Performance and Reliability Evaluation for DSRC Vehicular Safety Communication. Xiaoyan Yin

Transcription

1 Performance and Relablty Evaluaton for DSRC Vehcular Safety Communcaton by Xaoyan Yn Department of Electrcal and Computer Engneerng Duke Unversty Date: Approved: Kshor S. Trved, Supervsor Benjamn C. Lee Jeffrey H. Derby Loren W. Nolte Xaomn Ma Dssertaton submtted n partal fulfllment of the requrements for the degree of Doctor of Phlosophy n the Department of Electrcal and Computer Engneerng n the Graduate School of Duke Unversty 213

2 ABSTRACT Performance and Relablty Evaluaton for DSRC Vehcular Safety Communcaton by Xaoyan Yn Department of Electrcal and Computer Engneerng Duke Unversty Date: Approved: Kshor S. Trved, Supervsor Benjamn C. Lee Jeffrey H. Derby Loren W. Nolte Xaomn Ma An abstract of a dssertaton submtted n partal fulfllment of the requrements for the degree of Doctor of Phlosophy n the Department of Electrcal and Computer Engneerng n the Graduate School of Duke Unversty 213

3 Copyrght by Xaoyan Yn 213

4 Abstract Inter-Vehcle Communcaton (IVC) s a vtal part of Intellgent Transportaton System (ITS), whch has been extensvely researched n recent years. Dedcated Short Range Communcaton (DSRC) s beng serously consdered by automotve ndustry and government agences as a promsng wreless technology for enhancng transportaton safety and effcency of road utlzaton. In the DSRC based vehcular ad hoc networks (VANETs), the transportaton safety s one of the most crucal features that needs to be addressed. Safety applcatons usually demand drect vehcle-to-vehcle ad hoc communcaton due to a hghly dynamc network topology and strct delay requrements. Such drect safety communcaton wll nvolve a broadcast servce because safety nformaton can be benefcal to all vehcles around a sender. Broadcastng safety messages s one of the fundamental servces n DSRC. In order to provde satsfactory qualty of servces (QoS) for varous safety applcatons, safety messages need to be delvered both tmely and relably. To support the strngent delay and relablty requrements of broadcastng safety messages, researchers have been seekng to test proposed DSRC protocols and suggestng mprovements. A major hurdle n the development of VANET for safety-crtcal servces s the lack of methods that enable one to determne the effectveness of VANET desgn mechansm for predctable QoS and allow one to evaluate the tradeoff between network parameters. Computer smulatons v

5 are extensvely used for ths purpose. A few analytc models and experments have been developed to study the performance and relablty of IEEE 82.11p for safety-related applcatons. In ths thess, we propose to develop detaled analytc models to capture varous safety message dssemnaton features such as channel contenton, backoff behavor, concurrent transmssons, hdden termnal problems, channel fadng wth path loss, mult-channel operatons, mult-hop dssemnaton n 1-Dmentonal or 2- Dmentonal traffc scenaros. MAC-level and applcaton-level performance metrcs are derved to evaluate the performance and relablty of message broadcastng, whch provde nsghts on network parameter settngs. Extensve smulatons n ether Matlab or NS2 are conducted to valdate the accuracy of our proposed models. v

6 Contents Abstract... v Lst of Tables... xv Lst of Fgures... xv Acknowledgements... xx 1. Introducton Overvew of DSRC Safety Communcaton Research Problems Addressed Contrbutons of the Dssertaton Outlne of the Dssertaton Background DSRC-related background MAC Layer Protocol Descrpton Modelng Methods Markov Models Sem-Markov Process Model Queung Models Fxed-pont Iteraton Method Broadcast Safety Messages Evaluaton Motvaton System Assumptons v

7 3.3 Analytc Models SMP Model Servce Tme Computaton Fxed-pont Equaton Exstence, Unqueness and Convergence of Fxed-pont Iteraton Exstence Unqueness Convergence Performance Indces Mean Transmsson Delay Packet Delvery Rato Packet Recepton Rato Numercal and Smulaton Results Numercal Vs. Smulaton Results Comparson wth Prevous Models Impact Comparson between Concurrent Transmsson and Hdden Termnals Conclusons and Future Work Perodc Beacon Messages Evaluaton Motvaton System Assumptons Analytc Models v

8 4.3.1 Overall Model SMP Model Servce Tme Computaton Fxed-pont Iteraton Performance Indces MAC-level Performance Metrcs Mean Transmsson Delay Packet Delvery Rato Packet Recepton Rato Normalzed Channel Throughput Applcaton-level Performance Metrcs Node Recepton Probablty T-wndow Relablty Applcaton-level Delay Awareness Probablty Average Number of Invsble Neghbors Numercal Results Numercal Results for MAC-level Performance Metrcs Smulaton Descrpton Analytc Vs. Smulaton Results Compare wth Prevous Models Analytc-Numercal Results for Applcaton-level Performance Metrcs v

9 Analytc-numercal Results for Fxed Network Parameters Analytc-numercal Results for Dfferent Network Parameters VANET Applcatons Evaluaton Applcaton Requrements Case Studes for VANET Applcatons Emergency Vehcle Warnng Slow Vehcle Indcaton Rear-end Collson Warnng Conclusons Multple Types of Servces Evaluaton Motvaton System Descrpton and Assumptons Analytc Models SMP Model for AC Servce Servce Tme Computaton Fxed-pont Iteraton MAC-level Performance Metrcs Mean transmsson delay Varance of the transmsson delay Packet Delvery Rato (PDR) Packet Recepton Rato (PRR) Numercal Results x

10 5.5.1 Influence of Packet Arrval Rate Influence of Packet Length Influence of Backoff Wndow Sze Influence of Channel Sensng Tme Influence of Carrer Sensng Range Hdden termnal Vs. Concurrent Transmsson One Vs. Multple Types of Servces Preemptve Prorty Strct Prorty Varance of Transmsson Delay GI/G/1 Queue Extenson SMP Model Servce Tme Computaton Fxed-pont Iteraton MAC-level Performance Metrcs Mean Transmsson Delay Varance of Transmsson Delay PDR PRR Numercal Results A Case Study Conclusons x

