Estimation of DOA for Noncircular Signals via Vandermonde Constrained Parallel Factor Analysis

Lin, Heyun; Yuan, Chaowei; Du, Jianhe; Hu, Zhongwei

doi:https://doi.org/10.1155/2018/4612583

International Journal of Antennas and Propagation

On this page

Abstract Introduction System Model Conclusions Conflicts of Interest Acknowledgments References Copyright Related Articles

Research Article | Open Access

Volume 2018 | Article ID 4612583 | https://doi.org/10.1155/2018/4612583

Estimation of DOA for Noncircular Signals via Vandermonde Constrained Parallel Factor Analysis

Heyun Lin,¹Chaowei Yuan,¹Jianhe Du,²and Zhongwei Hu¹

Academic Editor: Pierfrancesco Lombardo

Received08 Jun 2017

Revised21 Aug 2017

Accepted12 Oct 2017

Published15 Jan 2018

Abstract

We provide a complete study on the direction-of-arrival (DOA) estimation of noncircular (NC) signals for uniform linear array (ULA) via Vandermonde constrained parallel factor (PARAFAC) analysis. By exploiting the noncircular property of the signals, we first construct an extended matrix which contains two times sampling number of the received signal. Then, taking the Vandermonde structure of the array manifold matrix into account, the extended matrix can be turned into a tensor model which admits the Vandermonde constrained PARAFAC decomposition. Based on this tensor model, an efficient linear algebra algorithm is applied to obtain the DOA estimation via utilizing the rotational invariance between two submatrices. Compared with some existing algorithms, the proposed method has a better DOA estimation performance. Meanwhile, the proposed method consistently has a higher estimation accuracy and a much lower computational complexity than the trilinear alternating least square- (TALS-) based PARAFAC algorithm. Finally, numerical examples are conducted to demonstrate the effectiveness of the proposed approach in terms of estimation accuracy and computational complexity.

1. Introduction

Direction-of-arrival (DOA) estimation of signals impinging on an array of sensors is an important and fundamental issue in array signal processing due to its wide applications in wireless communications, geophysics, radar, and so on [1–3]. In this context, many DOA estimators have been developed to solve this problem, such as propagator method (PM) [4], maximum likelihood (ML) methods [5, 6], tensor-based method [7], estimation of signal parameters via rotational invariance technique (ESPRIT) algorithm [8], and multiple signal classification (MUSIC) algorithm [9].

Unfortunately, all of the above works [4–9] ignore the characteristics of the incident signals. It has been shown that the accuracy of DOA estimation can be enhanced by taking advantage of the noncircular (NC) property of the impinging signals [10]. Examples of the noncircular signals include binary phase-shift keying (BPSK) and amplitude-modulated (AM) signals which are widely applied in wireless telecommunication systems. While the property of NC signals is utilized, the array aperture is virtually doubled in [10], which yields a better DOA estimation performance than that in [9]. Besides taking the noncircularity of signals into account, the maximum number of sources can potentially exceed the number of array elements. Consequently, solid researches on NC signals for DOA estimation have appeared in many literatures [11–14]. NC ESPRIT algorithm and NC Unitary ESPRIT algorithm for DOA estimation were presented in [11, 12], respectively. The work [13] developed a low-complexity noncircular rational invariance propagator method (NC-RI-PM) for angle estimation, which is used for uniform linear array (ULA). However, the performance of the NC-RI-PM became worse rapidly in the case of low signal-to-noise ratio (SNR). The authors in [14] utilized the parallel factor (PARAFAC) analysis [15] to acquire the two-dimensional (2D) angle estimation of NC signals for uniform rectangular array (URA). The main drawback in [14] is that it has large computational load and is sensitive to the iterative initial value.

