1. 1 The Importance of Being Wireless Romit Roy Choudhury

2. 2 The Context The edge of the internet becoming wireless  167,000 hotspots by 2008 end [GartnerSurvey06]  75 million user base  Mesh extensions offered VoIP in rural regions  Future predictions are unanimously positive Mobile phone sales will soon surpass computers OLPC + Wireless will bridge the digital divide The Vision  Let’s make wireless like electricity Let everyone take it for granted

3. 3 Wireless Networking & Mobile Computing RFID, Sensors Networks Personal Networks Personal Networks Mesh Networks Internet Offering information access anytime, anywhere

4. 4 Miles To Go … PHY MAC / Link Network Transport Security Application Localization Channel fluctuations Spatial Reuse Mobility Energy Savings Eavesdropping Loss Discrimination Privacy Interference Mgmt. Mobile Social Apps

5. 5 Webpage http://synrg.ee.duke.edu SyNRG

6. 6 PHY MAC / Link Network Transport Security Application Our Research Exploiting PHY Layer Capabilities (bottom up) Exploiting PHY Layer Capabilities (bottom up) Application Driven Research (top down) Application Driven Research (top down)

7. 7 Selected Projects Spotlight Shuffle Micro-Blog Mingle Wireless Networking Mobile Computing

8. 8 Selected Projects Spotlight Shuffle Micro-Blog Mingle Wireless Networking Mobile Computing

9. 9 Spotlight: Exploiting Smart Antennas for Wireless Networks

10. 10 Wireless Networking & Mobile Computing RFID, Sensors Networks Personal Networks Personal Networks Mesh Networks Internet Offering information access anytime, anywhere

11. 11 Internet Omnidirectional Antennas

12. 12 IEEE 802.11 with Omni Antenna D S X M K silenced Y Data ACK D silenced E A C F B G `` Interference management `` A crucial challenge for dense multihop networks

13. 13 Managing Interference Several approaches  Dividing network into different channels  Power control  Rate Control … Recent Approach … Exploiting antenna capabilities to improve the performance of wireless multihop networks

14. 14 From Omni Antennas (Bulbs) … D S X M K silenced Y D E A C F B G

15. 15 To Beamforming Antennas (Spotlights) D S X M K Y D E A C F B G

16. 16 To Beamforming Antennas (Spotlights) D S X M K Y D E A C F B G

17. 17 However … Sending Rate (Kbps) Throughput (Kbps)

18. 18 Selected Projects Spotlight Shuffle Micro-Blog Mingle Wireless Networking Mobile Computing

19. 19 Shuffle: A New Way to Cope with Wireless Interference

20. 20 Successive Interference Cancellation (SIC) State of the art allows only one reception  The stronger one SIC enables a receiver to receive both signals  Stronger signal decoded and subtracted  Residual signal decoded from the residue

21. 21 SIC based WLANs Existing schemes require SINR >   Game of out-shouting each other SIC offers payoff if transmitter  Either out-shouts or whispers  Fundamental changes for protocol design

22. 22 MIM + SIC Ongoing work on GNU radios  SIC + MIM implementation can enable protocols

23. 23 Wireless BEN Questions / thoughts  Should BEN venture into the wireless world Or is that something to build once we have the wired infrastructure  Wireless networks have several interdependences with wired backend Important to take these issues during system design BEN has the potential to become one such system  Wireless network management critical Wired network management tools/expertise is a new opportunity  Can BEN be the bridge between research prototype and real deployments Disaster relief Rural internet access Healthcare and education …

24. 24 Selected Projects Spotlight Shuffle Micro-Blog Mingle Wireless Networking Mobile Computing

25. 25 Mingle: Exploiting Social Behavior for Routing in Delay Tolerant Networks

26. 26 Delay Tolerant Networks Exploit mobility as an opportunity  When wireless connection unavailable

27. 27 Selected Projects Spotlight Shuffle Micro-Blog Mingle Wireless Networking Mobile Computing

