Upload
rana-basheer
View
2.220
Download
4
Tags:
Embed Size (px)
DESCRIPTION
Slides from my dissertation defense. Talks about the error in localizing a transmitter by measuring the signal strength. In addition, it presents new techniques for localization using cross-correlation of fading.
Citation preview
Real Time Localization Systems Using Receiver Signal Strength Indicator
Mohammed Rana Basheer Advisor: Dr. Jag Sarangapani
2
Publications Refereed Journal Papers
M.R. Basheer, and S. Jagannathan, "Enhancing Localization Accuracy in an RSSI Based RTLS Using R-Factor and Diversity Combination", in review, International Journal of Wireless Information Networks
M.R. Basheer, and S. Jagannathan, "Receiver Placement Using Delaunay Refinement-based Triangulation in an RSSI Based Localization“, revised and resubmitted, IEEE/ACM Transactions on Networking
M.R. Basheer, and S. Jagannathan, "Localization of RFID Tags using Stochastic Tunneling", accepted, IEEE Transactions on Mobile Computing
M.R. Basheer, and S. Jagannathan, "Localization and Tracking of Objects Using Cross-Correlation of Shadow Fading Noise", minor revision, revised and resubmitted, IEEE Transactions on Mobile Computing
M.R. Basheer, and S. Jagannathan, "Placement of Receivers for Shadow Fading Cross-Correlation Based Localization", to be submitted
3
Publications (contd.) Refereed conference papers
M.R. Basheer, and S. Jagannathan, "R-Factor: A New Parameter to Enhance Location Accuracy in RSSI Based Real-time Location Systems," Sensor, Mesh and Ad Hoc Communications and Networks, SECON '09. 6th Annual IEEE Communications Society Conference on , pp. 1-9, 22-26 June 2009.
M.R. Basheer, and S. Jagannathan, "A New Receiver Placement Scheme Using Delaunay Refinement-Based Triangulation," Wireless Communications and Networking Conference (WCNC), 2010 IEEE, pp.1-6, 18-21 April 2010.
M.R. Basheer, and S. Jagannathan, "Localization of objects using stochastic tunneling," Wireless Communications and Networking Conference (WCNC), 2011 IEEE, pp.587-592, 28-31 March 2011.
M.R. Basheer, and S. Jagannathan, " Localization of Objects Using Cross-Correlation of Shadow Fading Noise and Copulas," Global Communication Conference (GLOBECOM), 2011 IEEE, 6-8 Dec 2011.
M.R. Basheer, and S. Jagannathan, " Placement of Receivers for Shadow Fading Cross-Correlation Based Localization," Submitted to IEEE Local Computer Networks (LCN) 2012.
4
Presentation Outline Introduction and Background
Paper 1: Enhancing Localization Accuracy in an RSSI Based RTLS Using R-Factor and Diversity Combination
Paper 2: Receiver Placement Using Delaunay Refinement-based Triangulation in an RSSI Based Localization
Paper 3: Localization of RFID Tags using Stochastic Tunneling
Paper 4: Localization and Tracking of Objects Using Cross-Correlation of Shadow Fading Noise
Paper 5: Placement of Receivers for Shadow Fading Cross-Correlation Based Localization
Conclusions
Future Work
5
Used for locating or tracking assets in places where GPS signals are not readily available
Methodologies Time of Arrival (ToA), Time Difference of Arrival (TDoA), Angle of Arrival (AoA) or Received Signal Strength Indicator (RSSI)
Boeing factory floor**http://www.ce.washington.edu/sm03/boeingtour.htm
Real Time Location Systems (RTLS)
Friis Transmission Equation RSSI vs. Distance
)log(rnARSSI
RTLS using RSSI Uses signal strength of radio signals to locate objects
Classified into Range Based Range Free
RSSI Profile of ERL 114
6
7
RTLS Receiver
RTLS Tag MST Mote
Localization Hardware
IEEE 802.15.4 transceiver from XBee
Operating frequency 2.45 GHz with 100 MHz Bandwidth
8051 variant microcontroller
8KB RAM and 128 KB code space
Spatial diversity with 2 antennas
8
Motivation for RTLS using RSSITime and Angle based methods are costly and require dedicated
hardware
RSSI information easily accessible through API
Localization can be easily deployed on existing wireless infrastructure as a software upgrade
Time and Angle based localization achieves better accuracy under LoS condition
Coarse grained localization
Periodic radio profiling of target area under range free methods or calibration of parameters under range based method is essential
9
Goal and Objectives of the Dissertation Goal—Given a location error threshold, determine the location of a
transmitter and track it in a workspace by placing the appropriate number of receivers at the right position in a workspace.