In this paper, we present a Vandermonde constrained PARAFAC method to improve the DOA estimation accuracy by taking advantages of the property of NC signals and the array manifold matrix with Vandermonde structure. In order to construct a PARAFAC data model, we first build an extended matrix by concatenating the received data and its conjugated component. Then, the extended data matrix can be formulated as a PARAFAC model with Vandermonde constrained factor matrix. Finally, a closed-form solution can be utilized to acquire the angle information by taking the Vandermonde structure into account.

The contributions of this paper can be summarized as follows: (1) We combine the property of NC signals and the array manifold matrix with Vandermonde structure, based on which a Vandermonde constrained PARAFAC method is proposed to acquire the DOA of NC signals for uniform linear array. (2) Compared with trilinear alternating least square- (TALS-) based PARAFAC method, the proposed method provides a better angle estimation as well as a lower computational complexity. (3) The proposed approach also has a higher estimation accuracy than the ESPRIT method [8], NC ESPRIT method [11], and NC-RI-PM method [13].

The rest of this paper is structured as follows. In Section 2, we introduce the system model for ULA. Section 3 presents the Vandermonde constrained PARAFAC method for DOA estimation of noncircular signals. The complexity analysis and Cramér-Rao Bound (CRB) of NC signal DOA estimation are given in Section 4. Numerical results are provided in Section 5. Finally, Section 6 concludes the paper.

Notations: scalars, vectors, matrices, and tensors are denoted by lowercase letters, lowercase boldface letters, uppercase boldface letters, and calligraphic letters, respectively. The operators and denote the conjugate, transpose, conjugate transpose, pseudoinverse, and Frobenius norm, respectively. stands for a identity matrix. is the diagonal matrix with diagonal elements given by . ⊙, ∘, and ⊕ represent the Khatri-Rao product, outer product, and Hadamard product, respectively. is the rank of a matrix .

2. System Model

Let us consider a uniform linear array consisting of omnidirectional sensors. The distance between two adjacent sensors is . As shown in Figure 1, the first sensor is regarded as the reference point for the array, assuming that there are uncorrelated narrowband far-field signals impinging on this array from distinct directions We also assume that the number of sources is known a priori or has been estimated by the methods shown in [16, 17]. Under such a scenario, the received signal vector can be modeled as where is the array manifold matrix with size of , is the wavelength; is the signal vector at snapshot . stands for the additive Gaussian noise vector with zero-mean and covariance . From (2), we know that is a Vandermonde matrix with distinct nonzero generators In this paper, the strictly second-order noncircular signals [18] are considered. Thus, we have and

It is assumed that the DOAs of the sources are constant in snapshots. We collect snapshots corresponding to ; the received signal matrix is given by where stands for the source matrix and contains the samples of the additive noise. Due to the NC property, we can decompose the source matrix as [18], where is a real-valued matrix, is a diagonal matrix, and is referred to as the phase shift of the source. Then (3) can be rewritten as

3. Estimation of DOA via Vandermonde Constrained PARAFAC Decomposition

3.1. PARAFAC Model

Since the noncircular signals are considered, we construct the extended data matrix as follows [12, 14]: where is the exchange matrix with ones on its antidiagonal and zeros elsewhere. is the extended noise matrix. and is a matrix. is the diagonal matrix with diagonal elements given by the row of . According to Harshman [15], the signal part of (5) can be seen as a matrix representation of a third-order PARAFAC model whose element is given by where and represents the typical element of the tensor . denotes the element of , and similarly for the others. Using the reshaping operations, we can obtain the other two equivalent matrix representations of as follows:

Till now, we successfully link the expanded data matrix with a tensor model which satisfies the PARAFAC decomposition. According to (5), the PARAFAC decomposition can be accomplished by solving the following unconstrained optimization problem. where , and represent the columns of the matrices , , and , respectively. This is most often done by means of the TALS method. As its programming simplicity, the TALS algorithm has been successfully employed to estimate the parameters of several tensor models [7, 19, 20]. The basic idea of the TALS algorithm for fitting the proposed PARAFAC model is stated as follows:

