Safety and Infotainment Communications for Vehicles Opportunities and Challenges Mahbub Hassan School of Computer Science and Engineering University of New South Wales, Sydney, Australia NSERC DIVA Distinguished Lecture Series, University of Ottawa, 12 June 2012 New Frontiers for Wireless Networking From Home and Office To the Road 1
Vehicular Networking and Road Safety A real opportunity to save lives Wireless networking could save many lives (1 million people die each year from road accidents) Very attractive proposition for Governments (new spectrum already allocated) Pump vehicular data out 10 times a second Simple Idea, but Challenging to Realize Too many cars, too many beacons on the air Reception probability could be as low as 30% on a 8-lane road 2
INFOTAINMENT on the Road Mobile data is a rapidly growing market More users will access contents from inside a vehicle But high-speed vehicular mobility makes content delivery very challenging Our research SAFETY (Smart Beacon Management) Smart repetition of beacons (improve beacon reliability) Adaptive position update Leveraging beacons to improve positioning accuracy INFOTAINMENT Mapping wireless bandwidth to road networks Smart use of bandwidth maps (content streaming) 3
Smart Repetition of safety beacons Zhe Wang and Mahbub Hassan, Blind XOR: Low-Overhead Loss Recovery for Vehicular Safety Communications", IEEE Transactions on Vehicular Technology, 61(1) January 2012. Why Beacons Are Lost? Large number of wireless devices in a linear topology is recipe for hidden terminal and packet collision No luxury of collision avoidance (e.g., RTS/ CTS) due to strict timeconstraint and distributed nature of the network 4
What Could Happen When Beacons are Lost? An Intersection Crash 23m Scenario 23m 3m Car-1 Car-2 Violating Car (72km/h) (72km/h) unexpectedly crashes with a red-light violating car Driver of Car-1 sees the crash happening right in front of his eyes Could the drivers of Car-1 and Car-2 escape from possible crash & injury? Two important variables: Injury Speed Threshold 1 crashing with speed more than 2.34m/s (5mph) Driver Reaction Time 2 from the time driver noticed danger to the time brake is applied (1 sec is not uncommon) 1 B. Henderson. Putting the 5mph threshold to the test, Personal Injury Law Journal, September 2006 2 P. Olson, P. L. and M. Sivak. "Perception-Response Time to Unexpected Roadway Hazards." Human Factors: The Journal of the Human Factors and Ergonomics Society 28(1): 91-96, 1986. Relying on Brake Lights (1 sec later) 23m Car-1 Car-2 Car-1 crashes with at just before the driver had a chance to apply brakes (effect of driver reaction time) Car-1 driver is injured (crashing speed > injury speed threshold) Car-1 s brake light is lit just at the moment of crash Car-2 driver saw the brake light, but will he escape crash/injury? 5
Relying on Brake Lights (2 sec later) Car-1 Car-2 Car-2 crashes at just when it was about to brake Car-2 driver is injured Brake lights failed to prevent injuries for Car-1 & Car-2 drivers Now Consider VC-based Crash Warning 23m 23m Car-1 Car-2 broadcasts a warning message immediately after the crash The wireless message is received by both Car-1 and Car-2 simultaneously (wireless signals travel at the speed of light; packet processing time is negligible) 6
VC-based Crash Warning (after 1 sec) 23m Car-1 Car-2 Car-1 crashes with at injuring the driver (effect of driver reaction time) But Car-2 starts decelerating (at 8m/s 2 ) VC-based Crash Warning (after 3.2 sec) Car-1 Car-2 2m/sec Car-2 crashes with Car-1 at 2m/sec in contrast to 20m/s if relied on brake light only Driver is NOT injured (crashing speed < injury speed) 7
Impact of Packet Loss on the effectiveness of VC-based Crash Warning 23m 23m Car-1 Car-2 broadcasts a warning message immediately after the crash, but it gets lost Neither Car-1, nor Car-2 receives the warning message Impact of Packet Loss (after 0.1 sec) 21m 23m Car-1 Car-2 regenerates another message, which is received successfully by Car-1 and Car-2 Both Car-1 and Car-2 drivers are now warned 8
Impact of Packet Loss (after 1 sec) 23m Car-1 Car-2 None of the drivers applied brakes just yet (effect of driver reaction time) Car-1 crashes with at injuring the driver Impact of Packet Loss (after 1.1 sec) 21m Car-1 Car-2 Car-2 starts deceleration 9
Impact of Packet Loss (after 3.1 sec) Car-1 Car-2 3m/sec Car-2 crashes with Car-1 at 3m/sec injuring the driver (crashing speed exceeds injury speed threshold) Packet loss delayed deceleration of Car-2 by 0.1s, which pushed the crashing speed over the injury threshold Comparison of Crashing Scenarios Safety System Car-1 Car-2 No Crash Warning Crash Warning (no packet loss) Crash Warning (first packet lost) Crashes at 20m/s Crashes at 20m/s Crashes at 20m/s Crashes at 2m/s (below injury speed threshold) Crashes at 20m/s Crashes at 3m/s (above injury speed threshold) 10
