Vehicular Networking: Modeling, Simulation, and Optimization

Transcription

1 Vehicular Networking: Modeling, Simulation, and Optimization Hannes Hartenstein, joint work with Th. Mangel, J. Mittag, and T. Tielert PRESERVE and EIT Summer School, Ulm, September 18, 2012 Research Group Decentralized Systems and Network Services - DSN KIT University of the State of Baden-Wuerttemberg and National Research Center of the Helmholtz Association

2 Vehicular networking: key features studied in the last 10+ years Unique communication environment and behavior Modeling and Simulation IEEE p Analysis, Comparison, Optimization Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

3 State of the art collected in 2010 VANET Vehicular Applications and Inter- Networking Technologies Content Overview on C2X- State of the Art Within academic environment Within industry Current standardization efforts w.r.t. IEEE p and IEEE 1609 Modeling and Simulation of vehicular networks Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

4 on August 22, 2012 The government s safety regulators have said that vehicle-to-vehicle communication could help avoid or reduce the severity in an estimated 80 percent of vehicle crashes involving unimpaired drivers. The technology is there, and now we need to deploy it intelligently, said David Cole, a founder of the Center for Automotive Research in Ann Arbor. There is so much value to be gained, from saving lives to reducing insurance costs Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

5 The spectrum of application domains Safety Efficiency Convenience Unique environment does not mean uniform environment Various applications with different requirements Various wireless interfaces (UMTS/LTE, IEEE p, others) Different studies require different methodology Particularly, different modeling Required accuracy depends on goal of the study see next slide Applications can combine safety, efficiency and convenience Example: Virtual traffic lights Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

6 Increase in accuracy Increase in scalability Accuracy levels for modeling/understanding (1) from the perspective of a communication engineer Information Close to application Packets Bits Symbols or analog signal Close to physical communication Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

7 Increase in accuracy Increase in scalability Accuracy levels for modeling/understanding (2) from the perspective of a vehicle/traffic engineer Macroscopic Close to application Mesoscopic Microscopic Vehicular Dynamics Close to physical dynamics Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

8 Principal component analysis * *Where computer networking people typically are looking at. Information Packet Bit Symbol Safety Efficiency Convenience Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

9 Where is research? Information Packet Bit Complete and/or extend this picture Make sure that models on different levels match Symbol Safety Efficiency Convenience Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

10 Where is DSN research?* *Examples Can V2X Communication Help Electric Vehicles Save Energy?, ITST 2012 Information Packet Safety-critical communication A Change in Perspective: Information-Centric Modeling of Inter-Vehicle Communication, ACM VANET 2011 Bit Symbol Safety Efficiency Convenience Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

11 Scope of this talk Information Packet Safety-critical communication Bit Symbol Safety Efficiency Convenience Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

12 The technology is there, true: IEEE p is there. But: Does it properly work in all radio/channel environments? There is a wide variety of radio/channel conditions, how can we check whether p works properly? Do we have proper models, e.g. for non-line-of-sight conditions? Could we do better than p? ACM VANET 2012 panel discussion: what about a VANET 2.0? How can we deploy p intelligently? Congestion control with p Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

13 Structure of this talk Combined PHY/MAC/NET modeling and simulation Non line of sight models (urban intersections) How well does CSMA work? Congestion control with IEEE p Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

14 Combined PHY/MAC/NET modeling/simulation Information Packet Safety-critical communication Bit Symbol Safety Efficiency Convenience Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

15 Combined PHY/MAC/NET modeling/simulation Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

16 Combined Channel, Physical Layer and Network Simulation Starting point: open source NS-3 network simulator Higher Layers Smallest Unit Simulation Approach MAC Layer Packet 1:1 Simulation PHY Layer Signal Packet Statistical Abstraction Emulation 5.9 GHz Wireless Channel Signal Packet Statistical Abstraction Emulation New physical layer includes Orthogonal Frequency Division Multiplexing Convolutional Encoder / Viterbi Decoder Signal Detection Channel Estimation Packet Capture New channel model library includes Rayleigh / Rician Tapped Delay Line (TDL) based models Geometry-based stochastic model Open source, released under GPL Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

