Sie sind auf Seite 1von 11



Root-MUSIC-Based Azimuth-Elevation Angle-of-Arrival Estimation with Uniformly Spaced but Arbitrarily Oriented Velocity Hydrophones
Kainam Thomas Wong, Member, IEEE, and Michael D. Zoltowski, Fellow, IEEE
Abstract This novel underwater acoustic azimuth-elevation source localization scheme realizes the eigenstructure-based polynomial rooting procedure for an L-shaped uniformly spaced array of diversely oriented and possibly spatially co-located velocity hydrophones and an optional pressure hydrophone. A velocity hydrophone measures a Cartesian component of the acoustic particle velocity vector of the incident waveeld. At each uniformly spaced array grid, one or more co-located and diversely oriented velocity hydrophones and/or a pressure hydrophone are placed, with the number and orientations of velocity hydrophones possibly varying from grid position to grid position in some known prearranged manner. The diverse orientation of the velocity hydrophones, however, disrupts the Vandermonde array manifold structure in each of the two uniform-linear-array legs of the L-shaped array. Nonetheless, ingenuous mathematical manipulations proposed in this paper restore the disrupted Vandermonde algebraic structure, thereby permitting once again the use of polynomial rooting to estimate the directions of arrival. A proposed pairing procedure matches each sources x-axis direction cosine estimate with its corresponding y -axis direction cosine estimate. Simulation results verify the efcacy of the proposed scheme. Index Terms Array signal processing, blind estimation, estimation of direction of arrival, sonar arrays, sonar tracking, underwater acoustic arrays.

I. INTRODUCTION A. Velocity Hydrophones HIS PROPOSED underwater acoustic source localization scheme deploys regularly spaced1 but arbitrarily oriented and possibly spatially co-located velocity hydrophones and/or a pressure hydrophone. A velocity hydrophone meaManuscript received November 24, 1997; revised April 22, 1999. Parts of this work were published in the 1996 IEEE Midwest Symposium on Circuits and Systems. This work was supported by the National Science Foundation under Grant MIPS-9320890, the Air Force Ofce of Scientic Research under Contract F49620-95-1-0367, and the U.S. Army Research Ofce Focused Research Initiative under Grant DAAH04-95-1-0246. The associate editor coordinating the review of this paper and approving it for publication was Prof. Moeness Amin. K. T. Wong is with the Chinese University of Hong Kong, Department of Electronic Engineering, Shatin, Hong Kong (e-mail: M. D. Zoltowski is with the School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907-1285 USA (e-mail: Publisher Item Identier S 1053-587X(99)09188-6. 1 While the following algorithmic development will assume that all velocity hydrophones are congured in a uniform linear geometry, the virtual array interpolation method presented in [14] accommodates any nonuniform spacing amongst the velocity hydrophones.

sures one Cartesian component of the incident underwater acoustic waveelds three-dimensional (3-D) acoustic particle velocity vector, whereas a pressure hydrophone samples the acoustic pressure as a scalar. Velocity hydrophone technology has been available for some time [1] and represents the subject of recently renewed interest [2], [28]. Many different types of velocity hydrophones are available [2], [28] and have been constructed using a variety of technologies, with designs ranging from mechanically based [5] to optically based [6] to derivative based [18]. The velocity hydrophones recognition of the vector-eld nature of the underwater acoustic waveeld distinguishes this method from more customary pressure hydrophone methods treating the underwater acoustic waveeld as merely a scalar intensity-eld. From another perspective, a velocity hydrophone array exploits the directional diversity amongst the impinging sources more completely than the discrete-time spatio-temporal FIR ltering approach of a pressure hydrophone array, where the signals directional diversity is encapsulated only in the spatial phase-factors among spatially displaced pressure hydrophones. The inherent directivity of velocity hydrophones allows simultaneous estimation of both the azimuth and the elevation arrival angles with only a one-dimensional (1-D) (linear) array of diversely oriented velocity hydrophones. DSpain et al. [8], [11] have built and sea-tested arrays of co-located groups of velocity hydrophones and pressure hydrophones. Nehorai and Paldi [18] have derived the performance bounds and identiability and uniqueness issues for vector hydrophone triads, each of which consisting of three orthogonally oriented and colocated velocity hydrophones. Nehorai and Paldi [18] have also investigated in great details a 3-D array of diversely oriented velocity hydrophones, each occupying a distinct but arbitrary location in the array. Hawkes and Nehorai derived the Cram erRao bound for such an array and studied the tradeoff between the velocity hydrophones built-in directivity and its inherently lower SNR. B. Arbitrarily Oriented Velocity Hydrophones This proposed angle-of-arrival estimation scheme allows complete arbitrary orientation for each of the velocity hydrophones in the array. This diverse orientation is assumed to be known and time-invariant but needs to be neither orthogonal nor identical at the arrays various grid points. This orientational exibility contrasts with the more rigid orthogonal

