Direction Of Arrival Estimation Based On Tensor With Electromagnetic Vector-sensor Arrays

Direction of Arrival (DOA) is one of the most important applications in array signal processing. As a technology of detecting signal polarization information, electromagnetic(EM) vector sensor array is able to perceive complete electromagnetic information of incident signals, and therefore outperforms the traditional scalar array in estimation performance and has broad application prospect.Due to the multi-dimensional characteristics of electromagnetic vector sensor array outputs, traditional processing methods based on matrix analysis need to combine some dimensions and transform the multi-dimensional outputs into two-dimensional matrix, which may have certain limitations. Therefore, multi-dimension data processing tools based on tensor are playing an increasingly important role. Among all tensor models, canonical polyadic (CP) decomposition model is the most popular. This paper studies DOA estimation based on tensor with electromagnetic vector sensor array, the main contributions of this paper are summarized as follows:●Firstly, we studied third-order CP model of EMVS array and fourth-order CP model of EMVS array which was constructed via multi-invariance. Subsequently, a class of algorithms combined with signal statistic information are proposed for estimating DOA of these CP model outputs. The algorithms utilizes the independence of the signals. After unfolding tensor, Independent Component Analysis (ICA) is used to extract the independent signals, and the Khatri-Rao structure of mixing matrix can be restored by means of Joint Diagonalization (JD) algorithm. Finally, polarization steering vectors can be derived by the rank-1approximation, from which we can get DOA of signals. Simulations results show that the algorithms have advantages in resolving signals with near incident angles and polarization parameters. When multiple incident signals have near incident angles and polarization parameters, the proposed algorithms can obtain higher angular resolution. In estimating circular polarization signals under fourth-order CP model, the algorithm can have stronger identification capacity and higher estimation accuracy.●Secondly, we studied spatially "stretched" tri-pole plus tri-loop EMVS array CP model. Since the outputs are no longer a single tenor but two tensors with the same incident signals, we proposed a class of joint tensor decomposition algorithms for such array’s observed signals. The algorithm is extended form Alternating Least Squares (ALS) and JD respectively. Simulations show that joint tensor decomposition algorithms outperform single tensor decomposition algorithms in many aspects. The proposed algorithms can obtain more accurate DOA estimation in white noise, more robust with colored noise. Besides, JCP algorithms are able to identify linear polarization signals, and have better DOA identification performance with small deviations.