Predictable Packet Delivery from Wireless Channel Measurements. Daniel Halperin Wenjun Hu, Anmol Sheth, David Wetherall

Transcription

1 Predictable Packet Delivery from Wireless Channel Measurements Daniel alperin Wenjun u, Anmol Sheth, David Wetherall

2 Wi-Fi technology Fast Mbps in n represents a 300x speedup in 12 years Reliable - vehicular speeds, extended range, stable hardware and software Ubiquitous - few dollars per chip allows integration everywhere Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 2

3 Wi-Fi technology Fast Mbps in n represents a 300x speedup in 12 years Reliable - vehicular speeds, extended range, stable hardware and software Ubiquitous - few dollars per chip allows integration everywhere New, exciting apps on the horizon Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 2

4 New apps stress network Wireless Display Wireless nput Mobile Wireless Daniel alperin, SGCOMM 2010, 3

5 New apps stress network All-wireless ome Daniel alperin, SGCOMM 2010, 3

6 New apps stress network All-wireless ome Performance really matters Daniel alperin, SGCOMM 2010, 3

7 Performance in theory Channel Measurements 39 Mbps Textbook Algorithms Rate Selection Daniel alperin, SGCOMM 2010, 4

8 Performance in theory Channel Measurements n practice, this has never worked! 39 Mbps Textbook Algorithms Rate Selection Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 4

9 Performance n practice Statistics-based Adaptation 65 Mbps? 65 Mbps? 65 Mbps? 52 Mbps? 13 Mbps? Nope! Nope! Nope! Nope! Okay! Problem: Convergence time Dynamic environments Large search spaces >300 tx configs in n Combined rate & power Both are trends Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 5

10 Goals: Bridging Theory and Practice Accurately predict performance over real channels Agile response to changing channels Leverage measurements available in real NCs Extend to n and more applications Key: an accurate channel metric Daniel alperin, SGCOMM 2010, 6

11 Goals: Bridging Theory and Practice Accurately predict performance over real channels Agile response to changing channels Leverage measurements available in real NCs Extend to n and more applications Key: an accurate channel metric Daniel alperin, SGCOMM 2010, 6

12 Today s talk Why it s hard to predict performance with RF measurements today Our solution: an accurate channel metric using Effective SNR Evaluation of Effective SNR in Wi-Fi Networks Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu

13 Today s talk Why it s hard to predict performance with RF measurements today Our solution: an accurate channel metric using Effective SNR Evaluation of Effective SNR in Wi-Fi Networks Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu

14 SNR based on RSS 100 Received Signal Strength ndicator Measures total power received in packet With Noise, gives SNR for packet Treated as if directly reflects performance E.g., NC manufacturers list per-rate sensitivity PRR Packet level SNR (db) n practice, SNR at which a rate starts to work can vary more than 10 db for real links Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 9

15 802.11: OFDM and MMO Orthogonal Frequency Division Multiplexing Multiple-nput Multiple-Output Power Frequency Frequency-selective fading Spatial diversity Daniel alperin, SGCOMM 2010, 10

16 802.11: OFDM and MMO Orthogonal Frequency Division Multiplexing Multiple-nput Multiple-Output Power Frequency Key: Different subchannels Frequency-selective fading have different SNRs Spatial diversity Daniel alperin, SGCOMM 2010, 10

17 Packet SNR for 4 faded links 35 SNR (db) Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 11

18 Packet SNR for 4 faded links 30 db 35 SNR (db) Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 11

19 Packet SNR for 4 faded links 30 db 35 SNR (db) db Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 11

20 Packet SNR for 4 faded links 30 db 35 Packet SNR SNR (db) db Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 11

21 Packet SNR for 4 faded links 30 db 35 Packet SNR SNR (db) Errors db Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 11

22 Packet SNR for 4 faded links 30 Fundamental db SNR mismatch 35 Packet SNR SNR (db) Errors db Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 11