28. 28 Virtual Information Telescope

29. 29 Context Next generation mobile phones will have large number of sensors Cameras, microphones, accelerometers, GPS, compasses, health monitors, …

30. 30 Context Each phone may be viewed as a micro lens Exposing a micro view of the physical world to the Internet

31. 31 Context With 3 billion active phones in the world today (the fastest growing comuting platform …) Our Vision is …

32. 32 Internet A Virtual Information Telescope

33. 33 One instantiation of this vision through a system called Micro-Blog - Content sharing - Content querying - Content floating

34. 34 Content Sharing Virtual Telescope Cellular, WiFi Cellular, WiFi Visualization Service Web Service People Physical Space Phones

35. 35 Content Querying Virtual Telescope Cellular, WiFi Cellular, WiFi Visualization Service Web Service Phones People Physical Space Some queries participatory Is beach parking available? Some queries participatory Is beach parking available? Others are not Is there WiFi at the beach café? Others are not Is there WiFi at the beach café?

36. 36 Content Floating [on physical space] superb sushi Safe@ Nite? Safe@ Nite?

37. 37 If designed carefully, a variety of applications may emerge on Micro-Blog

38. 38 Applications Tourism  View multimedia blogs … query for specifics Micro Reporters  News service with feeds from individuals On-the-fly Ride Sharing  Ride givers advertize intension w/ space-time sticky notes  Respond to sticky notes once you arrive there  Negotiate deal on third party server Virtual order on physical disorder  Land in a new place, and get step by step information on your mobile

39. 39 Micro-Blog Beta live at http://synrg.ee.duke.edu/microblog.html

40. 40 Prototype

41. 41 Thoughts Micro-Blog: Rich space for applications and services But where exactly is the research here ???!!**

42. 42 Several research challenges and opportunities 1.Energy-efficient localization 2.Symbolic localization through ambience sensing 3.Location privacy 4.Incentives 5.Spam 6.Information distillation 7.User Inerfacing … Our Research

43. 43 Problem I Energy Efficient Localization (EnLoc)

44. 44 To GPS or not to GPS GPS is popular localization scheme  Good error characteristics ~ 10m Apps naturally assume GPS  Shockingly, first Micro-Blog demo lasted < 10 hours

45. 45 Cost of Localization Performed extensive measurements  GPS consumes 400 mW, AGPS marginally better  Idle power consumption 55 mW

46. 46 Alternate Localization WiFi fingerprinting, GSM triangulation  Place Lab, SkyHook … Improved energy savings  WiFi 20 hours  GSM 40 hours At the cost of accuracy  40m +  400m +

47. 47 40 Tradeoff Summary: 20 Research Question: Can we achieve the best of both worlds 400

48. 48 Given energy budget, E, Trace T, and location reading costs, e gps, e wifi, e gsm : Schedule location readings to minimize avg. error Given energy budget, E, Trace T, and location reading costs, e gps, e wifi, e gsm : Schedule location readings to minimize avg. error Formulation L(t 0 )L(t 1 ) L(t 2 ) L(t 3 ) L(t 4 ) L(t 6 ) L(t 7 ) Error t0t0 t1t1 t2t2 t3t3 t4t4 t5t5 t6t6 L(t 5 ) t6t6 Accuracy gain from GPS Accuracy gain from WiFi GPS WiFi

49. 49 Dynamic Program Minimize the area under the curve  By cutting the curve at appropriate points  Number of (GPS + WiFi + GSM) cuts must cost < budget

50. 50 Offline optimal offers lower bound on error Online algorithm necessary Online optimal difficult Need to design heuristics

51. 51 Our Approach Do not invest energy if you can predict (even partially)

52. 52 Predictive Heuristics Prediction opportunities exist  Exploit habitual mobility patterns / inertia  Population distribution can be leveraged Prediction also incorporated into Dynamic Program  Optimal computed on a given predictor Error t0t0 t1t1 t2t2 t3t3 t4t4 t5t5 t6t6 t6t6 Prediction generates different error curve

53. 53 1. Simple Interpolation Take GPS reading, followed by WiFi  Extrapolate GPS in the direction of WiFi  Reset prediction after threshold time, take GPS again