With the consideration of (5), while and are fixed, we update using the following least squares (LS) fitting: where stands for the estimation of . In the same way, while and are fixed, can be updated by the following LS fitting: where stands for the estimation of . Similarly, we obtain an updated via LS fitting as where stands for the estimation of . Repeat updating process of the three matrices until convergence, and then the estimation matrices and can be obtained. It is well known that the TALS algorithm is sensitive to the initial value as well as the array manifold, which may cause the TALS algorithm converging slowly. In the next subsection, a more efficient algorithm is applied to obtain the estimation of DOA for noncircular signals.

3.2. DOA Estimation with Vandermonde Constrained PARAFAC Method

From the previous analysis, we know that the matrix is a Vandermonde matrix with distinct nonzero generators Therefore, by taking the Vandermonde factor into account, the model can be seen as a tensor which admits the PARAFAC decomposition with Vandermonde constraint. And then, a more efficient algorithm is applied to fitting the PARAFAC model with Vandermonde constraint.

According to (6), we know that If , we can conclude that the matrix does not have full-column rank, which implies that is less than . Then, the rank of is also less than . Consequently, the spatial smoothing technique can be used to handle with this problem [21]. The detailed process of spatial smoothing is presented as follows:

Denote as the subarray of such that , where is the smoothing factor. The integers and must be chosen such that Taking the advantage of the Vandermonde structure of , we have where Then, denote we can obtain where denotes the first rows of . Note that where denotes the first rows of .

Therefore, (14) can be rewritten as

In this way, the matrix representation of the tensor , which involves a rank-deficient matrix (while ), is replaced by a matrix representation which only involves two full-column rank matrices and The matrix can be seen as a matrix representation of a constrained PARAFAC decomposition model [22–24] with three factors and When the matrix is obtained, a more efficient algorithm can be used to fit the Vandermonde constrained PARAFAC model [23, 24].

While the matrices and have full-column rank, we perform the compact singular value decomposition (SVD) of , which is given by where is a diagonal matrix consisting of the largest nonzero singular values. and are matrices composed of left and right singular vectors associated with the largest nonzero singular values, respectively. Since the factor matrices and have full-column rank, we can conclude that the subspace spanned by the column vectors of is the same as the subspace spanned by the column vectors of the matrix , so we have

Therefore, there exists a unique nonsingular matrix such that

Using the inherent Vandermonde structure of , we define the following two submatrices: where and , and the relationship between , , and is that

Therefore, there exists a diagonal matrix such that where with generators , .

According to (19), (20), (21) and (22), we can obtain that

From (23), we have where Assuming that the matrix has full-column rank, we can conclude that the matrices and have full-column rank and can be rewritten as

Therefore, the generators , , can be obtained from the eigenvalue decomposition (EVD) of . Then, the DOA of the source can be recovered via where stands for the estimation of and angle(·) is used for obtaining the phase. The proposed Vandermonde constrained PARAFAC algorithm for NC signals DOA estimation is presented in Algorithm 1.

Step 1. Construct the tensor model with the matrix representation by exploiting the property of NC signals.
Step 2. Choose the pair () subject to and construct the matrix representation via spatial smoothing.
Step 3. Perform the compact SVD of .
Step 4. Construct , .
Step 5. Perform the EVD of , with generators , .
Step 6. Compute , .

Remark 1. The proposed method in this paper is developed for noncircular signals. Therefore, we assume that the circular signals are not present in this paper. When the mixed noncircular and circular signals are considered, the DOAs can be estimated by the algorithms reported in [25, 26].

3.3. Identifiabilitity Analysis

The most remarkable characteristic of the PARAFAC decomposition is its essential uniqueness property, which makes PARAFAC decomposition a key tool for signal separation. In this subsection, we discuss the identifiabilitity condition for the proposed PARAFAC method.

