Diagnosing Wireless Packet Losses in : Collision or Weak Signal?

Transcription

1 Diagnosing Wireless Packet Losses in : Collision or Weak Signal? Shravan Rayanchu Arunesh Mishra Dheeraj Agrawal Sharad Saha Suman Banerjee Wisconsin Wireless and NetworkinG Systems (WiNGS) Lab University of Wisconsin Madison INFOCOM 2008

2 The Goal Distinguishing between collision and weak signal Consider a wireless link:

3 The Goal Distinguishing between collision and weak signal Consider a wireless link:

4 The Goal Distinguishing between collision and weak signal Consider a wireless link: Q. What caused the packet loss?

5 The Goal Distinguishing between collision and weak signal Wireless Errors

6 The Goal Distinguishing between collision and weak signal Collision Wireless Errors

7 The Goal Distinguishing between collision and weak signal Wireless Errors Collision Weak Signal

8 The Goal Distinguishing between collision and weak signal Wireless Errors Collision Weak Signal Q. Can we discern between these two?

9 Collision vs. Weak Signal Q. Why is it important to distinguish between errors?

10 Collision vs. Weak Signal Q. Why is it important to distinguish between errors?

11 Collision vs. Weak Signal Q. Why is it important to distinguish between errors?

12 Collision vs. Weak Signal Q. Why is it important to distinguish between errors?

13 Collision vs. Weak Signal Inferring the cause of error Collision Detection is hard! Given an error packet, can we conduct a post-mortem?

14 Collision vs. Weak Signal Inferring the cause of error Collision Detection is hard! Given an error packet, can we conduct a post-mortem?

15 Collision vs. Weak Signal Inferring the cause of error Collision Detection is hard! Given an error packet, can we conduct a post-mortem?

16 Collision vs. Weak Signal Inferring the cause of error Collision Detection is hard! Given an error packet, can we conduct a post-mortem? Example

17 Collision vs. Weak Signal Inferring the cause of error Collision Detection is hard! Given an error packet, can we conduct a post-mortem? Example

18 Collision vs. Weak Signal Inferring the cause of error Collision Detection is hard! Given an error packet, can we conduct a post-mortem? Example

19 A Simple Approach

20 A Simple Approach

21 A Simple Approach Metrics used to discern the cause: Received signal strength (RSS) Bit error rate (BER) Error rate per symbol (EPS) Symbol error rate (SER) Symbol error burst length (S-Score)

22 A Simple Approach Metrics used to discern the cause: Received signal strength (RSS) Bit error rate (BER) Error rate per symbol (EPS) Symbol error rate (SER) Symbol error burst length (S-Score)

23 A Simple Approach Metrics used to discern the cause: Received signal strength (RSS) Bit error rate (BER) Error rate per symbol (EPS) Symbol error rate (SER) Symbol error burst length (S-Score)

24 A Simple Approach Metrics used to discern the cause: Received signal strength (RSS) Bit error rate (BER) Error rate per symbol (EPS) Symbol error rate (SER) Symbol error burst length (S-Score)

25 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error

26 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error

27 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error Collision:

28 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error Collision:

29 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error (Higher in collision?) Received Signal Strength (RSS) RSS (S+I/n)

30 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error (Higher in collision?) Received Signal Strength (RSS) RSS (S+I/n)

31 Metrics : Intuition Bit Error Rate (BER) Percentage of total bits in error (Higher in collision?) Received Signal Strength (RSS) RSS (S+I/n) (Lower in weak signal?)

32 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error

33 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error

34 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error

35 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) SER = 5/10 = 0.5

36 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error

37 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error

38 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error

39 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error (Higher in collision?)

