trnsmit dt. The proposed CBT my co-exist with the existing p. For exmple, CBT cn be engged initilly for smll-sized cluster, nd if mny VNs re competing for VC, p cn be invoked subsequently. The protocol primitives to be dded in order to switch over between these two schemes should include threshold in terms of the number of collisions for compete-for-vc (CFV) messge. In the future works, the proposed CBT cn be extended by considering different trffic types with priorities. For exmple, rel-time trffic with higher-priority should possess more privilege to cquire time slots thn non-rel-time trffic with lower-priority. REFERENCES 1. C. Wng, B. Li, nd L. Li, A new collision resolution mechnism to enhnce the performnce of IEEE DCF, IEEE Trnsctions on Vehiculr Technology, Vol. 53, 2004, pp B. Shresth, D. Niyto, Z. Hn, nd E. Hossin, Wireless ccess in vehiculr environments using BitTorrent nd brgining, in Proceedings of IEEE Conference on Globl Telecommunictions Conference, 2008, pp F. Yu nd S. Bisws, Self-configuring TDMA protocols for enhncing vehicle sfety with DSRC bsed vehicle-to-vehicle communictions, IEEE Journl on Selected Ares in Communictions, Vol. 25, 2007, pp T. K. Mk, K. P. Lberteux, R. Sengupt, nd M. Ergen, Multichnnel medium ccess control for dedicted short-rnge communictions, IEEE Trnsctions on Vehiculr Technology, Vol. 58, 2009, pp H. Su nd X. Zhng, Clustering-bsed multichnnel MAC protocols for QoS provisionings over vehiculr d hoc networks, IEEE Trnsctions on Vehiculr Technology, Vol. 56, 2007, pp R. T. Goonewrdene, F. H. Ali, nd E. Stipidis, Robust mobility dptive clustering scheme with support for geogrphic routing for vehiculr d hoc networks, IEEE Journls on Intelligent Trnsport Systems, Vol. 3, 2009, pp Z. Y. Zydoun nd S. M. Mhmud, Towrd strongly connected clustering structure in vehiculr d hoc networks, in Proceedings of IEEE Conference on Vehiculr Technology, 2009.

18 Z. Wng, L. Liu, M. C. Zhou, nd N. Ansri, A position-bsed clustering technique for d hoc intervehicle communiction, IEEE Trnsctions on Systems, Mn, nd Cybernetics, Prt C: Applictions nd Reviews, Vol. 38, 2008, pp X. Zhu, L. Shen, nd T. S. P. Yum, Husdorff clustering nd minimum energy routing for wireless sensor networks, in Proceedings of IEEE Trnsctions on Vehiculr Technology, Vol. 58, 2009, pp Y. C. Chu nd N. F. Hung, Delivering of live video streming for vehiculr communiction using peer-to-peer pproch, in Proceedings of IEEE Conference on Mobile Networking for Vehiculr Environments, 2007, pp C. Guo, R. Hekmt, nd P. Pwelczk, Anlysis nd optimiztion of energy efficient cluster forming for wireless sensor networks, in Proceedings of IEEE Conference on Vehiculr Technology, 2007, pp P. Fn, Improving brodcsting performnce by clustering with stbility for intervehicle communiction, in Proceedings of IEEE Conference on Vehiculr Technology, 2007, pp K. Sito nd M. Nkym, The sptil reuse slot lloction with rellocting neighbor links in the TDMA-bsed wireless mesh network, in Proceedings of IEEE Conference on Vehiculr Technology, 2007, pp J. Du nd W. Shi, App-MAC: An ppliction-wre event-oriented MAC protocol for multimodlity wireless sensor networks, IEEE Trnsctions on Vehiculr Technology, Vol. 57, 2008, pp T. A. Kosts, A. H. Hddd, nd C. C. Lee, Performnce nlysis of dynmic bndwidth lloction lgorithm for mixed trffic TDMA network, IEEE Trnsctions on Vehiculr Technology, Vol. 56, 2007, pp J. M, K. Yng, nd S. Ou, Nimble nd dptive time-division multiple ccess control phse lgorithm for cluster-bsed wireless sensor networks, IET Communictions, Vol. 1, 2007, pp H. Su, X. Zhng, nd H. H. Chen, Cluster-bsed multi-chnnel communictions protocols in vehicle d hoc networks, IEEE Journls on Wireless Communictions, Vol. 13, 2006, pp R. Ding nd Q.-A. Zeng, A clustering-bsed multi-chnnel vehicle-to-vehicle (V2V) communiction system, in Proceedings of the 1st Interntionl Conference on Ubiquitous nd Future Networks, 2009, pp Tsng-Ling Sheu () received the Ph.D. degree in Computer Engineering from the Deprtment of Electricl nd Computer Engineering, Penn Stte University, University Prk, Pennsylvni, USA, in From September 1989 to July 1995, he worked with IBM Corportion t Reserch Tringle Prk, North Crolin, USA. In August 1995, he becme n Associte Professor, nd ws promoted to Full Professor in Jnury 2006 t the Deprtment of Electricl Engineering, Ntionl Sun Yt-Sen University, Kohsiung, Tiwn. His reserch interests include mobile communictions, wireless networks, nd multimedi net-

19 A CBT SYSTEM FOR INTER-VEHICLE COMMUNICATIONS 231 working. He ws the recipient of the 1990 IBM Outstnding Pper Awrd. Dr. Sheu is senior member of the IEEE, nd the IEEE Communictions Society. Yu-Hung Lin () received the B.S. degree from the Deprtment of Informtion Engineering, I-So University, Kohsiung, Tiwn, in 2007, nd the M.S. degree from the Deprtment of Electricl Engineering, Ntionl Sun Yt-Sen University, Kohsiung, Tiwn, in During his M.S. study, his reserch interest ws in the vehiculr d-hoc networks (VANET).