1. Downlink Traffic Scheduling in Green Vehicular Roadside Infrastructure
Abdulla A. Hammad
Department of of Electrical and Computer Engineering
McMaster University
Hamilton, Ontario, CANADA

2. Outline Introduction Energy Efficient Downlink Scheduling
VBR scheduling
CBR scheduling

3. VANET Applications

4. VANET Architecture V2V, V2I, hybrid and I2I communications
Roadside Unit (RSU)
I2I
V2V, V2I, hybrid and I2I communications
Roadside Unit (RSU) is fixed infrastructure that enables vehicle-to-infrastructure (V2I) communications.

5. VANET overview Applications Architecture: OBU and RSU
Safety (e-brakes, accidents)
Infotainment(VoD, file transfer, internet)
Traffic (Monitoring, management)
Architecture: OBU and RSU
Difference from MANET
Fast link disconnection
Temporal and spatial changing traffic density
Movement confined to road network
Energy Limits
RSU role
Increase network connectivity
Store delay tolerance contents
Internet connectivity
Safety: accidents and curvy road alerts

6. Green Vehicular Infrastructure?
S. Peirce and R. Mauri, “Vehicle-Infrastructure Integration (VII) Initiative: Benefit-Cost Analysis: Pre-Testing Estimates”, Intelligent Transportation Systems Joint Program Office, United States Department of Transportation, Washington, DC., March 30, 2007.
Includes cost projections for an initial national vehicular infrastructure deployment (1B USD expenditure)
40% of all rural freeway roadside infrastructure would be solar powered. Unavailable power grid connectivity.
over 63% of roadside unit costs consumed by solar provisioning costs, e.g., solar panel, battery and associated electronics. power savings ↑ energy provisioning cost ↓
Objective is to decrease energy use at the RSUs as aggressively as possible.
How much energy can we save using energy aware downlink scheduling?

7. V2I Roadside Unit (RSU)
When a vehicle v arrives into RSU coverage, it communicates its Rv bit (delay tolerant) request. The RSU schedules its downlink response for some future time.
Vehicles have unlimited energy reserves. RSU is energy constrained → focus on reducing downlink energy use.

8. Air Interface Assumptions
Constant bit rate (CBR)
Power control is used to adapt to channel conditions.
Packet or slot based
Schedulers try to minimize the downlink transmission energy needed to process vehicle requests.

9. Air Interface Assumptions
Variable bit rate (VBR)
fixed transmit power. The bit rate is adapted to channel conditions.
Fixed size packet lengths at different bit rates
Schedulers try to minimize the amount of downlink transmission time required to process vehicle requests.

10. Delay-Tolerant CBR Example
(Distance dependent exponential path loss, e.g., α = 3)
For example:
Choice of Vehicle i communication time, e.g., t2 vs t1.
Huge energy savings when communicating over shorter distances.
Requires delay tolerance! Requires energy cost estimates.

11. Another Example
Vehicle w is moving away from energy favourable locations. Vehicle v is moving towards energy favourable locations.
Therefore, we should serve Vehicle w first!

12. Another Example
How to schedule when there are different vehicular speeds and communication requirements?
It seems that faster moving vehicles with higher communication requirements should be given scheduling priority.

13. Part 1: Variable Transmission Power - CBR
Model Assumptions
Vehicles can store (buffer) stream more than it needs currently
No vehicle to vehicle communication
RSUs coverage do not overlap
BW is velocity independent

14. Mixed Integer Linear Program Formulation
Offline Energy Bounds
Packet-Based Scheduling:
Modeled as single machine scheduling with deadlines
NP hard
Timeslot-Based Scheduling
Mixed Integer Linear Program Formulation

15. Minimum Cost Flow Graph Formulation

16. Online Time slot-based Algorithms
Greedy Minimum Cost Flow (GMCF)
Event driven
Network Flow formed and Minimum Cost computed
Seeks global minima for all vehicles in range
High computation cost
Static Scheduler (SS)
Seeks local minima for all vehicles in range
Two phases: weight computation, scheduling
Weight computation: happens once, based on preferred time slots
Scheduling rule: static, based on highest weight first
Average computation cost
Weakness: bad performance under high variation in speed
Nearest Fastest Set (NFS) Scheduler
Event driven, preferred slots selected upon arrival
Scheduling:dynamic: only resolves contention when they actually happen
Strength: Low computation cost, handles high variation in speed
Weakness: High energy cost, due to delayed contention resolution