11 6. Mult-channel Operaton Evaluaton Motvaton System Assumptons Analytc Models Overall Method Descrpton SMP Model Servce Tme Computaton Fxed-pont Iteraton Performance Metrcs MAC-level Performance Metrcs Mean Transmsson Delay Node Recepton Probablty (NRP) Packet Recepton Rato (PRR) Packet Delvery Rato (PDR) Applcaton-level Performance Metrcs Applcaton-level Delay T-wndow Relablty Awareness Probablty Average Number of Invsble Neghbors Numercal Results Smulaton Descrpton Numercal Results x

12 6.5.3 Impacts of Channel Swtchng and Channel Fadng Conclusons Mult-hop Dssemnaton Evaluaton Motvaton System Descrpton and Assumptons Analytc Models Channel Fadng Model Rebroadcast Probablty Message-centrc rebroadcast probablty Recever-centrc rebroadcast probablty Average Rebroadcast Dstance Average Number of Hops to Reach a Dstance Average Rebroadcast Delay Average Delay to Reach a Dstance Metrcs Related to Message Vanshng Numercal Results Smplfcaton of f(x,m) m s a constant over [, d] m s a pecewse functon over [, d] Conclusons Two-Dmensonal Network Evaluaton Motvaton x

13 8.2 System Assumptons Analytc Models SMP Model Servce Tme Computaton Fxed-Pont Iteraton Performance Metrcs Mean Transmsson Delay Packet Delvery Probablty Packet Recepton Rato Numercal Results Conclusons Summary Bblography Bography x

14 Lst of Tables Table 1.1: Summary of chapters Table 3.1: DSRC communcaton parameter Table 3.2: Mean delay E[D] comparsons Table 3.3: PDR comparson Table 3.4: PRR comparson Table 4.1: Network parameter settngs Table 5.1: DSRC parameters for EDCA mechansm Table 5.2: Parameters for packet arrval rate nfluence evaluaton Table 5.3: Parameters for packet length nfluence evaluaton Table 5.4: D1 for backoff wndow sze Table 5.5: D2 for backoff wndow sze Table 5.6: D1 for channel sensng tme Table 5.7: D2 for channel sensng tme Table 5.8: D1 for carrer sensng range Table 5.9: D2 for carrer sensng range Table 5.1: Parameters for relablty factors evaluaton Table 5.11: Parameters for one type of servce Table 5.12: Parameters for multple types of servces usng EDCA Table 5.13: Parameters for preemptve prorty Table 5.14: Parameters for varance computaton xv

15 Table 5.15: Parameters for GI/G/1 queue Table 5.16: Parameters for GI/G/1 queue Table 5.17: Safety applcatons over the control channel Table 5.18: Parameters settng for case study Table 6.1: Input parameter settngs Table 7.1: Network parameter Table 7.2: Non dstance-based metrcs Table 8.1: DSRC communcaton parameter xv

16 Lst of Fgures Fgure 1.1: U.S. DSRC channel allocaton... 2 Fgure 1.2: Protocol stack for DSRC communcaton... 3 Fgure 2.1: Flow chart of DCF functon Fgure 3.1: Models capturng nteractons between vehcles Fgure 3.2: SMP model for 82.11p broadcast Fgure 3.3: SMP model for servce tme computaton Fgure 3.4: Abstracton of the packet transmsson tme Fgure 3.5: Import graph for fxed pont teraton Fgure 3.6: PRR computaton Fgure 3.7: Mean delay... 6 Fgure 3.8: PDR... 6 Fgure 3.9: PRR... 6 Fgure 3.1: Mean delay comparson Fgure 3.11: PDR comparson Fgure 3.12: PRR comparson Fgure 3.13: Impact on PDR Fgure 3.14: Impact on PRR Fgure 4.1: Model capturng nteractons between vehcles Fgure 4.2: Import graph for the overall method Fgure 4.3: SMP model for 82.11p beacon broadcast xv

17 Fgure 4.4: SMP model wth an absorbng state Fgure 4.5: Conceptual servce tme dstrbuton Fgure 4.6: Channel sensng deference Fgure 4.7: Node recepton probablty computaton Fgure 4.8: Smulaton flow chart Fgure 4.9: Mean transmsson delay Fgure 4.1: Packet delvery rato Fgure 4.11: Packet recepton rato Fgure 4.12: Normalzed channel throughput Fgure 4.13: Comparson of mean delay Fgure 4.14: Comparson of PDR Fgure 4.15: Comparson of PRR Fgure 4.16: Node recepton probablty (NRP) and awareness probablty (PA) wth dfferent packet requrement n Fgure 4.17: Applcaton-level delay Fgure 4.18: Average no. of nvsble neghbors Fgure 4.19: Node recepton probablty (NRP) Fgure 4.2: Awareness probablty PA(n=3) Fgure 4.21: Applcaton-layer delay Fgure 4.22: Number of nvsble neghbors Fgure 4.23: Emergency vehcle warnng applcaton-level metrcs wth parameters Rd=24, τ=.2, PL= xv

18 Fgure 4.24: Slow vehcle ndcaton applcaton-level metrcs wth parameters Rd=24, τ=.2, PL= Fgure 4.25: Rear-end collson avodance applcaton layer metrcs wth parameters Rd=24, τ=.2, PL= Fgure 5.1: SMP model for AC message Fgure 5.2: SMP model wth absorbng state for servce tme computaton Fgure 5.3: Influence of packet arrval rate Fgure 5.4: Influence of packet length Fgure 5.5: Influence of backoff wndow sze Fgure 5.6: Influence of channel sensng tme AIFS Fgure 5.7: Influence of carrer sensng range Fgure 5.8: Hdden termnal Vs. concurrent transmsson Fgure 5.9: One Vs. multple types of servces Fgure 5.1: Preemptve prorty results Fgure 5.11: Strct prorty results Fgure 5.12: Varance of the transmsson delay Fgure 5.13: Output measures for GI/G/1 queue Fgure 5.14: Output measures for M/G/1 queue Fgure 5.15: Applcatons wth lower bound packet arrval rates Fgure 5.16: Applcatons wth upper bound packet arrval rates Fgure 6.1: Import graph for the overall method Fgure 6.2: SMP model for a BSM transmsson Fgure 6.3: A BSM transmsson durng CCH nterval xv

