Network Embedded Systems Sensor Networks. Localization. Marcus Chang,

Transcription

1 Network Embedded Systems Sensor Networks Localization Marcus Chang, 1

2 Localization Localization Where am I? Navigation Where do I go? Tracking Where is my stuff? 2

3 Terminology Infrastructure Transmitters/receivers at known locations Mobile device Receives/transmit signal from/to infrastructure Dead Reckoning Tracking using internal sensors 3

4 Triangulation Time-of-Arrival (TOA) Use speed of medium to calculate distance Use intersection from multiple sources for location Round-trip Time-of-Flight (RTOF) Passive reflector: Radar/Sonar Active reflector: processing delay limits accuracy Limited by time synchronization Requires Line-of-Sight 4 Survey of Wireless Indoor Positioning Techniques and Systems, Liu et al.

5 Triangulation Phase-of-Arrival (POA) Phase difference together with wavelentgh, yields distance Signal s wavelength must be larger than cube diagonal Req. LOS Angle/Direction-of-Arrival (A/DOA) Req. LOS 5 Survey of Wireless Indoor Positioning Techniques and Systems, Liu et al.

6 Sensor Network-Based Countersniper System Gyula Simon, Miklós Maróti, Ákos Lédeczi, György Balogh, Branislav Kusy, András Nádas, Gábor Pap, János Sallai, and Ken Frampton, 2004 MICAZ nodes with DSP for sound processing Nodes measure Time-of-Arrival Data collected at central PC Shooter position triangulated 6

7 Experimental Setup 7

8 8

9 Scene Analysis RADAR: An In-Building RF-based User Location and Tracking System Paramvir Bahl and Venkata N. Padmanabhan WIFI Fingerprinting Use the known Access Point locations as landmarks Empirical Systematically collect RSS/SNR measurements throughout building Mobile receiver does a k-nearest neighbor search through database Analytical Use signal propagation models to predict RSS/SNR at different locations Accuracy: 2-3 meters 9

10 Fingerprinting Indoor Localization Without the Pain Chintalapudi et al. Problems Empirical data requires detailed survey of RSS in the area Signal strength varies over time due to people and object moving Question: Can we with only a couple of fix points and a comparatively smaller survey get similar results? 10

11 Main Idea Measure relative AP signal strength Use Log Distance Path Lost Model to solve localization eq. Measure relative location between AP and receiver 11

12 Fingerprint Crowdsourcing Mobile users upload RSS measurement vectors on-the-go System improves over time Solutions to the Log Distance Path Loss model can be scaled, translated, rotated and/or reflected versions of the true locations Use known fix points to solve ambiguity and fix map Accuracy: 2-4 meters Worse than RADAR but less setup and maintenance 12

13 FM-based Indoor Localization Yin Chen, Dimitrios Lymberopoulos, Jie Liu, and Bodhi Priyantha Thursday s paper Use FM radio stations for fingerprinting instead of WIFI Room level accuracy Fingerprint more stable over time Less maintenance 13

14 Reverse Scene Analysis Use infrastructure to collect signals from mobile devices Smartphones WIFI Bluetooth Track customers movement when shopping 14

15 15 GPS

16 GPS (Simplified) Location based on time-of-flight from (at least) 4 known satellite locations Satellite transmissions include very accurate timestamp Solve for unknown (x,y,z,t) t B : receiver clock offset 16 Souce:

17 GPS Hardware 32 satellites Approximately 20,200 km altitude milliseconds signal delay GHz CDMA (Code Division Multiple Access) Each satellite encodes signal with unique 1023 chip code Signal Timestamp Almanac (valid for 180 days) Coarse trajectory and status of all satellites Ephemeris (valid for 4 hours) Precise parameters for the (one) satellite s orbit 17

18 GPS Data Data Packet 50 bits per second Five 300 bits frames Repeats every 30 second Can deduce millisecond signal propagation Timestamp every 6 second Ephemeris every 30 second Almanac every 12.5 minutes 18 Image: Energy Efficient GPS Sensing with Cloud Offloading, Liu et al.

19 GPS Signal Modulation Data packet encoded with 1023 bit C/A code at 1023 kbps C/A Code repeats every 1 ms Final signal at GHz Can deduce nanosecond signal propagation 19 Image: Energy Efficient GPS Sensing with Cloud Offloading, Liu et al.

20 GPS Satellite Tracking Doppler shift 800 m/s satellite has a 4.2 khz Doppler shift GPS receivers collects frequency bins Code Phase shift Over sample 1023 bps code signal 8 MHz oversample 8000 bins Satellite Tracking Search frequency bins Search code phase bins 20 Image: Energy Efficient GPS Sensing with Cloud Offloading, Liu et al.