17 Combined Channel, Physical Layer and Network Simulation Validation Channel models are based on channel measurement campaigns Physical layer validated against off-the-shelf Atheros radios Wireless emulator testbed, CMU Pittsburgh A/D Converters Setup 1 Transmitter + 1 Receiver Different packet sizes Different data rates Different channel setups Performance metric Packet Reception Ratio w.r.t. Signal-to-Noise Ratio (SNR) Non-linear offsets due to emulator issues Laptop Laptop Channel setup 1: No Fading DSP Antenna Outputs Laptop Laptop Linear offset at most 1 db Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

18 Combined Channel, Physical Layer and Network Simulation Validation Channel models are based on channel measurement campaigns Physical layer validated against off-the-shelf Atheros radios Wireless emulator testbed, CMU Pittsburgh A/D Converters Setup 1 Transmitter + 1 Receiver Different packet sizes Different data rates Different channel setups Performance metric Packet Reception Ratio w.r.t. Signal-to-Noise Ratio (SNR) Laptop DSP Laptop Antenna Outputs Channel setup 2: Rayleigh Fading Laptop Laptop Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

19 Combined Channel, Physical Layer and Network Simulation Benefits of new modeling approach Brings Wireless Research Communities Together Increases Level of Detail / Accuracy Enables Usage of Complex Channel Models Enables Advanced Signal Processing Algorithms Allows to Perform and Evaluate Crosslayer Optimizations There is a price to pay: runtime performance Significant slowdown up to a factor of 10 5 Can effectively be reduced to 10 3 with GPU-based acceleration Benefit of GPU-based processing depends on scenario configuration on the number of nodes: the more, the better on channel models used Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

20 Structure of this talk Combined PHY/MAC/NET modeling and simulation Non line of sight models (urban intersections) How well does CSMA work? Congestion control with IEEE p Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

21 Non Line of Sight Models (urban intersection) Information Packet Safety-critical communication Bit Available online soon Symbol Safety Efficiency Convenience Hannes Prof. Dr. H. Hartenstein, PRESERVE and and EIT EiT Summer School, School, Ulm Ulm

22 CROSS TRAFFIC ASSISTANCE 22 Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

23 CROSS TRAFFIC ASSISTANCE: CHALLENGES Is NLOS reception relevant? Warn driver 3 seondes before potential collision: 42m distance at speed of 50 km/h Yes, but depends on environment Does NLOS reception matches requirements? We provide dedicated 5.9 GHz NLOS model Non-Line-Of-Sight (NLOS) reception? 23 We check feasibility via simulation of urban intersections under realistic traffic and communication conditions Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

24 Street 3 Street 1 NLOS reception: test setup. Goal: Reproducibility and comparability Street 4 Street Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

25 Google Maps Test Intersection Selection. Main Case / Wider Suburban / Urban. ID = 1 Suburban, 23m+21m ID = 2 Suburban, 23m+19m ID = 3 Suburban, 21m+21m ID = 10 Urban, 24m+21m ID = 11 Urban, 27m+21m ID = 20 Urban, 30m+30m Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

26 NLOS measurements: results (1) Results for a single run Map-based visualizations - One second averages - Center: reception rate - Donut: Reception signal strength x-y plots - Signal strength of messages - Averaged for specific distance: signal strength and reception rate Result - Reliable NLOS reception. - Reception rate for a singe packet> 0.5 when warning driver Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

27 NLOS measurements: results (2) Comparison of intersections Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

28 Modeling NLOS: VirtualSource11p (1) Characteristics Data can be nicely fitted. Switches between urban and suburban Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

29 Modeling NLOS: VirtualSource11p (2) Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

30 Use of the NLOS model in NS-2 in a realistic traffic scenario at an intersection Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