Objectives Develop algorithms for localization and tracking of wireless devices from
radio signal strength signals
Develop algorithms for placing wireless receivers around the workspace so that the error in locating a transmitter at any point in this workspace is less than a predefined threshold
Demonstrate the efficacy of the placement and localization algorithms analytically, in simulation environment and experimentally through hardware
Both range-based and range free methods are developed
10
Cohesion of completed work
Localization Using RSSI
Range-Based
Cross-Correlation
Paper 1. M.R. Basheer, and S. Jagannathan, "Enhancing Localization Accuracy in an RSSI Based RTLS Using R-Factor and Diversity Combination", under review at International Journal of Wireless Information Networks
Paper 2. M.R. Basheer, and S. Jagannathan, "Receiver Placement Using Delaunay Refinement-based Triangulation in an RSSI Based Localization", Revised and resubmitted to IEEE/ACM Transactions on Networking
Paper 3. M.R. Basheer, and S. Jagannathan, "Localization of RFID Tags using Stochastic Tunneling", Accepted at IEEE Transactions on Mobile Computing
Paper 4. M.R. Basheer, and S. Jagannathan, " Localization and Tracking of Objects Using Cross-Correlation of Shadow Fading Noise", Minor revision, revised and resubmitted, IEEE Transactions on Mobile Computing,
Paper 5. M.R. Basheer, and S. Jagannathan, "Placement of Receivers for Shadow Fading Cross-Correlation Based Localization", to be submitted,
11
Paper 1: Enhancing Localization Accuracy in an RSSI Based RTLS Using R-Factor and Diversity Combination
Friis Transmission Equation
Euclidean distance equation
affected by outliers
Weighted least square to the rescue
Weights are the radial distance variance and are called the R-Factor
Transmitter
Receivers
Base station
Transmitter location estimation happens in base station from Signal
Strength
12
Paper 1: Enhancing Localization Accuracy in an RSSI Based RTLS Using R-Factor and Diversity Combination Objectives
Derive a statistical parameter called the R-factor to grade radial distance estimates to a transmitter from RSSI values
Non-coherent diversity combination techniques that can improve radial distance estimation
Previous efforts involved Proximity in Signal Space (PSS), a heuristic algorithm that uses signal strength to
classify receivers for localization accuracy [Gwon 04]
Chi-square test to classify Line of Sight (LoS) condition at the receiver into Ricean or Rayleigh [Lakhzouri 03]
Binary classification of receivers into good or bad based on a test for Gaussian distribution [Venkatraman 02]
13
Assumptions
Received signal amplitude random variable X is Ricean distributed with PDF given by
Radial distance random variable R is related to RSSI X as
where n is the path loss exponent and l0 accounts for antenna geometry, wavelength etc.
202
22
22
2exp,|
XXXXX
AxI
xAxAxf
n
X
lXgR
1
20)(
14
where is the Confluent Hyper-geometric Function (CHF) and is the ratio of the power in the deterministic LoS component to the NLoS energy or the signal to noise ratio for localization.