1053587X/99$10.00 1999 IEEE



orientation requirements in direction nding algorithms [8], [11], [16], [18], [20][26], [30], [31], [34] that require the arrays constituent velocity hydrophones to be grouped into vector hydrophone triads, each of which consisting of three orthogonally oriented and co-located velocity hydrophones. Instead, this present scheme allows any number of (one, two, or more2) co-located and diversely oriented velocity hydrophones and/or a pressure hydrophone at each array grid point, and the number and orientations of velocity hydrophones may vary from grid position to grid position in some known manner. The vector hydrophones of DSpain et al. [8], [11], Shchurov et al. [16], Nehorai and Paldi [18], Hawkes and Nehorai [20], [21], [26], [33], Wong and Zoltowski [22], [23], [25], [27], [30], [31], Hochwald and Nehorai [24], and Wong [34] thus represent a special case of the more general array conguration accommodated by the present algorithm. In the present scheme, the data outputs from all hydrophones in a co-located group are to be added (through certain summing weights) prior to further signal processing. Each co-located group thus functions as a single sensor, whose spatial response is shaped by the choice of summing weights. It would be possible to tailor these weights, either in a predetermined or adaptive manner, for specic signal/interference spatial beamforming scenarios, as was accomplished by Cox and Zeskind [13]. This summing of data within each co-located group also reduces computational complexity but potentially at the expense of performance. C. Root-MUSIC-Based Direction Finding (Multiple Signal Classication) [3] A highly popular eigenstructure-based suboptimal parameter estimation method has partly supplanted the optimal and more traditional maximum likelihood (ML) approach because MUSIC is computationally more efcient and offers estimation performance comparable with that of ML methods in low SNR or few snapshot situations. MUSIC is eigenstructure based in that it forms a null spectrum with the noise-subspace eigenvectors of the data correlation matrix and then searches iteratively for nulls in this spectrum. MUSIC, in this spectral form, needs to perform a computationally expensive -dimensional extrema of a highly nonlinear scalar iterative search for parameters of all incident function to estimate the sources. Whether this iteration converges to the global (rather than local) optimum and the speed of convergence depend on the availability of initial parameter estimates close to the actual global optimum. In contrast, Root-MUSIC [4], which is a variant of spectral MUSIC and applicable to a linear array of uniformly spaced identical sensors, exploits the Vandermonde structure in such arrays and substitutes a polynomial rooting procedure for spectral MUSICs more costly iterative search, as various computationally efcient schemes have been proposed to root a polynomial [19]. Root-MUSIC needs no a priori initial parameter estimates to start of the iteration. Root-MUSIC also produces more accurate asymptotic estimates than spectral MUSIC because Root-MUSIC is unaffected by radial estima2 Or

tion errors on the complex -plane [7]. There exist other noise subspace parameter estimation methods (such as the Pisarenko method, the eigenvector method, and the minimum norm method) that differ from MUSIC only in what sub-subspace of the noise-subspace they each use [29]. Polynomial forms of these other methods may be analogously constructed and have advantages similar to those offered by Root-MUSIC over spectral MUSIC. A 1-D Root-MUSIC-based underwater acoustic direction nding and ranging algorithm has previously been developed by Weiss and Friedlander [15] using an array of pressure hydrophones. This proposed method will instead adopt the Root-MUSIC algorithm to the more complicated case of an array of diversely oriented velocity hydrophones, where the velocity hydrophones diverse orientation destroys the Vandermonde structure that exists in the uniform linear array of identical omni-directional pressure hydrophones of [15]. D. Problems with Adapting Root-MUSIC to Array of Arbitrarily Oriented Velocity Hydrophones identical sensors A uniformly spaced linear array of aligned along the -axis is characterized by the simple Vandermonde-structured array manifold

. . .


affect where the elevation and azimuth angles the array manifold only through the phase factor . Each element in embodies a . This permits the use power of the phase factor nulls in the of polynomial rooting to estimate the , where the columns of span null spectrum signies the null space of the data correlation matrix, denotes the the conjugate transpose operation and Frobenius norm. However, the arbitrary orientation of the velocity hydrophones complicates this simple Vandermonde structure. The array manifold of an array of diversely oriented velocity hydrophones is inuenced by the variable sensor response from each constituent velocity hydrophone to the impinging waveeld, in addition to the spatial phase delays among the spatially displaced velocity hydrophones and/or pressure hydrophones