19 Fgure 6.4: DIFS channel sensng durng CCH nterval Fgure 6.5: MAC-level mean transmsson delay Fgure 6.6: Packet transmsson relablty Fgure 6.7: Awareness probablty wth dfferent packet requrements Fgure 6.8: Applcaton-level delay Fgure 6.9: Average no. of nvsble neghbors... 2 Fgure 6.1: Impacts of channel swtchng and channel fadng on NRP Fgure 7.1: Message rebroadcast n a hop Fgure 7.2: NRP and recever-centrc rebroadcast probablty Fgure 7.3: Number of hops to reach a dstance Fgure 7.4: Average transmsson delay Fgure 7.5: Change order of double ntegral Fgure 7.6: Fadng parameter as a pecewse functon Fgure 8.1: Hdden termnals problem for 1-D network Fgure 8.2: Hdden termnals problem for 2-D network Fgure 8.3: SMP model for IEEE broadcast Fgure 8.4: Embedded DTMC of the SMP model for the servce tme Fgure 8.5: 2-D MANET model for performance analyss Fgure 8.6: Illustraton of S1 area calculaton Fgure 8.7: Illustraton of S2 area calculaton Fgure 8.8: PRR wth W= Fgure 8.9: Mean transmsson delay wth W= xx

20 Fgure 8.1: Impact of Nakagam fadng on PRR of DSRC broadcast wth network parameters W=15, λ=1 packets/s, Rd=24Mbps Fgure 8.11: PRR and PDP of DSRC broadcast wth network parameters W=15, Rd=24Mbps, β=1/(πr 2 ) Fgure 8.12: PDP wth network parameters λ=1 packets/s, W=15, Rd=24Mbps, β=1/(πr 2 ) Fgure 8.13: Impact of Communcaton range on PRR of DSRC broadcast wth network parameters W=15, λ=1 packets/s, Rd=24Mbps xx

21 Acknowledgements I would lke to express my deepest grattude to my advsor, Dr. Kshor S. Trved, for hs gudance, patence, and encouragement durng my research at Duke Unversty. Dr. Kshor S. Trved s a brllant, extraordnary, and nsprng professor, who taught me nvaluable research sklls, and helped me overcome many dffcultes. He s always there for me whenever I needed advce. The completon of ths dssertaton would be mpractcal wthout hs contnuous academc and passonate support. I wll also forever be thankful to Prof. Xaomn Ma, who gave me nsghtful suggestons and nvaluable comments on my work. Durng our research cooperaton, he has been extremely helpful n provdng hs scentfc advce and knowledge through many frutful meetngs and dscussons. I sncerely treasure the collaboraton experences wth Prof. Ma. I would lke to thank my commttee members durng my research work, Dr. John A. Board, Dr. Krshnendu Chakrabarty, Dr. Loren W. Nolte, and Dr. Matt Reynolds for ther kndness to serve on my commttee and provde constructve suggestons to my research. I would lke to thank the generous fnancal support for my research work from Natonal Scence Foundaton, Duke Unversty Graduate School and Electrcal and Computer Engneerng Department, General Motors and NEC Corporaton. xx

22 I thank my colleagues at Duke Unversty, Javer Alonso, Ermeson Andrade, Rahul Ghosh, DongSeong Km, Jae Shk Lm, Amta Devaraj, Fumo Machda, Arpan Roy, Ruofan Xa, and Yang Zhao for ther generous help and the wonderful workng envronment that they created. I would lke to thank my farther Huasheng Yn, mother Zhyng Lu, elder sster Dayan Yn and younger brother Jnzhong Yn for ther uncondtonal love and encouragement durng the pursut of my PhD degree. I am deeply grateful to my husband Le Zhang, who has been a true and great supporter durng my good and bad tmes. It s also a great pleasure and a wonderful lfe experence to have my beloved daughter Jen Zhang born durng ths unque perod of tme. xx

23 1. Introducton 1.1 Overvew of DSRC Safety Communcaton Vehcle safety s an mportant ssue for our socety. Although severty ameloraton technologes such as ar bags, seat belts, and automatc brakng system (ABS) have been appled for years to provde passve protecton to vehcle occupants, nearly 6.2 mllon polce-reported motor vehcle crashes occur annually n the Unted States (.e., one every 5 seconds). On the average, a person s njured n a polce-reported motor vehcle crash every 12 seconds, and someone s klled every 12 mnutes. The related economc loss due to crashes s $23.6 bllon annually. The development of Intellgent Transportaton System (ITS) [113] s to progress towards safe and smooth drvng wthout excessve delay. Vehcular ad hoc network (VANET) s one of the key enablng technologes n ITS. Dedcated Short Range Communcaton (DSRC) rado technology beng standardzed as IEEE 82.11p [1][112] s projected to support lowlatency wreless data communcatons between vehcles and from vehcles to roadsde unts. Such a communcaton technology s beng serously consdered by automotve ndustry and government agences, and the rado devces are expected to be nstalled n future vehcles and work wth sensors for enhancng transportaton safety and effcency of road utlzaton. Currently, DSRC s under actve development n the Unted States, Europe, Japan and other countres. Varous safety and non-safety applcatons wll be enabled 1