The mean and variance of the radial distance estimate by a receiver to a transmitter using Friis transmission equation based estimator under Ricean environment is given by
KMKKM
l
ln
n
KM
lKRE
n
X
X
n
X
X ,1,2
1
41
,1,21
222
,1,21
2),|( 2
11
22
0
02
2
1
22
02
KMKKM
l
lnKRVar
n
X
XX ,1,
2
1
41
,1,21
28),|( 2
12
22
0
02
22
Mean & Variance of Radial Distance Estimate
22 2 XAK ,,M
15
Localization Receiver and R - Factor A receiver for RSSI based RTLS, is called a localization receiver if the signal to noise ratio for the received signals is greater than 9
Mean and Variance of radial distance estimate is given by
22 2 XAK
0200
22,| r
Kn
nrKrRE
Kn
r
Kn
AlKrRVar
nn
2
20
2
42
00
22),|(
R-Factor (Receiver Error Factor) measures the variance in radial distance estimate by a localization receiver
20
2
02
20
02
20 2
),|(2
1X
bX
n
crln
rKrRVar
Kn
r
MSE is proportional to R-Factor
10821082 222
20 nnnnKn
rMSE
Bias
16
Localization Error and R-Factor
Theorem 1: (Comparison of localization accuracy under LoS and NLoS) For the same amount of NLoS energy at a localization receiver and a receiver under NLoS conditions, the MSE of the radial distance estimate for the localization receiver is lower than that of the receiver under the NLoS condition
Theorem 2: (R-factor and localization accuracy) The upper bound of the localization error decreases with R-factor in a Ricean environment for a RSSI based RTLS
Theorem 3: (Localization accuracy and receiver count) Localization accuracy using w+1 receivers is better in comparison with deploying w receivers in an RSSI based RTLS system when the maximum R-factor is kept the same in both cases
17
Channel Diversity and R-Factor Diversity is a method to improve certain aspects of the received
signal by using two or more communication channels
Two commonly used diversity schemes are Spatial Diversity using multiple antennas Frequency Diversity
For RTLS using RSSI only non-coherent combination is possible
Diversity channels were combined using one of the following methods Selection Combination: Best signal out from all channels Averaging: Mean of signals from all channels Root Mean Square: Compute RMS of signal from all channels
18
Simulation of R-Factor vs. Diversity Count
LoS Conditions NLoS Conditions
Variation of R-Factor with increasing diversity channel count
From simulations, combining RSSI from diversity channels using RMS produced the lowest R-Factor and consequently the best localization accuracy
19
Receiver layout in ERL 114
CDF of Localization Error
*Gwon et al. 2004
Localization Experiment
Localization MethodLocalization Error (cm)
Mean Median 90th percentile Std. dev
TIX + PSS 342 298 432 62.81TIX with R-factor 267 214 335 40.32
TIX with R-factor and Spatial Diversity 254 210 329 40.15
Summary of Localization Error Levels
22% decrease26%
decrease
28% decrease30%
decrease
22% decrease24%
decrease
20
Conclusions and Contributions
Conclusions Existing localization schemes can
use R-Factor to identify subset of receivers that will result in better location estimation
R-Factor combined with RMS channel diversity was shown theoretically and experimentally to improve localization accuracy
RMS diversity combination was shown to have better localization performance than averaging and selection diversity combination
Contributions A novel parameter called R-
Factor to identify receivers with low range estimation errors was presented
R-factor for selection combination, averaging and root mean square diversity combination were derived
21
Paper 2: Receiver Placement Using Delaunay Refinement based Triangulation in an RSSI Based Localization
Where do I place these receivers?
Shopping Mall Layout
Delaunay refinement placement is the solution
Transmitter being localized with guaranteed accuracy
Possible Applications• Locate cellphones to track foot traffic• Coupons for visiting shops• Theft prevention
22
Objective is to find a receiver layout that will locate a transmitter with error less than a preset threshold with least number of receivers
Euclidean equation that relates the transmitter location to radial distance
between transmitter and receiver is non-linear in xt and yt
Previous effort involved Delaunay Triangulation based placement that combines heuristics and wireless
coverage requirements [Wu 07] Receiver position based on minimizing the condition number of a linear
equation [Isler 06] Radial errors are assumed to be Gaussian distributed and then receiver
positions are selected based on minimization of Fisher information determinant [Martinez 05]
Paper 2: Receiver Placement Using Delaunay Refinement based Triangulation in an RSSI Based Localization
22
22222iiitt
titi
ryxyxyyxx
23
Wireless Propagation Model
Under far-field conditions between transmitter and receiver
However measured signal strength involves noise Pi = Pi*+ei
Estimate of di from Pi , represented as ri is
24
Multi-Lateration Using CWLS Constrained Weighted Least Squares provides a
method to linearize a non-linear equation and solve for parameters in a linear least square sense
Niryxyx
yyxx iiitttiti ,2,1;
22
22222
2 – Parameter Non-Linear Estimation Problem
NiryxR
yyxx iiistiti ,2,1;
22
222
Constraint22tts yxR
3 – Parameter Linear Estimation Problem
25
Localization Error Under CWLS Theorem 1: For an RTLS setup with N receivers the
localization error in estimating the position of the transmitter at location η using CWLS is given by
where λ1, λ2 and λ3 are the eigenvalues of the matrix ,
are the R-factors and ξ≥0 is the
Lagrange multiplier and as the cost of violating 22tts yxR
26
Receiver Placement Quality Metric Maximum localization error at location η occurs when ξ=0
Receiver placement quality metric is the maximum localization error throughout the workspace G
Objective is to attain with least number of receivers
27
Optimal Unconstrained Receiver Placement
Theorem 2: A receiver placement strategy whose objective is to span the largest area under localization coverage with least number of receiver while ensuring no coverage holes exists within the placement grid, will have all its receivers placed in an equilateral triangular grid with grid spacing equal to the communication range of the wireless device
Bounding walls around a workspace prevents equilateral grid placement of receivers !!!!