17. Performance Evaluation
Two sets of results
Best case scenario: accurate prediction of speed and position based on deterministic path loss scenario
Average case scenario: including shadowing effect (minimal in HW environment because of LoS)
Traffic Model
Difference between urban and HW environment
Tendency to maintain speed
Lane speed limits (different speed classes)
Arrival model: Poisson Distribution
Traffic direction

18. Performance Evaluation
Effect of speed
Two Classes:
18m/s
18,23,28,33 m/s
No shadowing
Light and medium load
arrival rate:1/22 v/s
Light Load: GMCF close
to Bound SS
More energy
Less comp. intensive

19. Effect of demand
No shadow
3 Classes
Fig.7: same speed
Fig.8:18,24,33 m/s

20. Shadowing: 4,12 dB vs increasing demand

21. Platooning
Fig.12: same speed
Fig.13: Three different speeds

22. Publications
AA Hammad, GH Badawy, TD Todd, AA Sayegh, D Zhao, IEEE Global Telecommunications Conference (GLOBECOM 2010)
Hammad, A.A.; Todd, T.D.; Karakostas, G.; Dongmei Zhao, "Downlink Traffic Scheduling in Green Vehicular Roadside Infrastructure," IEEE Transactions on Vehicular Technology

23. Part2: Variable Bit-rate
Model Assumptions
RSU with Constant Transmission power
Limited, non-overlapping coverage area
RSU battery powered with renewable power source
Highway environment traffic (constant velocity and platooning)
Vehicle announce requirement, velocity and location upon arrival
Vehicle demand can be satisfied by different combinations of time slots between which the received bit rate can be differ.
The objective is to minimize the number of time slots during which the AP is transmitting, while satisfying all vehicle demands.

24. Integer Programming Model

25. Analysis of VBR Scheduling Problem

27. Problem Relaxation
Relaxation can be done by relaxing the binary condition.
Problem: unbounded Integrality gap
Solver can assign 1 time slot to m vehicles each assigned 1/m of the slot

28. Offline Energy Bounds Generalized Flow Graph
Dynamic Network Topology Graph

29. Generalized Flow Graph
Fast computations based on special combinatorial algorithms
Ease of approximation to integral solution due to high percentage of flows are integers

31. Dynamic Network Topology Graph
GF drawbacks:
Time slots inversely proportional to velocity
Depending on slot size: large period of times may encounter no bit rate change.
DNTG: includes time in model

32. DNTG: Time Expanded Graph

33. Online Scheduling Algorithms
Bounds provide theoretical optimum energy usage (but assumes future arrivals knowledge)
Online Scheduling Algorithms
FCFS
Fastest First
Greedy GF Algorithm
Greedy DNTG Algorithm

34. FCFS First Come First Serve
Vehicles are assigned best possible time slots if they were not reserved for earlier arriving vehicles
Static Assignment: no decision revoking
Simple to implement
Fairness questionable

36. Fastest First Algorithm
Faster moving vehicle spend less time inside the RSU range than slower ones.
To increase fairness, faster vehicles are assigned slots prior to slower ones.
Not static: decisions can be revoked.

38. Greedy GF Algorithm Use GF design used to calculate the Bound
No future knowledge of arrival
Event based: activates upon vehicle arrival
Integer approximation upon execution

39. Greedy GF Algorithm

40. Greedy DNTG Algorithm
Greedy Implementation of the Dynamic Network Topology Graph
Event-based: triggered upon arrival of new vehicles
Outline similar to Greedy GF Algorithm outlined before

41. Performance Comparison
Highway environment
Vehicles maintain velocity/platoon
Poisson Arrival
Vehicles announce velocity,location and requirements
Dropping is allowed
Deterministic exponential path-loss and Log-normal shadowing component

42. Throughput (no shadowing random component)
Two classes of vehicles (18, 30 m/s)
Arrival rate 1/28 v/s
Platoon 10%
No Dropping

43. Demand Drop

44. Online algorithms under high demand
Two classes: 18, 30 m/s
Platooning 10%
Arr.rate: 1/30 v/s
Throughput saturates

45. Dropping under high demand

46. Energy vs Speed

47. Shadowing: throughput

48. Shadowing: Dropping

49. Jain’s Fairness Index
As dropping is allowed, fairness across algorithms needs to be measured

50. Jain’s Fairness Index

51. Thank you