MECHANICAL and hydraulic components in vehicles

Transcription

1 2160 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 5, JUNE 2009 Message Schedulig for the FlexRay Protocol: The Dyamic Segmet Ece Gura Schmidt ad Klaus Schmidt Abstract The FlexRay commuicatio protocol is expected to be thede facto stadard for high-speed, i-vehicle commuicatio. I this paper, we formally ivestigate the schedulig problem for the dyamic segmet (DS) of FlexRay. We take the bouds o the geeratio times ad the timig requiremets of the sigals ito cosideratio to propose a reservatio-based schedulig approach that preserves the flexible medium access of the DS. To obtai efficiet schedules, we formulate a oliear iteger programmig problem (NIP) that miimizes the required duratio of the DS. This NIP is the decomposed ito two liear biary iteger programmig problems to facilitate the computatio of feasible message schedules. A experimetal study illustrates our message schedulig approach for the DS of FlexRay. Idex Terms FlexRay, iteger programmig, real time, schedulig, vehicular commuicatio etworks. NOMENCLATURE Notatio Explaatio T c FlexRay cycle duratio. T SS,T DS Static segmet, dyamic segmet duratio). τ bit,t MS bit time, miislot duratio. N DS Number of miislots i the dyamic segmet. N Set of odes o the FlexRay etwork. M S Sporadic messages of ode N. M S All sporadic messages. Mm Message m of ode. pm m,dm m Period ad deadlie of Mm. lm m Legth of Mm. R =(,rp,w,l) Reservatio for ode. rp,w,l Reservatio period, offset, legth. R, R Set of reservatios for ode, all reservatios. r : M S R Map of messages to reservatios. L j Cycle load of a FlexRay cycle j. R j All reserved DYS o FC j. RP least commo multiple (lcm) of all reservatios periods. Mauscript received February 24, 2008; revised August 6, First published October 31, 2008; curret versio published May 11, The review of this paper was coordiated by Dr. P. Li. E. G. Schmidt is with the Departmet of Electrical ad Electroics Egieerig, Middle East Techical Uiversity, Akara 06531, Turkey ( egura@metu.edu.tr). K. Schmidt is with the Uiversity of Erlage-Nuremberg, Erlage, Germay ( klaus.schmidt@rt.eei.ui-erlage.de). Color versios of oe or more of the figures i this paper are available olie at Digital Object Idetifier /TVT f : R F Map of reservatios to FIDs. B = (l/rp) Badwidth reservatio for ode per FC. R R B = N B Badwidth reservatio for all odes. =1 Gq M S Message group q for ode. g (Gq ) Reservatio for Gq. G = N =1G All message groups. G S G Selected groups. I. INTRODUCTION MECHANICAL ad hydraulic compoets i vehicles have bee replaced by electroic compoets sice the 1970s. These i-vehicle electroic systems employ electroic cotrol uits (ECUs), which are embedded systems with, e.g., a microcotroller, sesors, ad actuators. Commuicatio etworks eable the iformatio exchage amog ECUs to support most of their tasks. These days, more tha 70 ECUs exchage aroud 2500 sigals i luxury cars [1], [2]. The commuicatio etworks i vehicles trasmit sigal data that are ecapsulated i messages. Most of these messages are real-time messages, i.e., their timely delivery must be guarateed. Techically, precomputed message schedules have to be supplied to meet such timig requiremets. I additio, cosiderig the fast growth i the umber of ECUs ad sigals i automotive electroics, the commuicatio must be efficiet to provide system extesibility. Oe of the first i-vehicle commuicatio etworks for automotive systems is the cotroller area etwork (CAN) [2], [3]. It ca provide bouded delay commuicatio amog ECUs at data rates betwee 125 kb/s ad 1 Mb/s ad is curretly the most widely used i-vehicle etwork. However, it is ot suitable for ovel applicatios such as electroic compoets of the power trai or x-by-wire applicatios, which are hard real time i ature, ad require high-speed, robust, ad predictable commuicatio. The first attempts to meet these demads are time-triggered CAN (TTCAN [4]), time-triggered protocol (TTP [5], [6]), ad ByteFlight [7]. TTCAN ad TTP are time-triggered techologies with predictable medium access, whereas ByteFlight is based o flexible time-divisio multiple access (FTDMA), which aims at efficiet badwidth use. FlexRay i-vehicle commuicatio etworks was fouded as a idustry cosortium by BMW, Daimler-Chrysler, Philips, ad Freescale i 2000 [8], [9]. Curretly, there are more tha 150 members i the cosortium, ad the first series productio car with FlexRay was o the road i FlexRay has a static segmet (SS) with time-divisio multiple access (TDMA) /$ IEEE Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

2 SCHMIDT AND SCHMIDT: MESSAGE SCHEDULING FOR THE FLERAY PROTOCOL: THE DYNAMIC SEGMENT 2161 operatio ad a dyamic segmet (DS) with FTDMA operatio. It is expected to be the ew de facto stadard, combiig the advatages of time-triggered ad FTDMA commuicatio [2]. It provides two chaels with a badwidth of 10 Mb/s each, eablig applicatios that were ot realizable with CAN, such as x-by-wire. The DS of FlexRay is desiged to accommodate sporadic real-time messages that are geerated by evet occurreces ad have to be trasmitted before their deadlie. To this ed, it is required to fid feasible message schedules that meet the timig requiremets. Previous work o the FlexRay DS mostly provides methods to test if a give schedule is feasible [10], [11]. I additio, oe study [12] aalyzes ad evaluates deadlie mootoic schedulig. Differet from the previous work, we propose a method for sythesizig efficiet ad feasible message schedules. Based o a formal problem descriptio, our approach determies the required system parameters such that the sporadic messages are delivered o time. Adoptig ideas from our work i [13], we cosider the bouds o the message geeratio times ad the timig requiremets for message delivery of the sporadic messages to reserve badwidth for each message while maitaiig the beefits of the FTDMA operatio of the FlexRay DS. I our framework, we defie