televisions video privacy and powerline
play

Televisions, Video Privacy, and Powerline Electromagnetic - PowerPoint PPT Presentation

Televisions, Video Privacy, and Powerline Electromagnetic Interference Miro Enev , Sidhant Gupta, Yoshi Kohno, & Shwetak Patel Security & UbiComp Labs @ UW Smart Home = Smart Devices + Smart Sensors The Picture Today Utility power


  1. Televisions, Video Privacy, and Powerline Electromagnetic Interference Miro Enev , Sidhant Gupta, Yoshi Kohno, & Shwetak Patel Security & UbiComp Labs @ UW

  2. Smart Home = Smart Devices + Smart Sensors

  3. The Picture Today Utility power sensor 3 rd Parties

  4. The Picture Today Utility power sensor 3 rd Parties UbiComp

  5. Emerging Trends: Sophisticated electrical sensors are already deployed in homes to help achieve new efficiency and utility goals. Benefactors:

  6. Research Q: What private information is available from the powerline?

  7. Research Q: What private information is available from the powerline? Answer : Modern TVs leak substantial information on the power line that is indicative of the screen content

  8. Research Q: What private information is available from the powerline? Answer : Modern TVs leak substantial information on the power line that is indicative of the screen content Information can be collected using a single sensor installed anywhere on power line

  9. What’s the BIG deal? U.S. Video Protection & Privacy Act of 1998 states that video viewing records be kept private

  10. What’s the BIG deal? U.S. Video Protection & Privacy Act of 1998 states that video viewing records be kept private We hope our work can inform future discourse about the directions of Powerline sensing technologies

  11. Smart Sensor Power measurements reveal private information about homeowner’s activity [Markham 2010]; concurrent work also looks at TVs [Greveler 2011] In-line power measurement of a PC can reveal web browsing habits [Clark 2011]

  12. Power Based Measurement In-line with device

  13. Voltage Based Measurement Home’s Powerline Infrastructure ... Parallel with device

  14. S witched M ode P ower S upply

  15. Switching Circuits generate high frequency Electro Magnetic Interference (EMI) Load Supply Inductor

  16. Switching Circuits generate high frequency Electro Magnetic Interference (EMI) Load Supply Inductor

  17. Switching Circuits generate high frequency EMI which couples onto the powerline Load Supply Inductor Home’s Power Line Infrastructure

  18. Switching Circuits generate high frequency EMI which couples onto the powerline we record the EMI Load Supply Inductor Home’s Power Line Infrastructure

  19. EMI : Summary ElectriSense ( 2010 Gupta et al. ) Many modern devices produce EMI! Frequency (kHz) EMI can be collected with a single sensor! Signals live in the frequency domain!

  20. EMI @ Home

  21. TV EMI Video Movie Link

  22. Our 8 TVs Samsung 58-A 58” Samsung 58-B Plasma, 2010 Panasonic 42-A Panasonic 42-B 42” Sharp 42 LCD, 2007 - 2009 Sharp 32-A 32” Sharp 32-B Samsung 32

  23. Research Questions? Q1: Do TVs produce repeatable EMI given repeated screen content ? Q2: For a given TV, does different screen content produce different EMI ? Q3: Is EMI consistent across TVs from the same model family ? Q4: Can we use EMI to determine what is being watched on TV? Q5: Can we match lab EMI to EMI recorded in various home settings ?

  24. Research Questions? Q1: Do TVs produce repeatable EMI given repeated screen content ? Q2: For a given TV, does different screen content produce different EMI ? Q3: Is EMI consistent across TVs from the same model family ? Q4: Can we use EMI to determine what is being watched on TV? Q5: Can we match lab EMI to EMI recorded in various home settings ?