. . .


none, if the resulting array manifold remains unambiguous.

for , in general. That where in the array manifold of diis, the dependence on versely oriented velocity hydrophones exists through not only but as well. The resulting array manifold no longer explicitly exhibits the Vandermonde structure, on which the applicability of RootMUSIC depends. It might thus appear that the use of such diversely oriented velocity hydrophones would necessitate a



reversion to the computationally expensive iterative method of spectral MUSIC. One primary contribution of the present work lies in the ingenious mathematical manipulation allowing the use of polynomial rooting without an explicit Vandermonde structure in the array manifold. II. MATHEMATICAL DATA MODEL This paper considers scenarios involving multiple uncorrelated narrowband underwater acoustic plane-waves that, having traveled from the far-eld through a homogeneous isotropic medium, impinge upon an L-shaped array with grid positions uniformly spaced at not more than a half grid positions uniformly wavelength along the -axis and spaced along the -axis. At each individual grid position, there may lie zero or one or more co-located and diversely oriented velocity hydrophones and/or pressure hydrophones. The number and orientations of the velocity hydrophones may vary from grid position to position in some known manner. A. Data Model for an Arbitrarily Oriented Velocity-Hydrophone An underwater acoustic point source would produce the acoustic particle velocity vector [18]

thus be identied as arriving from

, where

(5) , then , and , If as expected. If the th velocity hydrophone is aligned along an arbitrary , then the velocity hydrophone scalar response equals axis3 denotes a 3 (6), shown at the bottom of the page, where 1 row vector. Without loss of generality, assume that there diversely oriented velocity hydrophones co-located exist th at the th array grid, and the scalar response from the ) velocity hydrophone is weighted by (where , whereas the optional pressure the complex coefcient . hydrophone at this th array grid is weighted by Suppose that these are then summed to produce one overall scalar response at this th array grid point

. . .

. . .

(3) denotes the th sources elevation angle where denotes measured from the vertical -axis, and the th sources azimuth angle measured from the positive -axis. The th velocity hydrophone, if aligned along the -axis, would produce the scalar response

(7) embodies a 1 4 row where are assumed to be vector. preset and known. The superscript will be added to the above notations when referring to the -axis leg of the L-shaped array, and the superscript will be attached for the -axis leg. A MUSIC-based iterative algorithm that determines these weights adaptively for a (possibly) irregularly spaced array of velocity hydrophone triads (called vector hydrophones) is presented in [31]. B. Data Model for L-Shaped ULA of Arbitrarily Oriented Velocity Hydrophones number of uncorrelated narrowband4 Assume that planewaves impinge upon the array. The data
3 To avoid dealing directly with the nonrectilinear motion measured by a vertical and a horizontal velocity hydrophone pair, the velocity hydrophones may be arranged to lie only on the (x; y; 0)-plane. The cardioid [13] is formed using two orthogonally and horizontally oriented velocity hydrophones co-located with a pressure hydrophone. 4 These incident signals are narrowband in that their bandwidths are very small compared with the inverse of the wavefronts transit time across the array. The case involving broadband signals may be reduced to a set of narrowband problems via the use of a comb of narrowband lters.

(4) and similarly if aligned along the - or the -axis. For any arbitrarily oriented set of Cartesian coordinates , there always exists relating to another Cartesian coordinate system of cardinal orientation through the following tworotation about the -axis and then 2) a step operation: 1) a rotation about the newly constructed -axis. [Note that the order of these two steps are not commutable. in itself does not uniquely relate to ]. would An incident signal from





produced by the L-shaped array5 at time

equals (8)

from the

data set

equals the signals wavelength, denotes their where represents the th signals uniformlypropagation speed, and refer, distributed random carrier phase, and the data sub-vector respectively, to the symbolizes produced by the -axis and -axis legs, and vector of complex-valued zero-mean additive a white noise (AWN). Furthermore (9) ..

where denotes the data samples collected at the -axis leg, and refers to the data set produced by the -axis leg of the L-shaped array. III. 2-D DIRECTION FINDING WITH ARBITRARILY ORIENTED VELOCITY HYDROPHONES VIA POLYNOMIAL ROOTING A. Eigen-Decomposition of the Data Correlation Matrix In eigenstructure (subspace) direction-nding methods (such as MUSIC and Root-MUSIC), the overall sample correlation matrix (which embodies a maximum likelihood (ML) estimate of the true sample correlation matrix if the additive noise is Gaussian) for the entire L-shaped array is decomposed into a -dimensional signal subspace and -dimensional noise subspace. MUSIC derives a a null spectrum (or, alternately, a signal-subspace spectrum) parameterized by the direction cosines and then identies the deepest nulls in this null spectrum (or peaks for the signalsubspace spectrum) to estimate the arrival angles. To estimate parameters of all incident sources, MUSIC would the extrema of the null perform an -dimensional search for (or signal-subspace spectrum). symbolize the matrix composed of Let the eigenvectors corresponding to the largest eigenvalues , and let denote the of matrix composed of the remaining eigenvectors of (14)

. . .

