The Impact of Using Multiple Antennas on Wireless Localization Konstantinos Kleisouris Computer Science Department Rutgers University Joint work with: Prof. Yingying Chen, Jie Yang, Prof. Richard P. Martin (advisor)
Localization Office Floor Technology allows a large � variety of computing devices to communicate wirelessly Radio can be used not only � for communication but for localization of devices in 2D Localization and 3D (X, Y)
Localization Background Office Floor Many localization algorithms � use landmarks and a training set landmark Landmark: monitors packet � S1 traffic at known positions Training set: offline measured � S2 radio properties and locations landmark S1’ Properties: Received Signal � Strength (S i ), Angle of Arrival [X, Y, S1, S2, S3] S3 (AoA), Time of Arrival (ToA) S2’ fingerprint Fingerprint: a set of Signal � [X?, Y?, S1’, S2’, S3’] Strengths (S i ) measured at some location landmark S3’
Using RSS Indoors � Received Signal Strength (RSS) is affected indoors by environmental effects E.g. reflection, diffraction, scattering � � Difficult to associate signal strength to location � Can we alleviate the impact of RSS variability on the performance of localization algorithms?
Our Approach � Investigated signal strength variability when employing multiple antennas � Investigated the effects of using multiple antennas on RSS-based localization algorithms
Contributions � Multiple antennas can average out environmental effects on RSS indoors � Multiple antennas can improve the localization accuracy and stability of different algorithms
Talk Outline Introduction � Introduction Introduction � � � Methodology � RSS Variability Study � Stability & Accuracy Results � Conclusions & Future Work
RSS Indoors Reflection, diffraction and scattering of RSS leads to multipath � fading effects RSS can vary by 5-10 dB with small changes (a few wavelengths) in � location Granularity of a localization system is usually much larger (2-3 m) � Multiple receivers spaced on the order of a few wavelengths present � an opportunity to smooth out these effects Multiple receivers can be realized by multiplexing between multiple � antennas at a given landmark location
Testbed Infrastructure 802.11 (Wi-Fi) testbed � Experiments were � conducted in the yellow area 10 landmarks at 5 (red � stars) different locations 169 ft 2 per location � Three 7 dBi Omni � antennas per landmark location (1-2 ft from each other) Green dots: 101 testing � spots where we collected 219 ft data
Placements of transmitter Placement Coordinates (in feet), Description Floor (x, y, 0) Center (x, y, 3) East (x-1, y, 3) West (x+1, y, 3) Desk North (x, y+1, 3) South (x, y-1, 3) Vertical (x, y, 3), monitor vertical to the floor Parallel (x, y, 3), monitor parallel to the floor Shoulder (x, y, 5.16) Transmitter: Dell Laptop running Linux with an Orinoco silver card � 9 placements around a testing spot � 7 at the desk level (3 ft) � 3 along the z-axis (0 ft, 3 ft, 5.16 ft) � 2 rotations (vertical, parallel) � Collected 9 fingerprint data sets �
Metrics Accuracy: Euclidean distance between the location estimate � obtained from a localization system and the actual location This distance is called localization error � Stability � Measures how much the location estimate moves in the physical space � in response to small-scale movements of a mobile device Euclidean distance between the location estimate of a mobile at its � original position p 1 and the localization results when it is moved to locations p 2 , p 3 , …, p n We study CDFs for both metrics �
Talk Outline Introduction � Introduction Introduction � � Methodology � Methodology Methodology � � � RSS Variability Study � Stability & Accuracy Results � Conclusions & Future Work
Impact on Free Space Models Do multiple antennas “smooth out” the effects of small-scale � variations on signal strength? Smooth out: RSS does not vary much with a change in location � Metric: Examined the goodness of fit of RSS data from multiple � antennas to a theoretical propagation model = + Free Space Model S b b log( D ) 0 1 Goodness of fit is observable as the coefficient of � determination R 2
Goodness of fit For A, B, C, D averaging � the RSS for all 3 antennas (3-antenna-avg) achieves the best fit Adding multiple antennas � does improve the data fit to a simple free-space model