24 through nformaton exchange usng Vehcle-to-Vehcle (V2V) communcaton, and Vehcle-to-Infrastructure (V2I) communcaton. Compared to non-safety applcatons (e.g., toll collecton, traffc ndcaton, commercal servces), safety applcatons (e.g., collson avodance, emergency vehcle warnng, slow vehcle ndcaton) are more crtcal to prevent collsons on the road and hence save thousands of lves. U.S. Department of Transportaton (DOT) has estmated that V2V safety communcaton based on DSRC can assst drvers n preventng 76 percent of the crashes on the roadway, thereby reducng fataltes and njures that occur each year. Fgure 1.1: U.S. DSRC channel allocaton In U.S., Federal Communcatons Commsson (FCC) has allocated 75 MHz of spectrum n the 5.9 GHz band, whch conssts of one control channel (CCH) and sx servce channels (SCHs), for DSRC to be used by ITS as shown n Fg. 1.1 [45]. DSRC s a secure, hgh speed (3 Mbps~27 Mbps), short range (1 m~1 m) wreless nterface between vehcles and surface transportaton nfrastructure that enables rapd communcaton of vehcle data and other content between On Board Equpment (OBE) and OBE, and between OBE and Road Sde Equpment (RSE). 2

25 DSRC standards for varous layers are under development n U.S. to support DSRC communcatons. Fg. 1.2 shows the protocol stack [45] for DSRC communcaton. Accordng to the updated verson of the DSRC standard IEEE 82.11p [44], the DSRC physcal layer follows the same frame structure, modulaton scheme and tranng sequences specfed by IEEE 82.11a physcal layer standard wth mnor changes; MAC layer of the DSRC s equvalent to the Enhanced Dstrbuton Coordnaton Access (EDCA) 82.11e that has four dfferent access classes (ACs). IEEE [69] as an extenson for MAC layer provdes channel swtchng for mult-channel operatons, whereas [7] and [71] deal wth network servces and securty servces respectvely. Internet protocols for Network and Transport layer are also supported n DSRC communcaton, whch are manly used for non-safety applcatons. Fgure 1.2: Protocol stack for DSRC communcaton 3

26 In the DSRC based vehcular ad hoc networks (VANETs), the transportaton safety s one of the most crucal features that need to be addressed. Safety applcatons usually demand drect V2V ad hoc communcaton due to a hghly dynamc network topology and strct delay requrements. Such drect safety communcaton wll nvolve a broadcast servce because safety nformaton can be benefcal to all vehcles around a sender. Broadcastng safety messages s one of the fundamental servces n DSRC. The safety messages can be categorzed nto two: basc safety message (BSM) and eventdrven safety message (ESM). The BSMs are perodcally sent and also referred as perodc beacon messages. The BSM contans nformaton related to the status of vehcle (e.g., poston, velocty and drecton) and s perodcally broadcast by each vehcle to announce other vehcles about ther exstence. Neghborng vehcles utlze such messages to become aware of ther surroundngs out of sght and avod potental dangers (e.g., rear-end collson warnng, slow vehcle ndcaton, emergency vehcle warnng etc.) [58]. The ESMs are generated and broadcast by a vehcle to warn neghbors around when an abnormal condton or an mmnent danger s detected (e.g., road hazard warnng, traffc condton warnng, sgnal volaton warnng etc.) [58]. In order to provde satsfactory qualty of servces (QoS) for varous safety applcatons, safety messages need to be delvered n both relably and tmely manner. 4

27 1.2 Research Problems Addressed To support the strngent delay and relablty requrements of broadcastng safety messages, researchers have been seekng to test proposed DSRC protocols and suggestng mprovements. A major hurdle n the development of VANET for safetycrtcal servces s the lack of methods that enable one to determne the effectveness of VANET desgn mechansm for predctable QoS and allow one to evaluate the tradeoff between network parameters. Hence, we am to evaluate mportant performance and relablty metrcs under varous traffc scenaros to assess the effectveness of message dssemnaton schemes for safety applcatons. Several types of methods have been utlzed n the lterature to analyze the performance and relablty for safety communcaton. Computer smulatons are extensvely used, whch usually take long tme to collect suffcent data for accurate performance analyss. Several real world experments are also conducted to capture more practcal network dynamcs. However, due to the hgh equpment cost, the expermental testbed usually only conssts of a very few vehcles. Even though some technques are used to create a large scale vrtual communcaton network usng a small number of vehcles, the resultng vrtual network s stll only capable of capturng sparse network scenaros. Analytc modelng s a more attractve alternatve due to lower cost of solvng the model whle coverng a large network parameter space. However, there are very few accurate analytc models developed for safety applcatons evaluaton. Our 5

28 man goal for ths dssertaton s to propose comprehensve and hgh fdelty analytc models for the performance and relablty analyss under varous safety communcaton scenaros. The accuracy and effcency of the analytc models are valdated through extensve smulatons. 1.3 Contrbutons of the Dssertaton In ths dssertaton, the followng contrbutons are made: (1) Developed comprehensve and hgh fdelty analytc models. In V2V safety communcatons, vehcles on the road compete for the channel resource to transmt ther own safety messages. To reduce the complexty for developng and solvng monolthc models, we utlze the model decomposton technque and propose nteractng stochastc models based on the sem-markov process (SMP) to capture vehcles channel contenton and backoff behavor. Due to the nteractons between vehcles, fxed-pont teraton s used to obtan converged solutons, based on whch mportant performance and relablty metrcs are further derved. (2) Evaluated dfferent types of safety messages. As mentoned earler, the safety messages can be categorzed nto two: basc safety message (BSM) and event-drven safety message (ESM). A smplfed SMP model s frst developed for the ESM evaluaton, and a more precse SMP model s developed for the BSM evaluaton. The analytc-numercal results for these two approaches are also compared n Chapter 5. The obtaned results show that the performance and relablty metrcs for BSM and 6