. . .

. . .



. . .

. . .

. . .


denotes a diagonal matrix whose diagonal where symbolizes an entries equal the largest eigenvalues, and diagonal matrix whose diagonal smallest eigenvalues. entries contains the thus embodies an matrix whose -dimensional noise subspace columns span the . of Eigendecomposition performed for data collected at only the -axis leg would produce (15) denotes a diagonal matrix whose diagonal where largest eigenvalues of , and entries equal the symbolizes an diagonal matrix whose smallest eigenvalues. diagonal entries contains the Similar entities may be dened for the -axis leg. With denoting a matrix of all zero elements and signifying identity matrix a (16) (17)

(12) (13) represents the inter-element spacing for the -axis where refers to the -axis leg velocity/pressure hydrophones leg, and . preset summing weights, and similarly for snapshots taken at , With a total of the present direction nding problem6 is to determine all
5 Hua et al. [9] has determined that the L-shaped array geometry offers signicantly better estimation performance than the cross-shaped geometry. 6 Although the proposed algorithm is developed below in the batch processing mode, real-time adaptive implementations of this present algorithm may be readily realized for nonstationary environments using fast recursive eigendecomposition updating methods, such as that in [17].



Fig. 1. Estimation standard deviation versus SNR, with direction-cosine estimates paired using a priori source azimuth/elevation matching information.

but the columns of (18) . Thus, only span only a subspace of the column space of one eigendecomposition needs to be performed for the two polynomial rooting procedures (one for the -axis leg and one for the -axis leg). B. Application of MUSIC to a Velocity-Hydrophone ULA Invoking the MUSIC algorithm [3], the arrival angles can be estimated as (19) where the superscript denotes an estimate, the superscript denotes the conjugate transposition, and denotes the Frobenius norm. The minimization in (19) involves computationally 2-D minima of a highly expensive iterative searches for nonlinear scalar cost function. This iterative optimization could converge to local minima instead of the global minimum.

Convergence speed and convergence to the global minimum depend on availability of a priori coarse estimates of to start off the iterative search. All these problems are avoided in Root-MUSIC by applying polynomial rooting. C. Adaptation of Root-MUSIC to a Diversely Oriented Velocity Hydrophone ULA To recast (19) to a mathematical form suitable for polynomial rooting (20) This and the following subsections will focus on the leg along the -axis. The mathematics for the -axis leg are similar; simply substitute the subscripts and superscripts by corresponding ones signifying , as shown bearing in (21) and (22), shown at the bottom of the page, recalling is determined by . Note that is in that only through . quadratic form and depends on The bracketed term above embodies a real nonnegative scalar, is orthogonal which will be minimized to zero when





Fig. 2. Estimation bias versus SNR with direction-cosine estimates paired using a priori source azimuth/elevation matching information.

to all columns of . Equivalently, this zero-value minimum is singular will be reached in the asymptotic case iff and the nonzero vector lies in the null space of . (and only those ) rendering the scalar Hence, any cost function to zero makes and singular. It is true that for any producing det and , no corresponding may exist to give det and . Nonetheless, for [i.e., corresponds to an actual source], there do exist , by denition, to make and . Moreover in asymptotic cases, for all , must lie on the unit circle of the complex plane. This is not generally true that happens to render both for any and to singularity because the roots of and det generally are det arbitrarily located all over the complex plane. Probability equals zero that some , satisfying and det , also happens to det lie on the unit circle. In other words, represents sufcient, although not necessary, condition for . There might exist a lying in the null space of a nonzero and in the null space ; however, such a pair of spurious roots of a nonzero , as compared to roots corresponding to the incident sources, is relatively unlikely to lie on or close to the unit circle of the -plane, especially under high SNR situations.

In other words, may be found by such that det , and identifying the value of plays no role in this derivation of . That is, come to affect the minimization in (19) and (21) only through the phase factor . This new insight is original to this paper and is fundamental to the successful adoption of Root-MUSIC [4] to the present diversely oriented velocity hydrophone azimuth-elevation directional-nding problem. The following mathematical manipulation aims to express explicitly as a polynomial in in order det . to apply polynomial rooting to derive 4 matrix Toward this end, note that each element in the 4 embodies a polynomial in . The th element of , may be expressed as

(23) where th column of the 4 matrix dened in (10); diagonal matrix formed with as its main diagonal; column vector formed from the main diag. onal of The identity in (23) coalesces all the constant coefcients into constant matrix sandwiched an . The by a pair of Vandermonde-structured vectors term in the polynomial expression of coefcient for the equals the sum of all elements on the th det



Fig. 3.

