Abstract
Electromagnetic vector sensors (EMVS) have attracted growing attention in recent years. However, the mutual coupling effects in practical EMVS arrays may seriously degrade the parameter estimation performance. In order to solve this problem, a novel array configuration consisting of two parallel sparse dipole arrays is proposed. Based on the spatially rotational invariance property between the two parallel arrays and the interdipole spacing inside each array, highly accurate but ambiguous direction-cosine estimates, coarse direction-of-arrival (DOA) estimates, and polarization parameter estimation can be obtained jointly. The coarse DOA estimates are then employed to disambiguate the phase ambiguities in the fine estimates. Compared with collocated EMVS, the proposed array overcomes the mutual coupling problem. Moreover, the DOA estimation accuracy is promoted due to the sparse array aperture extension. Simulation results demonstrate the effectiveness of the proposed algorithm.
1. Introduction
Source localization and polarization estimation using electromagnetic vector sensors (EMVS) have received considerable attention over the past decades. Generally, an electromagnetic vector sensor consists of three orthogonal electrically short dipoles and three orthogonal magnetically small loops [1–4]. All of the six components are spatially collocated in a point-like geometry, and no spatial phase-factor exists among them. Many advanced algorithms have been developed for direction finding, polarization estimation, and tracking schemes [5–9]. Particularly, a lot of eigenstructure-based direction finding and polarization estimation schemes have been presented in recent years. In [10], a temporal-invariance version estimation of signal parameters via rotational invariance technique (ESPRIT) [11] has been applied for a single collocated six-component vector sensor to estimate the arrival angles and polarization states of multiple pure tones. In [12], a spatial-invariance version of ESPRIT has been used for some collocated EVMS’s deployed in a sparse rectangular grid without incurring cyclic ambiguity in the final estimates of the sources’ direction parameters. In [13, 14], a spatial-invariance version of ESPRIT has been proposed for arbitrarily spaced EVMS’s, whose locations are unknown. In [15], based on a spatially collocated EMVS, the author uses vector cross-product to perform direction finding for multiple noncooperative wideband fast frequency-hop signals.
In the above-mentioned literature, the algorithms mostly presume that all the six components of EMVS are utilized. However, the responses of electric field and magnetic field may be different between each other in practical systems. Therefore, in [16] Wong proposed to deploy only one triad to mitigate this problem. In [17], the authors utilized a triad of collocated and perpendicular electrically short dipoles (or of electrically small loops) and derived the Cramér-Rao lower bound accounting for the possibility of failure in any individual dipoles (or loops). Recently, aiming at the low radiation efficiency of such short dipoles or small loops, the authors developed a triad of electrically large circular loops for the estimation of an incident emitter’s direction-of-arrival (DOA) or polarization [18]. Similarly, in [19], the authors proposed a triad of electrically “long” dipoles for direction finding and polarization estimation. However, both of them adopted a spatially collocated array structure. Therefore, the mutual coupling effect of these collocated triad structure is strong, and the developed algorithms require the prior knowledge of polarization information, which is often invalid in practical applications.
Note that the mutual coupling effect across the collocated dipoles/loops is neglected in the above-mentioned literature. However, the mutual coupling may degrade the parameters estimation performance and increase the hardware cost of the vector sensors. Therefore, a vector cross-product direction finding algorithm is proposed in [20, 21] for six orthogonally oriented but spatially noncollocating dipoles/loops. However, the algorithm developed there requires a temporal invariance of the incident sources, which is often invalid in practical. Moreover, the array structure proposed is spatially displaced along one direction; thus it cannot only provide high accuracy and unambiguous DOA estimates. In [22], the author developed a sparse array composed of three noncollocating dipole (or loop) triads. Based on the ESPRIT principle, the high accuracy and unambiguous estimates of DOA and polarization parameters are obtained. However, the direction-cosines estimates along and axes are obtained from two independent eigendecomposition operations; thus they require additional parameter matching procedure. In [23], the author advanced spatially spread dipole/loop quads/quints for direction finding and polarization estimation. However, it assumes that the source signals are pure tones which may be invalid in practical scenarios. In [24], the authors proposed a spatially spread quint of only dipoles or loops, for DOA and polarization estimation. Based on the centrosymmetric array, the authors develop a DOA and polarization estimator via vector-cross-product. However, it requires a 2D search process for conducting coarse 2D-DOA estimation, which is computationally inefficient. In [25–27], the authors utilize different types of spatially separated dipoles/loops to constructs a sparse planar array. This kind of structure can provide high accuracy DOA estimates of multiple sources. However, the direction-cosines estimates along and axes are obtained from two independent eigendecomposition operations; thus they require additional parameter matching procedure.
