a distributed time-difference of arrival algorithm for acoustic … · 2018-01-10 · p. w....
TRANSCRIPT
MIT Lincoln Laborator y
A Distrib uted Time-Diff erence of Arriv alAlgorithm for Acoustic Bearing Estimation
Peter Boettc herGary Shaw
Jeff Sherman
SensIT PI MeetingJanuary 15-17, 2002
This work is sponsored by DARPA under A/F Contract #F19628-00-C-0002. Opinions, interpretations, recom-mendations and conclusions are those of the authors and are not necessarily endorsed by the Department ofDefense.
TDOA-1PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-2PWB January 15, 2002
Motiv ation
MIT Lincoln Laborator y
Algorithm that produces instantaneous position estimates
Tracker that operates outside the field of sensor s
Algorithm that is invariant to analog gain
Algorithm that assumes no target dynamics and does not constrain targetmotion
Tracker whose accurac y scales with additional sensor s
TDOA-3PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm– TDOA Estimation– Localization algorithms– TDOA Performance Simulations– TDOA Results from SITEX02
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-4PWB January 15, 2002
Acoustic Time-Diff erence of Arriv al
MIT Lincoln Laborator y
Estimate dominant frequenc y on each node
Downsample dominant frequenc y series, share among nodes
Cross-correlate to determine time delays
PSfrag replacements
Time
Em
itter
Fre
quen
cy
(Hz)
PSfrag replacements Nod
eN
ode
Time
Time
locationTargetTDOANetwork
communication generationestimation
frequencyestimation
Dominant
samplingAcoustic
TDOA-5PWB January 15, 2002
Acoustic Time-Diff erence of Arriv al
MIT Lincoln Laborator y
Estimate dominant frequenc y on each node
Downsample dominant frequenc y series, share among nodes
Cross-correlate to determine time delays
PSfrag replacements
Time
Em
itter
Fre
quen
cy
(Hz)
PSfrag replacements Nod
eN
ode
Time
Time
locationTargetTDOANetwork
communication generationestimation
frequencyestimation
Dominant
samplingAcoustic
TDOA-5-aPWB January 15, 2002
Acoustic Time-Diff erence of Arriv al
MIT Lincoln Laborator y
Estimate dominant frequenc y on each node
Downsample dominant frequenc y series, share among nodes
Cross-correlate to determine time delays
PSfrag replacements
Time
Em
itter
Fre
quen
cy
(Hz)
PSfrag replacements
�����
� ���
����� �
Nod
e
�
Nod
e
�
��
� �� �
�������Time
Time
locationTargetTDOANetwork
communication generationestimation
frequencyestimation
Dominant
samplingAcoustic
TDOA-5-bPWB January 15, 2002
Acoustic Time-Diff erence of Arriv al
MIT Lincoln Laborator y
Estimate dominant frequenc y on each node
Downsample dominant frequenc y series, share among nodes
Cross-correlate to determine time delays
PSfrag replacements
Time
Em
itter
Fre
quen
cy
(Hz)
PSfrag replacements
�����
� ���
����� �
Nod
e
�
Nod
e
�
��
� �� �
�������Time
Time
locationTargetTDOANetwork
communication generationestimation
frequencyestimation
Dominant
samplingAcoustic
TDOA-5-cPWB January 15, 2002
Bearing solution
MIT Lincoln Laborator y
Pairwise TDOA exchang e gives two ambiguous bearings
wavefront
wavefront
sensor �sensor �� ����
�! �"
With several sensor pair s, the correct estimate appear s consistentl y, pic kaverage bearing
PSfrag replacementsPick median bearing
SensorsIncorrect bearingCorrect bearing
TDOA-6PWB January 15, 2002
Bearing solution
MIT Lincoln Laborator y
Pairwise TDOA exchang e gives two ambiguous bearings
wavefront
wavefront
sensor �sensor �� ����
�! �"
With several sensor pair s, the correct estimate appear s consistentl y, pic kaverage bearing
PSfrag replacementsPick median bearing
SensorsIncorrect bearingCorrect bearing
TDOA-6-aPWB January 15, 2002
Hyperbolic location theor y
MIT Lincoln Laborator y
The hyperbola is the set of points at a con-stant range-difference ( # ) from two foci
Each sensor pair gives a hyperbola on whic hthe emitter lies
Location estimation is inter section of all hy-perbolas
Hyperbola ofconstantrange−differance
21PSfrag replacements
$&% $&'
SensorsLocation estimate
Hyperbola from (1,2)
Hyperbola from (1,3)
Hyperbola from (2,3)3
21PSfrag replacements
TDOA-7PWB January 15, 2002
Location on Conic Axis (LOCA) solution
MIT Lincoln Laborator y
With three sensor s, rewrite hyperbolic equations to form mathematical dual .
Sensor s on a conic with emitter at a focus.
Foci ambiguity:– Conic is ellipse: the ‘wr ong’ focus gives negative time-dela ys.
– Conic is hyperbola: the ambiguity is unresolv able
SensorsLocation estimates
conic axis
foci
SensorsLocation estimates
coni
c ax
is
foci
coni
c ax
is
conic axis
foci
foci
SensorsLocation estimates
Hyperbola − unremovable ambiguityEllipse − removable ambiguity
TDOA-8PWB January 15, 2002
Location on Conic Axis (LOCA) solution
MIT Lincoln Laborator y
With three sensor s, rewrite hyperbolic equations to form mathematical dual .
Sensor s on a conic with emitter at a focus.
Foci ambiguity:– Conic is ellipse: the ‘wr ong’ focus gives negative time-dela ys.
– Conic is hyperbola: the ambiguity is unresolv able
SensorsLocation estimates
conic axis
foci
SensorsLocation estimates
coni
c ax
is
foci
coni
c ax
is
conic axis
foci
foci
SensorsLocation estimates
Hyperbola − unremovable ambiguityEllipse − removable ambiguity
TDOA-8-aPWB January 15, 2002
Location on Conic Axis (LOCA) solution
MIT Lincoln Laborator y
With three sensor s, rewrite hyperbolic equations to form mathematical dual .
Sensor s on a conic with emitter at a focus.
Foci ambiguity:– Conic is ellipse: the ‘wr ong’ focus gives negative time-dela ys.
– Conic is hyperbola: the ambiguity is unresolv able
SensorsLocation estimates
conic axis
foci
SensorsLocation estimates
coni
c ax
is
foci
coni
c ax
is
conic axis
foci
foci
SensorsLocation estimates
Hyperbola − unremovable ambiguityEllipse − removable ambiguity
TDOA-8-bPWB January 15, 2002
LOCA solution
MIT Lincoln Laborator y
With four or more sensor s, conic type irrele vant
Each triad of sensor s gives a line (conic axis) containing emitter
– Number of conic axes = Number of triads =(*) )
Least-squares solution finds the point closest to all the lines
Feasib le rang e-diff erence bivector correction as prefilter
SensorsLocation estimate
TDOA-9PWB January 15, 2002
Clustered netw orks
MIT Lincoln Laborator y
Intra-c luster distance is much less than inter -cluster distance .
Compute TDOAs within cluster s to minimiz e comm unication rang e
Each cluster pub lishes results
Users and other cluster s subscribe with long distance comm unication andfuse results
Inter−cluster communication
Long range publish/subscribeUser/hub
SensorsClusters
Comm.
RemoteComm.
Local
On−node processing
TDOA-10PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm– TDOA Estimation– Localization algorithms– TDOA Performance Simulations– TDOA Results from SITEX02
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-11PWB January 15, 2002
Cluster comm unication constraints
MIT Lincoln Laborator y
PSfrag replacements
x (meters)y
(met
ers)
Local communication
x (meters)
y(m
eter
s)
Global communication
x (meters)
y(m
eter
s)
x (meters)
y(m
eter
s)
x (meters)
y(m
eter
s)
Sch
mid
tLO
CA
Sph
er.
Inte
rp.
Bea
ring
Inte
rp.
0 500 1000
0 500 10000 500 1000
0 500 10000 500 1000
0
200
400
600
800
1000
0
100
200
0
200
400
600
800
1000
0
20
40
0
200
400
600
800
1000
0
100
200
0
200
400
600
800
1000
0
20
40
0
200
400
600
800
1000
0
100
200
TDOA-12PWB January 15, 2002
Geometric precision sim ulations
MIT Lincoln Laborator y
PSfrag replacements
x (meters)
y(m
eter
s)+
rms ,
x (meters)
y(m
eter
s)
+rms ,
x (meters)
y(m
eter
s)
+rms ,
x (meters)
y(m
eter
s)
+rms ,
0 500 10000 500 1000
0 500 10000 500 1000
0
200
400
600
800
0
50
100
150
200
0
200
400
600
800
0
50
100
150
200
0
200
400
600
800
0
50
100
150
200
0
200
400
600
800
0
50
100
150
200
TDOA-13PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm– TDOA Estimation– Localization algorithms– TDOA Performance Simulations– TDOA Results from SITEX02
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-14PWB January 15, 2002
SITEX02 Frequenc y Estimation
MIT Lincoln Laborator y
PSfrag replacements
Fre
q(H
z)
50
60
70
80
90
100
110
120
130
140
150
TDOA-15PWB January 15, 2002
SITEX02 2-Node Delay Estimation
MIT Lincoln Laborator y
origin1
PSfrag replacements
Del
ay(s
)
Time (s)0 200 400 600 800 1000 1200 1400
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
TDOA-16PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-17PWB January 15, 2002
Time sync hronization
MIT Lincoln Laborator y
Inter -node time sync hronization is essential for many algorithms– Localization error is related to percenta ge error in time delay estimates.
– Using nodes spaced at 100m, 2.9ms is 1% error
– Nodes might be spaced more closel y
Crystal oscillator s drift at 10-100 ppm, equiv alent to accum ulated error of36-360 ms per hour of operation.
External time reference is necessar y.
NTP is a free program for sync hronizing computer s over a netw ork
Also sync hroniz es cloc ks from reference signals (GPS, radio signal, etc)– GPS provides a timing signal accurate to 100ns.
– Not availab le on Sensoria hardware (firmware/software upgrades will correct this)
NTP netw ork sync hronization used for SITEX02. Accurac y often better than10ms, depending on netw ork load.
TDOA-18PWB January 15, 2002
Single Node Time Skew
MIT Lincoln Laborator y
PSfrag replacements
UTC
Tim
esk
ew
(s)
SH4 Time
DSP Time
21:45 21:50 21:55 22:00 22:05 22:10 22:15-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
TDOA-19PWB January 15, 2002
Time Skew
MIT Lincoln Laborator y
PSfrag replacements
UTC
Tim
esk
ew
(s)
DSP Time
SH4 Time
21:45 21:50 21:55 22:00 22:05 22:10 22:15-0.15
-0.1
-0.05
0
0.05
0.1
TDOA-20PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-21PWB January 15, 2002
GPS averaging
MIT Lincoln Laborator y
Handheld GPS used at SITEX02for node sur veying
Single GPS measurement is ac-curate to 10m
GPS averaging over severalhour s significantl y impr oves theaccurac y.
PSfrag replacements
mm
-10 0 10-10
0
10
PSfrag replacements
Meters
Met
ers
-200 -100 0 100 200
-100
0
100
200
300
TDOA-22PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y
TDOA-23PWB January 15, 2002
Reference coherent bearing estimates
MIT Lincoln Laborator y
5-element acoustic sensor
Coherent bearing estimation & classification
Provides additional ground truth for later analysis
PSfrag replacements
Nov 14 2001, UTC
Bea
ring
tota
rget
18:30 19:00 19:30 20:00 20:30-200
-100
0
100
200
TDOA-24PWB January 15, 2002
Outline
MIT Lincoln Laborator y
Motiv ation
Time-Diff erence of Arriv al Algorithm
Time sync hronization
GPS averaging
Reference coherent bearing estimates
Summar y– Contrib utions– Future work– Publications
TDOA-25PWB January 15, 2002
Contrib utions
MIT Lincoln Laborator y
SITEX02: TDOA software ran in real-time over ISI Diffusion routing
TDOA perf ormance sim ulations- demonstrate benefits of cluster -basedprocessing
Enhanced SITEX02 recor ded data by implementing time sync h/GPSaccurac y impr ovements
Collected ground-truth bearing data
Demonstrated bearing fusion and used Declarative Routing Protocol in For tBenning Tanks Under Trees demonstration, Oct 2001.
Released/documented DRP software
TDOA-26PWB January 15, 2002
Future work
MIT Lincoln Laborator y
TDOA enhancements– Impr ove frequenc y trac ker for high-band width data
– Extend to multi-tar get trac king
– Integrate with trac king/displa y system
Dynamic cluster formation
Acoustic multipath mitigation
On-line refinement of sensor locations and/or time sync hronization
NTP and Lin ux fix es to allo w GPS time sync hronization
TDOA-27PWB January 15, 2002
Publications
MIT Lincoln Laborator y
P. W. Boettc her and G. A. Shaw, “A Distrib uted Time-Diff erence of Arriv alAlgorithm for Acoustic Bearing Estimation”, in Proc. 2001 InternationalConf . on Information Fusion, vol. 1, Montreal, August 2001.
P. W. Boettc her, J. A. Sherman, and G. A. Shaw, “Target localization usingacoustic time-diff erence of arriv al in distrib uted sensor netw orks”, to appearin SPIE AeroSense , Orlando, FL, April 2002.
Project repor ts– Declarative Routing Protocol project repor t
– TDOA sim ulation project repor t
TDOA-28PWB January 15, 2002
Bearing example
MIT Lincoln Laborator y
PSfrag replacements
Local time (hh:mm)
Bea
ring
tota
rget
(deg
rees
)
08:12 08:14 08:16 08:18-200
-150
-100
-50
0
50
100
150
TDOA-29PWB January 15, 2002