Research Article  Open Access
Youssef Fayad, Caiyun Wang, Qunsheng Cao, Alaa ElDin Sayed Hafez, "Direction of Arrival Estimation Accuracy Enhancement via Using Displacement Invariance Technique", International Journal of Antennas and Propagation, vol. 2015, Article ID 135753, 10 pages, 2015. https://doi.org/10.1155/2015/135753
Direction of Arrival Estimation Accuracy Enhancement via Using Displacement Invariance Technique
Abstract
A new algorithm for improving Direction of Arrival Estimation (DOAE) accuracy has been carried out. Two contributions are introduced. First, Doppler frequency shift that resulted from the target movement is estimated using the displacement invariance technique (DIT). Second, the effect of Doppler frequency is modeled and incorporated into ESPRIT algorithm in order to increase the estimation accuracy. It is worth mentioning that the subspace approach has been employed into ESPRIT and DIT methods to reduce the computational complexity and the model’s nonlinearity effect. The DOAE accuracy has been verified by closedform CramérRao bound (CRB). The simulation results of the proposed algorithm are better than those of the previous estimation techniques leading to the estimator performance enhancement.
1. Introduction
Direction of arrival Estimation (DOAE) for an antenna array is an important process because DOA is the creator of the tracking gate dimensions (the azimuth and the elevation). Accurate DOAE for the moving target leads to the reduction of the angle glint error which affects the accuracy of the tracking radars.
Doppler frequency shift is a significant phenomenon induced by the target movement. Lack of attention to the influence of Doppler shift on the ESPRIT method leads to DOAE process with poor accuracy.
The ESPRIT and its extracts have been widely studied in onedimensional (1D) DOAE for uniform linear array (ULA) [1–4], nonuniform linear array (NULA) for spatial multiresolution [5–8], or temporal multiresolution [9–11] and also extended to twodimensional (2D) DOAE [12–18]. All of these ESPRIT methods have been developed to upgrade the accuracy of DOAE with low calculation costs. However, these works did not pay attention to the effect of the Doppler frequency of the moving target on the DOAE accuracy. On the other hand, some other great efforts have been done to estimate Doppler frequency via using the fast Fourier transform (FFT), but it has a high order computation time for a large number of samples [19]. Another method applying ESPRIT technique to compute Doppler shift via employing the rotational factor resulted from time delay sampling [7, 8, 20]. However, these methods require intensive matrix computations or iterative optimization techniques.
This paper presents firstly a novel method to detect the target movement via estimating its Doppler frequency by comparing the displacement invariances of transmitter and receiver arrays, which enables the measurement of the sensor array displacement invariance fluctuation that resulted from wavelength change induced by the target movement. Secondly, for target moves with uniform acceleration, Doppler frequency correction has been embedded into the ESPRIT algorithm to refine the DOA estimated value. Subspace concept has been applied to reduce the model’s nonlinearity effect and to realize parallel processing which leads to the enhancement of the estimation accuracy with low computational load [5, 6].
The paper is organized as follows. In Section 2, the twodimensions time multiresolution ESPRIT (2D TESPRIT) DOAE technique has been introduced, the time multiresolution displacement invariance technique (TDIT), and the Doppler effect of the moving target have been incorporated into the estimation algorithm. In Section 3, the simulation results are presented, and Section 4 is the conclusion.
2. Proposed Algorithm
2.1. The Measurement Model
In this model, the transmission medium is assumed to be isotropic and nondispersive, so that the radiation propagates in straight lines, and the sources are assumed to be as a far field away the array. Consequently, the radiation impinging on the array is a summation of the plane waves. The signals are assumed to be narrowband processes, and they can be considered to be sample functions of a stationary stochastic process or deterministic functions of time [3]. Considering that there are narrowband signals and the center frequency is assumed to have the same , the th signal can be written aswhere is the signal of the th emitting source at time instant , is the carrier phase angles that are assumed to be random variables, each uniformly distributed on and all statistically independent of each other, and is the incident electric field that can be written as components form. As a general expression, we omit the subscript, and thenwhere and are the horizontal and the vertical components of the field, respectively.
Define as the auxiliary polarization angle and as the polarization phase difference, and then,The incident field can be also expressed in Cartesian coordinate system
Figure 1 shows that a planar antenna array has elements indexed as , along and directions, respectively. For any pair (), its coordinate is , where , , and and are reference displacements between neighbor elements along  and axes. The array elements are oriented in plane, and the space phase factors along and directions are expressed aswhere denote the th source elevation angle and azimuth angle, respectively, and is the wavelength of the th signal. The measurement vector can be expressed aswherestands for the additive white Gaussian noise (AWGN).