28
Constrained Receiver Placement Requirements Equilateral triangular grid wherever possible
Near bounding walls triangular grids that are as close to equilateral triangle as possible
Placement should satisfy localization error constraint
Should complete in linear time
29
Delaunay Refinement Triangulation
Originally developed to generate mesh for Finite Element Modeling and Computer Games
Delaunay Refinement satisfies all our receiver placement requirements
However, boundary walls results in sub-optimal receiver count
Delaunay meshing of a 3D object*
* G E O M E T R I C A (http://www-sop.inria.fr/geometrica/ )
30
Receiver Count Under Delaunay Refinement Placing
Theorem 3 (Upper Bound for Receiver Count): For a given workspace G, and a localization error threshold (ϵu), the receiver count generated using Delaunay refinement triangulation on G is suboptimal and is upper bounded by the receiver count for an optimal triangulation of the above receiver placement problem as,
where
31
Local Feature Size and Receiver Count
Local feature size can be described approximately as a measure of the feature (segments and vertices) density of a graph
Removing shorter segments in an input layout resulting in the bound getting tighter
Shorter segments in a layout are those segments that are less than twice wavelength
32
Bounded Receiver Count
Shopping mall layout Airport layout
33
Experimental Result
Layout using (DR) (11 receivers) Layout using DT* (16 receivers)
*Wu 07
Localization area is ERL 114 that measures approx. 12m x 12m
Upper threshold for localization error set at 1m
34
Localization Error (m)Layout Method
Mean Median 75th
percentile Std. dev
DT 1.137 1.038 1.589 0.786DR 0.808 0.678 1.189 0.657
CDF plot of Localization Error
Summary of Localization Error
Localization Accuracy Results
Test Points
28% decrease
34% decrease
25% decrease
35
Conclusions and ContributionsConclusions CWLS Multi-lateration on
receivers placed using Delaunay Refinement achieved better localization accuracy
Better performance of DR due to more triangular regions that are close to equilateral triangle than comparable method
Receiver count though sub-optimal was lower than comparable placement algorithm
Contributions CWLS Localization error was
derived
Relationship between R-factor and localization error under RSSI based RTLS using CWLS
A sub-optimal receiver placement algorithm with guarantees on localization accuracy presented
Paper 3: Localization of RFID Tags using Stochastic Tunneling Multipath fading and shadow fading noise are
the primary cause for large localization error in an indoor environment
Tx
Rx
Tx
Rx
Multipath Fading Shadow Fading
36
Spatial Correlation in Fading Noise
37
38
RFID Basics Typically passive device that are energized by radio waves from a tag
reader
RFID Tag varies the Radar Cross Section (RCS) to communicate its unique identification to the tag reader
Passive RFID system overview1
1Nikitin et al. 2006.
13.56MHz Passive RFID Tag
39
Application Scenario
Tag Reader
Anchor nodes placed around the
reader
Container with RFID tags
Displays tag ID and location
Movement causes multipath noise
Similarity in fading noise experienced by neighboring
RFID tags is exploited to localize them
40
Objective
Localizing RFID tags in a container
Objective to derive a RSSI localization method that works under fading noise
Localize multiple RFID tags simultaneously from a common transmitter
Anchor nodes provide localization correction and reorient the generated location to a global coordinate
Past Work Multi-Dimensional Scaling (MDS) [Ji 04] Local Linear Embedding (LLE) [Costa 06]
41
RFID Tag Localization Flow Chart
42
Assumptions In-phase and Quadrature-phase of backscattered signal
amplitudes are normally distributed
Distribution of backscattered energy around the tag reader is given by a circular normal distribution called von-Mises Distribution
Backscattered signal concentration
43
Theorem 1: Joint PDF of backscattered RSSI values measured by a tag reader from any two RFID tags separated by radial distance r12 is given by
Backscattered Signal PDF
where P1 and P2 are the backscattered RSSI random variables from tag 1 and 2 respectively with p1 and p2 being their realizations, µ1 and µ2 are their average values, 0≤ρ12≤1 and are the backscattered RSSI correlation parameters and I0(◦) is the zeroth order modified Bessel function of the first kind
where Θ12 is the azimuth orientation of the tag reader, δθ12 is the concentration of
multipath signals around the tag reader orientation Θ12 , , λ is operating wavelength and In(◦) and Jn(◦) are the modified and ordinary Bessel functions respectively of the first kind and order n
44
Localization from Correlation Coefficient Theorem 2: The large sample approximate PDF of ρ12 is given by
where is the indicator function that restricts the support of this PDF between [0, 1], and and are the PDF and CDF respectively of a standard normal distribution
Pseudo-likelihood method is used to create an approximate likelihood function for M RFID tags from their pair wise PDF
45
Algorithm called LOCUST (Localization Using Stochastic Tunneling)
8 RFID tags and 8 anchor nodes in a 20m x 20m x 20m workspace
Wireless tags were positioned randomly i.e. xi, yi and zi of the wireless tags are random variables with continuous uniform distribution in the domain [-10, 10] for i є {1,2,…,m}
Total of 50 simulation trials were done to determine the mean, median, standard deviation and 90th percentile of localization errors.