23 An n opportunity n provides detailed channel measurements Used for advanced MMO techniques Channel State nformation (CS) measures MMO and OFDM! Matrix captures per-antenna paths One matrix per subcarrier Can we use it to predict packet delivery? n theory? n practice? Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 12

24 Today s talk Why it s hard to predict performance with RF measurements today Our solution: an accurate channel metric using Effective SNR Evaluation of Effective SNR in Wi-Fi Networks Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu

25 Effective SNR ntroduced by Nanda and Rege in 1998 Packet SNR: total power in the link Effective SNR: useful power in the link 35 SNR (db) Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 14

26 Effective SNR ntroduced by Nanda and Rege in 1998 Packet SNR: total power in the link Effective SNR: useful power in the link 35 SNR (db) Subcarrier index Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 14 Effective SNR

27 Using Effective SNR Channel Measurements 39 Mbps Textbook Algorithms Rate Selection Daniel alperin, SGCOMM 2010, 15

28 Using Effective SNR Channel State nformation (MMO & OFDM) 39 Mbps Textbook Algorithms Rate Selection Daniel alperin, SGCOMM 2010, 16

29 Using Effective SNR Channel State nformation (MMO & OFDM) Effective SNR Model 39 Mbps Rate Selection Daniel alperin, SGCOMM 2010, 17

30 Using Effective SNR Channel State nformation (MMO & OFDM) 1x65 1x52 2x26 3x13 Effective SNR Model Working Configurations; Application Decision Daniel alperin, SGCOMM 2010, 18

31 Using Effective SNR Channel State nformation (MMO & OFDM) 1x65 1x52 2x26 3x13 Effective SNR Model Working Configurations; Application Decision Daniel alperin, SGCOMM 2010, 19

32 Obtaining CS RX measures CS from packet preamble NCs do this for MMO/OFDM operation For every received frame Measures all antennas + subcarriers used A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3-antenna Link 3x3 Matrix One matrix per Subcarrier Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 20

33 Using Effective SNR Channel State nformation (MMO & OFDM) 1x65 1x52 2x26 3x13 Effective SNR Model Working Configurations; Application Decision Daniel alperin, SGCOMM 2010, 21

34 Computing Effective SNR CS A A B A B A B B D D E D E D E E Compute SNRs per symbol SNRs Single antenna link (1x1) CS gives the per-symbol SNR Multiple RX antennas (1xN) Maximal-ratio combining MMO link (MxN) Minimum mean-square error (MMSE) Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 22

35 Computing Effective SNR CS A A B A B A B B D D E D E D E E SNRs Modulation BPSK BER(ρ) Q 2ρ Compute BERs QPSK QAM-16 QAM-64 Q ( ρ) Q ρ/5 Q ρ/21 Textbook per symbol BERs formulas Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 23

36 Daniel alperin, SGCOMM 2010, BEReff Computing Effective SNR A B D E A B D E A B D E A B D E A B D E A B D E A B D E A B D E 24 Average: Effective BER SNRs BERs CS

37 Computing Effective SNR CS A A B A B A B B D D E D E D E E SNRs Convert back SNReff to SNR BEReff BERs Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 25

38 Using Effective SNR Channel State nformation (MMO & OFDM) 1x65 1x52 2x26 3x13 Effective SNR Model Working Configurations; Application Decision Daniel alperin, SGCOMM 2010, 26

39 Predicting Packet Delivery Effective SNR thresholds for each rate Threshold per NC implementation, not per NC or per channel Adds flexibility to handle real NCs ard vs soft decoding Other special techniques e.g., use optimal Maximum Likelihood receiver only for small modulations Daniel alperin, SGCOMM 2010, 27

40 Example Applications Rate/MMO/Channel width selection: What is the fastest configuration for this link? A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 28

41 Example Applications Rate/MMO/Channel width selection: What is the fastest configuration for this link? 1x3 A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 28