From (3) and (4), we got For receiving the array, the whole receiving factors in subspaces matrix are included in ; that is,where denotes the Kronecker product. SoThe receiving model can be rewritten as,where and, .
The subspace approach not only decreases the computational load as a result of shrinking the matrix dimensions, but also reduces the influence of nonlinearity when dealing with the signal inside each small time step as a linear part. Using the TESPRIT method, the whole data is divided into snapshots at each time second with sampling frequency . Then, it picks up enough data enclosed by each snapshot with time period as short as possible. So, from (11), each receiving signal measurement value through th subspace is given asThe index runs as snapshots. Therefore, the whole spacetime steering data matrix can be expressed as where .
The dimension of [] is . For the th subspace, data matrix can be expressed as
For TESPRIT scheme, the ESPRIT algorithm is used in an appropriate picked data represented in (13) for each () subspace, shown in (14), in parallel for the same sampling accuracy; thus, reducing the calculations load and consequently saving time is achieved.
2.2. Coarse DOAE Based on TESPRIT
The ESPRIT algorithm is based on a covariance formulation [2]; that is,where is the correlation matrix of the array output signal matrix and is the autocorrelation matrix of the signal. The subscript denotes the complex conjugate transpose.
The correlation matrix of can be done for eigenvalue decomposition as follows:where the eigenvalues are ordered: The eigenvectors for larger eigenvalues span the signal subspace, and the rest smaller eigenvalues span the noise subspace which is orthogonal to the signal subspace. Therefore, there exists a unique nonsingular matrix , such that In (10), let and be the first and the last rows of , respectively; they differ by the factor along the direction. So, , where is the diagonal matrix with diagonal elements . Consequently, and will be the first and the last submatrices formed from . Then, the diagonal elements of are the eigenvalues of the unique matrix that satisfiesSimilarly, two submatrices and consist of the rows of numbered and , respectively; they differ by the space factors along the direction; . Then, , where is the diagonal matrix with diagonal elements . Consequently, forms the two submatrices, and . Then, the diagonal elements of are the eigenvalues of the unique matrix that satisfiesNote that, for multiple signals processing, the ordering of the eigenvectors of is generally permuted relative to the ordering of the eigenvectors of . So, it is essential to pair the eigenvalues of and . Thus, for all possible permutations, the eigenvalue of corresponding to , the eigenvalue of , is the element from the set which realizes the following minimization: for .
Therefore,And the coarse estimation of the arrival angles can be calculated as
2.3. Target Movement Detection
2.3.1. TDIT for Doppler Frequency Estimation
In TDIT, we apply TESPRIT method also on a signal () generated from a transmitted signal sample and has its phase; then, we compare its displacement invariance with the displacement invariance of the received signal, which enables the Doppler frequency shift estimation via measuring the wavelength change.
Due to target movement, the arrival arguments of the displacement invariances for the colocated planar array at the receiver should be calculated as where , are the fluctuated wavelength components of the received wave into antenna plane and they differ from the transmitted wavelength because of the Doppler frequency caused by the target velocity [21–23]. From Figure 2, it is obvious that the wavelengths and have expressions caused by the velocity components and :Substituting into (21), where We define as a wave velocity in free space, and for a target moves with uniform acceleration is the angle between the direction of wave propagation and the target velocity vector . Its value changes according to target maneuver. sign depends on that indicates the target direction toward or away from the antenna positiontarget maneuver, as declared in Table 1.

For the transmitted signal, where there is no Doppler effect, the departure arguments are calculated aswhere departure and arrival angles are almost equal for the colocated case. From (26), (27), and (23)–(25), and due to time multiresolution as TESPRIT used in DOAE, the Doppler shift inside each time subspace for each target () can be estimated as follows:And is the coarse estimated value of the elevation angle.
2.3.2. Target Direction Estimation
Obviously, it has been found from Figure 3 that target direction estimation depends on direction and value as indicated in Table 2.

It is clear that our main problem is to estimate sign to determine the target direction. By dividing (27) on (24), we have the following:So,and for stationary target.
2.4. Fine DOAE Based on Modified TESPRIT
The moving target echo signal is shifted by the Doppler effect. The more accurate TESPRIT algorithm should consider the effect of the Doppler frequency shift due to the target movement as mentioned in the previous section. Substitute by the Doppler frequency estimated value to refine angles estimated values as follows:So,It is obviously found from (33) that if , it will realize a stationary target indicated in (20). Figures 4(a) and 4(b) show the use of TDIT to refine DOAE process.