Estimation standard deviation versus SNR using the proposed pairing method.

diagonal of . The indexing convention here th diagonal equals the loweris that the matrix , the left element of the th diagonal refers to the main diagonal, and the th diagonal equals the upper-right element. as the vector whose Further dening th entry equals the th diagonal sum of the matrix

the elements



(27) (24) Successive co-factor expansion would express as a polynomial equation. For example, when only velocity hydrophones are deployed, (now a 3 3 matrix) may be found the determinant of via the co-factor expansion

(28) signies an th-order polynomial where in , and the symbol denotes the convolution operation, whereby the coefcients of the products of two polynomials may be computed. The diversely oriented velocity hydrophone 2-D direction nding problem has thus been successfully reduced to a polynomial rooting problem; hence, computationally expensive multidimensional searches are avoided. The above development for the -axis leg should be similarly applied to the -axis leg of the -shaped array. Note that have not been restricted to zero, nor is elevation angles or presumed as long as any particular choice of and both remain full-rank.

(25) represents the th element in , where signies the submatrix formed by deleting the th and . may be related to row and th column of



Fig. 4.

Estimation bias versus SNR using the proposed pairing method.

D. Pairing of the -Axis and -Axis Direction-Cosine Estimates Application of the above polynomial rooting procedure to , the -axis leg produces the estimates and similarly, application of the above polynomial rooting procedure to the -axis leg produces the estimates . However, these two sets of estimates are not yet paired. A simple pairing procedure is outlined below. and lie in The key insight is that 1) both with the vector space spanned by and similarly lying in the vector space spanned for any and by ; 2) however, and cannot both lie in . the vector space spanned by That is, there does not exist any set of complex coefcients , not all of which equal zero, that satisfy both equalities

fact that (29) and (30) cannot both be true implies that (31) where denotes an all-zero matrix of the appropriate size. In other words, only the correctly paired estimates would produce a set of steering vectors orthogonal to the noise subspace. Under more realistic conditions with noise and nite number of data samples, the above zero identity becomes only approximate. Based on the foregoing analysis, the estimates and are to be paired to minimize (32) . Other pairover all possible permutations of ing approaches are available in the literature [32]. E. Expressions for the Angle Estimates The azimuth and elevation angle estimates, respectively, equal if



if Because columns of and because columns of

, for

and the span the -dimensional vector space is orthogonal to all under noiseless or asymptotic conditions, the

if (33)



Fig. 5. Fraction of runs in Figs. 1 and 2 resulting in direction-cosine estimates mismatches.

where number

denotes the principal argument of the complex in the range of to rad. IV. SIMULATIONS

Simulation results in Figs. 15 demonstrate the proposed algorithms efcacy. An L-shaped array is deployed, with 12 uniformly half-wavelength spaced but diversely oriented velocity hydrophones on the -axis leg and another 12 similarly on the -axis leg. The rst, sixth, seventh, and tenth velocity hydrophones on each leg are nominally aligned along the -axis; the second, fth, eighth, and 12th velocity hydrophones on each leg are nominally aligned along the -axis; the third, fourth, ninth, and 11th velocity hydrophones on each leg are nominally aligned along the -axis. To investigate how velocity hydrophone orientation affects direction nding performance, another velocity hydrophone array is simulated analogously to the one above but with the following [for a denition, please refer to (5)] for both legs:

This arbitrary velocity hydrophone orientation is assumed known and time-invariant, and the correctional procedure in (6) is added to the signal processing algorithm to account for

these nonorthogonal orientations. Two closely spaced equalpower uncorrelated narrowband sources with random initial phases emit from and . That is, the rst source has and , and the second source has and . Eighty discrete-time samples are used in each Monte Carlo simulation run. Five hundred independent Monte Carlo runs contribute to each data point on each gure. Polynomial rooting is performed by the MATLAB root subroutine. Additive white Gaussian noise is present, and the SNR gures are calculated relative to each signals power. The RMS estimation standard deviation plotted in these gures is dened as the root-mean-square of the standard deviations of , and , with the RMS estimation bias similarly dened. Figs. 1 and 2 plot the direction-cosine estimation error versus SNR for the proposed diversely oriented velocity-axis hydrophone Root-MUSIC algorithm when the direction-cosines are forced to be paired perfectly with their -axis counterparts using a priori source directional information. (Figs. 35 will show simulation data with pairing performed by the procedure in Section III-D) Fig. 1 plots the RMS standard deviation and the Cram erRao bound, whereas Fig. 2 shows the corresponding estimation bias. Estimation errors depend on the particular choice of numerical procedure to root the polynomials; thus, lower errors may be realized with the use of more sophisticated rooting numerical algorithms. The RMS standard deviation, being roughly an order of magnitude larger than the RMS bias, represents the dominant