appropriate performace metrics to measure the efficiecy of each schedule. The, iteger programmig is applied to select the most efficiet feasible schedule. The structure of this paper is give as follows. I Sectio II, we describe the operatio of FlexRay ad itroduce our otatio for the system parameters. Sectio III addresses differet issues related to message schedulig for the DS. Our idea of message groupig to reduce the badwidth reservatio is first discussed i Sectio IV ad is the employed i Sectio V to fid optimal message schedules. Sectio VI presets experimetal results. Coclusios are give i Sectio VII. II. FLERAY PROTOCOL The FlexRay protocol defies two chaels that operate at a badwidth of C =10Mb/s, each leadig to a bit time of τ bit = 0.1 μs. I this paper, we cosider message trasmissios o oe FlexRay chael. A. Descriptio of the FlexRay Cycle (FC) The operatio of each FlexRay chael is based o a fixedduratio repeatedly executed FC that is time slotted [8]. The FC comprises a SS, ads, asymbol widow (SW), ad the etwork idle time (NIT). The SW ad the NIT provide time for the trasmissio of iteral cotrol iformatio ad protocolrelated computatios. The duratio of the FC, SS, ad DS, which are fixed durig the cofiguratio of a give system, are measured i millisecods ad are deoted as T c, T SS, ad T DS, respectively. A geeric FC is depicted i the upper part of Fig. 1. The SS is similar to TTP [6] ad employs the TDMA approach. The ivestigatio of the SS is ot i the scope of this paper ad ca be studied as a idepedet schedulig problem. Fig. 1. FlexRay cycle descriptio. We refer the reader to the compaio paper [14] for a detailed descriptio. The DS is similar to ByteFlight [7] ad employs the FTDMA approach. The smallest time uit i the DS is the miislot (MS) with a duratio of T MS (i millisecods), ad the DS cotais a fixed umber of N DS MS, where N DS N DS,max = The DS cosists of cosecutive dyamic slots (DYS) that are superimposed o MS. If a message is trasmitted i a DYS, the the legth of the DYS is equal to the umber of MS eeded for message trasmissio. Otherwise, the legth of the DYS is oe MS. Each ode maitais a slot couter to follow the progress of the DS. It is iitialized to 1 at the begiig of each FC ad is icremeted i every DYS. The arbitratio procedure esures that oly frames with a frame ID (FID) that equals the curret value of the slot couter ca be trasmitted [8]. Therefore, we iterchageably use the otio FID to express the frame ID ad the value of the slot couter i the remaider of this paper. The DS i Fig. 1 cosists of 20 MS. I the first FC, messages are trasmitted i the secod, fifth, ad sixth DYS, whereas the legth of, e.g., the secod DYS is 6 MS. B. Messages We cosider a commuicatio system that cosists of N odes (ECUs) that are coected by FlexRay, where the set of odes is N = {1,...,N}. The odes exchage periodic ad sporadic real-time messages that are trasmitted i FlexRay Frames. We assume that all periodic messages are scheduled i the SS as studied i [14]. I this paper, we ivestigate the trasmissio of sporadic messages i the DS. Our represetatio of the timig properties of sporadic messages follows the lies of related work i [12], [13], ad [15] [17]. For each sporadic message, there is a deadlie, which is the largest tolerable time iterval betwee the geeratio ad the trasmissio of the message. I our work, the deadlie icludes the message trasmissio time as well as the maximum jitter of the message as defied i [10]. I additio, the recurrece of a sporadic message is described by its miimum iterarrival time deoted as period, which characterizes the miimum time iterval betwee two cosecutive message geeratios. The sporadic messages of a ode N costitute a set M S = {M 1,...,MS }, ad the etire set of sporadic messages is deoted as M S := N =1 M S. Each sporadic message Mm M S has a period pm m ad a deadlie dm m, where dm m pm m.thelegth lm m (i MS) of Mm ca be computed as i [8], icludig the sigal data s m i multiples of twobyte words, the FlexRay framig overhead s m 4 bits + O F, ad the commuicatio-free DYS idle phase as l m = (s m 16 bits + s m 4 bits + O F ) τ bit /T MS. (1) Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

3 2162 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 5, JUNE 2009 Fig. 3. FlexRay software architecture. Fig. 2. (a) Message set. (b) DM schedulig. C. DS Schedulig: Issues ad Previous Work The costructio of the FC requires the offlie computatio of several system parameters. The FC duratio T c has to be chose cosiderig the badwidth ad delay requiremets of the messages that are scheduled i both the FlexRay SS ad DS, ad the duratios T SS of the SS ad T DS ofthedshave to fulfill T SS + T DS T c. Hece, it is desirable to keep T DS as small as possible to accommodate the bouds give by the SS ad the FC time, which deped o the message properties i the SS ad the DS. Furthermore, a uique assigmet of FIDs to odes has to guaratee that the messages are trasmitted before their deadlies. Sice FlexRay is fairly ew compared to the legacy bus stadards such as CAN, the literature o the DS is limited. I particular, most approaches assume that the above parameters have already bee determied ad suggest a schedulability aalysis to verify if all message deadlies are met. Cea ad Valezao [18] performed a basic aalysis of FTDMA (based o ByteFlight) ad provided guidelies to fid the maximum FID such that each message is scheduled o the bus cycle from which it arrives. Pop et al. [10] coducted a schedulability aalysis for the DS, give the above parameters. As a extesio of that work, Hagiescu et al. [11] modeled the DS usig service curves ad ivestigated service bouds for messages. The sythesis of feasible schedules is first addressed i [12], where a deadlie-mootoic (DM) approach for assigig FIDs to messages is proposed to compute T DS. I this paper, the respose times of the sporadic messages are miimized sice feasible schedules are achieved