54. 54 2. Mobility Profiling Build logical mobility tree per-user  Each link an uncertainty point (UP)  Sample location only when uncertain  Location predictable between UPs Exploit acclerometers  Predict traffic turns  Periodically localize to reset errors Home Gym Vine/Mich intersection Vine/Mich intersection Library Grocery Office 8:00 8:15 8:30 12:00 8:05 12:05 3:30 5:30 6:00

55. 55 3. Exploit Group Behavior Characterize how masses behave at uncertainties  Example: At traffic intersections  Predict individual mobility based on mass behavior Goodwin & Green U-TurnStraigh t RightLeft E on Green 00.8810.0390.078 W on Green 000.5960.403 N on Goodwin 00.6400.3590 S on Goodwin 00.51300.486

56. 56 Buy Accuracy with Energy Comparison of optimal with simple interpolation  GPS clearly not the right choice

57. 57 Thoughts Localization cannot be taken for granted  Critical tradeoff between energy and accuracy Substantial room for saving energy  While sustaining reasonably good accuracy However, physical localization  May not be the way to go …  Several motivations to pursue symbolic localization

58. 58 Problem/Opportunity 2 Symbolic localization via ambience sensing (SurroundSense & AAMPL)

59. 59 Symbolic Localization Services may not care about physical location  Symbolic location often sufficient  E.g., coffee shop, movie, park, in-car … Physical to Symbolic conversion possible  Lookup location name based on GPS coordinate  However, risky WalmartStarbucks GPS Error range

60. 60 Our Approach Build symbolic localization algorithms Use low accuracy physical localization as baseline Low accuracy conserves energy

61. 61 SurroundSense Sense ambient light, sound, colors …  Combine sensor readings to generate soft fingerprint For localization, gather fingerprint from mobile  Match with database of fingerprints Of course, fingerprint may not be globally unique  Use rough physical localization as a pre-filter GSM physical location says “you are in the mall” SurroundSense augments that with “you are in Apple Store”

62. 62 SurroundSense Design Prototype on Tmote Invent sensors  Sound and light sensors  Low acoustic frequency range [20, 250] Hz  Currently porting on Nokia N95 phones

63. 63 Fingerprint Extraction Light intensity and sound signals recorded  Fourier transform on sound  Overlapping frequency blocks generated  Each block = 23 bands, 10 Hz each Extract simple features  Each 10 Hz band one feature  Variance of each band another feature  Normalized light intensity another feature  Total - 48 features Train the system with half the data  48 dimensional fingerprint

64. 64 Fingerprint Generation and Matching Match test fingerprint with trained database  Use “Nearest Neighbor” algorithm for classification Location Feature Extract Feature Extract 48 features

65. 65 Results SurroundSense offers consistent localization  Database contains nearby shops in Duke campus  Both sensors > sound > light Pairwise Similarity

66. 66 Symbolic localization can be augmented with phone accelerometers Additional benefits in activity recognition

67. 67 Hypothesis Movement partially indicative of location  People sit in cafes  Run in gyms  Walk up/down aisles in grocery stores  Time duration spent may depend on location Augment location accuracy with acc. signatures  Enable activity recognition as well  E.g., Advertize shoes to users running in the gym

68. 68 AAMPL Classification

69. 69 Evaluation Gathered acc. signatures from many restaurants  Classified with AAMPL, compared with Google Restaurant: Chais Verde Rockfish Fast Food: Chipotle Jimmy John’s Store: Apple Wholefoods Journeys Solstice

70. 70 Evaluation AAMPL classified each location  Compared corresponding location with Google

71. 71 Thoughts Main Idea is that the surrounding is a fingerprint. Effective for separating out nearby contexts. In reality, Spatially clustered shops are diverse by design Aids AAMPL and SurroundSense

72. 72 Problem 3 Location Privacy (CacheCloak)

73. 73 (3) Location Privacy Location information reveals context  Thin line between utility and privacy Pseudonymns  Effective only when infrequent querying from mobiles  Else, spatio-temporal patterns enough to deanonymize Romit’s Office John LeslieJack Susan Alex

74. 74 Location Privacy K-anonymity  Convert location to a space-time bounding box  Ensure K users in the box  Location Service (LS) replies to the boxed region Issues  Not real-time  Poor quality of location  Degrades in sparse regions You Bounding Box K=4

75. 75 Confusion = Privacy Mixing or Cloaking users in space-time  Can offer good QoL Issues  Users need to be present in same space-time location  Else, cloaking needs to be performed post priori A A B B ? ? ? ?