Simulation Results
46
Summary of Localization Error LevelsMethod F
(MHz)
Localization Error (m)
Mean Median90th
percentile
Std. dev.
LOCUST
20.0
0.454 0.429 0.676 0.172
LLE 2.764 2.67 4.095 0.949
MDS 2.272 2.136 3.378 0.778
LOCUST
15.0
0.343 0.331 0.518 0.127
LLE 1.009 0.969 1.507 0.351
MDS 0.935 0.889 1.429 0.375
LOCUST
10.0
0.233 0.230 0.307 0.056
LLE 0.248 0.245 0.326 0.06
MDS 0.194 0.192 0.263 0.05
LOCUST
5.00
0.201 0.189 0.322 0.09
LLE 0.270 0.260 0.396 0.10
MDS 0.194 0.186 0.308 0.086LOCUST
2.50
0.195 0.191 0.283 0.066
LLE 0.187 0.180 0.272 0.063
MDS 0.202 0.195 0.286 0.062
LOCUST
1.00
0.111 0.103 0.177 0.048
LLE 0.198 0.191 0.291 0.07
MDS 0.127 0.117 0.197 0.062
LOCUST
0.06
0.105 0.099 0.164 0.048
LLE 0.202 0.197 0.289 0.066
MDS 0.177 0.165 0.281 0.072
Method F(MHz)
Localization Error (m)
Mean Median90th
percentile
Std. dev.
LOCUST20.0
1.359 1.259 1.943 0.485LLE 7.90 7.318 11.382 2.652MDS 6.019 5.609 8.486 2.238
LOCUST15.0
0.850 0.804 1.179 0.268LLE 2.866 2.831 3.818 0.874MDS 2.92 2.566 4.923 1.490
LOCUST10.0
0.696 0.702 1.067 0.286LLE 1.684 1.657 2.509 0.599MDS 1.722 1.652 2.383 0.513
LOCUST5.00
0.274 0.243 0.469 0.135LLE 0.542 0.500 0.791 0.201MDS 0.477 0.434 0.786 0.207
LOCUST
2.50
0.236 0.227 0.323 0.066LLE 0.198 0.179 0.287 0.061MDS 0.192 0.192 0.256 0.059
LOCUST
1.00
0.131 0.114 0.278 0.060LLE 0.189 0.185 0.177 0.059MDS 0.118 0.112 0.159 0.041
LOCUST
0.06
0.154 0.170 0.216 0.057LLE 0.213 0.189 0.327 0.081MDS 0.178 0.173 0.261 0.062
Accuracy degrades with increasing frequency
Accuracy degrades with LoS
f = 20MHz f = 10MHz
47
Localization Error and Anchor Node CountAnchor Node Count
Localization Error (m)
Mean Median 90th percentile Std. dev.