Theorem 1 [7, 15, 27, 28]. Let be a tensor with matrix representation , where and , if then the three factor matrices , , and are unique up to permutation and scaling of columns, meaning that any other triple () is related to () via where and are the [27, 28] of the matrices , , and , respectively. is a permutation matrix. , , are diagonal scaling matrices satisfying

If the three factor matrices have full rank, then condition (26) becomes

In this paper, we assume , then the identifiable condition (28) becomes . Therefore, the maximum number of sources that can be handled by the PARAFAC algorithm is no more than .

When is a tensor that admits the Vandermonde constrained PARAFAC decomposition, we have the following uniqueness result.

Theorem 2 [23, 24]. Let be a tensor with matrix representation , where is a Vandermonde matrix with distinct generators, and . Consider the matrix representation with . If for some , then , and the Vandermonde constrained PARAFAC decomposition of is unique. Generically, condition (29) is satisfied if and only if

When we assume and , the condition (30) becomes . Therefore, the maximum number of sources that can be identified by the Vandermonde constrained PARAFAC algorithm is . If , the Vandermonde constrained PARAFAC method can identify more sources than the PARAFAC method under the same conditions. Notably, the scaling does not involve the Vandermonde factor matrix in a PARAFAC model with Vandermode constraint. This means that the estimated matrices and are related to , , and via where are diagonal scaling matrices satisfying

Remark 2. The proposed method can be seen as a generalized ESPRIT method. In line with the experiments for ESPRIT in [29, 30], we can choose the pair () such that the dimensions of the matrices and are close, with the inequality satisfied. According to [29, 30], we know that such a direct method will yield a very good estimation result. Therefore, in order to reduce the computational complexity, is fixed to 3 and in this paper. Although it may not yield the best estimation result with the parameters being set as and , a good estimation result can be still achieved. Note that simulations in Section 5 show that the DOA performance of such a direct approach still outperforms than other methods.

4. Performance Analysis

4.1. Computational Complexity Analysis

In this subsection, we discuss the computational complexity of the proposed Vandermonde constrained PARAFAC method. The main computational tasks of the Vandermonde constrained PARAFAC method are computing the compact SVD of which requires [31], pseudoinverse operation of matrix which requires and the EVD of which requires Therefore, the main computational complexity of the proposed Vandermonde constrained PARAFAC method is The TALS-based PARAFAC method is also illustrated for comparison. The computational cost of TALS-based PARAFAC method is [32], where is the number of iterations. We summarize the main computational complexity of the proposed Vandermonde constrained PARAFAC method and the TALS-based PARAFAC method in Table 1.

4.2. Cramér-Rao Bound (CRB)

From [33], we can derive the Cramér-Rao Bound (CRB) of noncircular signal DOA estimation for ULA as follows: where is the noise power, is the row of , and with being the column of .

Remark 3. Since the proposed Vandermonde constrained PARAFAC solution can be seen as a generalized ESPRIT method, hence it could be better optimized by weighting subspace fitting methods such as [34, 35].

5. Simulation Examples

The simulations are presented to evaluate the DOA estimation performance of the proposed method in this section. The root mean square error (RMSE) is used to evaluate the DOA estimation performance, which is defined as where is the estimated values of in the trial, . is the number of sources.

Figure 2 shows the DOA estimation result of the proposed Vandermonde constrained PARAFAC method. In this simulation, we set , , . 30 Monte Carlo trials are carried out. Moreover, we assume that there exist noncircular signals locating at angles and the noncircular phases are set as It can be seen from Figure 2 that our Vandermonde constrained PARAFAC method has a good DOA estimation performance even at a low SNR level. Therefore, the effectiveness of the proposed method is verified.

Figure 3 depicts the DOA estimation result of the proposed Vandermonde constrained PARAFAC method for close sources. In this simulation, we assume , , , , and The number of the Monte Carlo trials is set to 30. From Figure 3, it can be observed that the proposed Vandermonde constrained PARAFAC method can separate close sources effectively. Moreover, Figures 2 and 3 present that the DOA estimation performance could be improved with SNR increasing.