Aiming at addressing the above-mentioned problems, we propose in this paper a novel array configuration which consists of two parallel sparse dipole arrays. Based on the spatially rotational invariance property between the two parallel arrays and the interdipole spacing inside each array, highly accurate but ambiguous direction-cosine estimates, coarse DOA estimates, and polarization parameters estimation can be obtained jointly. The coarse DOA estimates are then employed to yield the final DOA estimates with high accuracy.
Compared with the traditional methods, the proposed scheme enjoys the following advantages: (1) The interdipole spacing and intersubarray spacing can be extended much greater than half a wavelength, which could reduce the mutual coupling effect. (2) High accuracy 2D DOA estimates can be obtained for the azimuth angle and the elevation angle concurrently, due to the extended aperture in both directions. (3) Only dipoles are utilized in the array, which avoids the responses mismatch of dipoles and loops. (4) The parameters are derived in closed-form and are therefore computationally efficient. (5) The DOA and polarization parameters are paired automatically. (6) The source signals are not required to be pure tones.
The rest of the paper is organized as follows: Section 2 introduces the signal model of the parallel spatially spread dipole arrays. The proposed algorithm is described in Section 3. In Section 4, simulations are conducted to validate the performance of our method. We conclude this paper in Section 5.
Throughout the paper, the complex conjugate, transpose, Hermitian transpose, and pseudoinverse are denoted by , , , and , respectively. represents a m×m identity matrix, and is a m×m zero matrix.
2. Signal Model
In order to obtain high accuracy elevation and azimuth DOA estimates, we construct an array composed of two parallel dipole arrays as shown in Figure 1. Note that each dipole array consists of four dipoles that are uniformly spread along the -axis, with the interelements spacing being . At the origin of a Cartesian coordinate system, a short dipole antenna element denoted as is placed along the -axis. The other three of the dipoles (denoted as , , and ) are orthogonally oriented along each of the -axis, -axis, and -axis. In addition, the displacement along the -axis between the two parallel arrays is .
It is well known that, for a completely polarized unit power transverse electromagnetic wave travelling through a homogeneous isotropic medium, it can be characterized by a 3×1 electric-field vector in the Cartesian coordinate system [1]:where is the signal’s elevation angle measured from the positive -axis, denotes the azimuth angle measured from the positive -axis, stands for the auxiliary polarization angle, and refers to the polarization phase difference.
Considering the spatial phase shift introduced by displacement along the -axis, the array manifold of the four spatially spread dipoles on the left-hand side can be expressed asHerein, denotes the Hadamard product, and represents the direction-cosine of the -axis. Note that the interdipole displacement along the -axis could be utilized to extend the array aperture and also reduce the mutual coupling.
To obtain the aperture expansion along the -axis, another dipole-quad array with the same structure is added in parallel, and the displacement along the -axis can be much greater than a half wavelength. Considering the spatial phase-factor introduced by the parallel array structure, the array manifold can be expressed asand represents the direction-cosine of the y-axis.
Assume that there are a total of signals impinging onto the proposed array from the far-field. Thus, the received signal data can be expressed as follows:where B is the array manifold, s() is the signal vector, and n() denotes the additive Gaussian white noise.
Given a total number of snapshots generated at the distinct instants : , the main problem to be addressed in this paper is to obtain high accuracy 2D DOA estimates from these snapshots, i.e., estimate the azimuth angle and elevation angle parameters of the impinging signals. Moreover, for beamforming purposes, it may be also useful to estimate the corresponding polarization parameters .
Unlike the traditional scalar array antenna model, the proposed parallel sparse dipole arrays in Figure 1 are polarized, sparsely spread, and composed of noncollocating orthogonally oriented dipoles. Therefore, a novel algorithm is investigated in the next section for the polarized antenna arrays.
3. Proposed Algorithm
In order to generate unambiguous 2D DOA and polarization parameters with high accuracy for the proposed polarized array, our algorithm creatively uses the ESPRIT technique through the following stages: (1) based on the spatial rotational invariance property between the two parallel arrays, ESPRIT is utilized to generate the eigenvalues and the eigenvectors; (2) the eigenvalues offer the estimates of high accuracy but ambiguous -axis direction-cosines; (3) the relationships between the elements of each source’s steering vector are employed to obtain ambiguous high accuracy but ambiguous -axis direction-cosine estimates, coarse 2D DOA estimates, and polarization parameter estimation; (4) the coarse DOA estimation results are employed to select a set of highly accurate and unambiguous direction-cosine estimates from the set containing the cyclically ambiguous estimates.
3.1. High Accuracy and Cyclically Ambiguous y-Axis Direction-Cosine Estimation
It can be seen from (3) that the first and the last four components of the steering vector satisfy the spatial rotational invariance, which corresponds to the two parallel dipole arrays, respectively. For the th source, this rotational invariance property can be expressed aswhere and are the selection matrices.
