xiaofan jiang, chieh -jan mike liang, kaifei chen, ben zhang, jeff hsu

31
Design and Evaluation of a Wireless Magnetic-based Proximity Detection Platform for Indoor Applications Xiaofan Jiang, Chieh-Jan Mike Liang, Kaifei Chen, Ben Zhang, Jeff Hsu Jie Liu, Bin Cao, and Feng Zhao Microsoft Research Asia 20120730-Neight

Upload: dean

Post on 24-Feb-2016

61 views

Category:

Documents


0 download

DESCRIPTION

Design and Evaluation of a Wireless Magnetic-based Proximity Detection Platform for Indoor Applications. Xiaofan Jiang, Chieh -Jan Mike Liang, Kaifei Chen, Ben Zhang, Jeff Hsu Jie Liu, Bin Cao, and Feng Zhao Microsoft Research Asia. 20120730-Neight. Outline. MOTIVATION - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Design and Evaluation of a Wireless Magnetic-based

Proximity Detection Platform for Indoor Applications

Xiaofan Jiang, Chieh-Jan Mike Liang, Kaifei Chen, Ben Zhang, Jeff Hsu

Jie Liu, Bin Cao, and Feng ZhaoMicrosoft Research Asia

20120730-Neight

Page 2: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

MOTIVATION PROXIMITY ZONE Empirical Definition EVALUATION OF EXISTING TECHNOLOGIES LIVESYNERGY PLATFORM EVALUATION OF LIVESYNERGY APPLICATION DEPLOYMENT CONCLUSIONS

Outline

Page 3: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

To make applications intuitive to human users,

the discovered objects in the environment must be within the personal interaction sphere

Computer automatically wake up Refrigerator change its user interface Many typical low power communication

technologies, (Bluetooth, ZigBee) have difficulties maintaining robust communication zones

Motivation

Page 4: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

propose methodologies and systematically

compare the proximity zones created by various wireless technologies(BLE, ZigBee, and RFID reader)

Design, Implement, and Evaluate a magnetic-induction based wireless proximity sensing platform

Deploying LiveSynergy in an real-world application

Contributions

Page 5: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Boundary sharpness: boundary of proximity zone should be binary Boundary consistency: detection should be consistent over time

PROXIMITY ZONE

Page 6: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Obstacle penetration: Beaconing node and listening node can be mobileand against obstructions Additional metrics:1. Range and geometric shape of zones2. Beaconing frequency achievable3. Power consumption 4. Form-Factor of the mobile tag5. Cost of overall system

PROXIMITY ZONE

Page 7: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Broadcasts at fixed frequency f (packets ) P = a point in space at a distance of (, , ) from

the beacon

Classification of Points

Page 8: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

white/grey boundary:{P | Color(P, t, t’) = white} {P | Color(P, t, t’) = grey} if x, if x’ represents the decision boundary grey/black boundary:if x, if x’

Classification of Zones

Page 9: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Three proximity

zones

Page 10: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Questions?

Proximity Zones

Page 11: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Use support vector machines (SVM) as the

classifierseeks maximum-margin hyperplane to separate two classes

w and b are the parameters to define the hyperplane to separate the two classes.

Classifier

Page 12: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Two user-definable parameters: Error tolerance:- Smooth boundary vs. non-smooth boundary Tradeoff between training loss and regularization Cost parameter C Strictness:-Expect the white zone and the black zone contain no grey points-Related to error tolerance but non-symmetry

Classifier

Page 13: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

• Cost parameter C: the cost of false positiveC’: the cost of false negative C’• Strictness parameter:

Classifier

Page 14: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

RBF kernel as the kernel function Classifier:

RBF kernel

Kernel Trick

Page 15: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Size:Size of the white and grey zone, which can be computed numerically based on the boundaries. Boundary sharpness:

Fitness:How well the zone boundaries fit the data, or a confidencemeasure of the proximity zone classification.

Matrix

Page 16: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Questions?

Classifier

Page 17: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Hardware setup:• TI CC2540 BLE dev boards (transmitting on 2.4

GHz at 0 dBm),• A pair of TelosB motes with 802.15.4-

compliant TI CC24240 radio(transmitting on 2.4 GHz at 0 dBm)

• A Impinj Speedway R1000 RFID reader (transmitting on 902 MHz at 8 dBm)

Boundary Sharpness and Consistency

Page 18: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Parameters:• packet reception data is collected over a

period of 200 seconds• WPRR using a windows size of 3 seconds and • Strictness parameter = 0.99 Results:

Boundary Sharpness and Consistency

Page 19: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Boundary Sharpness

and Consistency

Page 20: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

The user carries the receiver in the right pants

pocket- calculate PRR from 500 packets as the user changes the body orientation by 90 each round at each distance

Human Obstacle Penetration

Page 21: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Signal propagation and geometry:RFID antennas usually have a radiation angle less than 180 degrees Form Factor and Costs:RFID can produce a more consistent and smaller grey zone802.15.4 and BLE have advantages in both form factor and costs.

Additional Metrics

Page 22: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Questions?

Evaluation

Page 23: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Pulse Transmitter: (use AC power)Four primary hardware microcontroller (MCU) and radio magnetic transmitter tuned at 125kHz Energy metering mechanical relay for actuation.

LIVESYNERGY PLATFORM

Page 24: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Link Receiver: ( battery-powered)Three primary hardware 9.2cm ×5.8cm × 2.3cm enclosure• MCU and radio• 3D magnetic coil• wake up chip

LIVESYNERGY PLATFORM

Page 25: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Boundary Sharpness

and Consistency

Page 26: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

human body has very little impact on the MI

signal propagation

Body orientation vs. distance

Page 27: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Geometry: two dimensions extends to all directions, covering 360◦ Range: maximum range (i.e., radius) is around 5m

Additional Metrics

Page 28: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

APPLICATION DEPLOYMENT

Diners enter the cafeteria from the entrance at the lower left corner at different times

Page 29: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Each diner takes a different route and visits

various food counters on the way Recorded a video as the customers walk

around the cafeteria purchasing food.- Use video timestamps

Experment

Page 30: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Result

Page 31: Xiaofan  Jiang,  Chieh -Jan Mike Liang,  Kaifei  Chen, Ben Zhang, Jeff Hsu

Values:1. Propose methodologies and systematically

compare the proximity zones2. Deploying LiveSynergy in an real-world

application

Future?1. MI still can implement in mobile phone…

Summary