form of estimation error. Note that arbitrary nonorthogonal orientations of the velocity-hydrophones affect estimation performance only minimally when pairing correctness is not , the two a factor. Because incident sources are resolved with high probability when their RMS standard deviation and bias fall below 0.02. This means that the proposed scheme succeeds for all SNR above about 2 dB. Figs. 35 plot simulation results versus SNR when the direction cosines are matched with no a priori information using the pairing algorithm in Section III-D. For the particular arbitrary nonorthogonal orientation of the velocity hydrophone in these three gures, it appears to have insignicant inuence on the proposed schemes performance for the aforementioned SNR range above 2 dB; however, a comprehensive analysis of the effects of nonorthogonal orientation is being pursued by the rst author. Fig. 5 plots the percentage of the 500 Monte Carlo runs resulting in direction cosine mismatch. There exist no mismatch for SNR above 5 dB. The direction cosine pairing step is therein revealed as the major source of estimation error for moderate to low SNR. Alternate direction cosine pairing approaches [32] might offer better performance, and the rst author is presently investigating that. V. CONCLUSION This proposed method realizes polynomial rooting for diversely oriented velocity hydrophones (in units of ones, twos, or more) and/or pressure hydrophones uniformly spaced in an L-shaped array conguration, despite the non-Vandermonde array manifold structure. [In fact, one such linear array sufces for 2-D (i.e., azimuth and elevation) direction nding due to the velocity hydrophones inherent directivity.] This method exploits the vector nature of the impinging underwater acoustic waveeld but does not require the co-located orthogonally oriented triad subarray geometry in some prior closed-form algorithms that also exploit the waveelds vector nature. This proposed algorithm may also be used with nonuniform nonlinear arrays via the virtual array interpolation method [14]. Although the proposed algorithm is developed below in the batch processing mode, real-time adaptive implementations of this present algorithm may be readily realized for nonstationary environments using fast recursive eigendecomposition updating methods, such as that in [17]. Broadband signals may also be handled by predecomposing the baseband data by a bank of comb lters, thereby transforming the broadband problem into a set of narrowband problems. An analogous algorithm for the electromagnetic medium has been developed by the present authors using diversely polarized antennas [27]. Although developed for use with Root-MUSIC, the present scheme may be straightforwardly modied for the polynomial rooting forms of the Pisarenko method, the eigenvector method, or the minimum norm method [28]. REFERENCES
[1] C. B. Leslie, J. M. Kendall, and J. L. Jones, Hydrophone for measuring particle velocity, J. Acoust. Soc. Amer., vol. 28, no. 4, pp. 711715, July 1956.