Considering the far-field radiating sources received by the array, the spatial rotational invariance property can be generalized as the following matrix form:where diag[v] represents a diagonal matrix with its main diagonal elements being those of v.
It is noticeable that contains the -axis direction-cosine information. Therefore, to obtain the estimation of , we first generate the covariance matrix of the received data x(): where denotes expectation. By taking the eigenvalue decomposition (EVD) of , we have the following equation: where is a diagonal matrix containing largest eigenvalues and is a diagonal matrix associated with the remaining 8- ones. , which spans the signal subspace of , is composed of the vectors corresponding to the largest eigenvalues of . , which spans the noise subspace of , is composed of the remaining 8- vectors corresponding to the 8- small eigenvalues of .
Based on the subspace principle, the 8× signal subspace matrix can be expressed aswhere T denotes an unknown K×K nonsingular matrix to be determined. Therefore, according to (9) and (13), the following relationships can be derived:
Applying the least squares (LS) algorithm [11], we havewhere denotes the Moore-Penrose inverse. The eigenvalue decomposition of the matrix yields the estimates of T and , respectively. And the main diagonal elements of are composed of the eigenvalues of . Therefore, the direction-cosine along the -axis can be obtained bywhere the operator returns the phase of the complex number in its argument.
Note that when the displacement parameter between the two parallel dipole arrays is much greater than half a wavelength, these direction-cosines estimation values in (17) will be of high accuracy, due to the extend aperture size along the -axis. However, they will suffer from cyclical ambiguity problem at the same time.
3.2. High Accuracy and Cyclically Ambiguous x-Axis Direction-Cosine Estimation
Since the nonsingular matrix T has been determined in the above subsection, the manifold matrix of the left-hand side subarray can be estimated from the following expression:where represents the estimation of and c is an unknown complex constant.
According to the array structure in Figure 1 and (2), the array manifold of the left-hand side subarray can be estimated as
Based on the estimated steering vector , the direction-cosine of -axis can be calculatedwhere the operator returns the phase of the complex number in its argument and denotes the th entry of .
Similarly, when the interdipole spacing parameter is much greater than half a wavelength, these direction-cosines estimation values in (20) will be of high accuracy, due to the extend aperture size along the -axis. However, they will suffer from cyclical ambiguity problem at the same time.
3.3. Coarse and Unambiguous 2D-DOA Estimation
With the estimation of the steering vector , we can construct the following 2×1 vectorTherefore, the coarse 2D DOA estimates can be derived from the following closed-form formulas [23]:
Moreover, the polarization parameters are also readily to be obtained in the closed-form
Accordingly, the coarse direction-cosine estimation for the and axes can be calculated, respectively, as follows:
These 2D coarse direction-cosine estimates can then be applied to disambiguate the phase ambiguities in the fine estimates, which are induced from the interdipole and intersubarray spatial displacement.
3.4. Disambiguation Algorithm
Note that when and are much greater than half a wavelength, the estimated values and in (20) and (17) will be cyclically ambiguous, which is demonstrated as follows:Herein, the ⌈·⌉ operator returns the smallest integer greater than its argument, and the ⌊·⌋ operator returns the largest integer less than its argument. As such, this process generates a set of cyclically ambiguous estimations.
Then, the coarse 2D DOA estimation results are applied as reference values for choosing a set of highly accurate and unambiguous direction-cosine estimates from the set of ambiguous estimates. This is conducted by first defining the following highly accurate and unambiguous direction-cosines for the and axes, respectively. Note that the direction-cosines estimates in [22] require two independent eigendecomposition operations; thus they require additional parameter matching procedure. On the contrary, the direction-cosines estimates in the proposed scheme are obtained from only one eigendecomposition, and therefore, the fine and coarse direction-cosine estimates for the and axes are automatically paired without any extra operation.
These terms are then applied to obtain highly accurate and unambiguous 2D-DOA parameter estimations as follows:
Consequently, the polarization parameter estimations can be updated by substituting and into (24) and (25). Note that the DOA and polarization estimates are automatically paired without any extra operation.
3.5. Implementation of the Algorithm
Note that the exact covariance matrices and subspaces are utilized in the previous subsections, but the theoretical covariance matrix like in (11) is unavailable due to the limited number of snapshots. In practical, it can be estimated as
Consequently, the procedure of our proposed algorithm is summarized as follows.
Step 1. Estimate the covariance matrix using (36).
Step 2. Eigendecompose to generate its signal subspace as in (13).
Step 3. Partition into two 4× matrices and via (14) and (15).
Step 4. Eigendecompose to obtain its eigenvalues and eigenvectors.
Step 5. The direction-cosine along the y-axis can be estimated by (17).
Step 6. Estimate the manifold matrix of the left-hand side subarray according to (18).
