A Quadrilinear Decomposition Method for Direction Estimation in Bistatic MIMO Radar

Abstract

We investigate into the problem of joint direction-of-departure (DOD) and direction-ofarrival (DOA) estimation in a multiple-input multiple-output radar, and a novel covariance tensor-based quadrilinear decomposition algorithm is derived in this paper. By taking into account the multidimensional structure of the matched array data, a fourth-order covariance tensor is formulated, which links the problem of joint DOD and DOA estimation to a quadrilinear decomposition model. A quadrilinear alternating least squares (QALSs) technique is applied to estimate the loading matrices, and thereafter automatically paired DODs and DOAs are obtained via the LSs fitting strategy. The proposed QALS algorithm can be regarded as an alternative to the direct parallel factor (PARAFAC) algorithm, which is more flexible than the latter since it can be easily expanded to scenario with spatially colored noise. Moreover, the proposed algorithm has much lower computational complexity than PARAFAC, especially in the presence of large snapshot. Numerical simulations verify the effectiveness of the proposed algorithm.

Joint DOD and DOA Estimation for Bistatic MIMO Radar: A Covariance Trilinear Decomposition Perspective

“…A tensor is the higher-order analogue of a vector and a matrix [32]. Before given the details of the proposed algorithm, let us first introduce some necessary preliminaries concerning tensor, which can be find in our pervious work [32].…”

“…Moreover, approaches such as the direct PARAFAC decomposition method and the ML estimator could neither perform properly. Typical strategies to suppress the spatially colored noise are rely on the covariance output of the measurement [26]- [32]. Since the PARAFAC model in this paper is based on the covariance matrix of the samples, it can be easily extended to the colored noise scenario.…”

“…In combination with (21) one can observe that the proposed algorithm can identify M +N +MN −2 2 at most. in Table 3, we summarize the identifiability of our algorithm, ESPRIT [10], HOSVD and its covariance version [17] (marked with C-HOSVD), PARAFAC [18] as well as QALS [32]. To make a fair comparison with ESPRIT, the HOSVD and C-HOSVD in the following are equal to HOSDV-ESPRIT and C-HOSVD-ESPRIT, respectively, i.e., after the subspace have been obtained by HOSVD and C-HOSVD, the ESPRIT-like technique is adopted to obtain the DODs and DOAs.…”

“…It should be emphasized all these de-noising methods are relay on the array covariance (or cross-covariance). Moreover, as shown in the literature, the PARAFAC variant offer more accurate performance than the HOSVD version [32], since the former utilizes more degree-of-freedom (DOF). Nevertheless, the fourth-order PARAFAC estimator is quit sensitive to the initializations and may suffers from the slowness of the convergence steps, making it unsuitable for massive MIMO configuration [35], [36].…”

“…In [26]- [29], the spatial cross-correlation methods has been proposed, in which the transmit array in these methods is divided into several subarrays, the non-correlation characteristic of the matched noise associated with different transmit antenna is explored. Assume that the noise associated with various pulse are uncorrelated, the temporal cross-correlation algorithms have been derived in [30]- [32]. In these algorithms, the array measurement is divided into two groups according to the temporal index, the cross-correlation of the above two groups is exploited to suppress the spatial colored noise.…”