29 ESM are smlar when ther average message generaton ntervals are the same. In addton, we conclude that the smplfed SMP model for ESM evaluaton wth Posson message arrvals can be used to approxmate the BSM evaluaton wth perodc message arrvals. Many papers [6][8][41][52][62][117] have used Posson arrval process to approxmate BSM arrval process wthout any proof. (3) Evaluated multple servces over a sngle channel. The control channel (CCH) s the default channel for V2V safety communcaton. If dfferent types of safety messages need to be transmtted to support multple safety servces, t s possble that the control channel wll be shared. Enhance dstrbuted channel access (EDCA) s the access mechansm specfed n IEEE 82.11p protocol to support multple types of servces wth prortes. Therefore, analytc models are proposed for the performance and relablty analyss of three types of vehcular safety-related servces usng EDCA mechansm n the DSRC system on hghways. Several mportant observatons are drawn. The results show that EDCA mechansm, whch utlzes dfferent channel sensng tme and backoff counters for dfferent servces, can only provde servce dfferentaton n terms of transmsson delay, but cannot help mprove the relablty for hgher prorty servces. Another mportant concluson s that hgher prorty servce should choose shorter packet length, hgher date rate and larger carrer sensng range to ensure hgh packet transmsson relablty. 7

30 (4) Evaluated mult-channel operatons for the BSM. The control channel (CCH) s dedcated to support safety communcatons, and BSM s most lkely to be transmtted va CCH. The SCH channel 172 n Fg. 1.1 s reserved probably for crtcal V2V safety communcatons, and hence the ESM (whch s sent n presence of an emergency event and hence s more crtcal than BSM) can be transmtted on ths channel to avod the channel contenton wth the more frequently generated BSMs. In addton, other SCHs n Fg. 1.1 are most lkely to be used for non-safety applcatons. At the early stage of the DSRC deployment, a vehcle may only have a sngle-rado devce nstalled due to the cost constrants. To support both safety and non-safety applcatons, ths sngle-rado devce can swtch between dfferent channels, one channel at a tme. IEEE [69] s an extenson for MAC layer to provde channel swtchng mechansm for mult-channel operatons. Analytc models are developed to evaluate the mpact of such mult-channel operatons on the performance and relablty of BSMs transmtted va the CCH. The results show that channel swtchng mechansm can greatly degrade the performance and relablty of the BSM transmsson. (5) Evaluated mult-hop transmssons for the ESM. To ensure hgh relablty for ESMs, whch are more crtcal than BSMs, a channel (probably channel 172) may be reserved specfcally for ESM broadcastng. In addton, some event-drven safety applcatons (e.g., post-crash notfcaton, road hazard warnng) may be requred to 8

31 cover longer dstances than the one-hop communcaton range. Therefore, mult-hop ESM dssemnaton s necessary n such scenaros. A robust relay selecton strategy utlzng dstance-based tmers s proposed n ths dssertaton and an accurate analytc model s proposed to evaluate mult-hop propagaton of ESMs. Important conclusons are drawn to provde deeper nsght nto the ESM transmsson behavor from dfferent angles. (6) Evaluated 2-Dmensonal traffc scenaros. Most analytc models proposed n the lterature concentrate on 1-Dmensonal (1-D) hghway traffc to smplfy the model. Unfortunately, very few of network scenaros n real applcatons can be abstracted as 1-D models. Therefore, n ths dssertaton, we frst developed an analytc model on the performance and relablty evaluaton of 2-Dmensonal (2-D) traffc at open feld to capture more realstc message transmsson behavors. (7) Incorporated varous factors for the performance and relablty metrcs dervaton. In ths dssertaton, we consdered varous mportant factors that can nfluence the message transmsson ncludng channel contenton, backoff behavor, concurrent transmssons, hdden termnals problems and channel fadng wth path loss. The degree of mpact of concurrent transmsson, hdden termnals problem and channel fadng wth path loss s also assessed. The results presented n Secton 3.5.3, Secton and Secton show that concurrent transmsson has very 9

32 mnor mpact, whereas hdden termnals problem and channel fadng wth path loss have sgnfcant mpact on the performance and relablty. (8) Conducted smulatons for cross valdaton purposes. We conducted extensve smulatons n ether Matlab or NS2 for comparson purposes n each Chapter. The good match between analytc-numercal results and smulaton results under a large range of network parameters valdate the accuracy of our proposed models. In addton, the tme consumed to solve analytc models and for smulatons s compared, whch shows that the analytc models are much more effcent than smulatons. Our contrbuton can also be summarzed from dfferent aspects accordng to the nterest of people from varous areas: (1) Provde valuable nsghts for protocol development. Frst, we proposed an effcent and accurate approach to easly evaluate whether the gven network parameter can satsfy the QoS and hghlght how to tune the parameters n order to meet the QoS. In addton, EDCA protocol s proven n ths dssertaton that t does not support servce dfferentaton regardng to broadcast relablty. Such observaton may form the background to support other effectve protocols development for servce prortzaton. Moreover, even though IEEE 82.11p s the standardzed protocol analyzed n ths dssertaton, our proposed models are based on characterzng the operaton flow for the DCF functon, whch forms the basc medum access 1

33 mechansm n many other protocols. Therefore, our models can be easly extended to analyze other protocols. (2) Provde more effectve approach than smulaton and experment based methods. To evaluate the effectveness of our proposed analytcal models, we conducted smulatons n every chapter for comparson purposes. Our models can be easly solved wthn a few seconds or mnutes, whle smulatons usually take up to several hours. Hence, our models are much more effcent than smulaton methods and can delver valuable results much faster for the development of DSRC safety communcaton. Furthermore, most experments are lmted to a few vehcles due to extremely hgh cost of purchasng vehcles, whch results n the nsuffcency for capturng realstc and mportant factors that nfluence the safety communcaton, such as hdden termnal problem. Therefore, the advantage of our model s that t effectvely consders varous factors and can be easly extended to ncorporate more factors of nterest wthout any resource lmt. (3) Provde more accurate approach than other analytc models. Even though there are many analytc models proposed n the lterature to analyze the performance and relablty of safety communcaton, most of them are based on Banch s dscretetme Markov chan (DTMC) model [27]. The system s contnuous tme behavor for channel sensng and deferrng n message transmsson are completely gnored, whch results n naccuracy. Our model consders more accurate message 11

