DRDC STARFISH ACOUSTIC SENTINEL AND PHASE GRADIENT HISTOGRAM TRACKING

Transcription

1 DRDC-RDDC-5-N5 UACE5 - r Unerwater Acoutic Conference an Exhibition DRDC STARFISH ACOUSTIC SENTINEL AND PHASE GRADIENT HISTOGRAM TRACKING Carmen Luca, Garry J. Hear, Nico Pelava Unerwater Surveillance an Communication Group Defence Reearch an Development Canaa - Atlantic 9 Grove Street, Dartmouth, Nova Scotia, Canaa BA C5 Ph: 9-4- Ext. carmen.luca@rc-rc.gc.ca Abtract: DRDC Atlantic Reearch Centre, an agency of the Department of National Defence in Canaa, ha evelope a prototype multi-influence unerwater urveillance enor calle the Starfih. The Starfih ha acoutic, electric an magnetic fiel enor, an onboar ignal proceing for target etection an tracing. The acoutic enor are in the form of a even-element cro-ipole hyrophone array, with two hyrophone along each orthogonal axi irection an one at the array centre. Acoutic target etection i performe uing time-omain an frequency-omain Sentinel algorithm. The Sentinel algorithm ue parallel exponential filter, with the frequency-omain algorithm uing parallel filter in each frequency bin. A Phae Graient bearing etimation algorithm i ue to perform the acoutic target tracing. Thi i a frequency-omain algorithm that etimate the graient of the patial phae of the incoming acoutic ignal, an irectly compute the wavevector of the acoutic arrival in each frequency bin. A Hitogram Tracing proceure i ue to iplay an etermine the trac angle for the broa-ban acoutic target. Thi paper will icu the etail of the frequency-omain Acoutic Sentinel algorithm an the Phae Graient bearing etimation algorithm with Hitogram Tracing. Keywor: hyrophone, etection, bearing, hitogram. INTRODUCTION The DRDC Atlantic Reearch Centre ha ha a multi-year reearch an evelopment program to create a prototype unerwater multi-influence enor pacage for littoral urveillance application. The prototype enor, hown in Fig., are in their econ generation an are calle "Starfih" enor noe. The Starfih noe are eigne to be eploye in up to 5 m water epth. Each Starfih enor noe contain a low-noie, high enitivity tri-axial magnetometer, three Unerwater Electric Potential enor, a evenelement croe-ipole hyrophone array, a wie-ban hyrophone, an a combine D gravitational an magnetic orientation enor. Each Starfih noe contain a powerful onboar ARM proceor that run an embee verion of the Linux operating ytem. The proceor i powerful enough to perform real-time acquiition, torage, an ignal proceing of the enor ata. Target etection an tracing reult from the onboar proceing are relaye through a networ of unerwater acoutic moem to a urface gateway buoy, which ha a wirele raio frequency lin to hore. Each Starfih noe contain internal oli-tate memory torage where enor ata may alo be logge for later analyi. The Starfih noe may be eploye with or without cable to hore. When eploye without cable, each Her Majety the Queen in Right of Canaa, a repreente by the Miniter of National Defence, 5

2 UACE5 - r Unerwater Acoutic Conference an Exhibition Starfih noe can run autonomouly for more than five ay of continuou ue, or for everal wee in a wae-on-etection moe. Thi paper ecribe the acoutic proceing running onboar the Starfih proceor. The current uite of acoutic proceing ha everal tep: firt the acoutic Sentinel frequency-omain algorithm etermine if there i an acoutic target nearby, econ the acoutic tracing algorithm uing the Phae Graient bearing etimation algorithm with Hitogram Tracing etermine the target bearing an bearing change rate, an finally a compact NILUS protocol meage, evelope at FFI Norway, i generate an tranmitte to the uer over the communication networ. Fig. : Secon generation of the DRDC Starfih enor noe. FREQUENCY DOMAIN ACOUSTIC SENTINEL The acoutic algorithm that etermine if a target i etecte by the Starfih enor noe i calle Acoutic Sentinel. The frequency omain Acoutic Sentinel algorithm wor on FFT procee ata collecte from the centre hyrophone of the croe-ipole hyrophone array. Each channel of the hyrophone array i ample at a rate of 5 ample per econ, an i low-pa filtere with a 7 Hz cutoff frequency. The typical frequency ban for frequency omain Acoutic Sentinel proceing i Hz to Hz. The Sentinel algorithm ue a ban of exponential igital filter, two filter for each FFT frequency bin in the Sentinel proceing banwith. The input to each pair of filter i the average power pectrum value obtaine from the FFT proceing for that frequency bin. The general input-output relationhip of the exponential filter i given by Eqn., with n an x(n), y(n) = when n <. y( n) ( ) x( n) y( n ) () ( ) x( n ) In Eqn., x(n) an y(n) are the filter input an output repectively at ample n. The parameter α etermine the low-pa cut-off frequency of the filter, < α <. With each frequency bin we aociate two filter with ifferent α value. A een from the econ line of Eqn., the "low" filter, with larger α value, average over more previou input x(n) than the "fat" filter with the maller α value. The low filter output i referre to a the long-term - 7 -