if all respose times are smaller tha the respective message deadlies. Our approach directly sythesizes feasible schedules that miimize T DS by employig the kowledge about the message deadlies. We provide a simple example to demostrate drawbacks related to the DM assigmet of FIDs to messages. Example 1: Let the odes i N = {1,...,8} commuicate usig FlexRay with T c =5ms, ad let the message set M S = {M1 1,...,M1 8 } i Fig. 2(a) be give. Here, we require that pm 1 = dm 1 for =1,...,8. Furthermore, it is give that the legth of the DS is limited to N DS = 120 MS. Assume that all messages arrive at the begiig of the DS. Cosider the arrival of messages i Fig. 2(b). Although M1 8 arrives i FC 0, it caot be trasmitted before FC 4 ad, hece, misses its deadlie. This is due to the fact that because of the repeated arrival of M1 5, M1 6, ad M1 7, there is always a message with a smaller FID to be trasmitted, while the remaiig DS is shorter tha lm 8 1. Note that this arrival sceario ca be exteded such that M1 8 is ot scheduled idefiitely. The drawback of DM schedulig is that each message is trasmitted i the first DYS where it fits: Whe M1 6 arrives (FC 4), there are two more FCs util its deadlie (FC 6). However, M1 6 is scheduled i FC 4, ad M1 8 misses its deadlie. I this paper, we propose a schedulig policy that tackles this problem. We reserve badwidth to provide DYS with certai legths ad periodic recurrece. The, messages ca be assiged to these DYS such that they are oly trasmitted i their reserved DYS. Based o this reservatio idea, our schedulig approach determies feasible schedules such that a guarateed opportuity exists for each message to be trasmitted before its deadlie while the duratio T DS ofthedsis miimized ad the duratio T c of the FC is chose, respectig the joit requiremets of the SS ad the DS, as described i Sectio III-C. It is importat to ote that the FTDMA structure of the DS is preserved. If the messages to be trasmitted i a reserved DYS are ot ready at the time of the DYS, the the DYS is oly 1 T MS log, ad the ext DYS ca start immediately. D. Software Architecture As specified i [8], the compoets of each ode are a host ad a commuicatio cotroller (CC) that are coected by a cotroller host iterface (CHI) as show i Fig. 3(a). I our framework, the host provides the schedulig fuctioality for the sporadic messages i the DS, while the CC idepedetly implemets the FlexRay protocol services. Hece, the FIDs allocated to the odes do ot directly idicate the frames to be set but idicate the odes to trasmit i a give DYS. To support reservatios, as described above, the CHI of each ode cotais a buffer for each related FID [see Fig. 3(b)]. The host implemets a periodic schedulig table (PST) per allocated FID that idicates the FCs i which the FID is to be used for message trasmissio. For each used etry i such a PST, the host maitais a correspodig priority queue (PQ) that holds the messages to be set with the respective FID ad FC sorted by icreasig deadlie. The, i each reserved DYS, the highest priority sporadic message, i.e., the message with the smallest deadlie i the PQ associated to the curret FC is assiged to the buffer of the respective FID i the CHI. 1 1 Here, we assume that ties amog messages with the same deadlie are resolved accordig to a predetermied rule. Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

4 SCHMIDT AND SCHMIDT: MESSAGE SCHEDULING FOR THE FLERAY PROTOCOL: THE DYNAMIC SEGMENT 2163 each message i M S. Furthermore, if r(m m)=(,rp,w,l) for M m M S, the rp T c dm m ad l lm m must be satisfied such that the correspodig DYS ca accommodate the legth of M m ad M m meets its deadlie. Together, our goal is to determie R ad r while optimizig the performace metrics defied i Sectio III-D. Fig. 4. (a) DYS reservatio. (b) Worst-case deadlie miss. (c) Example 1. Fig. 3 shows the software architecture for a FlexRay ode that is cofigured to schedule sporadic messages with FIDs 2, 3, ad 5 (compare Fig. 1). I the PQ, the priority icreases toward the lower part of the figure. The PST for FID 2 chooses messages from two PQs i alterate FCs, while the PSTs for FIDs 3 ad 5 provide access for messages i oe PQ i each FC. I the first FC, the sporadic messages b ad e are assiged to DYS 2 ad 5, as idicated by the solid arrows, while o message is preset for the DYS 3. The message y withalower priority tha e has to wait util e is trasmitted, while x ca be trasmitted with FID 2 i the ext FC accordig to its locatio i the PST. III. MESSAGE SCHEDULE FOR THE DS A. Schedulig Issues for the DS I our schedulig policy, each message Mm is mapped to a specific DYS, which has at least lm m reserved MS at least every time period of dm m. Cosiderig that o the oe had, Mm ca be geerated durig the dm m iterval oly oce (dm m pm m), ad o the other had, the DYS reoccurs before the deadlie of Mm, it is sufficiet to trasmit Mm durig the reserved DYS such that Mm meets the deadlie. I the followig, we formalize this idea. A reservatio R for a ode is a 4-tuple (,rp,w,l) with the reservatio period rp N, theoffset w {0,...,rp 1} ad the reservatio legth l N. I our schedulig framework, R stads for l MS that are reserved at all FCs (z rp + w), z N 0, while 1 MS is reserved i the remaiig FCs. Hece, the badwidth reservatio per FC for a give R is B R = l/rp MS. Two example reservatios for a ode are depicted i Fig. 4(a), where R 1 =(, 2, 0, 5), ad R 2 =(, 3, 1, 7). The respective badwidth reservatio per FC evaluates to B R1 =5/2MS ad B R2 =7/3MS. Each reservatio for a ode provides a recurrig DYS for at least oe message i M S. Deotig the set of all reservatios for as R ad the overall set of reservatios as R := N =1 R, this assigmet is expressed by the map r : M S R, i.e., we require that there is oly oe reservatio for B. FID Assigmet To relate the reservatio idea to the software architecture i Sectio II-D, we ote that each reservatio has to be associated to a FID ad etries i the correspodig PST with their respective PQs. I Sectio V, we preset a optimizatio approach that guaratees that every FID assigmet that obeys the followig rules ca be used to schedule the sporadic messages. Let F be the set of FIDs assiged to each ode.themap f : R F that relates each reservatio to a FID has to fulfill the followig coditios. 