40 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error (Higher in collision?) S-Score Measure of number of consecutive symbols in error

41 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error (Higher in collision?) S-Score Measure of number of consecutive symbols in error

42 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error (Higher in collision?) S-Score Measure of number of consecutive symbols in error

43 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error (Higher in collision?) S-Score Measure of number of consecutive symbols in error

44 Metrics : Intuition Symbol Error Rate (SER) Percentage of symbols which are in error (Higher in collision?) Error Per Symbol (EPS) Percentage of bits in error averaged over the symbols which are in error (Higher in collision?) S-Score Measure of number of consecutive symbols in error S-Score = n i=1 B i 2

45 Summary of Approach

46 Experiment Design Causing errors due to weak signal Weak Signal Environment free of other transmissions Enabled reception of packets in error Client mobility induced errors due to dynamic channel conditions

47 Experiment Design Causing collisions Collisions Disabled backoffs, enabled reception of packets in error Packet logs at the receivers are synchronized using common packets Collisions are identified using overlap in packet transmission times

48 Empirical Results : BER 100 S( 24, 36,48 ) 80 CDF (%) C( 24,36,48 ) Bit Error Rate (BER) 98% of weak signal packets have a BER of 12% or less 26% of collision packets have BER of 12% or less Cutoff value of 12% BER: Detects 74% of collisions with 2% false positives

49 Empirical Results : EPS CDF (%) S( 24, 36, 48) C( 24, 36, 48) Error Rate Per Symbol (EPS) 98% of weak signal packets have an EPS of 22% or less 30% of collision packets have the same EPS of 22% or less.

50 A Metric-Vote Scheme Metric-Vote Output a collision if any of the metrics vote for a collision

51 A Metric-Vote Scheme Metric-Vote Output a collision if any of the metrics vote for a collision Performance Table: Accuracy for Collision/Weak Signal BER EPS S-Score Metric-Vote Collision Weak Signal Accuracy: % of weak signal (or collision) packets which are correctly identified

52 Some Observations Why is the accuracy low for collision packets?

53 Some Observations Why is the accuracy low for collision packets? Strong Capture Effect

54 Some Observations Why is the accuracy low for collision packets? Strong Capture Effect

55 Some Observations Why is the accuracy low for collision packets? Strong Capture Effect Colliding Packet Size

56 A Joint Metric : SER-EPS Symbol Error Rate (SER) Signal Collision (1400,1400) Collision (1400,200) Error Per Symbol (EPS)

57 Reference Implementation Platforms: Linux based laptop, Netgear SPH101 VoWiFi phone COLLision Inferencing Engine (COLLIE) AP relays the error packet back to the client Client performs collision inferencing COLLIE based Link Adaptation Enhanced Auto Rate Fallback to make it collision-aware

58 Results (1) Mobile Scenario Mobile Client, Presence of other traffic Throughput (kbps) w/ COLLIE w/o COLLIE Time (secs) Throughput improvement 30%

59 Results (2) Collision Scenario Static client, Presence of additional collision sources Throughput (kbps) w/o COLLIE w COLLIE Run1 Run2 Run3 Throughput improvement as high as 60%

60 Results (3) Voice call emulation Netgear SPH-101 VoWiFi phone using TI chipset and proprietary rate adaptation algorithm Avg Retransmissions (%) w/o COLLIE w COLLIE Slow Medium High Reduction in wasted retransmissions 40%

61 Summary and Future Work Summary We addressed the fundamental question of what caused a packet to be in error collision or weak signal? Distinguishing between errors lead to improvement in throughput, energy efficiency Future Work Design better metrics Design a low overhead protocol Study the impact of non interference sources Enhance/design link adaptation mechanisms

62 Questions?

63 Backup slides

64 Empirical Results : Other Metrics S-Score Cutoff value of 500: 98% of signal packets and 26% of collision packets RSS High variation Delivery probability is a function of S/(I + n) instead of (S + I)/n, receiver sensitivity

65 Empirical Results : RSS CDF (%) S( 24,36,48) C( 24,36,48 ) RSS 98% of packets in error due to weak signal have an RSS of about -73 dbm or less 10% of packets suffering collision have RSS of -73 dbm or less

66 Empirical Results : S-Score 100 CDF (%) S( 24, 36, 48) C(24, 36, 48) S Score 98% of the weak signal packets have an S-Score of 500 or less 26% collision packets have an S-Score of 500 or less

67 Multi-AP Assistance Collision Detection Accuracy Basic Multi AP Basic Multi AP Basic Multi AP Basic Multi AP High Capture Effect (Multi AP approach improves accuracy) Low Capture Effect (Basic approach works well) APs are synchronized (using opportunistic common packet receptions) Information about packet reception is aggregated at the COLLIE server