LightWave is a novel sensing technique that can convert ordinary CFLs into sensors of human proximity. This requires no modification of the bulb or instrumentation on the human body. A single sensor installed anywhere in the home converts all CFL lamps in a home into sensors. LightWave can detect human proximity, touches on the lamp shade or the metal base and on the bulb itself. In addition, it can also detect changes in ambient temperature, all with off-the-shelf unmodified CFLs.

1. LightWave: Using Compact
Fluorescent Lamps as Sensors
Sidhant Gupta, Keyu Chen, Matt S. Reynolds*, Shwetak N. Patel
UbiComp Lab dub
University of Washington
*Duke University

2. LightWave: Using Compact
Fluorescent Lamps as Sensors
Sidhant Gupta, Keyu Chen, Matt S. Reynolds*, Shwetak N. Patel
UbiComp Lab dub
University of Washington
*Duke University

3. turn an ordinary Compact
Fluorescent Lamp (CFL) into
a human proximity sensor

4. CFLs are becoming increasingly popular

6. throughout a home

7. specialized sensor at each
point of sensing

8. commodity CFLs with no modification

9. commodity CFLs with no modification
no instrumentation of the human body

10. commodity CFLs with no modification
no instrumentation of the human body
multiple CFLs with a single sensor

13. Theremin

14. Theremin

15. Theremin
Resonant Circuit
Resonant Circuit

16. Theremin
Resonant Circuit
Resonant Circuit

17. Theremin
Resonant Circuit
Resonant Circuit

21. C

22. C
detunes the
oscillator

23. Electric Field in CFLs
C1 Gas Filled
Lamp
LMP2
120 VAC
Startup
Bridge
Circuitry &
Rectier
LMP1
Oscillator
L1

24. Electric Field in CFLs
resonant
C1 Gas Filled
circuitry Lamp
LMP2
120 VAC
Startup
Bridge
Circuitry &
Rectier
LMP1
Oscillator
L1

25. Electric Field in CFLs
resonant
C1 Gas Filled
circuitry Lamp
LMP2
120 VAC
Startup
Bridge
Circuitry &
Rectier
LMP1
Oscillator
L1

26. Where is the antenna?

27. Where is the antenna?
conductive ionized gases act
as an equipotential electrode

28. Where is the antenna?
conductive ionized gases act
as an equipotential electrode
C
detunes the
oscillator

29. no sensors in lamp to detect
change in oscillation

30. ElectriSense
Switching Circuits generates high frequency Electro
Magnetic Interference (EMI)
Load
Supply
Inductor

31. ElectriSense
Switching Circuits generates high frequency Electro
Magnetic Interference (EMI)
Load
Supply
Inductor

32. ElectriSense
Switching Circuits generates high frequency Electro
Magnetic Interference (EMI)
Load
Supply
EMI
Inductor

33. EMI

34. EMI

35. Detect Presence or Absence of EMI
Time (s)
PC Lamp
Frequency (kHz)

36. de-tuning of oscillator increase energy in
odd harmonics

37. de-tuning of oscillator increase energy in
odd harmonics
odd
harmonic

38. Hover Gestures

39. Touch Gestures

40. Touch Gestures
Lamp Shade Bulb Lamp Base

41. Touch Gestures
One Finger
Three Finger

42. Touch Gestures
3-finger touch vs. 1-finger touch
One Finger
Three Finger

43. Signal Processing Pipeline

44. Signal Processing Pipeline
1 Data Acquisition & FFT
2 Frequency Range Detection
3 Summing & Smoothing Filter
4 Event Detection & Segmentation

45. 1 Data Acquisition & FFT
Digitizer
&
High Pass Filter
Frequency
Transform
Any electrical 1 MS/s
outlet 16,384 point FFT

46. 2 Frequency Range Detection

47. 2 Frequency Range Detection
switch lamp
on/off twice

48. 2 Frequency Range Detection
switch lamp
on/off twice
measure
fundamental
frequency

49. 3 Summing & Smoothing Filter
compute
running sum
in the range

50. 3 Summing & Smoothing Filter
1.43
1.25
1.07
0.89
Amplitude (dB)
0.71
0.54
0.36
0.18
0.00
−1.18
−0.36
0 2 4 6 8 10 12
Time (s)

51. 3 Summing & Smoothing Filter
1.43
1.25
1.07
very noisy, low
0.89
SNR signal
Amplitude (dB)
0.71
0.54
0.36
0.18
0.00
−1.18
−0.36
0 2 4 6 8 10 12
Time (s)