1) FIDs are uiquely assiged to odes as stated i [8]. 2) FID 1 is assiged. 3) FID assigmets are cosecutive. 4) The total umber of FIDs that are assiged to N odes o a give FlexRay DS is smaller tha N DS,max. As log as f satisfies 1) 4), FID assigmets are arbitrary for our schedulig policy. C. Choice of the FC Time There are costraits for choosig T c such that feasible schedules ca be costructed for ay give message set. If a give message Mm is restricted to be trasmitted i the DS, as discussed i Sectio II-B, ad if T c is chose larger tha dm m, trasmittig Mm multiple times i the same FC does ot guaratee that Mm meets the deadlie. The iterval betwee the last trasmissio of Mm i the DS of the previous FC ad the first trasmissio of Mm i the DS of the curret FC ca be loger tha dm m. Hece, it must hold that T c dm mi, where dm mi is the miimum deadlie amog all sporadic message deadlies. A additioal costrait is illustrated i Fig. 4(b), where shaded areas idicate the trasmitted messages. Assume that r(mm)=r 3 ad rp 3 = dm m.letr 1, R 2, ad R 3 R, with f (R 1 )=1, f (R 2 )=2, ad f (R 3 )=3. I the worst case, Mm arrivesatms3offci ad misses its reserved DYS by 1MS.Mm is the trasmitted i the ext FC j with the reserved DYS for Mm, i.e., j = i + rp 3. Suppose R 1 ad R 2 take up the idicated umber of MS i FC j, as show i Fig. 4(b). As f (R 1 ) <f (R 3 ) ad f (R 2 ) <f (R 3 ), the schedulig of Mm is delayed i FC j, ad Mm misses the deadlie. This deadlie miss is less tha T c, as it is due to the relative MS positios betwee arrival ad trasmissio of Mm i FC i ad j, respectively. This ca be preveted idepedet of the FID assigmets if (rp 3 +1) T c dm m. Followig this cocept, we deote the required message reservatio period for a message Mm by rpm m := dm m/t c 1. Havig determied the message reservatio period for each message i M S, a good choice for the FC time T c is the greatest Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

5 2164 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 5, JUNE 2009 commo divisor (gcd) of all message reservatio periods sice this choice eables reservatios with the maximum allowable schedulig periods. We deote this parameter as T c,ds. Furthermore, our result for the SS i [14] idicates that T c must be a iteger divisor of a parameter T c,ss. Hece, we propose the use of the FC time T c = gcd(t c,ss,t c,ds ). For otatioal coveiece, we express dm m, pm m, rpm m, ad rp i i the uits of T c for the rest of this paper. I particular, we ow have rpm m = dm m 1, i.e., it must be satisfied for a feasible schedule that if r(mm)=r, the rp rpm m. We revisit Example 1 i Fig. 4(c), where our reservatiobased schedulig is applied. Here, each message has r(m1)= i R i ad f(m1)=i i 1, with rp i = dm i 1 1 ad i =1,...,8. The shaded areas show the reservatios that are used to trasmit the messages with the arrival patter that is aalogous to Example 1. I our approach, M1 8 has a reserved DYS to be trasmitted every rp 8 =5 1=4 FCs. Although M1 7 with a smaller FID arrives i FC 2, it waits for its ext reserved DYS i FC 3. Thus, M1 8 is trasmitted before its deadlie i the reserved DYS i FC 2, ad schedulability is achieved with N DS = 120 MS. D. Performace Metrics We itroduce the cycle load L j of a FC j as the first performace metric. It deotes the maximum umber of MS that is reserved for message trasmissio i FC j for a arbitrary assigmet of FIDs, cosiderig that at most oe FID ca be assiged per message. L j icludes both the case where o message is trasmitted for a FID (duratio of 1 MS) ad the case where a message is trasmitted. Let R j Rbe the set of all reserved DYS for message trasmissio o FC j, i.e., R R j, z N 0 such that j =(z rp + w). The, L j is defied as follows: 2 L j = l+ M S 1 = S. (2) R R j R Rj R Rj(l 1)+ M As the reservatios are periodic, the reservatio patter of the DS repeats every rp FCs, where the DS reservatio period rp is the lcm of all reservatio periods. Hece, the FC with the maximum cycle load occurs withi rp FCs. We defie the maximum cycle load as L max = max j {1,...,rp} (L j ). The, we choose N DS = L max ad miimize L max i Sectio V-B to determie a feasible schedule with the shortest possible DS. The badwidth reservatio B = R R(l/rp) ad B = N =1 B idicate the umber of MS reserved per FC for each ode Nad for all of the odes, respectively. The reserved MSs are dedicated to trasmittig specific messages, ad they caot be used to trasmit ew messages. Hece, a low B for a give message set idicates the efficiecy of the badwidth reservatio ad the extesibility of the system. We propose to miimize B i Sectio V-C. 2 Here, M S deotes the umber of messages i M S. TABLE I SPORADIC MESSAGES FOR EAMPLE 2 IV. MESSAGE GROUPING A. Message Groupig: Example I our settig, the relevat timig properties of each message Mm are give as the deadlie dm m ad the period pm m, where, usually, dm m <pm m. This meas that ot ecessarily all reservatios for Mm are utilized to trasmit a message Mm. Similar to our work i [4], we propose to assig multiple messages to the same reservatio while preservig schedulability. As a result, more reservatios are utilized such that the badwidth reservatio B is miimized. The followig example message set for a ode N demostrates this approach. The properties of the sporadic message set M S = {M 1,...,M5 } are listed i Table I. Example 2: Assume that the messages M1 ad M3 i Table I have the reservatios R 1 = r(m1 )=(, 2,w 1, 40) ad R 3 = r(m3 )=(, 8,w 3, 48). I this case, a umber of (40/2) + (48/8) = 26 MS is reserved for R 1 ad R 3 per FC. Furthermore, depedig o the choice of