31 Publications by Thomas Mangel: 11/ GHz inter-vehicle communication at intersections: a validated non-line-of-sight path-loss and fading model T. Mangel, O. Klemp, and H. Hartenstein; in EURASIP Journal on Wireless Communications and Networking, Vol 2011, No 1, pp Open Access Publication 11/11 5.9GHz IEEE p Inter-Vehicle Communication: Non-Line-Of-Sight Reception under Competition T. Mangel and H. Hartenstein; in 3rd IEEE Vehicular Networking Conference (VNC 2011), Amsterdam, Netherlands, pp /11 A Validated 5.9 GHz Non-Line-Of-Sight Path-Loss and Fading Model for Inter-Vehicle Communication T. Mangel, O. Klemp, and H. Hartenstein; in 11th International Conference on ITS Telecommunications (ITST2011), Saint-Petersburg, Russia, pp Best Paper Award 03/11 Real-World Measurements of Non-Line-Of-Sight Reception Quality for 5.9GHz IEEE p at Intersections T. Mangel, M. Michl, O. Klemp, and H. Hartenstein; in Communication Technologies for Vehicles, ser. Lecture Notes in Computer Science, vol. 6596, Springer Berlin / Heidelberg, pp /11 Vehicular Safety Communication at Intersections: Buildings, Non-Line-Of-Sight and Representative Scenarios T. Mangel, F. Schweizer, T. Kosch, and H. Hartenstein; in 8th International Conference on Wireless On-Demand Network Systems and Services (WONS), Bardonecchia, Italy, pp Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

32 Complete set of data available plus nice tools Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

33 Applying the NLOS model: VANET 2012 Poster Feasibility of Virtual Traffic Lights in NLOS Environments By Till Neudecker, Natalya An, Ozan K. Tonguz, Tristan Gaugel, Jens Mittag Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

34 Structure of this talk Combined PHY/MAC/NET modeling and simulation Non line of sight models (urban intersections) How well does CSMA work? Congestion control with IEEE p Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

35 Dissertation of Jens Mittag Available online Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

36 CSMA Coordination Performance MAC layer of p is based on Carrier Sense Multiple Access Main principle: Listen before Talk If medium busy: do not transmit and wait until it s free again, use random back-off Case 1: coordination successful! Case 2: coordination failed! Node 1 p 1 t Node 1 p 1 t TX request random back-off TX request Node 2 p 2 t Node 2 p 2 t From the perspective of a single node Packet Level Incoordination (PLI) t Delay of incoordinated transmission Incoordination Delay Profile (IDP) P(r) What s the probability that one (or multiple) nodes within a certain range transmit while I am transmitting? P( t) Why are those nodes incoordinated to me? 1) Hidden terminals? 2) Simultaneous transmission?? 0m Considered range r 1000m Start of transmission t End of transmission Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

37 CSMA Coordination Performance Every 3rd packet will be interfered from within 400m if no CSMA is used approx. 700m deterministic communication range / CS range Hidden terminal zone Very strong fading introduces incoordination at small distances Incoordination starts to occur at approx. 500m Packet Level Incoordination (PLI) Probability that one (or multiple) nodes within a certain range are incoordinated to my own transmission. Simulation setup Highway scenario 80 vehicles/km 400 byte packets Close to zero incoordination from within 700m 10 Hz packet rate ~700m communication range ~70% channel saturation Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

38 CSMA Coordination Performance Bad luck : vehicles closer than 700m start to transmit at the same point in time CSMA; Deterministic CSMA; Shadowing CSMA; Rayleigh Vehicles further away are hidden terminals: uniform distribution of t [μs] Incoordination Delay Profile (IDP) Probability distribution of the starting time difference between my own and incoordinated packet transmissions. Simulation setup Highway scenario 80 vehicles/km 400 byte packets 10 Hz packet rate ~700m communication range ~70% channel saturation Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

39 CSMA Coordination Performance Bad luck : vehicles closer than 700m start to transmit at the same point in time Hidden terminal issue is intensified with Shadowing CSMA; Deterministic CSMA; Shadowing CSMA; Rayleigh Vehicles further away are hidden terminals: uniform distribution of t [μs] Incoordination Delay Profile (IDP) Probability distribution of the starting time difference between my own and incoordinated packet transmissions. Simulation setup Highway scenario 80 vehicles/km 400 byte packets 10 Hz packet rate ~700m communication range ~70% channel saturation Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

40 CSMA Coordination Performance Bad luck : vehicles closer than 700m start to transmit at the same point in time CSMA; Deterministic CSMA; Shadowing CSMA; Rayleigh and gets even worse with Rayleigh! Vehicles further away are hidden terminals: uniform distribution of t [μs] Incoordination Delay Profile (IDP) Probability distribution of the starting time difference between my own and incoordinated packet transmissions. Simulation setup Highway scenario 80 vehicles/km 400 byte packets 10 Hz packet rate ~700m communication range ~70% channel saturation Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