Figure 4 presents the RMSE of the proposed Vandermonde constrained PARAFAC method versus SNR. We assume that , , and . The TALS-based PARAFAC algorithm, ESPRIT algorithm in [8], NC-ESPRIT algorithm in [11], NC-RI-PM algorithm in [13], and the CRB are also illustrated for comparison in Figure 4. From Figure 4, we can see that the Vandermonde constrained PARAFAC method can achieve a better estimation performance than ESPRIT, NC-ESPRIT, and NC-RI-PM. Notably, by taking the Vandermonde structure into account, our Vandermonde constrained PARAFAC method outperforms the TALS-based PARAFAC approach.

The bias and variance of the estimator versus SNR are shown in Figures 5 and 6, respectively. We consider there exist noncircular signals locating at the angles of and the noncircular phases are set as The number of antennas is set as , the snapshot is fixed to 200, while SNR is varied from 0 to 30 dB. As expected, we can see from Figures 5 and 6 that the bias and variance of the estimator decrease as SNR increases, respectively.

Table 2 shows the comparison of average run time (s) between Vandermonde constrained PARAFAC method and TALS-based PARAFAC method. In this simulation, SNR is fixed at 15 dB, , , while is varied from 50 to 300. We can see that average run time of each method grows as the number of snapshots increases. It can be also observed that the Vandermonde constrained PARAFAC method is more computationally efficient than the TALS-based PARAFAC method.

At last, the impact of the number of sources is investigated. Figure 7 presents the RMSE performance of our Vandermonde constrained PARAFAC method versus SNR with different , when and . It is observed from Figure 7 that the estimation accuracy decreases as the source number increases.

6. Conclusions

The problem of DOA estimation of noncircular signals for uniform linear array was considered in this paper. Due to the property of noncircular signals, we constructed an extended matrix by concatenating the received data and its conjugated component. And then, a Vandermonde constrained PARAFAC model was derived from the extended matrix. Finally, estimation of the angle information was obtained via eigenvalue decomposition of two submatrices. Compared with some existing methods, the proposed method consistently provided a better DOA estimation performance. Simulation result illustrates the effectiveness of our method in terms of estimation accuracy and computational complexity. Utilizing the weighting subspace fitting to optimize the proposed method will be carried out in our future work. The case of mixed noncircular and circular signals will be also a topic for future research.

Conflicts of Interest

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

This work was supported by the National High Technology Research and Development Program of China (863 Program, nos. 2015AA01A705 and 2014AA01A701), the National Natural Science Foundation of China (no. 61601414), and the Science and Technology Project of State Grid Corporation of China (no. SGLNDK00KJJS1700200).