3 UACE5 - r Unerwater Acoutic Conference an Exhibition average, an the fat filter output a the hort-term average. The low filter output will trac input ignal uch a bacgroun noie that change lowly over time, an reject horter perio ignal. The fat filter output will repon more quicly to a horter perio target ignal. Comparing the two filter output in a ratio of the fat to low will give a meaure that can be ue to etect a target paage, while ajuting to graual change in the bacgroun ambient noie. In the Acoutic Sentinel algorithm in the Starfih noe we typically ue 4-point FFT with 5% overlap to compute the average pectrum. Typically a total of FFT are ue for the averaging, with the lat FFT being ue in the next group of FFT. Thi reult in an average pectral power value input to the parallel filter, in each frequency bin, once every 48 ample, or at an effective ample rate of f =. Hz. For the low an fat repone filter the parameter α ha the efault value of.999 an.9 repectively. The frequency repone of the filter with thee two α value i hown in Fig.. The cut-off frequency (- B point) of the low repone filter i 94 µhz, an for the fat repone filter i 9.4 mhz, when ata i ent to the filter at a rate of f =. Hz. Fig. : Normalize frequency repone of the two exponential igital filter ue in the Starfih Acoutic Sentinel algorithm The Acoutic Sentinel algorithm will trigger acoutic target etection when 75% of the frequency bin in the Hz to Hz ban excee a ratio threhol value of 5. A target lot meage i triggere when 75% of the bin fall below a ratio threhol value of. The FFT length an overlap value, the percentage of the bin above an below the threhol value, the threhol level, an frequency ban range are all algorithm parameter that are programmable in the Starfih noe. Fig. how an example of the filter repone an reulting ratio of the fat to low output for a real target pa, for one FFT bin.. PHASE GRADIENT BEARING ESTIMATION ALGORITHM The Phae Graient algorithm wa eigne to proce ata from a tri-axi cro-ipole, even-element hyrophone array an to etimate the bearing an elevation angle to an - 7 -

4 UACE5 - r Unerwater Acoutic Conference an Exhibition incoming acoutic plane wave []. The algorithm irectly etimate the three Carteian component of the incoming ignal wavevector from the etimate cro-pectra between the hyrophone element. It i robut againt hyrophone failure, an every hyrophone in the array i ue to etimate each component of the wavevector. The algorithm i applie in the frequency omain uing FFT proceing an i ue to trac broa-ban acoutic target by the Starfih noe. Fig. : Repone an reulting ratio of the two exponential igital filter output to a target pa, for one FFT bin. The input to the filter i the average pectral power for that FFT bin at each time winow... BACKGROUND We conier a planar acoutic wave incoming to a hyrophone array from a fixe irection of arrival, a hown in Fig. 4. We efine (t) a the time omain ignal preent at hyrophone # ue to the plane wave arrival, in the abence of any noie. We efine the vector ˆ a the unit vector that point in the irection of travel of the incoming wave. We efine xi(t), i =.. a the time omain ignal meaure by the hyrophone, incluing noie. The time omain ignal xi(t) meaure at the i th hyrophone, in term of (t) i given by ˆ r i xi ( t) t ni ( t). c () Here we have efine r i a the poition vector to the i th hyrophone, an c a the pee of oun in water. The term ni(t) repreent the noie preent on the i th hyrophone that i uncorrelate to the incoming ignal. The meaure hyrophone time erie xi(t) are elaye or avance verion of (t) with noie ae, the elay time epening on the irection of arrival of the incoming ignal

5 UACE5 - r Unerwater Acoutic Conference an Exhibition Fig. 4: The hyrophone array layout for which the Phae Graient algorithm wa eigne, with array iameter. If we tae the Fourier Tranform of Eqn., we obtain the frequency omain repreentation of the hyrophone output, given by i ( f ) S( f ) exp j ri Ni ( f ) () S( f ) exp j r js( f ) N ( f ). The frequency omain repreentation of the time ignal i enote by capital letter, an j i the imaginary unit. The Argument of S ( f ) i enote by S( f ). The incoming ignal f wavevector ˆ ˆ of Eqn. i a function of frequency. From Eqn. we can ee c that for a given frequency, the patial phae at any poition vector r of the incoming ignal, neglecting noie, i given by i ( r) r S( f ). (4) The phae angle S( f ) oe not epen upon the patial poition vector r, an will have zero graient. If we can etimate the graient of the patial phae (r ) we will have an etimate of the wavevector ( r), an thu have the arrival irection. i.. ORTHOGONAL ARRAY PHASE GRADIENT We etimate the graient of the patial phae given by Eqn. 4, for the cro-ipole array of Fig. 4, from the Argument of a moifie efinition of the cro-pectrum between two hyrophone. Here we write in Carteian coorinate x, y, z, an ue to enote expectation value. With the array iameter being enote by, we have for the x-axi irection a new cro-pectrum efine a new SS 4 co( y 5 ) co( z ) e j. 4 5 (5) - 7 -