41 CSMA Coordination Performance CSMA achieves its design objective if no channel fading is exhibited Incoordination primarily due to hidden terminals or nodes that accidently start to transmit simultaneously Fading channel effects Introduce incoordinated transmission from within the communication range But: Level of incoordination increases only slightly And: Negative impact of incoordination decreases Primary Issue: Scalability Level of incoordination depends on the offered load See next slides Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

42 Structure of this talk Combined PHY/MAC/NET modeling and simulation Non line of sight models (urban intersections) How well does CSMA work? Congestion control with IEEE p Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

43 PULSAR: rate adaptation Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

44 Objective of this work compare Evaluation Contrasting and demanding scenarios Protocol Analysis of subtle design decisions Design methodology Selection of Tx rate and Tx range Related work consolidate Design principles integrate Requirements for VSC congestion control Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

45 Design principles [2] Schmidt et al. (2010) [3] Torrent-Moreno et al. (2009) Decentralization Ad-hoc communication [4] Mittag et al. (2008) [5] Baldessari et al. (2010) [6] Khorakhun et al. (2008) [7] Torrent-Moreno et al. (2005) [8] Schmidt et al. (2010) [9] Huang et al. (2010) [10] Sepulcre et al. (2011) [11] Sepulcre et al.(2011) [17] Busche et al. (2010) Participation rule Local fairness Global fairness Deference to safety applications Contribution to congestion participation Fairness in cong. control ( sharing channel info) Physically close nodes should be similarly regulated by congestion control. Max-min fairness (best effort) Excessively throttled nodes can become safety risk Adjustments within limits provided Awareness by safety applications Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

46 Protocol design methodology Our approach: Select Tx range first*, then adapt Tx rate w.r.t. load Target channel load many combinations of Tx range and rate Computationally tractable way? VSC applications require Rx optimization within certain target range Receiver s point of view Depends on driving context Same channel load (CBR 0.6) *range is adapted, but more slowly Target range 200m Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

47 Protocol design methodology (ctd.) Our approach: Select Tx range first, then adapt Tx rate w.r.t. load Optimal Tx rate Optimal Tx range Depends on node density Does not depend on node density Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

48 Protocol Description: Overview Periodically Updated Load Sensitive Adaptive Rate control (PULSAR) Driving context Tx range Safety application(s) AIMD Target rate Tx rescheduling Minimum Tx rate Tx Rate adaptation Periodic trigger Channel load assessment Ch. Busy Ratio (CBR) Self averaging Local channel measurement Received channel info Maximum Tx rate 2-hop piggybacking Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

49 PULSAR: 2-hop piggybacking Participation principle: Everyone contributing to congestion should participate in congestion control. Boundary for contribution: Carrier sense range CS range 2x theoretical Tx range (CSTh = -95 dbm) Deterministic propagation:? A B C congestion Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

50 PULSAR: 2-hop piggybacking Non-deterministic propagation: (here: Rayleigh) Timing considerations: A B C Time 1 hop 1 hop Typically more than 1 vehicle reports congestion Please see paper for details Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

51 PULSAR: Performance Transmission rate [Hz] Node density Channel Busy Ratio Node position Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

52 Summary and Conclusions Step-by-step design Details matter Evaluation Contrasting scenarios to stress protocol mechanisms Benchmark scenarios Comparison against closely related approaches PULSAR Design methodology Design principles Subtle design decisions Design principle Decentralization Participation Local fairness Global fairness Deference to safety apps Translate principles into protocol mechanisms Protocol mechanism Distributed algorithm 2-hop piggybacking Target rate AIMD, target rate, participation Setting Tx range, min/max rate Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm

53 Summary of the talk today Does it properly work in all radio/channel environments? There is a wide variety of radio/channel conditions, how can we check whether p works properly? combined PHY/MAC/NET simulations Do we have proper models, e.g. for non-line-of-sight conditions? NLOS model for urban intersection (in Munich) Could we do better than p? ACM VANET 2012 panel discussion: what about a VANET 2.0? CSMA is simple, but works pretty well How can we deploy p intelligently? Congestion control with p Yes, we can. Thanks to Tessa, Jens, and Thomas for their great work Hannes Hartenstein, PRESERVE and EIT Summer School, Ulm