Localization Results � Algorithms RADAR: nearest neighbor matching in signal space � Bayesian Networks (BNs) M1, M2, M3: multilateration � � Results Accuracy � Stability � � (x, y) plane: Center, North, South, East, West, Vertical, Parallel � z-axis: Center, Floor, Shoulder � Center placement is always the original p 1 position
RADAR Accuracy Desk, Center 3-antenna-avg best case � Improvement on � Median: 12ft to 9.6ft (20%) � 90 th percentile: 30ft to 21.2ft (29%) �
RADAR Stability (x, y) plane z-axis 3-antenna-avg best case 3-antenna-avg best case � � Improvement on Improvement on � � Median: 19ft to 11ft (42%) Median: 19ft to 10.5ft (44%) � � 90 th percentile: 36.1ft to 25.2 (30%) 90 th percentile: 35.4ft to 24.7ft � � (30%)
BN, M2, Accuracy Desk, Center, No Train., Test.=51 3-antenna-noavg best case � Improvement on � Median: 22ft to 13ft (40%) � 90 th percentile: 54ft to 28ft (48%) �
BN, M2, Stability (x, y) plane, No Train., Test.=51 z-axis, No Train., Test.=51 3-antenna-noavg best case 3-antenna-noavg best case � � Improvement on Improvement on � � Median: 16ft to 9ft (43%) Median: 15ft to 9ft (40%) � � 90 th percentile: 36ft to 20ft 90 th percentile: 32ft to 21ft � � (44%) (34%)
Conclusions & Future Work Multiple antennas help � Average out small-scale environmental effects � Improve localization accuracy and stability in localization � Adding multiple antennas is easy and probably worth the cost for � landmarks, although the impact is not huge There is not a clear trend whether averaging or not averaging is � better for localization algorithms Study the improvements with more than 3 antennas per location and � what the limiting number is where improvements tail off
Thank you!
Related Work Localization � [Bahl’00] RADAR: An In-Building RF-Based User Location System � [Priyantha’00] The Cricket Location-Support System � [Ward’97] The Bat Ultrasonic Location System � [Niculescu’01] Ad Hoc Positioning System (APS) � [Fox’01] Particle Filters for Mobile Robot Localization � [Lorincz’06] Motetrack: Robust, Decentralized Location Tracking � Antennas � [Lim’06] Zero-Configuration, Robust Indoor Localization � [Lymberopoulos’06] An Empirical Analysis of RSS Variability in 802.15.4 � Using Monopole Antennas [Hashemi’93] The Indoor Radio Propagation Channel � [Godara’97] Applications of Antenna Arrays to Mobile Communications � [Barrett’94] Adaptive Antennas for Mobile Communications � [Barroso’94] Impact of Array Processing Techniques on Mobile Systems � [Chryssomallis’00] Smart Antennas �
Localization Applications Track devices like laptops, handheld devices and badges � Control access to information and utilities based on location � Provide location-specific information in museums � Track personnel in factories and hospitals � Provide monitoring and management of wireless networks � Localize wireless sensors used for environmental monitoring �
RADAR Accuracy (1) Desk, Center Gaussian 3-antenna-avg best case Same trends but worse � � performance when compared Improvement on � to real data Median: 12ft to 9.6ft (20%) � 90 th percentile: 30ft to 21.2ft (29%) �
RADAR Accuracy (2) Floor Shoulder 3-antenna-avg best case 3-antenna-avg best case � � Improvement on Improvement on � � Median: 10.7ft to 9.6ft (10%) Median: 18ft to 10ft (44%) � � 90 th percentile: 28 ft to 20ft 90 th percentile: 30.6ft to � � (28%) 21.7ft (29%)
ABP Accuracy (1) Desk, Center Gaussian 3-antenna-noavg best case Same trends when � � compared to real data Improvement on � Median: 7ft to 2ft (71%) � 90 th percentile: 16ft to 4ft � (75%)
ABP Accuracy (2) Floor Shoulder Trends similar to Desk, Trends similar to Desk, � � Center Center
ABP Stability (x, y) plane z-axis 3-antenna-noavg best case 3-antenna-noavg best case � � Improvement on Improvement on � � Median: 8ft to 2ft (75%) Median: 7.7ft to 2ft (74%) � � 90 th percentile: 16.4ft to 4.3ft (73%) 90 th percentile: 16.2ft to 4.2ft � � (74%) At 0ft: ≥ 100% improvement �
BN, M2, Accuracy (1) Desk, Center, Train.=100, Test.=1 Gaussian, Train.=100, Test.=1 Similar performance for all Averaging and not averaging � � curves the RSS has the same performance
BN, M2, Accuracy (2) Desk, Center, No Train., Test.=51 Gaussian, No Train., Test.=51 3-antenna-noavg best case Averaging and not averaging � � the RSS has the same Improvement on � performance Median: 22ft to 13ft (40%) � 90 th percentile: 54ft to 28ft � (48%)
Recommend
More recommend