(a)
(b)
3. Simulation Results and Comparisons
Considering the 2DDOAE process with the additive white Gaussian noise (AWGN), the parameters are given; = 25 MHz. Assume total 25 temporal snapshots, pickup enclosed data times, , and GHz. A received signal model is simulated using 200 independent Monte Carlo simulations. In order to validate the TESPRIT method, it has been used in the planar case with number of elements, such as , with displacement values , and with initial values of and . It is found that the computational load was reduced as a result of reducing the measurement matrix dimension to () instead of and employing the subspaces parallel processing concept. Table 3 represents the computational time and complexity of the proposed TESPRIT method in terms of number of flipflops. It is obvious that the computational load has been reduced as a result of a simultaneous processing for subspaces with each having snapshots instead of processing for one space that has a large number of snapshots , () snapshots.

Figure 5 plotted the RMSEs for stationary target with different numbers of elements. The results in Figure 5 indicate that the errors are getting closer to the CRB due to the increase in the number of elements.
Figure 6 showed the errors of the TESPRIT method for different maneuver cases according to . For simplicity, we pointed to the three cases by , , and due to the different three values of , where is the maneuver at , is the maneuver at , and is the maneuver at (the highest maneuver because is close to 90°). The displayed results showed that the lowest maneuver () RMSEs curve is the closest to CRB [24, 25] on the contrary to the highest maneuver () RMSEs curve which clarify the effect of the target movement on the estimation accuracy.
Results in Figures 5 and 6 indicate that the RMSEs of the stationary target DOAE are more close to the CRB even in low SNR, but, in case of moving targets as shown in Figure 6, the RMSEs of the maneuvering target tend to be closer to the CRB with higher SNR.
The TDIT method has been used to estimate Doppler frequency via measuring the fluctuation of the sensor array displacement invariance while the method used in [20] employed the rotational factor that resulted from time delay. Figure 7 showed that the TDIT algorithm results are better than those of [20] algorithm. This improvement was due to the decrease of the errors caused by the model nonlinearity effect. Additionally, in order to introduce the improvement in the computation time compared to estimate Doppler frequency via using the FFT method, Table 4 represents a comparison between the proposed method computation time and the different FFT methods mentioned in [26].

The results in Table 4 show that the computation time is reduced to 11.09% compared to the best computation time which has been realized via using PWZFFT method [26]. This improvement has been realized because of using the temporal subspaces concept which enables parallel processing.
Note that the Doppler correction for 2DDOAE may increase the total computational time which is consumed to obtain the 2DDOAE. The computational time of the 2DDOAE with Doppler correction implementation is (10.2 msec) as indicated in Table 3. However, the computational time is still 57.63% of the conventional ESPRIT computational time.
Figure 8 showed the TESPRIT errors with embedding at ( Hz). The accuracy improvement of the 2DDOAE using proposed algorithm (TESPRIT with embedding) has been verified by comparing the resulted RMSEs with the RMSEs of the 2D Beam ESPRIT, the quaternion ESPRIT algorithms, and the ESPRITLike used in [14], [15], and [16], respectively; Figure 8 shows this comparison results.
Comparison results displayed in Figure 8 show that the proposed algorithm has a better performance, especially at a low SNR. This upgrade has been realized firstly due to the increase of DOAE accuracy when using the TESPRIT algorithm applying the subspace approach which decreases the errors caused by the model nonlinearity effect and secondly due to Doppler correction, which reduces the DOAE uncertainty associated with the effect of the target movement.
The integrated analysis for Figures 6 and 9 showed us that the proposed algorithm not only enhanced the accuracy of the estimation process, but also illuminated the effect of target maneuver on the estimation accuracy, which revealed the Doppler frequency effect on the estimation process.
Thus, the 2DDOAE accuracy has been improved by embedding the Doppler effect that resulted from target movement into TESPRIT algorithm. For more validation, this algorithm has been applied for two targets moving with different courses to highlight the good performance of this method; the first target is moving away from the antenna then turns toward it with initial value of . On the contrary, the second target moving course started with moving toward the antenna then moving away from it with initial value of and for both of them. Comparison between true and estimated angles () of the moving two targets is shown at Figures 9 and 10, respectively. Figure 11 showed comparison between true and estimated flight course plots of the two moving targets. Clearly, we can figure out that the process of integration between DIT algorithm to estimate Doppler frequency and the modified ESPRIT algorithm makes the estimated flight course more close to the real one, which means less angle glint error (almost up to 0.15°–0.3°) as shown in Figures 9 and 10. Simply, we can say that the developed algorithm achieved success in increasing the DOAE accuracy with low computational load leading to the increase of the estimator efficacy.