34 transmsson behavor n MAC layer by usng sem-markov Process (SMP) model wthout dscretzng the tme. The comparson of our model wth several DTMC based analytc models presented n many chapters verfes that our model s more accurate than prevalent DTMC models. In addton, most analytc models only consder MAC-level performance metrcs, whereas our model also consders applcaton-level metrcs to better fulfll the safety applcaton s QoS requrement. Moreover, several mportant factors such as dstance-based channel fadng, hdden termnal problem and channel swtchng mechansm are not evaluated n vast analytc approaches, whle we constructed much more accurate and practcal model by takng nto account of those essental factors. 1.4 Outlne of the Dssertaton Ths dssertaton s organzed as follows: Chapter 2 descrbes the background on DSRC-related topcs and stochastc modelng. Snce the MAC layer channel access for safety communcaton follows dstrbuted coordnaton functon (DCF) for one type of safety message accordng to IEEE protocol, the detaled DCF access control technque s frst descrbed. We also ntroduce the related analytc modelng methods used throughout ths dssertaton. Chapter 3 presents a general analytc model to evaluate the performance of safety message broadcastng, whch s sutable for ESMs. Posson message arrval s assumed for event-drven message generaton. Infnte MAC-layer queue for each 12

35 vehcle s assumed snce the ESMs are too crtcal to be dropped. Hence, the generaton and servce of ESMs n each vehcle s modeled by a generalzed M/G/1 queue. The overall model s a set of nteractng M/G/1 queues, one queue for each vehcle. To produce a smplfed yet hgh fdelty analytc model, we decompose the overall model and use sem-markov process (SMP) model to capture shared channel medum s behavor from a sngle vehcle s perspectve. Due to the nteractons between vehcles, fxed-pont teraton s utlzed to obtan converged solutons. Important MAC-level performance and relablty metrcs are subsequently derved. Smulatons n Matlab are developed to valdate the accuracy and effcency of the proposed analytc model. Chapter 4 descrbes a more accurate analytc model to capture the BSM broadcastng n a channel (probably the control channel), where assocated features such as perodc message generaton, out-dated message replacement and no queung n MAC-layer are ncorporated. Such a model for BSMs s compared wth the model presented n Chapter 3 for ESMs. The results prove that the smplfed analytc model for ESMs can be used to approxmately evaluate the BSMs transmsson. Besdes MAC-level performance and relablty metrcs, applcaton-level metrcs are also evaluated. Smulatons n Matlab are conducted for the comparson purpose. Snce the control channel s the default channel for V2V safety communcaton, multple types of safety messages (ncludng the ESM and BSM) can be transmtted together n the control channel (CCH). Therefore, n Chapter 5, multple types of servces 13

36 transmssons over a sngle channel (e.g., control channel) are evaluated based on the extenson of the analytc model developed n Chapter 3. The EDCA mechansm specfed n the IEEE 82.11p protocol s shown to be neffectve to guarantee prortes for dfferent servces. Important conclusons are also drawn to tune network parameters n order to mprove the performance and relablty for hgh prorty servces. Smulatons are performed n Matlab. The U.S. FCC has allocated seven 1 MHz channels for DSRC: one control channel (CCH) and sx servce channels (SCHs). At the early stage of DSRC deployment, a vehcle may have only a sngle-rado DSRC devce nstalled. To support concurrent applcatons on dfferent channels, such a sngle-rado devce can swtch between dfferent channels, one channel at a tme, to access safety messages (e.g., BSMs) on the CCH and other servces on the SCHs. Therefore, the IEEE mult-channel operaton s consdered n Chapter 6 to evaluate the performance and relablty of the BSMs transmsson on the CCH. The mpacts of varous factors such as concurrent transmsson, hdden termnals problem, channel fadng wth path loss and channel swtchng mechansm are evaluated. Smulatons are developed n NS2 to valdate the accuracy and effcency of the proposed analytc model. Among the sx servce channels, the Channel 172 may be reserved for lfe crtcal safety applcatons wth no tme dvson [45]. Ths mpled that a vehcle nterested n both safety and non-safety applcatons requres two DSRC rados: one tuned to Channel 14

37 172 all the tme, and the other partcpates n IEEE channel swtchng [45]. Snce ESMs are lfe-crtcal messages, they are most lkely to be transmtted on Channel 172. In addton, snce some event-drven safety applcatons (e.g., post-crash notfcaton, road hazard warnng) may requre longer transmsson dstance than the one-hop communcaton range, mult-hop dssemnaton of ESMs s necessary. Hence, we ntroduce an accurate analytc model n Chapter 7 to evaluate mult-hop propagaton of ESMs. Extensve Matlab smulatons are performed. Important conclusons are obtaned to provde deeper understandngs of the performance and relablty of the ESMs transmsson behavor. The analytc models developed n Chapter 3-7 are all concentrated on 1-D hghway traffc to smplfy the model. Untl now, there s no analytc model on 2-D evaluaton. Nevertheless, very few of network scenaros n real applcatons can be abstracted as 1-D traffc. Therefore, n Chapter 8, we frst ntroduce an analytc model on the performance and relablty evaluaton of 2-D traffc at open feld (.e., battle feld) to capture more realstc message transmsson behavors. Smulatons n NS2 are carred out and the numercal results are compared wth the analytc-numercal results under a large range of network parameter settngs. Chapter 9 summarzes ths dssertaton and descrbes future research work on ths topc. Table 1.1 summarzes and compares the work presented n each chapter. 15

38 Table 1.1: Summary of chapters Chapter Message Type Multple Servce? Channel Swtch? 2-D? Multhop? Fxedpont? Smulaton 3 BSM, ESM No No No No Yes Matlab, NS2 4 BSM No No No No Yes Matlab 5 BSM, ESM Yes No No No Yes Matlab 6 BSM No Yes No No Yes NS2 7 ESM No No Yes No No Matlab 8 BSM, ESM No No No Yes Yes NS2 16