  25. Screen Content = 20 IMDB Top Grossing Movies Action Animation Comedy Documentary Drama

  26. Lab Setup (I)Isolation transformer - external EMI filter (A)Power Line Interface - custom voltage sensor (U) Universal Software Radio Peripheral - analog to digital convertor (A)Spectrum analyzer - visualization & logging

  27. Signal Extraction Time (secs) 0 30 100 200 Frequency (kHz)

  28. Signal Extraction Time (secs) 0 30 100 200 Frequency (kHz)

  29. Extracting TV EMI

  30. Extracting TV EMI

  31. Research Questions? Q1: Do TVs produce repeatable EMI given repeated screen content ? Q2: For a given TV, does different screen content produce different EMI ? Q3: Is EMI consistent across TVs from the same model family ?

  32. Q4 : Can we use EMI to determine what is being watched on a TV?

  33. Q4 : Can we use EMI to determine what is being watched on a TV? Approach: Match EMI to a DB

  34. Time (mins.) 60 0 Building an EMI Database EMI . . . . . .

  35. Multiple TVs Multiple Runs Same Content Pan42A 0 Time (mins.) 60 Pan42B

  36. Multiple TVs Multiple Runs Same Content Pan42A 0 Time (mins.) 60 Pan42B Query DB vs

  37. Query DB vs = ? TV Movie Q.Len • 1: 8 • 1:20 • 1:60 0 15 Time (mins.)

  38. Query DB vs = ? TV Movie Q.Len • 1: 8 • 1:20 • 1:60 0 15 Time (mins.)

  39. Query DB vs = ? TV Movie Q.Len • 1: 8 • 1:20 • 1:60 0 15 Time (mins.) Best match

  40. Possible Query Results • Hit: the search engine is confident * in the match (accept) and the match was the movie from which the query itself was extracted. (Success.) • Miss: the search engine is confident * in the match (accept) but there is a mismatch between the search engine’s best guess and the query origin. (Failure.) • Reject: the best match was a not a clear winner*, and the matching algorithm chooses not to respond. (Neither success nor failure.) * Confidence is the gap size between the top match and the runner up

  41. Query Length & Confidence Confidence is the gap size between the top match and the runner up Query Length 1 min 6 min 12 min 15 min Reject Reject Reject Reject

  42. Q4: Can we determine what is being watched from EMI?

  43. Q4: Can we determine what is being watched from EMI?

  44. Q5: Can we match lab EMI to home EMI?

  45. Home Details Style Year Built Size Single family Home 2003 3000 sq. ft Apartment 2009 657 sq. ft. Multi-family Home 1906 800 sq. ft.

  46. Home Setup

  47. Home Setup

  48. Q5: Can we match lab EMI to EMI recorded in home settings

  49. Extensions : Can we predict EMI from screen content without a TV?

  50. Predicting EMI Extract Train Predict Features Model EMI Model: Neural Net Input Features:  Brightness : cumulative sum of averaged RGB intensities  Flux : change in brightness between consecutive frames  Edge Intensity : pixelsum of a Canny Edge filter  FFT: slope of the best fit line to an FFT  Color: mean and standard deviation for Gaussians fitted to R, G, and B color histograms  Bitrate: kbits/second computed using FFMPEG Cross Validation: 10 Train, 5 Test

  51. Predicting EMI

  52. Predicting EMI

  53. TV EMI : Theory of Operation SMPS light use scene save AC @60Hz dark scene TV EMI Backlight (brightness) is key source of TV EMI

  54. Defenses and Tensions > Better filtering by power supplies > Signal Injection & Jamming (Energy Star) > Battery Masking [Mclaughlin 2011] > Infrequent sensor readings

  55. Takeaways > Devices produce EMI on the powerline > Single sensor tracks many devices > EMI can be > Growing attack surface

  56. Questions? miro@cs.washington.edu

  57. Does EMI matching scale?

  58. Does EMI matching scale?

  59. Sharp 32 and SNR

  60. Related Work “Our research shows that the analysis of the household’s electricity usage profile does reveal what channel the TV set in the household was displaying. “

  61. Q1 : repeatable EMI from repeated screen content ? A & B : Lion King C: Bourne Ultimatum Cross Correlation EMI Time (secs)

  62. TVs (1) 58” B Samsung 58-A

  63. TVs (2) 42” Panasonic 42-A B

  64. TVs (3) 32 ” Sharp 32-A B

Recommend


More recommend