21 GPS Power Consumption ma Comparable to the TelosB s 18 ma Signal reception time Almanac: 12.5 min. worst case Ephemeris: 30 seconds Tracking multiple satellites is power consuming 21

22 Reducing Reception Time Assisted GPS GPS Location Server Logs almanac and ephemeris Transmit data to GPS receiver over high(er) bandwidth link GPS receiver uses almanac and ephemeris as starting point for satellite search and tracking 22 Image: All About Symbian

23 Reducing Reception Time and Processing Energy Efficient GPS Sensing with Cloud Offloading Jie Liu, Bodhi Priyantha, Ted Hart, Heitor S. Ramos, Antonio A.F. Loureiro. Sensys Main idea Time synchronize with millisecond accuracy using other means Use elevation data to constrain search Collect and store raw GPS samples for online processing Reception time can be reduced to 2 milliseconds! 2 C/A Codes to guard against bit transitions in the Navigational Data 23

24 CO-GPS 24

25 Meanwhile in 1992 The TIDGET - A Low Cost GPS Sensor for Tracking Applications ION Satellite Division International Technical Meeting, Albuquerque, NM, Sept TrackTag GPS Store raw GPS measurements in FLASH Reconstruct route offline

26 Differential GPS Error sources: Satellite clock Receiver clock Atmospheric distortions Use known location to calculate more accurate satellite ranges Accuracy: ~10 cm 26 Oklahoma State University,

27 Low-cost Differential GPS High-Accuracy Differential Tracking of Low-Cost GPS Receivers Will Hedgecock, Miklos Maroti, Janos Sallai, Peter Volgyesi, Akos Ledecz Bluetooth GPS dongles Sub-meter accuracy 27

28 Argos Reverse GPS Polar Orbiting Environmental Satellites NOAA/Eumetsat 6 satellites Altitude: 850 km Orbit: 100 min/14.1 revs Passes: min. 4/day 28

29 Argos Use Doppler shift to determine when satellite passes Use first and last signal in each pass to constrain search Accuracy: m (depending on radio conditions) 29

30 Argos Developed at JHU APL! back in 1983! Today 22g Argos/GPS hybrid GPS location Argos communication $3950 Duty-cycle Every 5 days 5 GPS fixes per day 3 year lifetime Argos platform transmitter terminals, left to right: Early solar-powered PPT (APL), 30- and 20-g Nano PPTs (Microwave Telemetry, Inc.), and prototye solar-powered GPS/PPT (Microwave Telemetry, Inc.). 30

31 Inertial Measurement Unit 31

32 Dead Reckoning Inertial Measurement Unit: 3-axis accelerometer 3-axis gyroscope 3-axis magnetometer General Use measurements to calculate speed and location Pedestrians Distance: linear with step duration Use sensor location on body to infer motion 32

33 Dead Reckoning Problems Sensors and ADCs have inaccuracies Errors accumulate over time leading to larger and larger deviations Solution Hybrid systems with periodic known locations to calibrate IMU or reset position 33

34 IMU and WIFI Fingerprinting Pedestrian Navigation in Harsh Environments using Wireless and Inertial Measurements J. Prieto, S. Mazuelas, A. Bahillo, P. Fernandez, R. M. Lorenzo, and E. J. Abril Combine WIFI RSS and TOA with IMU Foot mounted IMU Use the stationary position to set IMU in known state Use Kalman filter to fusion location estimates 34

35 Walk multiple laps on the same floor 35

36 IMU and Particle Filtering Hybrid Positioning System: Combining Angle-based Localization, Pedestrian Dead Reckoning and Map Filtering P. Kemppi, T. Rautiainen, V. Ranki, F. Belloni, J. Pajunen Main idea: Indoor positioning with fixed beacons (exact location) Signal loss: switch to dead reckoning (relative location) Use particle distribution for most likely locations Use indoor maps to filter out illegal positions 36

37 Angle-of-Arrival Angle-based Indoor Positioning System for Open Indoor Environments by F. Belloni, V. Ranki, A. Kainulainen, A. Richter Mobile transmitters Transmit known pattern, 5 packets per second 1 receiver array of antennas (with dual polarity) Switch antennas during reception of 1 packet Sample phase (Time-of-Flight) and amplitude of received signals Calculate Angle-of-Arrival 37

38 Direction-of-Departure Reverse the process Mobile receivers with known height (z) Stationary transmitters with known position and direction Calculate Direction-of-Departure Cons: Only 2D localization High θ-resolution => high z 38

39 Accuracy Two beacons mounted to 20m high ceiling Covering 1300 m 2 Accuracy: <2.4 m 50%, <3.4 m 90% 39

40 Fusion Filter Use exact localization to construct/update probability distribution Use PDR to update each particles location Use map to filter out illegal movement 40

41 Results PDR with map filtering 630 m route 1 anchor room 41

42 Results Hybrid experiment 270 m route Cafeteria: No line-of-sight and no map filtering guidance 42

43 Bonus: Cameras 43

44 Kinect 2.0 Time-of-flight LEAP motion Stereoscopic triangulation 44

45 45

46 Schedule Week 1: Introduction and Hardware Week 2: Embedded Programming Week 3: Medium Access Control Week 4: Link Estimation and Tree Routing Week 5: IP Networking Week 6: JHU Special feat. Doug Carlson Week 7: (seminar, no lecture) Week 8: Energy Management and Harvesting Week 9: Review and Midterm Week 10: Time Synchronization Week 11: Localization Week 12: TBD Week 13: (seminar, no lecture) Week 14: TBD 46