Step 7. Obtain the direction-cosines along the x-axis via (20).
Step 8. Construct the 2×1 vector according to (21), and the coarse DOA estimates can be obtained from (22), (23).
Step 9. The coarse direction-cosine estimates are employed to disambiguate the ambiguities in the fine direction-cosine estimates.
Step 10. The unambiguous direction-cosine estimates are employed for 2D DOA estimation according to (35) and (36), and the polarization parameters are estimated according to (25) and (26).
4. Simulation Results
In this section, several numerical simulations are conducted to validate the performance of the proposed algorithm. The interdipole distance along the -axis is set as 10λ, and the intersubarray distance along the -axis is set as 10λ. It is assumed that the number of sources is known a priori in the following simulations. And the signal-to-noise ratio (SNR) of the -th source is defined as , where is the noise power and is the power of the -th source signal.
In the first experiment, we verify the effectiveness of the proposed algorithm. Consider that there are three uncorrelated equipowered sources impinging on the array. The DOA and polarization parameters are set as , , and , respectively. The number of snapshots is 100 and the SNR equals 20 dB. Figure 2(a) demonstrates the results of the azimuth and elevation angle estimation from 200 independent realizations. Figure 2(b) illustrates the results of the auxiliary polarization angle and polarization phase difference estimation from 200 independent realizations. According to Figure 2, it is evident that the estimated DOA and polarization parameters are correctly paired and almost equal to the true value. Thus, the proposed algorithm can provide high accuracy direction find and polarization estimation results. To further verify the parameter estimation performance versus the SNR and the snapshot number, two more experiments are carried out. The performance is measured in terms of root mean-square error (RMSE) for each individual parameter, which is defined aswhere stands for the parameter, denotes the estimation of in the -th trial, and is the number of Monte Carlo trials that are conducted in the following simulations, which is 1000.
(a) Azimuth and elevation angle estimation
(b) Polarization parameter estimation
In the second experiment, the estimation accuracy of the proposed algorithm versus SNR is explored. Consider that two uncorrelated equipowered sources are impinging on the array. DOA and polarization parameters are and , respectively. The SNR varies from 0 dB to 60 dB in steps of 5 dB. The number of snapshots is fixed at 100. Figure 3 illustrates the RMSEs of DOA and polarization estimation as a function of SNR. Note that, for the azimuth and elevation angle estimates, both the coarse estimation and the fine estimation results are presented for comparison. From Figures 3(a) and 3(b), it is obvious that the coarse DOA estimates have higher RMSEs since the extended array aperture is not fully utilized. In addition, as is expected, the fine estimates have superior estimation accuracy than the coarse ones after disambiguation. Moreover, the RMSEs of both the DOA and polarization estimates decrease as the SNR increases.
(a) RMSEs of azimuth angle estimates versus SNR
(b) RMSEs of elevation angle estimates versus SNR
(c) RMSEs of polarization parameter estimates versus SNR
In the third experiment, we investigate the RMSEs of the proposed algorithm with the variation of the number of snapshots. The parameter settings are same as that of the second experiment except that the SNR is set equal to 20 dB, and the number of snapshots varies from 10 to 1000. From Figures 4(a) and 4(b), we can draw a similar conclusion that the fine DOA estimates have superior estimation accuracy than the coarse ones after disambiguation. Moreover, it can be observed that the parameter estimation performance of the two sources improves as the snapshot number increases. This is because the fact that a larger sample support will produce better estimate of the covariance matrix for stationary data.
(a) RMSEs of azimuth angle estimates versus snapshot number
(b) RMSEs of elevation angle estimates versus snapshot number
(c) RMSEs of polarization parameter estimates versus snapshot number
5. Conclusions
This paper developed a joint DOA and polarization estimation algorithm for the proposed two parallel spatially spread dipole arrays. The ESPRIT algorithm is employed for conducting highly accurate and unambiguous coarse parameter estimation based on the spatial rotation invariance of the array structure. The spatially noncollocated array structure is conducive toward reducing the mutual coupling effect and the hardware cost of the electromagnetic vector-sensors array compared with spatially collocated array structures. Moreover, the interelement spacing could be much greater than half a wavelength, which promotes the 2D DOA estimation accuracy.
Data Availability
No data were used to support this study.
Conflicts of Interest
The authors declare no conflicts of interest.
Acknowledgments
This research was funded by the National Natural Science Foundation of China under Grant no. 61601372, Shenzhen Science and Technology Innovation Committee of Basic Research Projects under Grants nos. JCYJ20170306154016149 and JCYJ20170815154325384, China Postdoctoral Science Foundation under Grant no. 2017M613200, Natural Science Basic Research Plan in Shaanxi Province of China under Grant no. 2017JQ6068, and Shanghai Aerospace Science and Technology Innovation Fund under Grant no. sast2017-077.