6 0.486 0.432 0.713 0.253
7 0.354 0.378 0.693 0.173
8 0.293 0.278 0.492 0.142
9 0.223 0.244 0.454 0.119
10 0.215 0.210 0.431 0.113
11 0.220 0.216 0.441 0.121
12 0.236 0.225 0.469 0.136
Localization accuracy degraded after 10 anchor nodes due to the large dimension of the estimated variables
CDF Of localization error at f=20MHz
48
Conclusions and ContributionsConclusions A novel RFID tag localization algorithm
called LOCUST that estimates the position of RFID tags by measuring the correlation between RSSI values between co-located tags was presented
Above 10MHz the non-linear relationship between the correlation coefficient and radial separation results in LOCUST performing better than MDS and LLE
Localization error under LoS condition was larger in comparison to NLoS conditions primarily due to faster drop in correlation coefficient with distance under LoS conditions
Contributions A novel RFID tag localization
algorithm called LOCUST that estimates the position of RFID tags by measuring the correlation between RSSI values between co-located tags was presented
Joint distribution of backscattered power from adjacent RFID tags was derived
Functional relationship between backscatterd signal power correlation, radial separation and line of sight condition was derived
49
Paper 4: Localization and Tracking of Objects Using Cross-Correlation of Shadow Fading Noise Objectives
Increase the frequency of operation of cross-correlation based RSSI localization
Resilient to pedestrian or machinery traffic Improve the convergence speed of cross-correlation
based localization Past work
Network Shadowing [Agarwal 09] Large scale correlation model [Gudmundson 91] Multi-Dimensional Scaling (MDS) [Ji 04] Local Linear Embedding (LLE) [Costa 06]
50
Localization from Shadow Fading Correlation
Neighboring receivers experience similar shadow
fading noise
51
Flowchart of Shadow Fading Cross-Correlation Based Localization
52
Shadow Fading Wireless Channel Model Geometrically Based Single Bounce Elliptical Model (GBSBEM) Wireless Channel Model is assumed under shadow fading
Any radio signal that reaches the receiver after bouncing off of a scatterer in the localization region can affect signal fading if and only if its ToA satisfies
GBSBEM Wireless Channel Model
where r is the radial separation between the transmitter and receiver, c is the speed of radio and τm is the signal integration time at the reciever
IEEE 802.15.4 receivers integrate the signal for 128us before computing the signal strength resulting in τm = 128us
53
Shadow Fading Correlation Coefficient
Pedestrian traffic is modelled as Poisson process
Shadow fading attenuation is normally distributed
Theorem 1: Correlation coefficient under GBSBEM given by
Overlapping of scattering regions causing cross-correlation in shadow fading
where |·| is the area operator, S1 and S2 are the elliptical scatterer regions surrounding receivers R1 and R2 respectively, S12 is overlapping region between scattering regions S1 and S2.
54
Extraction of Shadow Fading Residuals Ornstein Uhlenbeck stochastic model usually applied for high volatility stock trading is used to extract shadow fading residuals from RSSI
Autoregressive Model (AR) for Xs(t) to separate path loss from shadow fading residuals
Generalized Auto Regressive Conditional Heteroskedasticity (GARCH) for to account for fast changes in pedestrian traffic
where ϵs(t)=σs(t)Zs
55
Localization Using Student-t Copula Copula function helps to create joint distributions from marginal CDFs
and their inter-dependency Gaussian & Student-t Copula models linear dependency Gumbel, Frank and Clayton Copulas model tail dependency
Theorem 2: For an M receiver localization system, Student-t copula was used since shadow fading correlation coefficient is a linear dependency
where is the inverse CDF or quantile function vector of a student-t distribution with degree of freedom v, is an M-variate student-t copula density with v degree of freedom, P is an MxM correlation coefficient matrix given by Ρ={ρkl}; k,l ϵ {1,2,…,M} and ρkl is the correlation coefficient between receiver k and l and
56
Tracking Using Divergence Divergence arise in classification problem when a measurement x has to be
categorized into two possible groups C1 or C2
Miss-classification occurs when x is assigned to C1 while it should have been in C2 or vice versa
α-Divergence is a measure of the upper bound in Bayes error in classification problems
where C1||C2 implies divergence operation between groups C1 and C2, f(x|Ci) is the
PDF of random variable X given that it belongs to group Ci;iϵ{1,2}, x is a single realization of random variable X and the integration is over the entire range of random variable
For velocity estimation the hypothesis being tested is that the RSSI values that a receivers measures is from a stationary transmitter
57