39 2. Background In ths chapter, we brefly ntroduce the background on DSRC-related topcs (ncludng DCF access mechansm) and stochastc modelng methods (ncludng Markov models, sem-markov models and queung models) used n ths dssertaton. 2.1 DSRC-related background Ths secton presents essental background of DSRC for analytc modelng of safety communcaton n VANETs MAC Layer Protocol Descrpton Fgure 2.1: Flow chart of DCF functon To model and assess the performance of safety message broadcastng, MAC layer behavor specfed by IEEE 82.11p has to be accurately evaluated. Hence, we brefly descrbe the MAC layer channel access for safety message broadcastng n ths secton. From [44], we know that the DSRC adopts IEEE MAC layer specfcaton based on the carrer sense multple access wth collson avodance (CSMA/CA) wth mnor modfcatons. In the MAC layer protocol [11], dstrbuted coordnaton functon (DCF) s the prmary medum access control technque for broadcast servces. 17

40 Fg. 2.1 descrbes n detal the basc access mechansm of DCF for broadcast n the context of the safety communcaton. Enhanced Dstrbuted Channel Access (EDCA) mechansm for multple servces wth prortes specfed n IEEE 82.11p [44] comples wth the DCF functon f only one type of servce s consdered n the channel. Accordng to Fg. 2.1, each vehcle n the network can generate messages and compete for the channel resource to transmt the message. If a vehcle does not have any message to transmt, t wll wat for a packet to be generated. Then, for a newly generated packet, the vehcle senses the channel actvty before t starts to transmt the packet. If the channel s sensed dle for a tme perod of dstrbuted nter-frame space (DIFS), the packet can be drectly transmtted. Otherwse, the vehcle contnues to montor the channel untl channel s detected to be dle for DIFS tme perod. Subsequently, accordng to the collson avodance feature of the protocol, the vehcle goes through the backoff process before transmttng the packet. It generates an ntal random backoff counter from a unform probablty mass functon (pmf) over the range [, CW], where CW represents the contenton wndow. The backoff tme counter s decreased by one f the channel s sensed dle for a tme slot of duraton σ. Otherwse, the counter s frozen and reactvated when the channel s sensed dle agan for more than DIFS duraton. The packet s transmtted as soon as the backoff counter reaches zero. After ths packet s transmtted, f there s no packet left n ths vehcle, the process wll start over agan and the vehcle wll wat for a new packet to be generated. Otherwse, f 18

41 there are packets left, the vehcle repeats the procedure startng wth sensng the channel for DIFS duraton and goes through the backoff procedure before transmttng the next packet. Accordng to the protocol, a vehcle must go through the backoff process between two consecutve packet transmssons even f the channel s sensed dle for the duraton of DIFS tme for the second packet. 2.2 Modelng Methods To evaluate the performance and relablty of safety communcaton n VANETs, state-space stochastc models are used n ths dssertaton to characterze the safety message broadcastng behavor. Compared to non-state-space models, state-space models can capture more complex dependences n the system under consderaton. Ths secton ntroduces some commonly used state-space models ncludng Markov models and sem-markov models Markov Models The stochastc process for Markov models s whose dynamc behavor for future development depends only on the current state and not on how the process arrved n that state, whch s well known as Markov property. The formal defnton for Markov process s: Defnton: A stochastc process {X(t) t } s a Markov chan f for any t < t1 < < tn < t, the condtonal probablty mass functon (pmf) of X(t) satsfes: 19

42 ( ( ) ( ), ( ),..., ( ) ) ( ( ) ( ) ) P X t = x X t = x X t = x X t = x = P X t = x X t = x (2.1) n n n 1 n 1 n n If the state space I s dscrete as above, the Markov process s known as a Markov chan. If the parameter space T s also dscrete, then we have a dscrete-tme Markov chan (DTMC). If the parameter space T s contnuous, then we have a contnuous-tme Markov chan (CTMC). The Markov chan {X(t) t } s sad to be (tme-) homogeneous f the state transton probablty depends only on the dfference of the two tme epochs that we are consderng. Otherwse, the Markov chan s a non-homogeneous Markov chan. Henceforth, we only consder the homogeneous case. The transent behavor of a CTMC satsfes the Chapman-Kolmogorov equaton n matrx form as follows: dπ( t) = π ( t) Q dt (2.2) where π(t) = [π1(t), π2(t),, πn(t)] s the state probablty vector, and n s the number of states n the CTMC. Q = [qj]nxn s the nfntesmal generator matrx, where qj ( j) s the transton rate from state to state j, and q = - j qj. In steady-state, Eq. (2.2) becomes: π Q = (2.3) For a DTMC, the state probablty vector after n-step transton s gven by: n p( n) = p () P (2.4) 2

43 where p() = [p1(), p2(),, pn()] s the ntal state probablty vector, and P = [pj]nxn s the one-step transton probablty matrx (pj s the transton probablty from state to state j). In steady-state, denote v = lm p ( n), then Eq. (2.4) becomes: n v = v P (2.5) Sem-Markov Process Model Sem-Markov process (SMP) model s a generalzaton of both contnuous and dscrete tme Markov chans whch permt arbtrary sojourn tme dstrbuton functon, possbly dependng on both the current state and the state to be vsted next [116]. For a better understandng, let us consder a stochastc process as follows [115]. Frst construct a k-state dscrete-tme Markov chan (DTMC) wth state transton probablty matrx P = [pj]; next construct a process n contnuous tme by markng the tme spent n a transton from state to state j have a dstrbuton functon Fj(t) such that tmes are mutually ndependent; at the end of the nterval, we have a pont event of type j. Such a stochastc process s called sem-markov process, whch s a generalzaton of both contnuous and dscrete tme Markov processes wth countable state spaces. A descrptve defnton of SMP [115] s that t s a stochastc process whch moves from one state to another state among a countable number of states wth the successve states vsted formng a dscrete-tme Markov chan, and that the process stays n a gven state for a random amount of tme, the dstrbuton functon of whch 21