[2] C. L. LeBlanc, Handbook of hydrophone element design technology, Naval Underwater Syst. Cent., Tech. Rep. 5813, 1978. [3] R. O. Schmidt, Multiple emitter location and signal parameter estimation, in Proc. RADC Spectral Estimation Workshop, Grifss AFBS, NY, 1979. [4] A. J. Barabell, Improving the resolution performance of eigenstucturebased direction-nding algorithms, in Proc. ICASSP, 1983, pp. 336339. [5] M. A. Josserand and C. Maerfeld, PVF2 velocity hydrophone, J. Acoust. Soc. Amer., vol. 78, no. 3, pp. 860867, 1985. [6] V. J. Hughes, J. G. Boulton, J. M. Coles, T. R Empson, and N. J. Kerry, Why an optically-based hydrophone works better, Sensor Rev., vol. 7, no. 3, pp. 123126, July 1987. [7] B. D. Rao and K. V. S. Hari, Performance analysis of Root-MUSIC, IEEE Trans. Acoust., Speech, Signal Processing, vol. 37, pp. 19391949, Dec. 1989. [8] G. L. DSpain, W. S. Hodgkiss, and G. L. Edmonds, The simultaneous measurement of infrasonic acoustic particle velocity and acoustic pressure in the ocean by freely drifting swallow oats, IEEE J. Ocean. Eng., vol. 16, pp. 195207, 1991. [9] Y. Hua, T. K. Sarkar, and D. D. Weiner, An L-shaped array for estimating 2-D direction of wave arrival, IEEE Trans. Antennas Propagat., vol. 39, pp. 143146, 1991. [10] J. C. Nickles, G. Edmonds, and R. Harriss, A vertical array of directional acoustic sensors, in Proc. IEEE Oceans, 1992, pp. 340345. [11] G. L. DSpain, W. S. Hodgkiss, and G. L. Edmonds, Initial analysis of the data from the vertical DIFAR array, in Proc. IEEE Oceans, 1992, pp. 346351. [12] A. J. van der Veen, P. B. Ober, and E. F. Deprettere, Azimuth and elevation computation in high resolution DOA estimation, IEEE Trans. Signal Processing, vol. 40, pp. 18281833, July 1992. [13] H. Cox and R. M. Zeskind, Adaptive cardioid processing, in Proc. Asilomar Conf., 1992, pp. 10581061. [14] B. Friedlander, Root-MUSIC algorithm for direction nding with interpolated arrays, Signal Process., vol. 30, no. 1, pp. 1529, Jan. 1993. [15] A. J. Weiss and B. Friedlander, Range and bearing estimation using polynomial rooting, IEEE J. Oceanic Eng., vol. 18, pp. 130137, Apr. 1993. [16] V. A. Shchurov, V. I. Ilyichev, and V. P. Kuleshov, Ambient noise energy motion in the near-surface layer in ocean wave-guide, J. Phys., vol. 4, no. 5, pt. 2, pp. 12731276, May 1994. [17] M. Moonen, F. J. Vanpoucke, and E. F. Deprettere, Parallel and adaptive high-resolution direction nding, IEEE Trans. Signal Processing, vol. 42, pp. 24392448, Sept. 1994. [18] A. Nehorai and E. Paldi, Acoustic vector sensor array processing, IEEE Trans. Signal Processing, vol. 42, pp. 24812491, Sept. 1994. [19] M. Lang and B.-C. Frenzel, Polynomial root nding, IEEE Signal Processing Lett., vol. 1, pp. 141143, Oct. 1994. [20] M. Hawkes and A. Nehorai, Bearing estimation with acoustic vectorsensor arrays, in Proc. Acoust. Velocity Focused Workshop, 1995, pp. 345348. [21] M. Hawkes and A. Nehorai, Hull-mounted acoustic vector-sensor arrays, in Proc. 29th Asilomar Conf., 1995. [22] K. T. Wong and M. D. Zoltowski, Orthogonal-velocity-hydrophone ESPRIT for sonar source localization, in Proc. MTS/IEEE Oceans Conf., 1996, vol. 3, pp. 13071312. , ESPRIT-based extended-aperture source localization using [23] velocity-hydrophones, in Proc. MTS/IEEE Oceans Conf., 1996, vol. 3, pp. 14271432; Extended-aperture underwater acoustic multisource azimuth/elevation direction-nding using uniformly but sparsely spaced vector hydrophones, IEEE J. Oceanic Eng., vol. 22, pp. 659672, Oct. 1997. [24] B. Hochwald and A. Nehorai, Identiability in array processing models with vector-sensor applications, IEEE Trans. Signal Processing, vol. 44, pp. 8395, Jan. 1996. [25] K. T. Wong and M. D. Zoltowski, Source localization by 2-D rootMUSIC with scalar triads of velocity-hydrophones, in Proc. 39th IEEE Midwest Symp. Circuits Syst., 1996, vol. 2, pp. 677680. [26] M. Hawkes and A. Nehorai, Surface mounted acoustic vector-sensor array beamforming, in Proc. IEEE ICASSP, 1996, pp. 31703173. [27] K. T. Wong and M. D. Zoltowski, Diversely polarized root-MUSIC for azimuth-elevation angle-of-arrival estimation, in Proc. IEEE Antennas Propagat. Soc. Int. Symp., 1996, pp. 13521355. [28] M. J. Berliner and J. F. Lindberg, Acoustic Particle Velocity Sensors: Design, Performance and Applications. Woodbury, NY: Amer. Inst. Phys., 1996. [29] M. H. Hayes, Statistical Digital Signal Processing and Modeling. New York: Wiley, 1996, pp. 459469.