From Table 3, it has been found that the proposed algorithm requires flops, while the conventional ESPRIT algorithm needs flops. The developed ESPRIT algorithm requires about 57.6% of the computational time compared with that of the classical ESPRIT algorithm. Simply, it has been found that the developed ESPRIT method achieved success in increasing the DOAE accuracy with low computational load leading to the increase of the estimator efficacy.
4. Conclusions
The DOA estimator performance is affected by target maneuver and nonlinear relationships between the measured quantities and the desired target coordinates.
In order to reduce the DOAE errors, the DOA estimated values should be corrected according to the target maneuver. From our point of view, Doppler effect is the best way that can be used to describe the effect of target maneuver on DOAE process because the correction can be applied according to the fluctuation of the echo signal wavelength change from the transmitted one induced by the target maneuver. Lack of attention to the influence of Doppler shift on the phase shift measurement in DOAE process causes angle glint error and consequently leads to DOAE process with poor accuracy.
In this paper, a new algorithm for 2DDOAE, which aspires to contribute to the increase of the estimation process accuracy and the decrease of the computational load, has been firstly carried out. It has introduced time multiresolution in ESPRIT method to realize temporal subspace approach (TESPRIT) that decreases errors caused by the model’s nonlinearity effect and to reduce the computational load via processing the temporal subspaces in parallel, which leads to the shrinkage of the covariance matrix dimension. Additionally, the subspace approach has been employed into DIT method in order to form TDIT technique to reduce the Doppler frequency estimation computational costs. Secondly, in order to curb target maneuver effect on the DOAE process, an equation that describes the effect of Doppler frequency on the ESPRIT method has been derived to realize the DOAE correction.
The proposed algorithm has less computational complexity; it requires flops. Also, it reduces 88.91% of the computational time of Doppler estimation via PWZFFT method and 42.37% of the 2DDOAE total computational time with Doppler correction implementation. Additionally, the proposed algorithm realizes less angle glint error (almost up to 0.15°–0.3°). Finally, it has been found that the estimation accuracy has been increased with low computational load, and it consequently enhances the estimator performance.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
The authors would like to acknowledge that this work was supported in part by the National Natural Science Foundation of China (no. 61301211) and the Key Laboratory of Radar Imaging and Microwave Photonics (Nanjing University of Aeronautics and Astronautics), Ministry of Education, Nanjing University of Aeronautics and Astronautics, Nanjing, China.
References
 R. Roy, A. Paulraj, and T. Kailath, “ESPRITestimation of signal parameters via rotational invariance techniques,” in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '86), vol. 11, pp. 2495–2498, Tokyo, Japan, 1986. View at: Google Scholar
 R. H. Roy, ESPRITestimation of signal parameters via rotational invariance techniques [Ph.D. thesis], Stanford University, Stanford, Calif, USA, 1987.
 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
 Z. I. Khan, R. A. Awang, A. A. Sulaiman et al., “Performance analysis for estimation of signal Parameters via Rotational Invariance Technique (ESPRIT) in estimating direction of arrival for linear array antenna,” in Proceedings of the IEEE International RF and Microwave Conference (RFM '08), pp. 530–533, IEEE, Kuala Lumpur, Malaysia, December 2008. View at: Publisher Site  Google Scholar
 Y. Fayad, C. Wang, A. E.D. S. Hafez, and Q. Cao, “Direction of arrival estimation using novel ESPRIT method for localization and tracking radar systems,” in Proceedings of the 11th International Bhurban Conference on Applied Sciences and Technology (IBCAST '14), pp. 396–398, IEEE, Islamabad, Pakistan, January 2014. View at: Publisher Site  Google Scholar
 Y. Fayad, C. Wang, Q. Cao, and A. E.D. S. Hafez, “A developed ESPRIT algorithm for DOA estimation,” Frequenz, vol. 69, no. 3, pp. 263–269, 2015. View at: Google Scholar
