1 / 40

Performance Enhancement of the EDCA Mechanism for DSRC Safety Communication

Performance Enhancement of the EDCA Mechanism for DSRC Safety Communication . Sarah Sharafkandi, Gaurav Bansal, John Kenney Toyota InfoTechnology Center, USA Presentation to IEEE 1609 WG June 18, 2012. Outline. Background Study of homogeneous case Study of heterogeneous case

maureen
Télécharger la présentation

Performance Enhancement of the EDCA Mechanism for DSRC Safety Communication

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Performance Enhancement of the EDCA Mechanism for DSRC Safety Communication Sarah Sharafkandi, Gaurav Bansal, John Kenney Toyota InfoTechnology Center, USA Presentation to IEEE 1609 WG June 18, 2012

  2. Outline • Background • Study of homogeneous case • Study of heterogeneous case • Isolation Technique • Virtual Division Technique

  3. Background • Focus of this study • Scope … • Goal …

  4. Focus: DSRC Safety • DSRC Safety Applications prevent collisions • In US, Ch. 172 used for safety • Dominated by period SAE Basic Safety Messages (BSMs) • May include other messages, e.g. SPaT, MAP, also likely to be periodic • Similar situation in Europe with Control Channel: • Cooperative Awareness Messages (CAMs) • DENM and other messages as well. • MAC protocol is 802.11p: CSMA/CA

  5. Scope: EDCA • Improve DSRC safety by reducing 802.11 MAC frame collisions for periodic traffic • Use standard Enhanced Distributed Channel Access (EDCA)

  6. Goals • Determine optimal strategy for EDCA on DSRC Safety Channel: • Parameter selection • Adaptive Inter-Frame Space Number: AIFSN • Contention Window minimum size: CWmin • Access Category use - Two cases: • Homogeneous: No classification of frames based on perceived differences in value of reception • Heterogeneous: Two or more classes of frames, with critical class comprising 10% or less of whole • Eventually standardize use of EDCA for DSRC Safety Channel (but not today)

  7. EDCA • Up to 4 queues, each associated with an Access Category (AC) • Head-of-line frame in a queue competes for channel access • Competes with frames in other STAs • Competes with other queues in same STA • Competition governed by: • Inter-Frame Space parameter (AIFSN) • Contention Window parameter (CWmin)

  8. EDCA: After busy period • When frame advances to head-of-line, choose countdown integer from [0, CWmin] • When channel becomes idle, decrement countdown 1 per slot time (13 μsec) • Channel considered idle at SIFS + (AIFSN * 13 μsec) after end of prior frame Lowest Priority = AC_Background AC0 (AIFSN =9, CW=15) Data AC1 (AIFSN =6, CW=7) Data AC_Best Effort AC2 (AIFSN =3, CW=3) AC3 (AIFSN =2, CW= 3) Data AC_Voice SIFS Prior Data Data Highest Priority = AC_Video Countdown while medium is idle Suspend countdown when medium becomes busy Defer access

  9. SIFS Interrupted Countdown AIFS Station A SIFS Data 4 3 2 1 0 AIFS AIFS Station B SIFS Busy medium Data 2 1 0 7 6 5 4 3 Countdown frozen at 3 during busy period

  10. Considerations • EDCA is designed for for “typical” LAN traffic. Well studied in that domain. • Video & Voice: delay sensitive, loss tolerant • Best Effort: delay tolerant, usually loss sensitive • Safety traffic is atypical • Dominantly periodic • Latency sensitive, but not like “real time” • Compare with usual video/voice assumptions • Channel access latencies on order of 10s msec OK • Loss sensitive, but not as strict as file transfer • Tolerable reception gaps on order 100s msec Specific Goal: minimize frame collisions while keeping queuing delay on order 10s msec

  11. Limitations • Assume initially all nodes are within reception range of each other

  12. Outline • Background • Study of homogeneous case • Study of heterogeneous case • Isolation Technique • Virtual Division Technique

  13. Simulation Setup • Simulation tools: • MATLAB, NS2 • Topology: • All the stations at the same location • Load: • 20 to 300 nodes each sending packets at the message rate of 10HZ

  14. Effect of AIFSN • All frames detect idle channel at same time (homogeneous case) • Increasing AIFSN just wastes time, reduces capacity

  15. Varying AIFSN • PER increases steadily with increasing AIFSN • Conclusion: Choose AIFSN = 2 (minimum allowed) for homogeneous case • No trade-offs involved

  16. Effect of CW • Low CW reduces average queue latency • Assume STA A and STA B enter countdown during same busy period • If they choose the same countdown value their frames will eventually collide • Probability of that is 1/(CW+1), so low CW increases chance of that collision • If 3 or more become active during a busy period, the probabilities are more complex • STA A’s frame can also collide with a STA that enters countdown during a different busy period • Analytical approach to computing collision probability is very difficult. Open research question.