NEAR FIELD RELIABILITY Zhe Wang and Mahbub Hassan, Network Coded Repetition: A Method to Recover Lost Packets in Vehicular Communications", IEEE ICC 2011, Kyoto, Japan. Repetition at Source existing solution for recovering a lost beacon Can we improve? Xu, Q., T. Mak, et al. (2004). Vehicle-to-vehicle safety messaging in DSRC. Proceedings of the 1st ACM international workshop on Vehicular ad hoc networks, VANET04. Philadelphia, PA, USA. 11
The Magic of XOR If you XOR m n-bit packets, it produces a single n-bit packet A single XORed packet can help recovery of different packets A A 1 B B 2 3 Node-1 receives A and B Node-2 receives only A Node-3 receives only B A B = A A B A 1 B A B 2 3 B A = B A B Node-1 retransmits a single packet Node-2 recovers B and Node-3 recovers A from the same retransmitted packet Blind XOR For XOR, retransmitting node needs knowledge of reception status at potential receivers In the previous example, XOR would not be effective if Node-2 and Node-3 both received Packet A, Packet B, Neither A nor B But it is difficult to acquire reception knowledge in vehicular comms (ACK/NACK not practical) Blind XOR: XOR decisions are to be made without reception knowledge We explore Blind XOR for vehicular comms 12
Simulation Parameters MAC protocol Transmission Range Data Rate Slotted Non-persistent CSMA 150m 6Mbps Slot Length 16µs Packet Size 300 Bytes Inter-vehicle Space (same lane) Number of lanes 6 12m Reception Failure Probability as a Function of k at a distance of 100 meter There is an optimum k (not a surprise) Blind XOR reduces both optimal k and the RFP Less retransmissions, more recovery 13
Average Improvement with Blind XOR same optimized k for all distances k = 8 for SR, but 4 for Blind XOR Half as much retransmissions, but several folds reduction in failure probability No Repetition Simple Repetition Repetition with Blind XOR 10-fold improvement FAR FIELD RELIABILITY Zhe Wang and Mahbub Hassan, Blind XOR: Low-Overhead Loss Recovery for Vehicular Safety Communications", IEEE Transactions on Vehicular Technology, 61(1) January 2012. 14
Why XORing Two Packets Could be Counterproductive Previous attempts failed to gain by XORing two packets Theorem: When m=2, coding gain is greater than 1 if and only if p 1 >0.5 and p 2 >0.5 CRP Distribution and Blind XOR Opportunity For Blind XOR to produce positive gain, CRP has to be greater than 0.5 We find that 93% of CRPs are greater than 0.5 (for 150m range) 15
Number of packets XORed is important Recovery can be maximized by carefully selecting the number of packets (m) to be XORed based on the conditional reception probability (p) Blind XOR Can Gain Even Under Strict Time Limits Of all the retransmissions with CRP greater than 0.5: Percentage where XOR assembly (2 or more packets) was possible Percentage where optimal assembly was possible 16
Blind XOR Assembly Distribution with 50ms Limit 90% of the cases, we need to XOR only 2-3 packets! Blind XOR Performance under 50ms Limit Basic Relay XOR Relay 17
Dealing with High-Speed Vehicular Mobility The Sydney Measurement, 2008 Jun Yao, Salil Kanhere, and Mahbub Hassan, Improving QoS in High-speed Mobility Using Bandwidth Maps", IEEE Transactions on Mobile Computing, 11(4) April 2012. Wireless Channel Dynamics on the Road 18
PHY Solution Use adaptive modulation and coding (AMC) 3G base stations switch modulation every few microseconds? AMC keeps bit error rate low at all times But bit rate (bandwidth) fluctuates Fluctuating bandwidth needs to be dealt with at the higher layers Measurement Architecture Probe Server @ UNSW Downlink Probe (Packet Train) Probe Trigger (every 200m) Internet Bandwidth is measured every 200 meters of a road Provider B (HSDPA) Provider A (HSDPA) Provider C (pre-wimax) Probe Client 38 19
Measurement Hardware/Software Off-the-shelf Hardware (Soekris) Totally user-driven (no support from service provider) 39 Routes Taken Two routes (inbound: 7Km & outbound: 16.5Km) Typical urban driving speed ~70-80Kmh 75 repeated trips spread over 8 months (Aug07 Apr08) Collectively 60 driving hours & 1600Km inbound outbound UNSW 40 20
Distribution of available bandwidth is sensitive to road locations 31 19 Location = 500 meter road segments (Our Sydney 3G data) 41 Content streaming to vehicular user Geo-TFRC Makes use of bandwidth knowledge per road segment TFRC No bandwidth knowledge 42 21