the offsets w 1 ad w 3, it holds that either L max =48MS or that L max =88MS. Here, at least 8/rpm 1 =4DYS have to be reserved for M1 withi eight FCs to guaratee its timely trasmissio. However, at most 8/pm 1 =2messages ca be geerated. Hece, at least two reserved DYS for M1 remai uused. Alteratively, we ca assig M1 ad M3 to the same reservatio R 1 such that r(m3 )=r(m1 )=R 1 = (, mi(rpm 1,rpm 3 ),w 1, max(lm 1,lm 3 )) = (, 2,w 1, 48). For every occurrece of R 1, oly oe message is trasmitted. If M1 ad M3 are ready at the same time, we always give M1 the higher priority as rpm 1 <rpm 3. Thus, M1 is trasmitted accordig to the software architecture i Sectio II-D. Nevertheless, the trasmissio of M3 withi rpm 3 =8FCs after it is geerated is guarateed because M1 ca oly be geerated twice durig eight FCs, leavig out at least two reserved but uused DYS for the trasmissio of M3. The beefits regardig the performace metrics defied i Sectio III-D are give as follows. Whe assigig M1 ad M3 to the same reservatio R 1, their cotributio to the badwidth reservatio B decreases to max(lm 1,lm 3 )/mi(rpm 1,rpm 3 )=24 MS/T c. The maximum cycle load for R 1 ow is L max = max(lm 1,lm 3 )= 48 MS. Cosequetly, such groupig ca, ideed, be used to improve the performace metrics as defied i Sectio III-D. It ca readily be observed that there is more tha oe assigmet of reservatios that icludes M1. For example, it is possible to assig M2 ad M1 to the same reservatio as oe DYS for M1 remais uused i rpm 2 = 4 FCs. However, ot all of such multiple assigmets improve our performace metrics. Usig separate reservatios for M1 ad M2, the badwidth reservatio is (lm 1 /rpm 1 )+ Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

6 SCHMIDT AND SCHMIDT: MESSAGE SCHEDULING FOR THE FLERAY PROTOCOL: THE DYNAMIC SEGMENT 2165 (lm 2 /rpm 2 )=45MS/T c, whereas whe assigig them to the same reservatio, the total badwidth reservatio becomes max(lm 1,lm 2 )/mi(rpm 1,rpm 2 )=50 MS/T c, which icreases B. Accordigly, such groups will be excluded i the optimizatio i Sectio IV-B. B. Message Groupig: Geeral Formulatio I this sectio, the costructio of message groups, i.e., assigmets of (multiple) messages to a reservatio for a ode N is geeralized. A message group Gq M S, q N correspods to the reservatio g (Gq )=(,rp,w,l), where g (Gq )=r(mm) for all Mm Gq. Here, we determie rp = mi M m Gq (rpm m) ad lq = max M m Gq (lm m), as discussed i Sectios III-C ad D. The costructio process of a message group Gq starts with a empty group, i.e., ay message Mm ca be added to Gq. The, Gq = {Mm}, g (Gq )=(, rpm m,w,lm m). For the further discussio, we defie RMq,l ad Pq,l as metrics to check if a ew message Ml ca be added to a oempty Gq. Let g (Gq )=(,rp,w,l). RMq,l deotes the least umber of remaiig DYS that ca be used for Ml withi a time period of rpm l if all messages with smaller schedulig periods i G q are geerated as frequetly as possible ad scheduled before Ml.Wehave RMq,l = rpm l /rp rpm l /pm m. M m G q,rpm m rpm l If RMq,l 1, the M l ca be scheduled i Gq, together with the already preset higher priority messages. Pq,l is the profit i B whe addig Ml to Gq compared to the case where Ml is scheduled separately, i.e., (3) P q,l =(l/rp + lm l /rpm l ) max (lm l,l) /rp. (4) If Pq,l 0, the addig M l to Gq decreases B. Cosiderig (3) ad (4), Mm fits ito Gq if RMq,l 1 ad P q,l 0. There are already two groups {M1 } ad {M1,M3 } that iclude M1 i Example 2. Next, we costruct additioal groups with M1. Cosider G1 = {M1 } with g (G1 )=(, 2,w 1, 40). We exted the group to G2 = {M1,M3 }, with g (G2 )= (, 2,w 2, 48), as discussed above. If we cosider M4, we see that RM2,4 =1 ad P2,4 =2. A ew group is G3 = {M1,M3,M4 }, with g (G3 )=(, 2,w 3, 48). Nomoreew groups ca be formed by extedig G3 as M4 uses the last available DYS. Aother possible message to exted G2 is M5, with RM2,5 =2ad P2,5 =2. The, G4 = {M1,M3,M5 }, with g (G4 )=(, 2,w 4, 48). Similarly, the set of all possible message groups ca be determied algorithmically. Algorithm 4.1 checks if Ml ca be added to a give group Gq while all existig messages i Gq are still trasmitted withi their schedulig periods. There are three possible results. If the result is o_fit, the Ml does ot fit ito Gq. If the result is last_fit, the Ml is added to Gq,but o other messages ca be added. If the result is more_fit, the more messages ca be added after Ml.IfM l ca be added to Gq, the Algorithm 4.1 geerates the update Gq = Gq {Ml } ad adds the ew Gq to the set G of all message groups for ode. Algorithm 4.1 (Check ad Add) Iput: M l, G q, G. Iit: result = more_fit if (P q,l < 0 or RM q,l < 1) result = o_fit else G q := G q {M l } ad G = G G q if (RM q,l =1) result = last_fit retur result Algorithm 4.2 uses Algorithm 4.1 to eumerate all possible groups Gq M S. Here, the ordered list LM S of all messages sorted by icreasig deadlie is used. Two operators ext(ml ) ad last(lm S) retur the message followig Ml ad the last message i LM S, respectively. The compariso Mk <M l returs true if Mk is located before M l i LMS. Before Algorithm 4.2 is ru for Mm M S, G q, G, ad Mc are iitialized to {Mm}, {{Mm}}, ad Mm, respectively. Algorithm 4.2 (Group) Iput: LM S, Mc, Gq, G (while Mc < last(lm S)) Mc = ext(mc )) tempgq = Gq result = Check ad Add(Mc,tempGq, G ) if (result = more_fit ad Mc last(lm S)) tempmc = Mc Group (LM S, tempmc, tempgq, G ) The messages i LM S are checked to fit i Gq.Ayremaiig capacity is idicated whe Check ad Add returs a more_fit value. I this case, a ew group is formed, which exteds Gq with the remaiig of the list of messages by ruig Group recursively. We apply Algorithm 4.2 to our example message set i Table I ad costruct the message groups for M1 with the schedulig period rpm 1 =2. The ordered list evaluates to LM S = M1,M2,M3,M4,M5. The step-by-step evaluatio of the algorithm is depicted i Table II. The etire set G obtaied by applyig Algorithm 4.2 for the rest of the messages i M S is G = {{M1 }, {M1,M3 }, {M1,M3,M4 }, {M1,M3,M5 }, {M1,M4 }, {M1,M4,M5 }, {M1,M5 }, {M2 }, {M2,M4 }, {M2,M5 }, {M3 }, {M4 }, {M5 }}. The set G has the followig characteristics that eed to be addressed. First, there are multiple groups that, for example, cotai the message M 1. However, i the fial schedule, exactly oe reservatio for M 1 is required. Thus, oe out of these groups has to be selected for trasmittig M 1. Secod, for each group G q G, the offset of the correspodig reservatio Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