  17. Effect of CWmin on PER Effect of CW goes away at high load • Clear PER advantage to higher CW for low-to-moderate load • Advantage diminishes with increasing CW (gap getting smaller) • Higher CW gives lower PER for low-to-moderate loading • Higher CW leads to higher latency • Therefore a trade-off exists

  18. Choosing CW • At very high load, the statistics of the nodes waiting in countdown are the same after every a busy period • Therefore, probability of colliding when node counts down to zero becomes independent of how long the node has been counting down, i.e. independent of CW (like slotted ALOHA)* • Every active queue counts down 1 per busy period • Max. queuing delay ≈ CW * busy period • For BSM-dominated channel, busy period ≈ 0.5 msec • For CW = 31, max. queuing delay on the order of 15 msec. Doubling to CW = 63 would double max delay with small additional PER benefit • Conclusion: For Homogeneous case, a single AC with AIFSN = 2 and CWmin = 31 is a good choice. * Congestion Control for Vehicular Safety: Synchronous and Asynchronous MAC Algorithms, Subramanian et al., 9th ACM VANET Workshop, June 2012

  19. Outline • Background • Study of homogeneous case • Study of heterogeneous case • Isolation Technique • Virtual Division Technique

  20. Classification • Now assume we classify the Safety Channel traffic into two or more ACs • Not concerned here with rules of classification. Just assume “high priority” AC has 10% - not a magic number, just convenient. • How to classify is hard problem, out of scope here: • Could be segregation of BSMs into small “critical” subset and large “regular” subset • Could involve a third class for SPaT, MAP, Security, Management, … • Take a look at performance using aggressive ACs, with 10% traffic in AC3 and 90% in AC2

  21. AC3 vs AC2 vs Homogeneous Heterogeneous Low Priority (LP) AC2 AIFSN = 3, CW = 3 Homogeneous AC3 AFISN = 2, CW = 3 Heterogeneous High Priority (HP) , AC3 AIFSN=2, CW=3 • Removing 90% of lower priority traffic from AC3 improves PER significantly • PER of that lower priority traffic is a bit higher than homogeneous case • Can we do better?

  22. Competition between AC3 and AC2 AC3 (High priority): AIFSN=2, CWmin=3 AC2 (Low priority): AIFSN=3, CWmin=3 AIFSN=3 AIFSN=2 SIFS Data Data CW=3 Slot reserved for high priority AC2 offers no competition in 3rd slot after SIFS

  23. Isolation Idea • High priority messages: • Further decrease the PER of HP class by completely isolating it from low priority messages • High priorities only collide within themselves • Lower number of high priority messages • Low priority messages: • Pros: Less intra-class collision • Cons: Less capacity

  24. Isolating Traffic Classes High priority AIFSN=2, CWmin=3 Low priority AIFSN=6, CWmin=3,7,… AIFSN=6 AIFSN=2 SIFS Data CW=3 Two classes won’t collide! AIFSN(LP) = AIFSN(HP) + CWmin(HP) +1 6 = 2 + 3 +1

  25. Isolating High Priority Traffic Blue and red are same as prior graph (2,3) and (3,3) LP with Isolation: PER improved for low/moderate load, somewhat higher for high load Note dramatic reduction in HP PER when isolation is applied

  26. Isolation for 3 ACs • Isolation can be applied to two ACs: • HP = (2,3) • LP = (6,x) x = 15 on prior slide • Isolation could be applied to 3 ACs: • HP = (2,3) • MP = (6,7) • LP = (14,x) • Isolation could be applied to 4 ACs, but AIFSN of 4th might lead to excessive latency

  27. Outline • Background • Study of homogeneous case • Study of heterogeneous case • Isolation Technique • Virtual Division Technique

  28. Observations from Isolation • Average PER of low and high priority traffic is a function of percentage of traffic at each priority class • In some cases, average PER can be even lower than the PER of homogeneous traffic with the “optimal” single-AC EDCA parameters • This relates to convexity of PER vs Load curve

  29. Convexity Example If 150-node traffic is divided into 3 ACs, the aggregate PER might improve

  30. Virtual Division • Idea: Within a set of traffic that is nominally of equal reception value, classify it into two or more ACs. We call this “virtual division,” since it is not based on a natural difference within the traffic. • How about fairness?: • Any given node assigns its traffic on a rotation basis to the set of virtual division ACs. Could be according to a deterministic schedule, or could be probabilistic. • We can subdivide to more than 2 categories

  31. Virtual Division: Homogeneous Virtual Classifier 3-way AC1 BSM sequence from a vehicle AC2 Alternating assignment among ACs with 1/3, 1/3, 1/3 distribution AC3

  32. Homogeneous: Virtual Division • Virtual Division offers general PER advantage with 1/3, 1/3, 1/3 split • More research needed to determine optimal split ratio

  33. Heterogeneous Traffic • What if we have “real” high priority class at 10% load? • Solution: • Keep EDCA parameters fixed for HP (CWmin=3, AIFSN=2) • Artificially Subdivide LP into two categories: • Class1: AIFSN=6, CW=7 • Class2: AIFSN=14, CW=15

  34. Virtual Division: Heterogeneous Classifier Virtual Classifier AC1 Alternating AC2 and AC1 BSM sequence from a vehicle AC2 Critical vs Regular Critical content BSMs AC3

  35. Heterogeneous: Virtual Division • Virtual Division offers general PER advantage with ½, ½ split among “regular” class • More research needed to determine optimal split ratio • PER of HP class unchanged

  36. IPG 95% quantile (Heterogeneous) • IPG = Inter-Packet reception Gap. Relates to application observability of remote vehicle • General IPG advantage with virtual division, especially at high load

  37. Conclusion IEEE 802.11p Default EDCA Parameter Set For Safety Channel: Use these defaults with virtual division • Determine if safety channel is homogeneous or heterogeneous, and proportions for latter • Determine optimal splits of virtual division

  38. A note about hidden nodes • In a hidden-node environment, advantages from EDCA shown above are muted • Reducing collisions among neighbor nodes leads to increased channel load for a given number of vehicles. Higher channel load increases probability of hidden node collisions • We have studied. Advantages are still present, but are not as dramatic.

  39. Interaction of EDCA and Congestion Control • Above work assumes constant rate traffic. • EDCA work and Congestion Control both trying to improve throughput – complementary • Some congestion control approaches adapt message rates – no longer constant • Further study warranted • Does improved EDCA suggest different target channel load? • Does adaptive message rate impact optimal splitting ratios?

  40. Thank you!

More Related