6 In Eqn. 5 the phae contribution from the y an z-axi irection combine to create a realvalue amplitue factor epening on y an z becaue of the array ymmetry, leaving only half the phae ifference between hyrophone # an # a the Argument of the complex exponential. Uing the new cro-pectrum enure that all the array hyrophone are ue in the etimate of the phae ifference. We can etimate the x-component of ) (r from. new x () For the y an z-axi irection we can perform imilar operation to Eq. 5 an, obtaining the remaining component of r ) ( an the irection of arrival... TETRA ARRAY The Phae Graient algorithm ha been aapte for ue with a 4-element tetraheral array hape []. Fig. 5 how the Tetra array geometry, where =. m i the itance from the origin to each x-y plane hyrophone, an z =.8 m i the height of the top hyrophone above the x-y plane. We etimate the arrival angle for the Tetra array from the component of the wavevector, which for the Tetra geometry are given by Eq. 7. Fig. 5: The Tetra hyrophone array for which the Phae Graient algorithm wa aapte. (7) Arg y Arg z z Arg x UACE5 - r Unerwater Acoutic Conference an Exhibition

7 UACE5 - r Unerwater Acoutic Conference an Exhibition Here we efine i, i =.. a being the complex FFT pectral value in a particular frequency bin for the i th hyrophone. We etimate the bearing an elevation angle from the wavevector component in each frequency bin from y arctan, x z arctan. (8) x y 4. HISTOGRAM TRACKING When proceing the hyrophone array ata with the Phae Graient algorithm we get a irect etimate of the horizontal bearing an elevation angle of the incoming ignal at each FFT bin in the frequency ban of interet. There i no beamforming an earching over arrival angle in the algorithm, o bearing etimate may be compute quicly. Thi algorithm i well uite for bearing etimation in embee ytem [] that have limite proceing power. Thi proceing typically reult in a large number of bearing etimate being generate rapily in time. The Hitogram Tracing proceure wa evelope to ue all the available bearing information being generate from the algorithm to etimate the horizontal bearing to a broa-ban target at each proceing time winow. For the Hitogram Tracing proceure we ivie the horizontal bearing range of -8⁰ to +8⁰ into a et of equally pace bearing bin of equal bearing range. At each time winow where we etimate the cro-pectra require by the algorithm we get a bearing etimate at each FFT frequency bin. For every FFT frequency bin, we a one count to the bearing bin whoe aociate bearing range contain the bearing etimate from the algorithm. The um of all the count over all the bearing bin equal the number of FFT frequency bin at each time winow. Fig. : Horizontal bearing from the Phae Graient algorithm, with color coe bearing a a function of time an frequency

8 UACE5 - r Unerwater Acoutic Conference an Exhibition Fig. an 7 how the reult of the Phae Graient proceing an Hitogram Tracing uing ata collecte from a Tetra array. The ata wa upplie by the German Feeral Defence Organization Bunewehr, uner the Next Generation Autonomou Sytem (NGAS) NATO Joint Reearch Project (JRP) agreement. The ata wa collecte at the NGAS engineering trial in Horten, Norway. The ata et contain a broa-ban acoutic target pa. For thi ata et the hyrophone ample rate i Hz. Data wa procee from 5 Hz to Hz uing -point FFT. The frequency ban upper limit i etermine by the Tetra array ize, the Phae Graient algorithm require the maximum array element eparation to be le than half a wavelength. The horizontal bearing range wa ivie into bearing bin, each ⁰ wie. A total of 4 FFT were ue for the pectral averaging with no overlap, reulting in 9 time winow each 4 econ long. In Fig. we ee the Phae Graient reult a color coe horizontal bearing a a function of time an frequency. Fig. 7 how the Hitogram Tracing reult, with color coe hitogram count. A three-element boxcar moothing operation wa performe over the bearing imenion. The broa-ban target trac for thi ata et i well efine uing the Phae Graient algorithm an Hitogram Tracing proceure. Fig. 7: Hitogram tracing of the horizontal bearing from the Phae Graient algorithm, with color coe hitogram count. REFERENCES. Luca, C., Hear, G., an Pelava, N., The Phae Graient Bearing Etimation Algorithm, Proceeing of Meeting on Acoutic (POMA),.. Datta, U., Otne, R., an Luca, C., Bearing etimation uing mall tetraheral paive hyrophone array, Ocean Conference, Seattle Wah.,