44 depends on ths state as well as the one to be vsted next but does not depend on whch states the system had been n before t got there. A formal defnton of SMP [116] s descrbed as follows. A SMP s the process Y={Yt; t } defned by: Yt = X N ( t) = X n, f Sn t < S n + (2.6) 1 for t, where N(t) s the countng process. From the SMP defnton, t should be observed that the process only changes state (possbly back to the same state) at the Markov regeneraton epochs Sn. To analyze the steady-state behavor or to calculate some expected values of a SMP model, there exsts a method called the two-stage method. It descrbes an SMP model usng the matrx P and the vector H(t), where P = [pj] s the one-step transton probablty matrx for the embedded Markov chan (EMC) of the SMP model, and H(t)=[H(t)] wth =1,,n, s the sojourn tme dstrbuton n state. Such a method consders SMP transtons as takng place n two stages: 1) In the frst stage, the system stays n state for some amount of tme, the mean sojourn tme n state s: ( 1 ( )) h = H t dt (2.7) 2) In the second stage, the system moves to state j wth probablty pj. When ths two-stage method s appled to the steady-state analyss of SMP model, we frst calculate the steady-state probablty vector of the EMC usng Eq. (2.5). Gven the 22

45 mean sojourn tme vector h=[h1, h2,, hn], the steady-state probablty vector π=[π1, π2,, πn] of the SMP can be wrtten as: π = n k = v h k v h k (2.8) whch s the rato between the average tme spent n state (vh) over the total average tme spent ( kvkhk) over all states Queung Models A queung system conssts of one or more statons (servers) that provde servces to arrvng customers (jobs). Customers who arrve to fnd all servers busy generally jon one or more queues (watng lnes) n front of the servers. The queung system can be characterzed by three mportant components: arrval process, servce process, and the number of servers. Assume that successve nter-arrval tmes Y1, Y2,, between jobs are ndependent dentcally dstrbuted random varables havng a dstrbuton FY. Smlarly, the servce tmes S1, S2,, are assumed to be ndependent dentcally dstrbuted random varables havng a dstrbuton FS. Let m denote the number of servers n the system. Therefore, we can use the notaton FY/ FS/m to descrbe the queung system. The followng symbols are used to denote the specfc types of nter-arrval tmes and servce tme dstrbutons [9]: M: (for memoryless) for the exponental dstrbuton D: for a determnstc or constant nter-arrval or servce tme 23

46 Ek: for a k-stage Erlang dstrbuton Hk: for a k-stage hyperexponental dstrbuton G: for a general dstrbuton GI: for general ndependent nter-arrval tmes Based on the above conventon, we can easly nterpret the queung system. The most frequent example of a queung system s M/M/1 queue, whch stands for a sngle-server queue wth exponentally dstrbuted nter-arrval tmes and exponentally dstrbuted servce tme. In ths dssertaton, M/G/1 queue and GI/G/1 queue notons are used. M/G/1 queue represents a sngle-server queue wth exponentally dstrbuted nter-arrval tmes and an arbtrary servce tme dstrbuton, and GI/G/1 queue represents a sngle-server queue wth general ndependent nter-arrval tmes and an arbtrary servce tme dstrbuton. If a queung system has lmted buffer space so that at most n jobs can be n the system, such a system can be denoted as FY/ FS/m/n usng ths Kendall notaton. Therefore, D/G/1/1 queue system presented n Secton 4.1 denote a sngle-server queue wth determnstc nter-arrval tmes dstrbuton, general servce tmes dstrbuton, and the number of jobs n the system s at most 1. Besdes the nature of the nter-arrval tme and servce tme dstrbuton, queue schedulng dscplne that descrbes how the server s to be allocated to the jobs watng for servce also needs to be specfed. In ths dssertaton, two schedulng dscplnes are used: Frst Come Frst Served (FCFS) and Last Come Frst Served (LCFS). 24

47 2.2.4 Fxed-pont Iteraton Method In numercal analyss, fxed-pont teraton s a method used to compute fxed ponts of terated functons. More specfcally, a soluton to the equaton f(x)=x s referred to as a fxed pont of the functon f. Geometrcally, the fxed ponts of a functon f(x) are the pont(s) of ntersecton of the curve y=f(x) and the lne y=x. Gven a pont x n the doman of f, successve substtuton xn+1 = f(xn), n=,1,2, s commonly used to determne the fxed pont. The sequence x, x1, x2,, s expected to converge to a pont x. The proofs for the exstence, unqueness and convergence of the fxed-pont teraton process are provded n Chapter 3. 25

48 3. Broadcast Safety Messages Evaluaton 3.1 Motvaton In the DSRC based vehcular ad hoc networks (VANETs), the transportaton safety s one of the most crucal features that needs to be addressed. Safety applcatons usually demand drect vehcle-to-vehcle ad hoc communcaton due to a hghly dynamc network topology and strct delay requrements. Such drect safety communcaton wll nvolve a broadcast servce because safety nformaton can be benefcal to all vehcles around a sender. Broadcastng safety messages s one of the fundamental servces n DSRC that s beng standardzed as IEEE 82.11p. In the lterature, uncast for IEEE has been extensvely nvestgated. Banch [27] proposed a smple yet accurate dscrete-tme Markov chan (DTMC) model to evaluate the performance of uncast mechansm under saturaton condtons. Hs paper has nspred many other researchers to develop analytc models of uncast based on DTMC. For example, multple types of safety messages are evaluated n [28][29] under saturaton condtons. Banch s work was later extended to the unsaturated case n [3]. Unque characterstcs of the broadcast make ts analyss dfferent from that of uncast. Frst, uncast can have handshake between the sender and the recever, whch provdes feedback to the sender for the possble need for retransmsson to enhance relablty. By contrast, broadcastng sender cannot obtan feedback nformaton from the 26