References

A. B. Gershman, M. Rbsamen, and M. Pesavento, “One- and two-dimensional direction-of-arrival estimation: an overview of search-free techniques,” Signal Processing, vol. 90, no. 5, pp. 1338–1349, 2010.
View at: Publisher Site | Google Scholar
B. R. Jackson, S. Rajan, B. J. Liao, and S. Wang, “Direction of arrival estimation using directive antennas in uniform circular arrays,” IEEE Transactions on Antennas and Propagation, vol. 63, no. 2, pp. 736–747, 2015.
View at: Publisher Site | Google Scholar
F. Gao and A. B. Gershman, “A generalized ESPRIT approach to direction-of-arrival estimation,” IEEE Signal Processing Letters, vol. 12, no. 3, pp. 254–257, 2005.
View at: Publisher Site | Google Scholar
S. Marcos, A. Marsal, and M. Benidir, “The propagator method for source bearing estimation,” Signal Processing, vol. 42, no. 2, pp. 121–138, 1995.
View at: Publisher Site | Google Scholar
W. H. Fang, Y. C. Lee, and Y. T. Chen, “Maximum likelihood 2-D DOA estimation via signal separation and importance sampling,” IEEE Antennas and Wireless Propagation Letters, vol. 15, pp. 746–749, 2016.
View at: Publisher Site | Google Scholar
M. P. Clark and L. L. Scharf, “Two-dimensional modal analysis based on maximum likelihood,” IEEE Transactions on Signal Processing, vol. 42, no. 6, pp. 1443–1452, 1994.
View at: Publisher Site | Google Scholar
N. D. Sidiropoulos, R. Bro, and G. B. Giannakis, “Parallel factor analysis in sensor array processing,” IEEE Transactions on Signal Processing, vol. 48, no. 8, pp. 2377–2388, 2000.
View at: Publisher Site | Google Scholar
R. Roy and T. Kailath, “ESPRIT-estimation of signal parameters via rotational invariance techniques,” IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 37, no. 7, pp. 984–995, 1989.
View at: Publisher Site | Google Scholar
R. O. Schmidt, “Multiple emitter location and signal parameter estimation,” IEEE Transactions on Antennas and Propagation, vol. 34, no. 3, pp. 276–280, 1986.
View at: Publisher Site | Google Scholar
H. Abeida and J. P. Delmas, “MUSIC-like estimation of direction of arrival for noncircular sources,” IEEE Transactions on Signal Processing, vol. 54, no. 7, pp. 2678–2690, 2006.
View at: Publisher Site | Google Scholar
A. Zoubir, P. Charg, and Y. Wang, “Non circular sources localization with ESPRIT,” in Proceedings of the 6th European Conference on Wireless Technology (ECWT 03), Munich, Germany, October 2003.
View at: Google Scholar
M. Haardt and F. Romer, “Enhancements of unitary ESPRIT for non-circular sources,” in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2004), pp. II-101–II-104, Montreal, Canada, August 2004.
View at: Publisher Site | Google Scholar
X. Chen, C. Wang, and X. Zhang, “DOA and noncircular phase estimation of noncircular signal via an improved noncircular rotational invariance propagator method,” Mathematical Problems in Engineering, vol. 2015, Article ID 235173, 12 pages, 2015.
View at: Publisher Site | Google Scholar
L. Zhang, W. Lv, X. Zhang, and S. Li, “2D-DOA estimation of noncircular signals for uniform rectangular array via NC-PARAFAC method,” International Journal of Electronics, vol. 103, no. 11, pp. 1839–1856, 2016.
View at: Publisher Site | Google Scholar
R. Harshman, “Foundations of the PARAFAC procedure: models and conditions for an “explanatory” multimodal factor analysis,” UCLA Working Papers in Phonetics, vol. 16, pp. 1–84, 1970.
View at: Google Scholar
L. Huang, T. Long, E. Mao, and H. C. So, “MMSE-based MDL method for robust estimation of number of sources without eigendecomposition,” IEEE Transactions on Signal Processing, vol. 57, no. 10, pp. 4135–4142, 2009.
View at: Publisher Site | Google Scholar
H. T. Wu, J. F. Yang, and F. K. Chen, “Source number estimator using Gerschgorin disks,” in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 94), pp. IV/261–IV/264, Adelaide, Australia, April, 1994.
View at: Publisher Site | Google Scholar
B. Picinbono, “On circularity,” IEEE Transactions on Signal Processing, vol. 42, no. 12, pp. 3473–3482, 1994.