7 2166 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 5, JUNE 2009 TABLE II MESSAGE GROUP CONSTRUCTION FOR M1 IN EAMPLE 2 TABLE III MESSAGE GROUPS FOR EAMPLE 3 depedig o the offset w, the cycle load of certai FCs is icreased by l. We propose a iteger programmig approach to determie G S ad w for the reservatio of each G q G S such that B ad L max are miimized, as discussed i Sectio III-D. We illustrate our optimal message schedulig approach with the followig example: Example 3: Let N = {1, 2}. We assume that the groups i G have bee computed as listed i Table III by ruig Algorithm 4.2 o a give set of messages M S. Our goal is ow to determie G S ad the offsets w i of the correspodig reservatios such that L max (ad, thus, the required duratio T DS of the DS) is miimized. g (G q ) has ot bee determied yet. I Sectio V, we employ the eumeratio of all possible message groups, as derived above, to fid the choice of groups ad reservatio offsets that optimizes the performace metrics i Sectio III-D. V. O PTIMAL SCHEDULING OF MESSAGES The schedule for M S establishes the umber of reserved DYS for each Mm M S. As we discussed i Sectio III-A, FIDs ca be assiged to reservatios arbitrarily. Hece, the parameters left to be determied are the selectio of message groups to be used ad the offsets for correspodig reservatios while miimizig the cycle load. A message Mm ca be icluded i multiple groups i G. Amog these groups, oe Gq G, with Mm Gq,mustbe selected such that oe ad oly oe reservatio for Mm is icluded i the schedule. Let G := N =1 G deote the set of all groups, ad let G S G deote the set of selected groups. Oce Gq is selected, the reservatio (,rp,w,l):=g (Gq ) is allocated to ode to trasmit messages i Gq, with a cotributio of l/rp to the badwidth reservatio B. Furthermore, A. Exact Formulatio We formulate iteger programmig problems with two compoets to fid the optimal message schedule. The first compoet addresses the selectio of the message groups ad the correspodig reservatios. The biary decisio variable g i {0, 1} takes the value of 1 if G i G S ad is 0 otherwise. The secod compoet is determiig the reservatio offsets. A reservatio R i ca have a offset w i {0...rp i 1}. The biary decisio variable x i,k {0, 1} takes the value of 1 if w i = k ad is 0 otherwise, where k =0,...,rp i 1. Furthermore, it ca readily be observed that the reservatio patter repeats after G RP FCs, where G RP = lcm Gi Grp i is the least commo multiple of the reservatio periods rp i correspodig to each group G i G. Hece, oly the FCs 0,...,G RP 1 eed to be take ito accout. Cosider a reservatio R i that correspods to G i G.The cotributio of R i to L j, j =1,...,G RP 1, is give as follows. 1) g i =0: The, G i G S ad R i does ot add to L j. 2) g i =1ad x i,k =1for k = j mod rp i : The, w i = k ad l i MS are reserved for R i i L j. 3) g i =1ad x i,k =0for k = j mod rp i : The, w i k ad oe MS is reserved for R i i L j. Accordigly, we ca express the cycle load L j as follows: L j = G i G g i (x i,k l i +(1 x i,k ) 1) (5) where k = j mod rp i. Ay FC j {0,...,G RP 1} ca have the maximum cycle load. Assumig without loss of geerality Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

8 SCHMIDT AND SCHMIDT: MESSAGE SCHEDULING FOR THE FLERAY PROTOCOL: THE DYNAMIC SEGMENT 2167 that L max = L 0, the expressio to be miimized is mi L 0 = mi g i x i,0 l i + g i (1 x i,0 ) 1 (6) G i G where is a vector with all variables g i ad x i,k.therequiremet that oly oe reservatio is selected for each Mm ad exactly oe offset is determied for each used reservatio is formulated by the costraits i (7) ad (8), respectively. Sice L max = L 0, there are o FCs j {1,...,G RP 1}, with L j >L 0. This costrait is stated i Mm, g i =1 (7) i,mm G i for i =1,..., G, rp i 1 k=0 x i,k = g i (8) for j =1,...,G RP 1, L j L 0. (9) The optimal message schedule is the solutio of the optimizatio problem with the objective fuctio i (6) ad the costraits i (7) (9). Note that it is a oliear iteger programmig problem (NIP), as the computatios i (5) ad (9) cotai products of the decisio variables g i ad x i,k. A feasible schedule for Example 3 has bee computed usig the Tomlab optimizatio eviromet [19]. As a result, g 2 = g 4 = g 9 = g 10 = g 11 =1 ad g i =0 for the remaiig values of i have bee foud. Hece, G S = {{M 1 1,M 1 2 }, {M 1 3 }, {M 2 1,M 2 4 }, {M 2 2 }, {M 2 3 }}. Furthermore, the correspodig offsets are w 2 =0, w 4 =2, w 9 =1, w 10 = 0, w 11 =1, ad the worst-case maximum cycle load is L max = 81 MS. Respectig the coditios i Sectio III-B, the FIDs ca ow be assiged to the reservatios of the selected message groups arbitrarily. However, a further aalysis of the NIP solutio ca reduce the umber of required FIDs ad, hece, lead to a shorter DS by assigig multiple reservatios of oe ode that ever appear i the same FC to the same FID. 3 A efficiet FID assigmet for Example 3 is f 1 (R 2 )=1, f 1 (R 4 )=2, f 2 (R 9 )=3, f 2 (R 11 )=4, ad f 2 (R 10 )=3,as show i Fig. 5(a). Here, the reservatios R 9 ad R 10 of ode 2 ca be assiged to the same FID, istead of choosig a separate FID such as f 2 (R 10 )=5, as they ever occupy the same FC. Hece, the two uused MSs that would appear i the logest FC (with R 2 ad R 10 ) i the latter case ca be elimiated. The resultig cycle load is reduced from 81 to 79 MS. Fig. 5(b) depicts the correspodig software architecture accordig to Sectio II-D. B. Two-Step Formulatio As solvig the offlie NIP is a hard problem, eve for small message sets, we propose to decompose the formulatio i (6) (9) ito two liear biary iteger programmig problems (BIPs) to eable the problem solutio also for large message Fig. 5. Optimal reservatios for Example 3. (a) NIP. (b) Software architecture. (c) Decomposed BIP. sets. I the first step, we suggest that the groups to be scheduled are selected such that our performace metric B is miimized. I the secod step, the offsets for these selected groups are computed to miimize L max. The followig objective fuctio is used to miimize B 4 : mi subject to the costraits B = mi M m M S, g i (l i /rp i ) (10) G i G i,m m G i g i =1. (11) Completig the first step yields G S G, where G i G S g i =1. I the followig step, L max is miimized. Here, we deote the umber of groups i G S as G S ad the lcm of their reservatio periods as G S,RP.Wehave L max = mi L 0 = mi subject to the costraits G i G S x i,0 l i +(1 x i,0 ) 1 (12) for j =1,...,G S,RP 1:L j L 0 (13) for i =1,...,G S : rp i 1 k=0 x i,k =1. (14) The BIP i (10) ad (11) has bee solved for Example 3 usig Tomlab [19]. As a result, g 2 = g 4 = g 8 = g 10 = g 12 =1 ad g i =0 for the remaiig values of i. Hece, G S = {{M1 1,M2 1 }, {M3 1 }, {M1 2,M3 2 }, {M2 2 }, {M4 2 }}. Note that, differet from the NIP solutio, M3 2 is grouped with M1 2 i G 8 sice the resultig badwidth reservatio B =55.1 MS is smaller tha for the NIP solutio with B = 55.7 MS. I the ext step, we solve the BIP i (12) ad (13) for Example 3. x 2,1, x 4,0, x 8,0, x 10,1, ad x 12,0 are foud to be 1, while the rest of the x i,k is 0. The worst-case maximum cycle load is L max =84ad ca be reduced to 82 MS by a efficiet FID assigmet, as depicted i Fig. 5(c). Although the resultig 3 A algorithmic approach to tackle such FID assigmet has bee developed but is ot i the scope of this paper. 4 Note that the fractioal coefficiets of the objective fuctio ca be coverted ito itegers by multiplyig B with G RP. Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

9 2168 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 5, JUNE 2009 TABLE IV SPORADIC MESSAGES OF THE SAE BENCHMARK DS is slightly larger compared to the NIP solutio, ow oly two BIPs have to be solved. Together, it ca be stated that there is a possible tradeoff betwee the badwidth reservatio B ad the maximum cycle load L max. Furthermore, the decomposed optimizatio ca both provide a upper boud for the miimum L max ad a good iitial feasible solutio for ruig the NIP. VI. EPERIMENTAL STUDY I this sectio, we preset a compariso of the NIP solutio i Sectio V-B ad the BIP solutio i Sectio V-C. Furthermore, we study the schedule costructio for large message sets. I all our experimets, we used the CPLE solver of Tomlab [19] to obtai the iteger programmig solutios, ad for each data poit, te sample rus have bee evaluated o a persoal computer with a dual-core Petium GHz processor ad 1 GB of radom access memory. Fig. 6. Cycle load. (a) SAE message set ad (b) costructed message set. Optimizatio time. (c) SAE message set ad (d) costructed message set. A. Society of Automotive Egieers (SAE) Bechmark Set The SAE bechmark set i [15] comprises 31 sporadic messages with data sizes smaller tha 8 bits whose deadlies ad periods are iteger multiples of 5 ms ad are trasmitted by five odes (see Table IV). With the choice of T MS =6.0μs ad O F =90bits (see [8]), each message fits ito a frame with ( ) 0.1/6.0 MS =3MS. For the SAE bechmark set, the NIP ad BIP formulatios, i combiatio with the efficiet FID assigmet, yield the same result of T DS =26T MS = 156 μs. B. Compariso Betwee the NIP ad the BIP Solutios Our compariso betwee the solutios of the NIP formulatio i Sectio V-B ad the BIP formulatio i Sectio V-C is based o the SAE message set i Table IV that represets a practical message set ad the example message set i Table III that was costructed to illustrate the potetial differet solutios for the NIP ad the BIP. As i the SAE message set, we employ five odes commuicatig o the FlexRay bus. Sice the umber of messages i both sets is ot sufficiet to geerate cosiderable traffic o FlexRay, we exted these sets by radomly choosig messages from the respective set ad assigig them to oe of five FlexRay odes util a give arrival rate is reached. Here, for a give message set M S, we deote the arrival rate as Mm M S lm/dm m. For both message sets, the NIP ca be solved for up to 16 messages i the DS, while the solver fails to fid a optimal solutio for larger message sets. Fig. 6(a) (SAE messages) ad Fig. 6(c) (costructed message set) plot the maximum cycle load L max, as computed i (12) agaist the arrival rate. It ca be see that the NIP ad BIP formulatios yield the same optimizatio Fig. 7. Large message sets. (a) DS legth. (b) Effect of groupig. results i all our experimets. This suggests that although cases exist where the BIP does ot give a optimal solutio (compare Example 3), BIP is suitable i practical examples. Moreover, solvig the BIP is much less computatioally expesive, as illustrated i Fig. 6(b) ad (d). C. Message Schedulig for Large Message Sets Further experimets were carried out to evaluate the BIP approach for larger message sets. Fig. 7(a) shows that more tha 270 messages of the SAE message set ca be scheduled i a DS with T DS 336 T MS =2.0 ms, while computatio timesoflesstha1harerequired. The beefit of the groupig idea is preseted i Fig. 7(b) by comparig the badwidth reservatio B [see (10)] eeded to schedule idividual messages ( idividual ) to the badwidth reservatio with groupig ( grouped ). I all cases, the badwidth required to schedule the give message set is reduced by about 20%. D. Discussio It has to be oted that allowig arbitrary FID assigmets i the optimizatio accordig to Sectio III-B potetially leads to suboptimal badwidth reservatios. However, it is readily Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.