 V. Vasylyshyn, “Direction of arrival estimation using ESPRIT with sparse arrays,” in Proceedings of the EUMA6th European Radar Conference, Rome, Italy, September 2009. View at: Google Scholar
 F. Gao and A. B. Gershman, “A generalized ESPRIT approach to directionofarrival estimation,” IEEE Signal Processing Letters, vol. 12, no. 3, pp. 254–257, 2005. View at: Publisher Site  Google Scholar
 G.M. Xu and J.G. Huang, “Multiresolution parameters estimation for polarization sensitivity array,” in Proceedings of the IEEE International Symposium on Knowledge Acquisition and Modeling Workshop (KAM '08), pp. 180–183, December 2008. View at: Publisher Site  Google Scholar
 A. N. Lemma, A.J. van der Veen, and E. F. Deprettere, “Joint anglefrequency estimation using multiresolution ESPRIT,” in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP '98), vol. 4, pp. 1957–1960, May 1998. View at: Google Scholar
 A. N. Lemma, A.J. D. van der Veen, and E. F. Deprettere, “Multiresolution ESPRIT algorithm,” IEEE Transactions on Signal Processing, vol. 47, no. 6, pp. 1722–1726, 1999. View at: Publisher Site  Google Scholar
 A. L. Swindlehurst, B. Ottersten, R. Roy, and T. Kailath, “Multiple invariance ESPRIT,” IEEE Transactions on Signal Processing, vol. 40, no. 4, pp. 867–881, 1992. View at: Publisher Site  Google Scholar
 Y. Zhang, Z. Ye, X. Xu, and J. Cui, “Estimation of twodimensional directionofarrival for uncorrelated and coherent signals with low complexity,” IET Radar, Sonar and Navigation, vol. 4, no. 4, pp. 507–519, 2010. View at: Publisher Site  Google Scholar
 C. P. Mathews, M. Haardt, and M. D. Zoltowski, “Performance analysis of closedform, ESPRIT based 2D angle estimator for rectangular arrays,” IEEE Signal Processing Letters, vol. 3, no. 4, pp. 124–126, 1996. View at: Publisher Site  Google Scholar
 Y. Li, J. Q. Zhang, B. Hu, H. Zhou, and X. Y. Zeng, “A novel 2D quaternion ESPRIT for joint DOA and polarization estimation with crosseddipole arrays,” in Proceedings of the IEEE International Conference on Industrial Technology (ICIT '13), pp. 1038–1043, Cape Town, South Africa, February 2013. View at: Publisher Site  Google Scholar
 F.J. Chen, S. Kwong, and C.W. Kok, “ESPRITlike twodimensional DOA estimation for coherent signals,” IEEE Transactions on Aerospace and Electronic Systems, vol. 46, no. 3, pp. 1477–1484, 2010. View at: Publisher Site  Google Scholar
 H. Jiang and G. Yang, “An improved algorithm of ESPRIT for signal DOA estimation,” in Proceedings of the International Conference on Industrial Control and Electronics Engineering (ICICEE '12), pp. 317–320, August 2012. View at: Publisher Site  Google Scholar
 X. Xu and Z. Ye, “Twodimensional direction of arrival estimation by exploiting the symmetric configuration of uniform rectangular array,” IET Radar, Sonar and Navigation, vol. 6, no. 5, pp. 307–313, 2012. View at: Publisher Site  Google Scholar
 V. C. Chen, The MicroDoppler Effect in Radar, Artech House, Norwood, Mass, USA, 2011.
 C. Yunhe, “Joint estimation of angle and Doppler frequency for bistatic MIMO radar,” Electronics Letters, vol. 46, no. 2, pp. 170–172, 2010. View at: Publisher Site  Google Scholar
 F. B. Berger, “The nature of Doppler velocity measurement,” IRE Transactions on Aeronautical and Navigational Electronics, vol. 4, no. 3, pp. 103–112, 1957. View at: Publisher Site  Google Scholar
 A. Nabavizadeh, M. W. Urban, R. R. Kinnick, and M. Fatemi, “Velocity measurement by vibroacoustic Doppler,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 59, no. 4, pp. 752–765, 2012. View at: Publisher Site  Google Scholar
 Alevel Physics (Advancing Physics), http://en.wikibooks.org.
 A. N. D'Andrea, U. Mengali, and R. Reggiannini, “The modified CramerRao bound and its application to synchronization problems,” IEEE Transactions on Communications, vol. 42, no. 2, pp. 1391–1399, 1994. View at: Publisher Site  Google Scholar
 S. T. Smith, “Statistical resolution limits and the complexified CramérRao bound,” IEEE Transactions on Signal Processing, vol. 53, no. 5, pp. 1597–1609, 2005. View at: Publisher Site  Google Scholar
 R. Tao, W.Q. Zhang, and E.Q. Chen, “Twostage method for joint time delay and Doppler shift estimation,” IET Radar, Sonar and Navigation, vol. 2, no. 1, pp. 71–77, 2008. View at: Publisher Site  Google Scholar
Copyright
Copyright © 2015 Youssef Fayad 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.