[30] K. T. Wong and M. D. Zoltowski, Closed-form underwater acoustic direction-nding with arbitrarily spaced vector hydrophones at unknown locations, in Proc. IEEE Int. Symp. Circuits Syst., vol. 4, pp. 25572560, 1997; IEEE J. Ocean. Eng., vol. 22, pp. 566575, July 1997. [31] K. T. Wong and M. D. Zoltowski, Self-initiating velocity-eld beamspace MUSIC for underwater acoustic direction-nding using irregularly spaced vector-hydrophones, in Proc. IEEE Int. Symp. Circuits Syst., 1997, vol. 4, pp. 25532556. [32] Y. Hua and K. Abed-Meraim, Techniques of eigenvalues estimation and association, Digital Signal Process., pp. 253259, 1997. [33] Hawkes and A. Nehorai, Acoustic vector-sensor beamforming and capon direction estimation, IEEE Trans. Signal Processing, vol. 46, pp. 22912304, Sept. 1998. [34] K. T. Wong, Adaptive geolocation and blind beamforming for wideband fast frequency-hop signals of unknown hop sequences and unknown arrival angles, in Proc. IEEE Wireless Commun. Networking Conf., 1999.

Kainam Thomas Wong (M97) received the B.S.E. (Chem.) degree from the University of California, Los Angeles, in 1985, the B.S.E.E. degree from the University of Colorado, Boulder, in 1987, the M.S.E.E. from Michigan State University, East Lansing, in 1990, and the Ph.D. degree in electrical engineering from Purdue University, West Lafayette, IN, in 1996. He worked as a Manufacturing Engineer at the General Motors Technical Center, Warren, MI, from 1990 to 1991. As a Senior Professional Staff Member at the Johns Hopkins University Applied Physics Laboratory, Laurel, MD, from 1996 to 1998, he developed next-generation radar and satellite communications technology. He was an Assistant Professor at Nanyang Technological University, Singapore, during part of 1998. He joined the Department of Electronic Engineering, Chinese University of Hong Kong, Hong Kong, in Fall 1998. He is a contributing author of about 70 articles for the telecommunications section of the inaugural edition of the CRC Dictionary of Pure and Applied Physics and the CRC Comprehensive Dictionary of Physics. His current research interests are signal processing for communications and sensor-array signal processing. Dr. Wong serves on the Technical Program Committees of the 1999 IEEE Wireless Communications and Networking Conference (WCNC99), the 1999 IEEE International Workshop on Intelligent Signal Processing and Communication Systems (ISPACS99), the EuroComm 2000 Conference, and the 2000 Spring IEEE Vehicular Technology Conference (VTC00 Spring). He also serves on the Organizing Committees of the 2000 IEEE International Symposium on Circuits and Systems (ISCAS00) and the Symposium 2000 on Adaptive Signal Processing, Communications, and Control (AS-SPCC). He also serves regularly as a peer reviewer for the IEEE TRANSACTIONS ON SIGNAL PROCESSING, IEEE COMMUNICATIONS LETTERS, IEEE COMMUNICATIONS MAGAZINE, IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS, IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, and Proceedings of the IEE: Radar, Sonar, and Navigation. He is a member of the Eta Kappa Nu and Phi Beta Delta honor societies.

Michael D. Zoltowski (F99) was born in Philadelphia, PA, on August 12, 1960. His received both the B.S. and M.S. degrees in electrical engineering (with highest honors) from Drexel University, Philadelphia, in 1983 and the Ph.D. degree in systems engineering from the University of Pennsylvania, Philadelphia, in 1986. From 1982 to 1986, he was an Ofce of Naval Research Graduate Fellow. In Fall 1986, he joined the faculty of Purdue University, West Lafayette, IN, where he currently holds the position of Professor of electrical and computer engineering. During 1987, he held a position of Summer Faculty Research Fellow at the Naval Ocean Systems Center, San Diego, CA. His present research interests include space-time adaptive processing and blind antenna array beamforming for all areas of mobile and wireless communications, radar, and GPS. He is a contributing author to Adaptive Radar Detection and Estimation (New York: Wiley, 1991), Advances in Spectrum Analysis and Array Processing, Vol. III (Englewood Clioffs, NJ: Prentice-Hall, 1994), and CRC Handbook on Digital Signal Processing (Boca Raton, FL: CRC, 1996). Dr. Zoltowski was the recipient of the IEEE Signal Processing Societys 1991 Paper Award in the Statistical Signal and Array Processing Technical Area. He was the recipient of the IEEE Outstanding Branch Counselor Award for 1990 to 1991 and the recipient of The Fred Ellersick MILCOM Award for Best Paper in the Unclassied Technical Program at the IEEE Military Communications (MILCOM98) Conference. He served as an Associate Editor for the IEEE TRANSACTIONS ON SIGNAL PROCESSING and is currently an Associate Editor for the IEEE COMMUNICATIONS LETTERS. Within the IEEE Signal Processing Society, he has been a member of the Technical Committee for the Statistical Signal and Array Processing Area and is currently a member of the Technical Committee for Communications and the Standing Committee on Education. In addition, he is currently a Member-at-Large of the Board of Governors and Secretary of the IEEE Signal Processing Society.