52. 3 Summing & Smoothing Filter
1.43
1.25
1.07
very noisy, low
0.89
SNR signal
Amplitude (dB)
0.71
0.54
0.36
0.18 Savitzky-Golay
0.00
−1.18 Filter
−0.36
0 2 4 6 8 10 12
Time (s)

53. 3 Summing & Smoothing Filter
after filtering
1.43 1.43
1.25 1.25
1.07 1.07
0.89 0.89
Amplitude (dB)
Amplitude (dB)
0.71 0.71
0.54 0.54
0.36 0.36
0.18 0.18
0.00 0.00
−1.18 −1.18
−0.36 −0.36
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time (s) Time (s)

54. 4 Event Detection & Segmentation
0.36
0.18
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)

55. 4 Event Detection & Segmentation
0.36
0.18
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)

56. 4 Event Detection & Segmentation
0.36
0.18
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)

57. 4 Event Detection & Segmentation
variation
0.36
during a
0.18
gesture
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)

58. 4 Event Detection & Segmentation
0.36
Gesture
0.18 Start
dB/dt energy
0.00
−0.18
Gesture
−0.36 End
−0.54
0 2 4 6 8 10 12
Time (s)

59. 4 Event Detection & Segmentation
0.36
0.18
Gesture
Start
Schmitt
Trigger
dB/dt energy
0.00
−0.18
Gesture
−0.36 End
−0.54
0 2 4 6 8 10 12
Time (s)

60. 4 Event Detection & Segmentation
0.36
0.18
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)
Gesture Detection Bit

61. 4 Event Detection & Segmentation
0.36
0.18
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)
Gesture Detection Bit

62. 4 Event Detection & Segmentation
0.36
0.18
dB/dt energy
0.00
−0.18
−0.36
−0.54
0 2 4 6 8 10 12
Time (s)
Gesture Detection Bit
0.6 < ∆T < 5 sec.
∆T

63. Can gestures be detected across
different people, brands and lamp types?

64. 10 Participants (3 females)
6 Different Lamp Types
5 Gestures on each lamp repeated 4 times
Sessions on different days
In total 2400 gestures collected
Gestures:
1. Hover
2. Bulb Touch 1-finger and 3-finger
3. Shade Touch 1-finger and 3-finger

65. In-home experiments in 2 homes
2 participants & 2 lamps
Gestures:
1. Hover
2. Bulb Touch 1-finger and 3-finger
3. Shade Touch 1-finger and 3-finger

66. Average hit rate across all
participants and lamps

67. Average hit rate across people and lamps
100
99.79 99.79
91.25 91.25
85.75
75
Hit Rate %
50
25
0
Hover Bulb 1F Bulb 3F Shade 1F Shade 3F
1F = 1 Finger
3F = 3 Fingers Touch

68. Average hit rate across people and lamps
100
99.79 99.79
91.25 91.25
85.75
75
Hit Rate %
50
25
0
Hover Bulb 1F Bulb 3F Shade 1F Shade 3F
1F = 1 Finger
3F = 3 Fingers Touch

69. Average hit rate for Lamp Shade Touch
Metal Metal
100
100 99
Glass
Plastic Metal 85
75
75
70
Hit Rate %
50
25
0
L1 L2 L3 L5 L6
1F = 1 Finger Lamps
3F = 3 Fingers

70. Average hit rate for Lamp Shade Touch
Metal Metal
100
100 99
Glass
Plastic Metal 85
75
75
70
Hit Rate %
50
25
0
L1 L2 L3 L5 L6
1F = 1 Finger Lamps
3F = 3 Fingers

71. Average hit-rate for in-home experiments
100
93.75 93.75 93.75
87.5 87.5
75
Hit Rate %
50
25
0
Hover Bulb 1F Bulb 3F Shade 1F Shade 3F
1F = 1 Finger
3F = 3 Fingers Touch

72. Works across different
people and types of lamps

73. We tested an additional 6
different brands of CFLs

74. Change in Proximity
1 1
Normalized Amplitude
Normalized Amplitude
0.95 0.95
0.9 0.9
0.85 0.85
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Time (s) Time (s)

75. Change in Proximity
hand moving towards vs. away
1 1
Normalized Amplitude
Normalized Amplitude
0.95 0.95
0.9 0.9
0.85 0.85
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Time (s) Time (s)

76. Detecting Ambient Temperature
Change
140
Increase in
145
ambient
150 temperature
155
Frequency(kHz)
160
165
170
175
180
185
10 20 30 40 50 60 70
Time (s)

77. LightWave is a sensing technique
that turns ordinary CFLs into
human proximity sensors

78. Questions?
sidhant@uw.edu
ubiCompLab.cs.washington.edu
www.sidhantgupta.com

79. Hover Gesture

81. EMI in Home