Tensor Model And Parameter Estimation For Vector Sensor Array

With years of development of array signal processing, the information on time and space have been fully exploited. In recent years, new researches are started on polarization. As a receiver which can perceive information on polarization while on space and time. In this array, the data can be received as high-dimensional structure. However, for an electromagnetic vector sensor array, the traditional approaches transform the data to structure of matrix for processing, which makes the high-dimensional structural information of the data be not taken advantage of.By use of tensor which is a professional tool for high-dimensional data, this paper studied modeling of BTD(Block term decompositions) model of uncompleted wave, parameter estimation method based on Tucker decomposition, parameter estimation method based on CP(Canoical Decomposition)decomposition, and DOA estimation method based on BTD decomposition. This paper focuses on how to make full use of high-dimensional vector information inherent in electromagnetic vector sensor array, and launched several parts of work as follows:Firstly, MUSIC algorithm based on tensor decomposition of Tucker is improved. By tensor decomposition, the orthogonality bettwen the estimated noise subspace and the true signal steering vector is improved, which enhances the resolution of electromagnetic vector sensor array. Compared to the matrix method for estimating the noise subspace tensor method, tensor method takes the various dimensions of the data into consideration, and can recover true value from data be covered by noise data. Simulation results show that the use of tensor decomposition estimated noise subspace MUSIC parameter estimation can improve to wave signal resolution. This part describes the tensor overall dimensions, namely high-dimensional feature of tensor data is a worthwhile use of information.Secondly, blind estimation method for an array based on tensor decomposition is improved. In this section, CP tensor decomposition is used. Compared to Tucker decomposition, CP decomposition not only reflect the integrity of the data between the various dimensions. More importantly, it can also blindly estimate the composition of each dimension. By the use of the blind estimation, structural information on each component can be used as a constraint to influence each iteration of CP decomposition. Namely, the feature of blind estimation creates the advantage for the use of structural information of the array itself. In addition, this paper proposed a method that adds structural constraints by parametric method. That is, in each iteration, blind estimated components is used to estimate the parameters, then the model and parameters are used to synthesize components which are subject to the structural constraints. Simulation results show that as the constraint structure information is added, the parameter estimation error decrease.Finally, the parameter estimation of uncomplete polarized wave based on BTD decomposition is studied. For the receiving model of electromagnetic vector sensor array, the wave in the same direction is equal to two uncorrelated signals multiply an array of size 6×2. In this paper,BTD decomposition is firstly used which is more suitable for uncomplete polarized wave data, while the condidition of uniquess for CP decomposition doesn’t meet. The BTD decomposition can blindly obtain the estimation of the spatial array steering vector and column space on polarization. In addition, this paper has studied the method for obtaining estimation of direction of arrival wave by use of the Poynting’s Theory in electromagnetic field. In the end, the space array steering vector and direction vector are merged to obtain the estimation of direction of arrival. The simulation results shows that for the receiving model of the uncomplete polarized wave, the method based on BTD decomposition can obtain more precise result of estimation.