View at: Publisher Site | Google Scholar
J. Du, C. Yuan, P. Tian, and H. Lin, “Channel estimation for multi-input multi-output relay systems using the PARATUCK2 tensor model,” IET Communications, vol. 10, no. 9, pp. 995–1002, 2016.
View at: Publisher Site | Google Scholar
D. Nion and L. De Lathauwer, “A block component model-based blind DS-CDMA receiver,” IEEE Transactions on Signal Processing, vol. 56, no. 11, pp. 5567–5579, 2008.
View at: Publisher Site | Google Scholar
N. D. Sidiropoulos and X. Liu, “Identifiability results for blind beamforming in incoherent multipath with small delay spread,” IEEE Transactions on Signal Processing, vol. 49, no. 1, pp. 228–236, 2001.
View at: Publisher Site | Google Scholar
G. Favier and A. de Almeida, “Overview of constrained PARAFAC models,” EURASIP Journal on Advances in Signal Processing, vol. 2014, p. 40, 2014.
View at: Publisher Site | Google Scholar
M. Sørensen and L. De Lathauwer, “Blind signal separation via tensor decomposition with Vandermonde factor: canonical polyadic decomposition,” IEEE Transactions on Signal Processing, vol. 61, no. 22, pp. 5507–5519, 2013.
View at: Publisher Site | Google Scholar
M. Sørensen and L. De Lathauwer, “Tensor decompositions with Vandermonde factor and applications in signal processing,” in Proceedings of conference record of the forty sixth Asilomar conference on signals, systems and computers (Asilomar), pp. 890–894, Pacific Grove, CA, USA, November 2012.
View at: Publisher Site | Google Scholar
F. Gao, A. Nallanathan, and Y. Wang, “Improved MUSIC under the coexistence of both circular and noncircular sources,” IEEE Transactions on Signal Processing, vol. 56, no. 7, pp. 3033–3038, 2008.
View at: Publisher Site | Google Scholar
J. Steinwandt, F. Roemer, and M. Haardt, “ESPRIT-type algorithms for a received mixture of circular and strictly non-circular signals,” in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2015), pp. 2809–2813, Brisbane, Australia, April 2015.
View at: Publisher Site | Google Scholar
J. B. Kruskal, “Three-way arrays: rank and uniqueness of trilinear decompositions, with application to arithmetic complexity and statistics,” Linear Algebra and its Applications, vol. 18, no. 2, pp. 95–138, 1977.
View at: Publisher Site | Google Scholar
N. D. Sidiropoulos and R. Bro, “On the uniqueness of multilinear decomposition of N-way arrays,” Journal of Chemometrics, vol. 14, no. 3, pp. 229–239, 2000.
View at: Publisher Site | Google Scholar
S. Vanhuffel, H. Chen, C. Decanniere, and P. Vanhecke, “Algorithm for time-domain NMR data fitting based on total least squares,” Journal of Magnetic Resonance, Series A, vol. 110, no. 2, pp. 228–237, 1994.
View at: Publisher Site | Google Scholar
J. M. Papy, L. De Lathauwer, and S. Van Huffel, “Exponential data fitting using multilinear algebra: the single-channel and multi-channel case,” Numerical Linear Algebra with Applications, vol. 12, no. 8, pp. 809–826, 2005.
View at: Publisher Site | Google Scholar
G. H. Golub and C. F. van Loan, Matrix Computations, Johns Hopkins University Press, Baltimore, 1996.
Y. Rong, S. A. Vorobyov, A. B. Gershman, and N. D. Sidiropoulos, “Blind spatial signature estimation via time-varying user power loading and parallel factor analysis,” IEEE Transactions on Signal Processing, vol. 53, no. 5, pp. 1697–1710, 2005.
View at: Publisher Site | Google Scholar
P. Stoica and R. L. Moses, Spectral Analysis of Signals, Pearson Prentice Hall, Upper Saddle River, NJ, USA, 2005.
M. Viberg and B. Ottersten, “Sensor array processing based on subspace fitting,” IEEE Transactions on Signal Processing, vol. 39, no. 5, pp. 1110–1121, 1991.
View at: Publisher Site | Google Scholar
A. L. Swindlehurst and T. Kailath, “Azimuth/elevation direction finding using regular array geometries,” IEEE Transactions on Aerospace and Electronic Systems, vol. 29, no. 1, pp. 145–156, 1993.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2018 Heyun Lin et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies