Noise cancellation and source localization - Semantic Scholar
Noise
cancellation
and
source
localization
Michael D. Collins, Nicholas C. Makris, and Laurie T. Fialkowski Naval ResearchLaboratory, Washington,DC 20375
(Received24 December1993;acceptedfor publication18 May 1994) A noise-canceling processoris developedfor matched-fieldprocessingproblemsinvolvinga signal buriedin noise.This processoris basedon modelingboth signaland noiseand searchingthe space of unknownparametersto achievethe best agreementbetweencovariances.The noise-canceling processorreducesto the Bartlettprocessorin the limit of high signal-to-noiseratio. The examples illustratethe localizationof a sourceobscuredby interferencefrom ambient noise or a second source.The noise-canceling processoris also appliedto localize a silent object using scattered ambient
noise.
PACS numbers: 43.30.Nb, 43.30.Wi
wherethevector p contains a snapshot 2of thecomplex pres-
INTRODUCTION
Matched-fieldprocessing (MFP) is an approachfor de-
termining thelocation of anacoustic source. 1Thistopichas beenan activearea of researchfor more than a decade.2 MFP methodsare basedon comparingacousticdata from an array of hydrophones with solutionsof the wave equationthat correspondto testsourcelocations.The unknownparametersof MFP usually consistof the coordinatesof the source.The MFP searchspace has recently been expandedto include
environmental parameters,, 3sourcepath parameters• 4andthe coordinates of additional sources. s In thispaper, weexpand the MFP searchspaceto includenoiseparameters. The noise-cancelingprocessor,which is describedin Sec. I, is based on modeling both signal and noise and searchingthe spaceof unknownparametersto achievethe best agreementbetweencovariances. This processorcan localize sources buried
in interference
due to loud sources or
sure(obtainedby takinga Fouriertransformover a relatively shorttimeinterval),thesuperscript asteriskdenotesthecomplex conjugatetranspose, andthe bracketsdenotean average of the outer productover a sequenceof snapshots.We assume the data are dominatedby n temporallyuncorrelated processesto obtain (2) j--1
whereKi is thecovariance matrixfor thejth process. Some or all of the covariancematricesdependon unknownparameters,suchas sourcecoordinates, sourcelevels, noiseparameters, andenvironmental parameters. To estimate theseparameters,the least-squares differencebetween the measured
and modeled
covariance
matrices
is minimized
over the parameterspace.We place the entriesof the covariancematricesappearingin Eq. (2) intovectorsthatcontain
ambientnoise and reducesto the Bartlett processorin the noise-freecase.Since the Bartlettprocessoris tolerantof the m2 entries, normalize thesevectors, anddefinethecostfuncdeviationsfrom ideal conditionsthat are prevalentin applition, cations,the noise-cancelingprocessorshouldperform well n 2 for applications.The developmentof the noise-canceling processorwas motivatedby the successof a noise-canceling E-=fi-• ej•,j , (3) j=l beamformer(which is basedon similartechniques) for data
froma towedarray. 6 Examplesare presentedin Sec. II to illustratethe localization of a sourceobscuredby interferencefrom ambient noise or a secondsource.The noise-canceling processoris also appliedto localize a silentobjectusingscatteredambient noise.There has recentlybeen a great deal of interestin
problems of thistype. 7-• I. A NOISE-CANCELING
either modeled or estimated from data. We define A to be the
m2Xn matrix whose jth column is•,j, andEq.(3)becomes
matrix for the acoustic data from
an array of m receiversis definedby
K=(pp*), J. Acoust. Soc. Am. 96 (3), September 1994
(4)
wheree contains theentriesej.
PROCESSOR
ances.
1773
corresponds toKj, andtheenergylevelsej areassumed to
beunknown. Toevaluate therightsideofEq.(3),the•,•are
E--I-gel
In this section,we derive a noise-cancelingprocessor that is a generalizationof the Bartlett processorand should thereforebe robustfor applications.Since it is difficult to model time seriessnapshots of noise,we work with covariThe rn X rn covariance
where theunitvector fi corresponds toK, theunitvector •,j
(1)
To eliminatethe energylevelsfrom the parameterspace, we minimize E over e while holding the other parameters constant.The least-squares minimum over the energylevels occurs for
e= (A*A)- 1A* fl.
(5)
Substituting this solutioninto Eq. (4), we obtain
minE= 1-B 2
(6) 1773
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
II
o
!ttl!
100
'
t
O0
00
,
3011
!
400
400
0
0
10tl
11tlt
O0
oo
30{}
300
400
l)
',
10
i
It
Ran_
1o
1•
II
Ran_•e(kin
FIG. 1. The Bartlettprocessor (top) andthe noise-canceling processor (bottom) for exampleA, which involvessourcesat (r,z)-(13 kin, 150 m) and (r,z)=(8 kin, 150 m). Redcorresponds to the mostlikely sourcelocations. Purple corresponds to the least likely sourcelocations.
Be=f*A(A*A)-•A*6.
$
(7)
Thenx n matrix A*A reduces totheidentity matrix if the•j
FIG. 2. The Bartlettprocessor (top)andthe noise-canceling processor (bottom) for exampleB, whichinvolvessourcesat (r,z)-(13 kin, 150 m) and (r,z)=(5.12 km, 150 m). Red corresponds to the most likely sourcelocations.Purple corresponds to the leastlikely sourcelocations.
geneouswater columnin which the soundspeedis 1500 m/s and a homogeneoussediment.The approachof Ref. 14 is used to model ambient
noise.
aremutually orthogonal. In thiscase, B2 is in theformof a
ExamplesA, B, and C involve a 25-Hz sourceat the the noise-canceling unknownlocation(r,z) = (13 kin, 150 m), a 400-m deep processor B reduces to theBartlettprocessor. AlthoughA *A ocean,and an array of 13 hydrophoneswith the jth hydrois not invertibleif the columnsof A are linearly dependentat phoneplacedat z= (- 15 + 30j)m. For theseexamples,the a point in the searchspace(thisoccursfor someof the exsedimentsoundspeedis 1700 m/s, the sedimentdensityis Bartlett processor.For the case n-1,
amples inSec.II), thesingularity inthebounded function B•'
1.5g/cm 3, andthesediment attenuation is 0.5 dB/h.An in-
is removable.
terferingsourceis placedat (r,z)=(8 kin, 150 m) for ex-
An estimatefor the unknownparametersis obtainedby maximizingB. If this ambiguityfunctiondependson a large
numberof parameters(suchas source,environmental, and noise parameters),this task may call for an optimization
method suchassimulated annealing, 1z13 whichwasusedin Refs. $ and 6. For a low-dimensionalparameterspace,it is often practicalto evaluatethe ambiguityfunctionon a dense subsetof the space.This approachis commonly used for MFP problems(includingthe examplesin Sec.II) in which the rangeand depthof a sourceare the unknowns.
ample A. The locationof this source,which radiatesat the same level as the other source, is assumed to be known a
priori. For thesesourcelocations,the signalsdue to the two sourcesare uncorrelatedon the array. Resultsfor exampleA generatedwith the Bartlett and noise-canceling processors appearin Fig. 1. The Bartlettprocessorprovidesan ambiguousestimateof the locationof the unknownsource.The main peak in the noise-canceling processor,which closelyresemblesthe Bartlett processorfor a singlesource,occursat the locationof the unknownsource. We also testedthe sensitivityof the noise-cancelingproces-
sorto mismatch • forthisproblem by computing thereplica II. EXAMPLES
In this section, we illustrate the performanceof the noise-cancelingprocessor.Each of the examplesinvolves a range-independent oceanenvironmentconsistingof a homo1774
J. Acoust. Soc. Am., Vol. 96, No. 3, September 1994
fields using an assumedsedimentsoundspeedof 1725 m/s. We found that this amount of mismatchcausesthe performance of the noise-cancelingprocessorto be only slightly
degraded(the main peak is slightlyreducedrelativeto the main sidelobes). Collinset aL: Noise cancellationand source localization 1774
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
0
- 0
-45
oO
0
45
;oo
4O0
90
o
-90
-45
ioo
oo
g
300
45
90
400
0
;
10
15
-90
20
-45
0
45
90
Azi•nuth (deg)
Range(kin) FIG. 3. The Bartlettprocessor (top)andthe noise-canceling processor (bottom) for exampleC, which involvesa sourceat (r,z)=(13 km, 150 m) buried in ambientnoise.Red corresponds to the most likely sourcelocations.Purplecorresponds to the leastlikely sourcelocations.
FIG. 4. The Bartlettprocessor (top) andthe noise-canceling processor (bottom) for exampleD, which involvesa 10-m spherethat scattersambient noise.Red corresponds to the most likely sourcelocations.Purple correspondsto the leastlikely sourcelocations.
ExampleB is identicalto exampleA with the exception thatthe interferingsourceis movedto (r,z)=(5.12 km, 150
which are placedin a squaregrid with a spacingof 2.5 m, is placedin the middle of the water column 100 m away from the sphereand orientedfacing the sphere.In the sediment,
m). For thislocation,the signalsdueto the two sourcesare correlatedon the array.Resultsfor exampleB generatedwith the Bartlettand noise-canceling processors appearin Fig. 2. For the Bartlettprocessor,the sidelobenearr = 9 km is comparable to the peak at the locationof the unknown source. The noise-canceling processorhas a large peak at the location of the unknown
source. The sidelobes
in the noise-
cancelingprocessorare lessprominentthan for exampleA. Example C involves ambientnoise that is 14 dB above the signaldue to the unknownsourceon the array.Results for example C generated with the Bartlett and noisecancelingprocessors appearin Fig. 3. For the Bartlettprocessor,thereare large peaksnear the oceansurface(where the noiseis generated)and near the oceanbottom(which resemblesthe ocean surface in the acoustic sense). The sourcepeak and the sidelobesshowup faintly in this ambiguity surface.However,the peakcorresponding to the source locationis dominatedby oneof the sidelobes. The mainpeak in the noise-canceling processorcorresponds to the source location.
Example D involves 300-Hz processingto localize a pressure-release spherethat scattersambientnoise.The approachof Ref. 11 is usedto solvethe scatteringproblem.The sphereis of radius 10 m and is centeredat z=50 m in a 100-m-deepocean.A 7x7 billboard array of hydrophones, 1775
J. Acoust. Soc. Am., Vol. 96, No. 3, September 1994
thesound speed is 1700m/s,thedensity is1.9g/cm 3,andthe attenuation is 0.4 dB/h.
The signal-to-noiseratio is -20 dB. The spherewas imagedusingplane-wavereplicafields.Resultsfor example D generatedwith the Bartlettand noise-canceling processors appearin Fig. 4. Little or no evidenceof the scattererappears in the Bartlettprocessor. The scattererand its reflectionfrom the oceansurfaceshowup clearlyin the noise-canceling processor.The reflectionfrom the oceanbottomalso showsup faintly.
III. CONCLUSION
The noise-cancelingprocessorlocalizessourcesburied in noiseby matchingthe covarianceof the data with a combinationof signaland noisecovariances. This processorreducesto the Bartlettprocessorin the limit of high signal-tonoiseratio andis relatedto a noise-cancellation techniquefor plane-wavebeamformingthat has been appliedsuccessfully to data. The noise-cancelingprocessorhas been testedfor problemsinvolving interferencefrom an additional source and ambientnoise.The noise-cancelingprocessorwas also applied to localize a silent object using scatteredambient noise.
Collinset al.' Noise cancellationand source localization 1775
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
•H. P.Bucker,"Useof calculated soundfieldsandmatched-field detection to locate sound sourcesin shallow water," J. Acoust. Soc. Am. 59, 368373 (1976).
2A.B. Baggeroer, W.A. Kuperman, andP.N. Mikhalevsky, "Anoverview
acousticglow," JASON Rep. JSR-85-108, MITRE Corp., McLean, VA (1985).
8K. Case,R. Davis,S. Flatt&andF. Zachiriasen, "Occultation study summary,"JASON Rep. JSR-86-108,MITRE Corp.,McLean,VA (1987).
of matchedfield methodsin oceanacoustics,"IEEE J. OceanEng. 18, 401-424 (1993).
øM. J. Buckingham, B. V. Berkhout, andS. A. L. Glegg,"Imagingthe
3M. D. CollinsandW.A. Kuperman, "Focalization: Environmental focus-
10j.R. Potter,"Acoustic imaging usingambient noise:Sometheoryand
ing and sourcelocalization,"J. Acoust.Soc.Am. 90, 1410-1422 (1993).
oceanwith ambientnoise," Nature 356, 327-329 (1992). simulationresults,"J. Acoust. Soc. Am. 95, 21-33 (1994).
4M. D. Collins,L. T. Fialkowski, W. A. Kuperman, andJ. S. Perkins, l•N. C. Makris,F. Ingenito, andW. A. Kuperman, "Detection of a sub"Environmentalsourcetracking," J. Acoust. Soc. Am. 94, 3335-3341 (1993).
M.D. Collins, L. T. Fialkowski, W.A. Kuperman, andJ.S.Perkins, "The
mergedobject insonifiedby surfacenoise in an oceanwaveguide,"J. Acoust.Soc. Am. 96, 1703-1724 (1994).
12N.Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller,andE.
multi-valuedBartlettprocessorand sourcetracking,"J. Acoust.Soc.Am. (submitted).
Teller, "Equationsof statecalculationsby fast computingmachines,"J. Chem.Phys.21, 1087-1091 (1953).
6M. D. Collins,J. S. Berkson, W. A. Kuperman, N. C. Makris,andJ. S.
•3S. Kirkpatrick, C. D. Gellatt,andM.P. Vecchi, "Optimization by simu-
Perkins,"Applicationsof optimaltime-domainbeamforming,"J. Acoust. Soc. Am. 93, 1851-1865 (1993).
7S. Flatt• andW. Munk, "Submarine detection: Acousticcontrast versus
1776
J. Acoust. Soc. Am., Vol. 96, No. 3, September 1994
latedannealing,"Science220, 671-680 (1983).
14W. m.Kuperman andF.Ingenito, "Spatial correlation of surface generated noise in a stratifiedocean," J. Acoust. Soc. Am. 67, 1988-1996 (1980).
Collins et aL: Noise cancellationand source localization 1776
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
cancellation
and
source
localization
Michael D. Collins, Nicholas C. Makris, and Laurie T. Fialkowski Naval ResearchLaboratory, Washington,DC 20375
(Received24 December1993;acceptedfor publication18 May 1994) A noise-canceling processoris developedfor matched-fieldprocessingproblemsinvolvinga signal buriedin noise.This processoris basedon modelingboth signaland noiseand searchingthe space of unknownparametersto achievethe best agreementbetweencovariances.The noise-canceling processorreducesto the Bartlettprocessorin the limit of high signal-to-noiseratio. The examples illustratethe localizationof a sourceobscuredby interferencefrom ambient noise or a second source.The noise-canceling processoris also appliedto localize a silent object using scattered ambient
noise.
PACS numbers: 43.30.Nb, 43.30.Wi
wherethevector p contains a snapshot 2of thecomplex pres-
INTRODUCTION
Matched-fieldprocessing (MFP) is an approachfor de-
termining thelocation of anacoustic source. 1Thistopichas beenan activearea of researchfor more than a decade.2 MFP methodsare basedon comparingacousticdata from an array of hydrophones with solutionsof the wave equationthat correspondto testsourcelocations.The unknownparametersof MFP usually consistof the coordinatesof the source.The MFP searchspace has recently been expandedto include
environmental parameters,, 3sourcepath parameters• 4andthe coordinates of additional sources. s In thispaper, weexpand the MFP searchspaceto includenoiseparameters. The noise-cancelingprocessor,which is describedin Sec. I, is based on modeling both signal and noise and searchingthe spaceof unknownparametersto achievethe best agreementbetweencovariances. This processorcan localize sources buried
in interference
due to loud sources or
sure(obtainedby takinga Fouriertransformover a relatively shorttimeinterval),thesuperscript asteriskdenotesthecomplex conjugatetranspose, andthe bracketsdenotean average of the outer productover a sequenceof snapshots.We assume the data are dominatedby n temporallyuncorrelated processesto obtain (2) j--1
whereKi is thecovariance matrixfor thejth process. Some or all of the covariancematricesdependon unknownparameters,suchas sourcecoordinates, sourcelevels, noiseparameters, andenvironmental parameters. To estimate theseparameters,the least-squares differencebetween the measured
and modeled
covariance
matrices
is minimized
over the parameterspace.We place the entriesof the covariancematricesappearingin Eq. (2) intovectorsthatcontain
ambientnoise and reducesto the Bartlett processorin the noise-freecase.Since the Bartlettprocessoris tolerantof the m2 entries, normalize thesevectors, anddefinethecostfuncdeviationsfrom ideal conditionsthat are prevalentin applition, cations,the noise-cancelingprocessorshouldperform well n 2 for applications.The developmentof the noise-canceling processorwas motivatedby the successof a noise-canceling E-=fi-• ej•,j , (3) j=l beamformer(which is basedon similartechniques) for data
froma towedarray. 6 Examplesare presentedin Sec. II to illustratethe localization of a sourceobscuredby interferencefrom ambient noise or a secondsource.The noise-canceling processoris also appliedto localize a silentobjectusingscatteredambient noise.There has recentlybeen a great deal of interestin
problems of thistype. 7-• I. A NOISE-CANCELING
either modeled or estimated from data. We define A to be the
m2Xn matrix whose jth column is•,j, andEq.(3)becomes
matrix for the acoustic data from
an array of m receiversis definedby
K=(pp*), J. Acoust. Soc. Am. 96 (3), September 1994
(4)
wheree contains theentriesej.
PROCESSOR
ances.
1773
corresponds toKj, andtheenergylevelsej areassumed to
beunknown. Toevaluate therightsideofEq.(3),the•,•are
E--I-gel
In this section,we derive a noise-cancelingprocessor that is a generalizationof the Bartlett processorand should thereforebe robustfor applications.Since it is difficult to model time seriessnapshots of noise,we work with covariThe rn X rn covariance
where theunitvector fi corresponds toK, theunitvector •,j
(1)
To eliminatethe energylevelsfrom the parameterspace, we minimize E over e while holding the other parameters constant.The least-squares minimum over the energylevels occurs for
e= (A*A)- 1A* fl.
(5)
Substituting this solutioninto Eq. (4), we obtain
minE= 1-B 2
(6) 1773
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
II
o
!ttl!
100
'
t
O0
00
,
3011
!
400
400
0
0
10tl
11tlt
O0
oo
30{}
300
400
l)
',
10
i
It
Ran_
1o
1•
II
Ran_•e(kin
FIG. 1. The Bartlettprocessor (top) andthe noise-canceling processor (bottom) for exampleA, which involvessourcesat (r,z)-(13 kin, 150 m) and (r,z)=(8 kin, 150 m). Redcorresponds to the mostlikely sourcelocations. Purple corresponds to the least likely sourcelocations.
Be=f*A(A*A)-•A*6.
$
(7)
Thenx n matrix A*A reduces totheidentity matrix if the•j
FIG. 2. The Bartlettprocessor (top)andthe noise-canceling processor (bottom) for exampleB, whichinvolvessourcesat (r,z)-(13 kin, 150 m) and (r,z)=(5.12 km, 150 m). Red corresponds to the most likely sourcelocations.Purple corresponds to the leastlikely sourcelocations.
geneouswater columnin which the soundspeedis 1500 m/s and a homogeneoussediment.The approachof Ref. 14 is used to model ambient
noise.
aremutually orthogonal. In thiscase, B2 is in theformof a
ExamplesA, B, and C involve a 25-Hz sourceat the the noise-canceling unknownlocation(r,z) = (13 kin, 150 m), a 400-m deep processor B reduces to theBartlettprocessor. AlthoughA *A ocean,and an array of 13 hydrophoneswith the jth hydrois not invertibleif the columnsof A are linearly dependentat phoneplacedat z= (- 15 + 30j)m. For theseexamples,the a point in the searchspace(thisoccursfor someof the exsedimentsoundspeedis 1700 m/s, the sedimentdensityis Bartlett processor.For the case n-1,
amples inSec.II), thesingularity inthebounded function B•'
1.5g/cm 3, andthesediment attenuation is 0.5 dB/h.An in-
is removable.