10 SCHMIDT AND SCHMIDT: MESSAGE SCHEDULING FOR THE FLERAY PROTOCOL: THE DYNAMIC SEGMENT 2169 observed that determiig a globally optimal FID assigmet is computatioally itractable sice its computatioal complexity is eve higher tha the NIP formulated i Sectio V-B, which ca oly be solved for small message sets, as show i Sectio VI-B. I this respect, the decompositio of the NIP ito two BIP eables the schedule costructio for large messages sets, as described i Sectio VI-C, while the experimets i Sectio VI-B idicate that for practical message sets, the NIP ad BIP formulatios lead to idetical results. Hece, our BIP approach geerates feasible schedules for large message sets, while esurig a miimal badwidth reservatio ad the shortest possible duratio T DS of the DS. VII. CONCLUSION This paper has addressed the message schedule costructio for sporadic real-time messages that are to be trasmitted i the DS of the FlexRay protocol. Our approach proposes to reserve badwidth such that each sporadic message ca meet its deadlie. Based o a formal descriptio of the schedulig problem, we determie a NIP to compute a optimal message schedule. Here, the badwidth reservatio ad the cycle load are employed as appropriate performace metrics that have to be miimized. To facilitate the problem solutio, we suggest a decompositio of the NIP ito two BIPs to approximate the optimal result. First, we fid a set of reservatios that miimizes the badwidth reservatio, ad the, we schedule the obtaied reservatios such that the cycle load is miimized. The performace of the proposed approach was evaluated i a experimetal study. It is verified that the NIP ad BIP formulatios yield idetical results for practical message sets. Furthermore, it was possible to costruct feasible schedules for large message sets. Together, our approach eables the algorithmic computatio of a optimal schedule for the sporadic messages i the FlexRay DS. REFERENCES [1] A. Albert, Compariso of evet-triggered ad time-triggered cocepts with regard to distributed cotrol systems, i Proc. Embedded World, 2004, pp [2] N. Navet, Y. Sog, F. Simoot-Lio, ad C. Wilwert, Treds i automotive commuicatio systems, Proc. IEEE, vol. 93, o. 6, pp , Ju [3] CAN i Automatio, May [Olie]. Available: org/ca/ [4] TTCAN, May [Olie]. Available: ca/ttca/ [5] Time-Triggered Protocol TTP/C, High-Level Specificatio Documet, Nov protocol ver [Olie]. Available: [6] H. Kopetz ad G. Bauer, The time-triggered architecture, Proc. IEEE, vol. 91, o. 1, pp , Ja [7] J. Berwager, M. Peller, ad R. Griegbach, A New High Performace Data Bus System for Safety Related Applicatios, [Olie]. Available: [8] FlexRay Commuicatio System, Ju Protocol specificatio, ver [Olie]. Available: [9] R. Makowitz ad C. Temple, FlexRay A commuicatio etwork for automotive cotrol systems, i Proc. IEEE It. Workshop Factory Commu. Syst., 2006, pp [10] T. Pop, P. Pop, P. Eles, Z. Peg, ad A. Adrei, Timig aalysis of the FlexRay commuicatio protocol, i Proc. 18th Euromicro Cof. Real-Time Syst., 2006, pp [11] A. Hagiescu, U. D. Bordoloi, S. Chakraborty, P. Sampath, P. V. V. Gaesa, ad S. Ramesh, Performace aalysis of FlexRay-based ECU etworks, i Proc. Des. Autom. Cof., 2007, pp [12] P. Pop, P. Eles, ad Z. Peg, Bus access optimizatio for distributed embedded systems based o schedulability aalysis, i Proc. Cof. Des., Autom. Test Eur., 2007, pp [13] K. Schmidt ad E. G. Schmidt, Systematic message schedule costructio for time-triggered CAN, IEEE Tras. Veh. Techol., vol. 56, o. 6, pp , Nov [14] K. Schmidt ad E. G. Schmidt, Message schedulig for the FlexRay protocol: The static segmet, IEEE Tras. Veh. Techol., vol. 58, o. 5, pp , Ju [15] Class C Applicatio Requiremet Cosideratios, SAE Hadbook, vol. 2, SAE, Warredale, PA, SAE Paper J2056/1 Jue 93, 1994, pp [16] K. Tidell, A. Burs, ad A. J. Welligs, Calculatig cotroller area etwork (CAN) message respose times, Cotrol Eg. Pract., vol. 3, o. 8, pp , Aug [17] J. Gamiz, J. Samitier, J. Fuertes, ad O. Rubies, Practical evaluatio of messages latecies i CAN, i Proc. IEEE Cof. Emergig Techol. Factory Autom., Sep. 2003, vol. 1, pp [18] G. Cea ad A. Valezao, O the properties of the flexible time divisio multiple access techique, IEEE Tras. Id. Iformat., vol. 2, o. 2, pp , May [19] Tomlab, Jul [Olie]. Available: Ece Gura Schmidt received the B.S. degree i electrical ad electroics egieerig from the Middle East Techical Uiversity, Akara, Turkey, i 1997 ad the M.S. ad Ph.D. degrees i electrical ad computer egieerig from Caregie Mello Uiversity, Pittsburgh, PA, i 2001 ad 2004, respectively. She is curretly a Faculty Member with the Departmet of Electrical ad Electroics Egieerig, Middle East Techical Uiversity. Her research iterests iclude high-speed etworks, etworked cotrol systems, ad vehicular commuicatio etworks. Klaus Schmidt received the Diploma ad Ph.D. (with distictio) degrees i electrical, electroic, ad commuicatio egieerig from the Uiversity of Erlage-Nuremberg, Erlage, Germay, i 2002 ad 2005, respectively. He is curretly a Postdoctoral Researcher ad the Chair of Automatic Cotrol with the Uiversity of Erlage-Nuremberg. His research iterests iclude cotroller sythesis for discrete evet systems, etworked cotrol systems, ad vehicular commuicatio etworks. Authorized licesed use limited to: ULAKBIM UASL - MIDDLE EAST TECHNICAL UNIVERSITY. Dowloaded o Jue 11, 2009 at 11:15 from IEEE plore. Restrictios apply.