α-Divergence for a Mobile Transmitter Theorem 3: For a mobile transmitter operating under GBSBEM wireless
channel model, α-divergence of RSSI measured between two time instances n and n-1 is given by
θn-1 is the azimuth angle of arrival of LoS radio signal at the receiver with respect to the direction of motion of the transmitter while rn-1 is the radial separation between the transmitter and receiver at time instance n-1, Δrn is the distance the transmitter travelled between time instances n-1 and n and rm=cτm and ω is the average scatterer density
Tracking an IEEE 802.15.4 Transmitter
58
Speed Estimation Using α-divergence For slow moving (≤1m/s) IEEE 802.15.4 transmitter, α-
Divergence is related to transmitter’s velocity as
Bhattacharyya coefficient where α=0.5 is used for velocity estimation
Fully functional dead-reckoning based tracking system can be realized from velocity and transmitter heading measured using either a gyroscope or an antenna array
59
Copula Smoothing Using Bayesian Particle Filter
Dead reckoning based tracking method results in incremental position error over time
Bayes particle filter is a stochastic filter that generates multiple random points or particles around the position estimated from dead reckoning method
Student-t Copula likelihood function computes the likelihood of each generated particle which forms the weight of that particle
The copula smoothed position is the weighted average of all the generated particles
Copula smoothing will generate an accurate solution as long as the dead-reckoning generated position is close to the global maxima of the Likelihood function
60
Static Localization Experimental Results
Transmitter Location
Localization Error (m)
Mean Median 90th Perc. Std. Dev
T1 2.458 2.329 3.962 1.727
T2 2.378 2.267 3.628 1.221
T3 3.537 3.496 5.234 2.377
T4 2.739 2.912 4.138 1.839
MethodLocalization Error (m)
Mean Median 90th Perc. Std. DevMDS 12.343 15.925 25.358 6.464
Proposed Method 2.778 2.751 4.2405 1.791
Localization Errors At Various Locations
Summary of Localization Errors
Layout of food court area with dark lines showing physical boundary
Localization area approx. 1250 sq. m with an average of 1000 people moving in this area during peak lunch hour traffic on a weekend between of 10AM and 1PM
8 Receivers R1 through R8 localizing a transmitter
77% decrease
82% decrease
83% decrease
61
Tracking Experimental Results Tracking experiment performed at ERL 114
8 wireless receivers R1 through R8 tracking a transmitter
3DM-GX2 Attitude Heading Reference System (AHRS) from Microstrain attached to the transmitter
Top view of ERL 114 with tracked points
Comparison of Tracking Errors
MethodTracking RMSE (m)
Mean Min Max Std. Devα-divergence 0.3859 0.0464 0.8652 0.2944
INS 0.2466 0.0025 0.6719 0.1972Copula
Smoothing 0.1777 0.0105 0.4379 0.1505
Summary of Tracking Errors
62
Conclusions and ContributionsConclusions Extended the operating frequency
range of cross-correlation based localization from 20MHz to 2.15GHz
At small velocities α-Divergence based velocity estimation performed better than accelerometer based velocity estimation
Contributions Derived the correlation in shadow fading
noise between adjacent receivers
Developed a stochastic filtering method to isolate shadow fading residuals from RSSI
Developed a transmitter velocity estimation technique that measures α-divergence of RSSI values
Copula smoothing algorithm using Bayesian particle filter was implemented to prevent the accumulation of tracking error over time
63
Paper 5: Placement of Receivers for Shadow Fading Cross-Correlation Based Localization Objectives
Provide a receiver placement algorithm for cross-correlation based localization
Resilient to pedestrian and machinery traffic
Past Work Sub-optimal receiver placement using Delaunay
Refinement [Basheer 10] Optimal receiver placement algorithm [Isler 06] Placement based on maximizing the condition number
[Martinez 05]
64
Shadow Fading Wireless Channel Model Geometrically Based Single Bounce
Elliptical Model (GBSBEM) Wireless Channel Model is assumed under shadow fading
Any radio signal that reaches the receiver after bouncing off of a scatterer in the localization region can affect signal fading if and only if its ToA satisfies
GBSBEM Wireless Channel Model
where r is the radial separation between the transmitter and receiver, c is the speed of radio waves, r/c is the ToA of LoS signal and τm is the signal integration time at the reciever
IEEE 802.15.4 receivers integrate the signal for 128us before computing the signal strength resulting in τm = 128us
65
Optimal Unconstrained Receiver Placement Theorem 1: (Equilateral Triangular Grid for Receiver
Placement) A receiver placement strategy whose objective is to span the largest area under localization coverage with least number of receiver while ensuring no coverage holes exists within the grid will have all its receivers placed in an equilateral triangular grid.