terferingsourceis placedat (r,z)=(8 kin, 150 m) for ex-
An estimatefor the unknownparametersis obtainedby maximizingB. If this ambiguityfunctiondependson a large
numberof parameters(suchas source,environmental, and noise parameters),this task may call for an optimization
method suchassimulated annealing, 1z13 whichwasusedin Refs. $ and 6. For a low-dimensionalparameterspace,it is often practicalto evaluatethe ambiguityfunctionon a dense subsetof the space.This approachis commonly used for MFP problems(includingthe examplesin Sec.II) in which the rangeand depthof a sourceare the unknowns.
ample A. The locationof this source,which radiatesat the same level as the other source, is assumed to be known a
priori. For thesesourcelocations,the signalsdue to the two sourcesare uncorrelatedon the array. Resultsfor exampleA generatedwith the Bartlett and noise-canceling processors appearin Fig. 1. The Bartlettprocessorprovidesan ambiguousestimateof the locationof the unknownsource.The main peak in the noise-canceling processor,which closelyresemblesthe Bartlett processorfor a singlesource,occursat the locationof the unknownsource. We also testedthe sensitivityof the noise-cancelingproces-
sorto mismatch • forthisproblem by computing thereplica II. EXAMPLES
In this section, we illustrate the performanceof the noise-cancelingprocessor.Each of the examplesinvolves a range-independent oceanenvironmentconsistingof a homo1774
J. Acoust. Soc. Am., Vol. 96, No. 3, September 1994
fields using an assumedsedimentsoundspeedof 1725 m/s. We found that this amount of mismatchcausesthe performance of the noise-cancelingprocessorto be only slightly
degraded(the main peak is slightlyreducedrelativeto the main sidelobes). Collinset aL: Noise cancellationand source localization 1774
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
0
- 0
-45
oO
0
45
;oo
4O0
90
o
-90
-45
ioo
oo
g
300
45
90
400
0
;
10
15
-90
20
-45
0
45
90
Azi•nuth (deg)
Range(kin) FIG. 3. The Bartlettprocessor (top)andthe noise-canceling processor (bottom) for exampleC, which involvesa sourceat (r,z)=(13 km, 150 m) buried in ambientnoise.Red corresponds to the most likely sourcelocations.Purplecorresponds to the leastlikely sourcelocations.
FIG. 4. The Bartlettprocessor (top) andthe noise-canceling processor (bottom) for exampleD, which involvesa 10-m spherethat scattersambient noise.Red corresponds to the most likely sourcelocations.Purple correspondsto the leastlikely sourcelocations.
ExampleB is identicalto exampleA with the exception thatthe interferingsourceis movedto (r,z)=(5.12 km, 150
which are placedin a squaregrid with a spacingof 2.5 m, is placedin the middle of the water column 100 m away from the sphereand orientedfacing the sphere.In the sediment,
m). For thislocation,the signalsdueto the two sourcesare correlatedon the array.Resultsfor exampleB generatedwith the Bartlettand noise-canceling processors appearin Fig. 2. For the Bartlettprocessor,the sidelobenearr = 9 km is comparable to the peak at the locationof the unknown source. The noise-canceling processorhas a large peak at the location of the unknown
source. The sidelobes
in the noise-
cancelingprocessorare lessprominentthan for exampleA. Example C involves ambientnoise that is 14 dB above the signaldue to the unknownsourceon the array.Results for example C generated with the Bartlett and noisecancelingprocessors appearin Fig. 3. For the Bartlettprocessor,thereare large peaksnear the oceansurface(where the noiseis generated)and near the oceanbottom(which resemblesthe ocean surface in the acoustic sense). The sourcepeak and the sidelobesshowup faintly in this ambiguity surface.However,the peakcorresponding to the source locationis dominatedby oneof the sidelobes. The mainpeak in the noise-canceling processorcorresponds to the source location.
Example D involves 300-Hz processingto localize a pressure-release spherethat scattersambientnoise.The approachof Ref. 11 is usedto solvethe scatteringproblem.The sphereis of radius 10 m and is centeredat z=50 m in a 100-m-deepocean.A 7x7 billboard array of hydrophones, 1775
J. Acoust. Soc. Am., Vol. 96, No. 3, September 1994
thesound speed is 1700m/s,thedensity is1.9g/cm 3,andthe attenuation is 0.4 dB/h.
The signal-to-noiseratio is -20 dB. The spherewas imagedusingplane-wavereplicafields.Resultsfor example D generatedwith the Bartlettand noise-canceling processors appearin Fig. 4. Little or no evidenceof the scattererappears in the Bartlettprocessor. The scattererand its reflectionfrom the oceansurfaceshowup clearlyin the noise-canceling processor.The reflectionfrom the oceanbottomalso showsup faintly.
III. CONCLUSION
The noise-cancelingprocessorlocalizessourcesburied in noiseby matchingthe covarianceof the data with a combinationof signaland noisecovariances. This processorreducesto the Bartlettprocessorin the limit of high signal-tonoiseratio andis relatedto a noise-cancellation techniquefor plane-wavebeamformingthat has been appliedsuccessfully to data. The noise-cancelingprocessorhas been testedfor problemsinvolving interferencefrom an additional source and ambientnoise.The noise-cancelingprocessorwas also applied to localize a silent object using scatteredambient noise.
Collinset al.' Noise cancellationand source localization 1775
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
•H. P.Bucker,"Useof calculated soundfieldsandmatched-field detection to locate sound sourcesin shallow water," J. Acoust. Soc. Am. 59, 368373 (1976).
2A.B. Baggeroer, W.A. Kuperman, andP.N. Mikhalevsky, "Anoverview
acousticglow," JASON Rep. JSR-85-108, MITRE Corp., McLean, VA (1985).
8K. Case,R. Davis,S. Flatt&andF. Zachiriasen, "Occultation study summary,"JASON Rep. JSR-86-108,MITRE Corp.,McLean,VA (1987).
of matchedfield methodsin oceanacoustics,"IEEE J. OceanEng. 18, 401-424 (1993).
øM. J. Buckingham, B. V. Berkhout, andS. A. L. Glegg,"Imagingthe
3M. D. CollinsandW.A. Kuperman, "Focalization: Environmental focus-
10j.R. Potter,"Acoustic imaging usingambient noise:Sometheoryand
ing and sourcelocalization,"J. Acoust.Soc.Am. 90, 1410-1422 (1993).
oceanwith ambientnoise," Nature 356, 327-329 (1992). simulationresults,"J. Acoust. Soc. Am. 95, 21-33 (1994).
4M. D. Collins,L. T. Fialkowski, W. A. Kuperman, andJ. S. Perkins, l•N. C. Makris,F. Ingenito, andW. A. Kuperman, "Detection of a sub"Environmentalsourcetracking," J. Acoust. Soc. Am. 94, 3335-3341 (1993).
M.D. Collins, L. T. Fialkowski, W.A. Kuperman, andJ.S.Perkins, "The
mergedobject insonifiedby surfacenoise in an oceanwaveguide,"J. Acoust.Soc. Am. 96, 1703-1724 (1994).
12N.Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller,andE.
multi-valuedBartlettprocessorand sourcetracking,"J. Acoust.Soc.Am. (submitted).
Teller, "Equationsof statecalculationsby fast computingmachines,"J. Chem.Phys.21, 1087-1091 (1953).
6M. D. Collins,J. S. Berkson, W. A. Kuperman, N. C. Makris,andJ. S.
•3S. Kirkpatrick, C. D. Gellatt,andM.P. Vecchi, "Optimization by simu-
Perkins,"Applicationsof optimaltime-domainbeamforming,"J. Acoust. Soc. Am. 93, 1851-1865 (1993).
7S. Flatt• andW. Munk, "Submarine detection: Acousticcontrast versus
1776
J. Acoust. Soc. Am., Vol. 96, No. 3, September 1994
latedannealing,"Science220, 671-680 (1983).
14W. m.Kuperman andF.Ingenito, "Spatial correlation of surface generated noise in a stratifiedocean," J. Acoust. Soc. Am. 67, 1988-1996 (1980).
Collins et aL: Noise cancellationand source localization 1776
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 18.38.0.166 On: Mon, 12 Jan 2015 18:30:52