76. 76 Our Objective Real-time, high QoL, entropy guarantees, even in sparse populations

77. 77 Our Approach Exploit mobility prediction to deliberately create mix zones Accurate locations can be revealed that are all confusing to adversary

78. 78 CacheCloak Assume trusted privacy provider  Reveal location to CacheCloak  CacheCloak exposes anonymized location to LS CacheCloak Loc. App1 Loc. App2 Loc. App3 Loc. App4

79. 79 CacheCloak Design User A drives down path P1  P1 is a sequence of locations  CacheCloak has cached response for each location User A takes a new turn (no cached response)  CacheCloak predicts mobility  Deliberately intersects predicted path with other path (P2)  Exposes predicted path to application  Application must reply to queries for entire path Application/adversary confused  New path emanates from both P1 and P2  Not clear where the user came from

80. 80 Example

81. 81 Quantifying Privacy City converted into grid of small sqaures (pixels)  Users are located at a pixel at a given time Each pixel associated with 8x8 matrix  Element (i, j) = probability that user enters i and exits j Probabilities diffuse  At intersections  Over time Privacy = entropy i j pixel

82. 82 Diffusion Probability of user’s presence diffuses  Diffusion gradient computed based on history  i.e., what fraction of users take right turn at this intersection When entropy needs to be increased  Generate spurious branches Time t 1 Time t 2 Time t 3 Road Intersection

83. 83 CacheCloak Benefits Real-time  Response ready when user arrives at predicted location High QoL  Responses can be specific to location  Of course, high overhead due to many responses Entropy guarantees  Entropy increases at traffic intersections  In low regions, desired entropy through false branching Sparse population  Can be handled with dummy users

84. 84 Evaluation Trace based simulation  VanetMobiSim + US Census Bureau trace data  Durham map with traffic lights, speed limits, etc.  Vehicles follow Google map paths  Performs collision avoidance 6km x 6km 10m x 10m pixel 1000 cars 6km x 6km 10m x 10m pixel 1000 cars

85. 85 Results High average entropy  Quite insensitive to user density (good for sparse regions)  Minimum entropy reasonably high

86. 86 Results Length of predictions  Remains reasonably short  Overhead proportional to this length

87. 87 Issues and Limitations CacheCloak overhead  Application replies to lots of queries  However, overhead on wired infrastructure  Caching reduces this overhead significantly CacheCloak assumes same, indistinguishable query  If user asks different query at each road segment  Overhead increases Adaptive branching & dummy users  Offer user-specified privacy guarantee

88. 88 Closing Thoughts Two nodes may intersect in space but not in time Mixing not possible Mobility prediction allows space-time intersections Enables better privacy

89. 89 Conclusion The Virtual Information Telescope  A generalization of mobile, location based, social computing Just developing apps  Not enough Many challenges  Energy  Localization  Privacy  Incentives, data distillation … Internet

90. 90 Conclusion Project Micro-Blog  Addressing the challenges systematically  Building a fully functional system with applications  The project snapshot as of today, includes: Micro-Blog: Overall system and application EnLoc: Energy Efficient Localization SurroundSense & AAMPL: Context aware localization CacheCloak: Location privacy via mobility prediction Micro-Blog: Overall system and application EnLoc: Energy Efficient Localization SurroundSense & AAMPL: Context aware localization CacheCloak: Location privacy via mobility prediction

91. 91 Stay tuned for more at http://synrg.ee.duke.edu Thank You