Equilateral grid is not possible near bounding walls
66
Receiver placement near bounding walls
Localization coverage hole near bounding walls
67
Theorem 2: Cramer-Rao Lower Bound for the variance in estimating the transmitter at Cartesian coordinate from receivers that are under localization coverage with a transmitter using cross-correlation of shadow fading residuals between receiver pairs is given by
Receiver placement quality metric is for workspace G
Objective is to attain with least number of receivers
Receiver Placement Quality MetricReceiver Placement Quality Metric
67
68
Flowchart of Receiver Placement Algorithm
Equilateral GridCoverage holesLocalization CoverageLocalization Error
69
Simulation Results Delaunay Refinement was used to
search for receiver positions in linear time
Receiver count for cross-correlation coverage placement was lower than using Delaunay Refinement search
Delaunay refinement generates more receivers near sharp edges
Improvement in receiver count was at the expense of search time
Receiver count vs. communication range
70
Simulation ResultsTx.
Location
No. of rx. in range
Localization Error (m)
Mean Median
90th Perc.
Std. Dev
T1 4 0.894 0.792 1.579 0.466
T2 3 0.926 0.883 1.526 0.472
T3 4 0.792 0.828 1.377 0.418
T4 4 0.779 0.698 1.534 0.481
T5 4 0.879 0.927 1.445 0.407
T6 3 0.955 1.100 1.693 0.562
T7 5 0.652 0.690 1.076 0.325
T8 6 0.677 0.550 1.401 0.484
T9 3 0.907 0.943 1.484 0.475
T10 4 0.712 0.762 1.167 0.360
71
Conclusions and Contributions
Contributions Derived the optimal unconstrained
placement for cross-correlation based localization
Derived the Cramer-Rao lower bound for transmitter location estimation variance under cross-correlation based localization method
A receiver placement algorithm was developed for the cross-correlation method that ensures the localization accuracy within the workspace is less than a pre-specified threshold.
Conclusions Developed a receiver placement
algorithm for cross-correlation –based localization
Average localization error was well under the designed 1m error
This method generated lower receiver count for a given communication range when compared with a Delaunay refinement based placement
72
Conclusions and Future Work--Dissertation
Conclusions Localization from cross-
correlation of RSSI is suited for multipath rich environment such as factory floor, food court etc.
Our technique takes advantage of the temporal correlation in RSSI that arise in co-located receivers due to the movement of people or machinery in its vicinity
Performance of our proposed algorithms were validated using hardware experiments on IEEE 802.15.4 receivers
Future work Explore techniques that can
measure transmitter heading from RSSI values so that the requirement for compass or gyroscope can be removed
Improve the execution time for receiver placement algorithm for cross-correlation based localization
Improve the accuracy of shadow fading correlation by better modeling of the shadow fading cross-correlation likelihood
73
Questions
Thank you!
References [Costa 06] J. A. Costa , N. Patwari, and A. O. Hero, “Distributed weighted-multidimensional scaling for node
localization in sensor networks,” ACM Trans. on Sensor Networks, vol.2, No.1, pp.39-64, Feb. 2006.
[Gwon 04] Y. Gwon and R. Jain, “Error characteristics and calibration-free techniques for wireless LAN-based location estimation,” Proc. of ACM MobiWac, pp. 2-9, October 2004.
[Isler 06] V. Isler, “Placement and distributed deployment of sensor teams for triangulation based localization,” In Proc. IEEE ICRA, pp. 3095-3100, May, 2006.
[Ji 04] X. Ji, and H. Zha, “Sensor positioning in wireless ad-hoc sensor networks using multidimensional scaling,” 23rd Annual Joint Conf. of the IEEE Computer and Commun. Soc. , vol.4, pp. 2652- 2661, Mar. 2004.
[Lakhzouri 03] A. Lakhzouri, E. S. Lohan, R. Hamila, and M. Renfors, “Extended kalman filter channel estimation for line-of-sight detection in WCDMA mobile positioning,” EURASIP Journal on Applied Signal Processing, vol. 2003, no. 13, pp. 1268-1278, 2003.
[Martinez 05] S. Martínez, and F. Bullo, “Optimal sensor placement and motion coordination for target tracking,” Proc. of the inter. Conf. on Robotics and Automation, Barcelona, Spain, pp. 4544-4549, April 2005.
[Venkatraman 02] S. Venkatraman and J. Caffery Jr., “Statistical approach to nonline-of-sight BS identification,” Proc. of the 5th International Symp. on Wireless Personal Multimedia Comm., vol. 1, pp. 296–300, Hawaii, USA, October 2002.
[Wu 07] C. Wu, K. Lee, and Y. Chung, “A Delaunay Triangulation based method for wireless sensor network deployment,” Computer Communications, Volume 30, Issue 14-15, pp. 2744-2752, Oct 2007.
74