42 Example Applications Rate/MMO/Channel width selection: What is the fastest configuration for this link? 2x3 A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 28

43 Example Applications Rate/MMO/Channel width selection: What is the fastest configuration for this link? 3x3 A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 28

44 Example Applications Rate/MMO/Channel width selection: What is the fastest configuration for this link? A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz 20 Mz 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 28

45 Example Applications Power Consumption: Which receive antenna is best to disable to save power? RX Antenna Selection A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 29

46 Example Applications Spatial Reuse: What is the lowest transmit power at which can support 100 Mbps bitrate? Power A A B A B A C B A C B A C B A C D B A C D B E C D B E C D F E C D F E D F E D F G E D F G E F G E F G F G G G G 3x3, 40 Mz Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 30

47 Today s talk Why it s hard to predict performance with RF measurements Our solution building a better metric using Effective SNR Evaluation of Effective SNR in Wi-Fi Networks Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu

48 mplemented in ntel NC ntel Wi-Fi Link 5300 NC (3x3, 450 Mbps) Two testbeds with > 200 widely varying links Linux ( rc3) open source iwlwifi driver Firmware debug mode: send CS to RX host Real-time computation: ~4 µs per 3x3 CS Daniel alperin, SGCOMM 2010, 32

49 Evaluation Questions Does Effective SNR accurately predict packet delivery? Does an Effective SNR rate selection algorithm perform well? More results in the paper Wireless link transition region Transmit power control Collisions Daniel alperin, SGCOMM 2010, 33

50 Predicting Optimal 3x3 Rate Daniel alperin, SGCOMM 2010, 34

51 Predicting Optimal 3x3 Rate 65 Rate / stream (Mbps) Packet SNR SNR (db) (db) Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 34

52 Predicting Optimal 3x3 Rate 65 Rate / stream (Mbps) Packet SNR SNR (db) (db) Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 34

53 Predicting Optimal 3x3 Rate 65 Rate / stream (Mbps) Effective SNR SNR (db) (db) Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 35

54 Rate control evaluation a: Does Effective SNR match related work? ESNR versus SampleRate, SoftRate, OPT n: Does Effective SNR extend to n? ESNR versus OPT Channel simulation over mobile trace to compare against related work & vary speed MATLAB simulation + SoftRate GNU Radio Effective SNR algorithm gets corrupted CS Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 36

55 Effective SNR for a Avg. delivered rate (Mbps) ESNR SampleRate SoftRate SampleRate SampleRate fixed fixed retry Packet trace index (x100) Matches or beats a algorithms All within 15% of OPT Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 37

56 Effective SNR for a Avg. delivered rate (Mbps) ESNR SampleRate SoftRate SampleRate SampleRate fixed fixed retry No rate fallback on retries: % performance gap Packet trace index (x100) Matches or beats a algorithms All within 15% of OPT Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 37

57 ESNR extends to MMO 200 Avg. delivered rate (Mbps) Packet trace index (x400) OPT Previous-OPT ESNR 80% accuracy, 10% overselection 24 rates vs 8, larger gap vs Previous-OPT Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 38

58 Related work Daniel alperin, SGCOMM 2010, 39

59 Related work a SoftRate (2009) AccuRate (2010) Error Estim. Codes (2010) Effective SNR Daniel alperin, SGCOMM 2010, 39

60 Related work a MMO & Ant Sel. TX Power Channel Width Real NCs SoftRate (2009) AccuRate (2010) Error Estim. Codes (2010) Effective SNR Daniel alperin, SGCOMM 2010, 39

61 Conclusions For the first time, we can use measurements available in real NCs to predict packet delivery over real channels Matches good performance of existing rate adaptation algorithms and extends to n Applies to a broad problem space and provides a simple, practical AP for protocols Lots more in the paper! Daniel alperin, SGCOMM 2010, dhalperi@cs.washington.edu 40

62 Thanks! Questions?